id
stringlengths 7
12
| sentence1
stringlengths 6
1.27k
| sentence2
stringlengths 6
926
| label
stringclasses 4
values |
|---|---|---|---|
train_95500 | This new speech resource will further support the development of speech and language technologies for under-resourced language. | they were reviewed in order to best meet the needs of ASR. | neutral |
train_95501 | In recent years, large databases of spoken Polish have been collected such as Jurisdict (Demenko et al., 2008), Diagest (Jarmołowicz et al., 2007) or Paralingua . | for the needs of studying cross-linguistic perception and production of prominence and phrasing, stability of rhythm metrics and rhythmic classification of Polish, we also recorded non-native speakers in Polish and in their L1s (see: Sec. | neutral |
train_95502 | Our other serious interest is focused on the DNN-based LVCSR for Slovak using Kaldi (Povey et al., 2011). | we report here the preliminary results (see Table 2.2.) | neutral |
train_95503 | The generated hypotheses are then manually corrected and completed by trained human annotators. | the new acoustic data will be appended again to the already expanded training data and the process of LVCSR system retraining and semi-automatic annotation will iteratively proceed until a sufficient amount is reached (usually hundreds of hours). | neutral |
train_95504 | 2 hours of spoken language recordings with transcription, and a set of basic language models like a pronunciation dictionary and a description of the phone inventory of the language. | additionally, the project adheres to and promotes bestpractice standards in the development of language data and tools. | neutral |
train_95505 | For the same reason, books are not read entirely. | we chose two chapters that were near the mean of overall alignment scores (hypothesized medium quality alignments), one chapter which was above the mean score (hypothesized good quality alignment) and a final chapter below mean score (hypothesized bad quality alignment). | neutral |
train_95506 | However, these metrics only provide global evaluation and are unable to evaluate MT systems' performance on specific aspects. | we further select only idioms whose translation is non-compositional, and create a blacklist for each idiom. | neutral |
train_95507 | One of the most popular semantic resources for English is WordNet (Fellbaum, 1998;Miller, 1995). | due to these problems, although we consider using WordNet in the future both as a resource and as a possible benchmark, we did not use it in the experiments presented in this paper. | neutral |
train_95508 | In practice, we repeat this process 5 − 6 times to build tokens representing longer ngrams. | we used English Wikipedia from June 2017, from which we used the meta pages archive which resulted in a text corpus with more than 9 billion words 3 . | neutral |
train_95509 | For example, one of the intransitive frames in the class Run-51.3.2 is shown in (1), with the final 4 lines making up the semantic representation: (1) The horse ran into the barn. | a similar sentence with an agent causing the motion, such as John herded the sheep into the barn, would add cause(agent, E) to the semantic representation in (1). | neutral |
train_95510 | This paper focuses on resource building, and the corpus of old Vedic is, more or less, restricted to the R . | the system has a remarkably high tokenization accuracy. | neutral |
train_95511 | • as...as construction. | all p values are < 0.001. ally. | neutral |
train_95512 | In this task, we corrected the errors in the processes, and performed the POS tagging and morphological analysis manually for 450 cases. | if a given target word in a sentence is disambiguated as a specific frame-semantics by the target identification task, its frame elements are identified if only it matched with the valence patterns. | neutral |
train_95513 | Finally, we also input as onehot vectors external lexicon information, constructed using the Lefff lexicon (Sagot, 2010). | these word embeddings can be initialized with pretrained embeddings and/or learned when training the model. | neutral |
train_95514 | As expected, the proposed approach is efficient for diacritics, apostrophes and 1 letter modifications (DIACR, APOST, MOD1C). | it is worth noticing that in our approach, as the lexicon used in the correction step is not exhaustive, we observe 80 added errors due to the fact that some words, which were correct in the raw text, but not present in the lexicon, have been erroneously modified into an in-vocabulary form. | neutral |
train_95515 | On the output layer, we use a softmax activation. | it is inefficient for agglutination AGGLU and SPLIT, and word error rates (lines four to six). | neutral |
train_95516 | We investigated two variations of this model: (i) a bidirectional LSTM model with sequential CRF layer on top (LSTM-CRF) treating the problem as a discrete classification task, (ii) a new bidirectional LSTM model with a sigmoid layer on top (LSTM-SIG) for a continuous prediction. | while the LSTM-CRF had a better (higher) F-score on the classification task, its RMS score was worse (higher) than the LSTM-SIG model, which may be due to the limitation of the model as discussed in Section 5. | neutral |
train_95517 | in our research for DHH users, we have found that an evaluation metric designed for predicting the usability of an ASR-generated transcription as a caption text for these users could benefit from word importance information (Kafle and Huenerfauth, 2017). | the LStM layer size was set to 300 in each direction for word-and 100 for character-level components. | neutral |
train_95518 | Taxi C, D and A would have force themselves through far more road construction works and traffic jams than taxi B. | in our experiment, we assessed trust, reliability, predictability, competence, acceptance and likability of the robot as dependent variables. | neutral |
train_95519 | The same procedure was repeated for the second task scenario. | we tested each gender-manipulated variable on perceived masculinity and femininity by the user. | neutral |
train_95520 | Given the variety of available objects they would need to compromise their choices and collectively decide where to place objects. | no significant difference was found on the same case for the listeners' gaze [t = -0.767, p = 0.444]. | neutral |
train_95521 | Then, the dialogue act of current utterance can be formally defined as: ( ) denotes the dialogue act of the i th utterance ( ) and , denotes jth candidate dialogue act of the ith utterance ( ), given a dialogue including n utterances ( 1, ). | (Bahou et al., 2008) used utterances semantic labeling based on the frame grammar formalism. | neutral |
train_95522 | This is particularly important to ensure that the principle of data minimization (see above) is respected. | it is necessary to periodically review the results of anonymization processes. | neutral |
train_95523 | The portal gives a web-based access, and for this it makes use of a database that holds information about access rights to digital objects, and information about users and their authorization records. | the aim was to explore how to best transfer the management of research data to these units for the longtime archiving of linguistic resources. | neutral |
train_95524 | The data centre cooperates closely with the CLARIN-D centre in Tübingen which is funded by the German Federal Ministry of Education and Research (BMBF). | there are differences in the use of ACLs, and where they are stored. | neutral |
train_95525 | We envision any process used to collect linguistic data as comprised of the interaction of several components: task, incentives, workforce, workflow and processing. | linguistic data is presented, selected and segmented, segments are classified and labels, annotation records are stored along with information concerning the source media, time and task and 4 https://lingoboingo.org contributors. | neutral |
train_95526 | A key feature of Wiki sites is to allow people from all over the world to add or modify articles anonymously and without consequence. | if high precision values are more desirable, then one-class classification should be used instead. | neutral |
train_95527 | It was created using best-worst scaling in a similar approach described earlier in this paper for obtaining emotion intensity scores. | we refer to the full list of words along with their real-valued scores of affect intensity as the NRC Affect Intensity Lexicon. | neutral |
train_95528 | Antonymous terms tend to co-occur with each other more often than chance, and are particularly problematic when one uses automatic co-occurrence-based statistical methods to capture word-emotion connotations. | a notable exception to this is the NRC Emotion Lexicon, which includes words that are associated with (or connotate) an emotion (Mohammad and Turney, 2013). | neutral |
train_95529 | Overall, we observe that the three negative emotions can be conveyed by words having a wide range of values for valence, arousal, and dominance. | sadness words tend to have slightly lower dominance scores than fear and anger words. | neutral |
train_95530 | We find that anger, fear, and sadness words, on average, have very similar VAD scores. | for each of the eight emotions, we created separate lists of terms that satisfied either one of the two properties listed below: • The word is already known to be associated with the emotion (although the intensity of emotion it conveys is unknown). | neutral |
train_95531 | For comparison against the human ceiling, we found eight pairs of emotion lexicons with partially overlapping entries distributed over four languages. | unsurprisingly, Valence displays a strong positive correlation with Joy and strong negative correlations with the remaining BE5 categories. | neutral |
train_95532 | The last dimension, Dominance, is sometimes omitted, leading to the VA model. | the lexicons as well as the source code for building them are publicly available (see Section 6.). | neutral |
train_95533 | This outcome led us to conclude that the relationship between VAD and BE5 ratings is not fully constant across different languages (respectively data sets). | this stream of work does not target the goal of predictive modeling (Stevenson et al., 2007;Pinheiro et al., 2017). | neutral |
train_95534 | Stress results in increased sympathetic activity and can be tracked for example through bodily reactions, such as an increase in heart rate, greater blood flow to extremities and an increase in the respiration rate etc. | a dyadic team corpus was used to examine the association between the "defusers" physiological signals following the "instructor's" questions to predict whether they occurred in a short or long turn-taking period of time. | neutral |
train_95535 | The notch of a box shows the confidence interval which is normally based on the median±1.15 * /√ (the interquartile range (IQR) is the 25 to 75 percentage). | recall will be significantly underestimated. | neutral |
train_95536 | We have compiled in a unsupervised way a large resource of pattern paraphrases that is available for academic purposes. | we can now estimate how many pairs were correct on average for each approach, and how many correct paraphrases were contributed to the pool of correct paraphrases by each approach. | neutral |
train_95537 | For this sentences we can compare the accuracy of the whole text, not only of the verb. | this is a direction that we would like to exploit further. | neutral |
train_95538 | Based on this observation, we developed a simple rule-based tokeniser in which 's was separated from the remaining part of the token if it was not a noun. | as expected, tagging accuracy (F1 score) was lower if the training data and test data were derived from different domains. | neutral |
train_95539 | DS errors are perhaps the hardest to agree on as each gold standard could have different annotation guidelines. | the Fleiss's κ agreement figures show that there is a high agreement between the human annotators (a value between 0.61 and 0.80 denotes substantial agreement, while a value between 0.81 and 1.00 denotes an almost perfect agreement). | neutral |
train_95540 | We selected 200 abstracts for annotation from PubMed, according to the criteria that each abstract should contain at least one of the relation types that we were looking for. | table 3: the averaged statistics per document for both abstracts and full papers in our corpus. | neutral |
train_95541 | The biomedical track at WMT 2017 (Jimeno Yepes et al., 2017) continued to offer new resources for French (EDP corpus), Portuguese and Spanish (extension of the Scielo corpus). | document parsing and processing. | neutral |
train_95542 | In this paper, we present the process of creating an annotated, phrase-level PICO (Population, Intervention, Comparison, Outcome) corpus and a sentence-level sentiment analysis corpus. | meanwhile, the non-experts had difficulties in identifying Populations that had no reference to a Person entity (e.g. | neutral |
train_95543 | Since there was no appropriate data available to create an automatic PICO approach or sentiment classifier, we created it ourselves with the help of expert and nonexpert annotators. | an Intervention that did not perform better than its Comparison, should be classified as negative. | neutral |
train_95544 | If to do so, Hazem and Daille (2014) use distributional approach as introduced in Subsection 2.2., here we use distributed models (Mikolov et al., 2013b). | given the multi-word candidate terms CCT 1 = (T 1 , E 1 ) and CCT 2 = (T 2 , E 2 ) and syn(CT 1 , CT 2 ) a synonym relation between the candidate terms CT 1 and CT 2 , the following inference rules are used: In rule R 1 , the heads are identical and the expansions are synonymous, while in rule R 2 heads are synonymous and expansions are identical. | neutral |
train_95545 | This work was supported by the EC's FP7 (FP7/2007(FP7/ -2013 under grant agreement number 610516: "QTLeap: Quality Translation by Deep Language Engineering Approaches". | for example, with θ ≤ 0.1 we restrict the MWE candidate list to contain only those items whose compositionality score is less than or equal to 0.1. | neutral |
train_95546 | The entries of these types of lexicons are mostly composed of MultiWord Expressions (MWEs). | the result of the previous step (Extraction and alignment of MWEs) is a list of alignment links candidates. | neutral |
train_95547 | In this example, the bilingual dictionary provides the alignments: "be/être", "decided/décidé" and "there/y". | to solve this program, we used two scoring functions. | neutral |
train_95548 | This approach is hybrid because it considers the global task of identification and alignment of MWEs as an optimization problem and it uses external linguistic resources: a seed single word bilingual dictionary and morpho-syntactic patterns. | to traditional approaches for MWEs alignment which consist in firstly identifying monolingual MWEs candidates and secondly applying alignment to find bilingual correspondences, our approach extracts and aligns MWEs in a one-step process. | neutral |
train_95549 | One of the points is to annotate not only expressions with strong temporality but also expressions with weak temporality, such as states and habits. | the major cause of the disagreements among the annotators is that there are multiple interpretations depending on the context and common sense, closely related to the writing style and theme of newspaper. | neutral |
train_95550 | The same applies to the e-e pairs, where out of 54 system output links, 44 are valid with 26 (59.09%) correctly classified. | solving this cases would require a system to apply inference mechanisms. | neutral |
train_95551 | The techniques could be integrated in for instance a reading aid, an activity mainly focused on receptive understanding. | the corpus is then used to retrieve paradigmatically analogous constructs, resolving the abstract slot, as seen in table 1. | neutral |
train_95552 | A total of 22 subjects participated to the recordings of the speech data: one native speaker for each language, and 10 learners for each language. | corpus-based studies can be used to determine how some morpho-syntactic phenomena are acquired in English as a foreign language (see for example the project English Profile, Hawkins and Buttery 2009), how students' pronunciation of L2 can be influenced by his L1 (I-PFc project, Racine et al. | neutral |
train_95553 | This kind of questionnaire presents the advantage to force the subjects to perform certain utterances for a specific given context. | the prosodic tier is completed manually, and the misc tier serves for remarks. | neutral |
train_95554 | In the last years, especially with the development of new methodologies and new technologies in corpus and computational linguistics, language corpora have become more and more common and needed in linguistic research. | it is a rather difficult task for beginners or intermediate learners. | neutral |
train_95555 | The FLAIR text classification is based on parsing information and on the official English language curriculum of schools in Baden-Württemberg . | indeed, the sentence level alignment, also presented by Kauchak (2013), was possible in only 28% of the sentences from the SEW (and in 4.25% of the sentences from EW). | neutral |
train_95556 | This is similar to SourceFinder's method, in which on-line newspapers and magazines are downloaded and processed as text sources (Passonneau et al., 2002;Sheehan et al., 2007). | in spite of all the effort to present readable and interesting texts to learners, those systems do not indicate how learners can improve their skills by using the indicated texts. | neutral |
train_95557 | The answer isn't trivial. | we only annotate referential bridging. | neutral |
train_95558 | in a case like http://www.ims.uni-stuttgart.de/ institut/mitarbeiter/roesigia/ guidelines-bridging-en.pdf where some people might consider this a bridging case, as the foreign secretary Mottaki is probably not interpretable alone for a typical WSJ reader without the mentioning of Iran first. | the annotation is done using the annotation tool Slate (Kaplan et al., 2012) 4 . | neutral |
train_95559 | For the single anaphor type categories (definite, indefinite, comparative) we have created separate columns, as well as one joint column which contains all the bridging links. | only a small subset of the annotated pairs contains truly anaphoric bridging anaphors, which is why annotated corpus resources like the one presented in our paper are still beneficial (c.f. | neutral |
train_95560 | Dominance of negative examples over positive is a typical situation in real texts, where most pairs are not coreferent. | each experiment (excluding experiment 3) was tested on a set of various different pre-selected connection threshold values: 0.5, 0.75, 0.85, 0.95, and 0.99. | neutral |
train_95561 | The data for our experiments, as for all previous configurations, come from the Polish Coreference Corpus (Ogrodniczuk, 2015, PCC), a large corpus of Polish general nominal coreference manually annotated over the texts of the National Corpus of Polish (Przepiórkowski et al., 2012) 1 and Rzeczpospolita Corpus (Presspublica, 2002). | the network takes an input vector of 1147 units and is passed through a fully connected network with a single output (a value between 0 and 1). | neutral |
train_95562 | A lány piros ruhát viselt. | figure 2 tells us that repetitions, hypernyms and adverbial anaphors are much more frequent in the student essays than in the newspaper articles. | neutral |
train_95563 | Results from SANAPHOR have not been reported because it has been built over the Stanford decoref that uses the CoNLL specific features and then cannot be compared with the actual models of Sanaphor++ and Stanford deep-coref. | it means that if an antecedent is not an empty antecedent and if the set of true antecedents is equal to empty, then this antecedent is likely to be a wrong pair; belonging to the hierarchy of type of the antecedent is in the set of the true types, then this antecedent is likely to be a good pair. | neutral |
train_95564 | We have presented a coreference resolution system that is able to capture the semantic of the entities into a single method based on two different systems that use divergent methods. | we decided to create a new threedimensional vector, one dimension for each semantic feature: i) if the entities of the two mentions are the same, ii) if the type of the two mentions are the same, and iii) if the type of one mention is included into the hierarchy of the other one. | neutral |
train_95565 | This version adds syntactic annotations in addition to the existing coreference and speech transcription ones. | 2016) through state-of-the-art automatic parsing techniques. | neutral |
train_95566 | The only resources for the language pair English-German that are known to us include the GECCo corpus (Lapshinova-Koltunski and Kunz, 2014), the ParCor corpus and the multilingual coreference corpus described by (Grishina and Stede, 2015). | moreover, negation in German can be expressed not only with the adverb nicht, but also with the indefinite pronoun kein. | neutral |
train_95567 | The coreference relation is shared across all languages. | the annotation scheme takes inspiration from the schemes used in all the three resources mentioned above (Kunz, 2012;Grishina and Stede, 2016). | neutral |
train_95568 | For studies on Japanese dialect, (Yoshino et al., 2016) collected a parallel database of Japanese dialects with voice recording in their well-equipped recording environment. | 1 https://github.com/mattdiamond/ Recorderjs Table 1: List of collected dialects. | neutral |
train_95569 | Vocabulary knowledge prediction is an important task in lexical text simplification for foreign language learners (L2 learners). | in this section, we describe the content and format of this resource. | neutral |
train_95570 | Training on larger annotated corpora showed only very small increases in performance. | previously studied methods that use hand-crafted rules based on one or two word features have had limited success. | neutral |
train_95571 | In case several transcripts have been proposed for a title area, they are all proposed in order to associate the best one or to correct the closest. | this is illustrated in Figure 1 by the red dotted arrow. | neutral |
train_95572 | By studying the theater that was excluded from the system of privileges, the Italian Theater, we analyze the questions of acculturation and institutionalization including the fusion of the Opera-Comique with the Italian Comedy. | the diversity and complexity of data contained in a daily page of Italian comedy require types and tags to be more specific and hold-back all levels of annotation. | neutral |
train_95573 | Crowd workers were instructed to inspect each image and then produce a plain text description of the fashion item. | recent years have seen a renewed interest in text-image multi-modality and have seen the emergence of tasks such as automatically generating textual captions for a given image (Karpathy and Fei-Fei, 2015;Lu et al., 2016), using plain text descriptions to query images (Socher et al., 2014), and matching lexical tokens or constituents to regions in an image (Ma et al., 2015). | neutral |
train_95574 | In German, a simplification corpus with a scale of 7,000 sentences (Caseli et al., 2009), In Italian the PaCCSS-IT (Brunato et al., 2016) with a scale of 63,000 sentences, In Spanish a simplification corpus do exist made by hand by the few rules (Mitkov andŠtajner, 2014;Stajner et al., 2015), and so on. | red, green, and blue were necessary to explain the concept of color.) | neutral |
train_95575 | The resulting data set will be made available through the EU (according to the H2020 Open Research Data Pilot) for research purposes after the end of the project, excluding the sentences sampled from Coursera as these were restricted by copyright to project-internal usage only. | (1) Agrippa 's Trilemma states that there are three options if we try to prove the truth . | neutral |
train_95576 | As Jiménez-Crespo (2017) observes, Facebook has used it for the translation of its social networking site and its user interfaces, Amara and TED for audiovisual practices, Kiva and the platform Trommons from the Rosetta Foundation for non-profit initiatives. | the subtitle genre contained spontaneous speech properties, truncated sentences, elliptical formations, disfluencies, repetitions, interjections and fillers. | neutral |
train_95577 | Presumably, due to the small training dataset, the SVR with an RBF kernel was overfitted, and its RMSE scores were worse than those in the case when linear kernels were used. | item response theory is used like a name of a field rather than a specific models. | neutral |
train_95578 | This is probably because 1) most of the tests are used to grade test takers, i.e., placement tests; thus, they are treated as private information that should not be leaked, and 2) the primary focus of most language-educators is how to measure their students' ESL vocabulary, rather than the test results of the other test takers. | we did not include English proficiency tests popular almost solely in Japan, such as the "Eiken" test, an English proficiency test popular with Japanese high-school students while we required our test takers to have a TOEIC test score, it did not matter when they took the test. | neutral |
train_95579 | Let the set of test takers (learners) be I and the set of problems be J. | this is not the case for developing educational software, in which participants are more diverse than in typical classrooms. | neutral |
train_95580 | Other interesting work such as (Chen et al., 2012) introduces a Deep Belief Network model that can handle highdimensional feature space (Lin et al., 2010) worked on a mixed model that combined feature-based and kernel-based models together. | besides, our model is not very language-sensitive, so we will extend our work to other languages. | neutral |
train_95581 | Question is the uncertainty category with the most microblogs, followed by Possible. | ), "如果" (if), and we also list the specific statistics of the agreement in each sub-class, as shown in Table 2. | neutral |
train_95582 | In this case, the uncertain semantics will be implicitly conveyed by the whole sentence rather than by the explicit cue phrases. | we found that in Sina microblogs, there were a lot of sentences expressing the uncertainty statement with expected semantics. | neutral |
train_95583 | Our future tasks include running a pilot survey to validate usability of survey format for our task and evaluate the adequacy and clarity of the questions for annotators. | merriam Webster dictionary defines spin doctor as "a person (such as a political aide) responsible for ensuring that others interpret an event from a particular point of view" 1 . | neutral |
train_95584 | The goal of those previous location estimation researches is to estimate the locations vaguely with best-effort approaches. | we observed the variation of accuracy across the diseases. | neutral |
train_95585 | Does not include "BOT", "Bot", "bot" in one or more of Twitter client name, account name, display name, and profile (1) of the above conditions is aimed at easily judging whether or not the result of estimating the position information from the tweet sentence is correct. | people have limited kinds of sources of information on diseases such as TV news and newspapers (including Internet news.) | neutral |
train_95586 | Overall, the corpus contains 192 documents with 1,140 annotated temporal expressions, and has been made publicly available to further boost research in temporal tagging. | this work has been partially funded by the EUCLIP RES project under the FESR Program of the Autonomous Province of Bolzano -South tyrol. | neutral |
train_95587 | Bethard, S., Savova, G., Chen, W.-T., Derczynski, L., Pustejovsky, J., and Verhagen, M. (2016). | the articles contain words and phrases which are not used in High German, including the temporal expression heuer which translates to "this year". | neutral |
train_95588 | The workshop will constitute a forum for sharing a common wordnet structure across languages». | this study cannot be considered an exhaustive review of the field seen through the lenses of the LREC workshops because for the time being just a few features have been taken into account. | neutral |
train_95589 | More thorough analyses can be carried out both on the research side for detecting the thematic trends and on the "societal" side for better defining this heterogeneous community constituted by the scholars who participated to the numerous LREC workshops. | guidelines and principles for comparing existing and acquiring new languages are of utmost importance in the field. | neutral |
train_95590 | Manually annotated corpora are one of the pillars of research in NLP. | files (or URL strings) can be added via drag and drop functionality and immediately enriched with metadata. | neutral |
train_95591 | Off-the-shelf database support for CoNLL TSV is currently limited to dialects with constant number of columns. | while, after conversion from CoNLL TSV, dependency labels are stored as literal values of conll:EDGE, it is easy to transform them into object properties using a simple SPARQL Update statement: As the example also illustrates, RDF resources can be defined in a way that their IRIs/URIs resolve against external resources, e.g., the dependency label nsubj yields the URL http://universaldependencies.org/ u/dep/nsubj and thus a human-readable definition. | neutral |
train_95592 | CoNLL Merge and CoNLL-RDF are designed to be applicable to every OWPL corpus format and thus depend on user input regarding the labels and types of columns. | we provide means for (i) rule-based rewrite operations, (ii) visualization and manual annotation, (iii) merging CoNLL files, and (iv) data base support. | neutral |
train_95593 | A loop -when an edge returns to the same token it started from -, is not part of the dependency formalism. | the thorough analysis of the errors can lead to new ideas and methods, and finally, to better results. | neutral |
train_95594 | In this category we identify errors due to: i) the slightly difference between some frame meaning (e.g., many errors are due to the prediction of the frame Measure duration instead of Calendric unit); ii) the frequent absence of related senses so that is not possible to trigger any frame, even for correct BabelFy annotations. | we apply the methodology from Biemann et al. | neutral |
train_95595 | In Figure 2 we draw the distribution of the counts of synsets across the number of frames that are triggered in the phase of candidate selection (step 1 of our WFD approach), i.e., in the XWFN profile, we counted ten thousand synsets bs for which |CF (bs)| = 1. | in the resource, each sense (e.g., read#VB#1) is defined by its related senses (quote#VB#0, recite#VB#0, and so on) which have been observed to co-occur with the sense. | neutral |
train_95596 | For the bag-of-words features, we experimented with different window sizes but found that using all words in a sentence provides best performance. | since we can consider our task as some type of text classification task, we should examine simple surface-based features, such as bag of words. | neutral |
train_95597 | (Sentences contain considerable fewer words than larger units of texts which are usually considered for text classification, such as paragraphs or documents.) | a more recent alternative to Brown clustering is the usage of word embeddings. | neutral |
train_95598 | Further, we replace the hard mutual exclusive constraint from McIntosh and Curran (2008), which does not allow term and patterns to belong to multiple categories, with the soft constraint captured in Eq. | intuitively, we aim for a relatively high value for λ 1 because we prefer fewer, high-quality expanded terms (i.e., the expansion should be cautious). | neutral |
train_95599 | Each premise provides a support in the form of logical reasoning for, or evidence in support of, the conclusion to which it is connected. | • Section 5. describes the SSS dataset that we create using opinions extracted a set of hotel reviews and SER. | neutral |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.