id
stringlengths 7
12
| sentence1
stringlengths 6
1.27k
| sentence2
stringlengths 6
926
| label
stringclasses 4
values |
|---|---|---|---|
train_99300 | Despite the a priori desiratum of assimilating the derivation and interpretation of wh-questions in Japanese to that in English, empirical evidence shows that overt displacement of wh-phrases in Japanese is not the same as overt wh-movement in English. | the examples in (6) clearly show that a non-wh-phrase object may be scrambled across the subject to a position other than SpecCP: (6) a. Dare-ga sakana-o tabeta no? | neutral |
train_99301 | Moreover sentence information and discourse information are dealt with equally at the time of evaluation as providing binding information, which allows for their seamless integration. | here we adopt an hPSG-based syntactic parser (Miyao and Tsujii, 2008), because it offers wide coverage and high accuracy, and provides detailed syntactic information that is sufficient for SCT. | neutral |
train_99302 | If these extra-syntactic mechanisms do not involve movement, they should also operate beyond LF in that LF movement is fundamentally a syntactic movement without phonological consequences. | the only cost accrued in the model is the presence of redundancy where earlier outputs are sometimes put back at a later level. | neutral |
train_99303 | (5) a. O-namae-wa nan-deshi-tak-ke. | their categories remain S\S Q or S\S f oc and cannot be S\(S\S) which seems to reflect their semantics better. | neutral |
train_99304 | In 18aand 18b, what is transferred to the RECIPIENT is the possession of a hole and the content of a song respectively. | 3 the malefactive sentence is only a slightly special case of the benefactive sentence with the BENEFICIARY coinciding with the subject (i.e. | neutral |
train_99305 | French, Japanese and English. | in another hand, links involving two or several languages are used to retrieve terms across those languages. | neutral |
train_99306 | We have examined 6, 510, 554 sentences from the encyclopedia articles in the Japanese Wikipedia, and for characteristics such as usage of plain or polite forms, the writing style appears to be quite consistent, despite the open nature of Wikipedia. | the difference between the encyclopedia articles and the other two page types is that the first has no specified reader, while the content of the other pages is more personal and perhaps thought as meant for other editors. | neutral |
train_99307 | To gain insight into the performance of individual feature, we first examine the experimental results using single features, as shown in Table 3. | since most of the existing researchers have focused on non-textual features, in this paper, we aim to explore the effectiveness of diverse textual features on the task of best answer detection. | neutral |
train_99308 | Although CQA tries to help users come up with the best answers with its human-oriented strategy, the quality of user-generated answers is sometimes difficult to control (Gyongyi et al, 2008). | it is no surprise that Web 2.0 system is trying to seek improvement in information foraging. | neutral |
train_99309 | (2009) also points to the fact that Japanese allows for some compound verbs which can encode both manner and path of motion, much like an equipollently-framed language, such as in (4a) below. | this differs from English in that Chinese allows for either manner or path of motion to be encoded in a sentence by simply adding one more particle, whereas an English preposition, which encodes path, cannot stand alone like a Chinese path of motion verb. | neutral |
train_99310 | Ideally, such experimentation would include low, mid, and high level learners, and would hopefully deal with languages form at least two different geographical locations. | the use of such compound verbs is also rather limited, as only a limited number of verbs can be combined in this manner, as in the incorrect example (4b), and some combinations have meanings other than what might be expected, as in (4c). | neutral |
train_99311 | In this case, we choose training to testing corpora of the ratio 9 to 1 to conduct following experiment to generate novel candidate list. | we use the linear chain form of CRF, in which x and y are the sequence texts and output labels respectively. | neutral |
train_99312 | For the purpose of studying behavior of novel terms in a special period, i.e. | in such case, it would not be tagged as true term in testing corpora; therefore, this term would not be retrieved. | neutral |
train_99313 | In their work, syntactic path refers to the path of syntactic functions of one NP. | the features they choose include surface word forms, POS tags, orthographic features and head-noun features. | neutral |
train_99314 | The experiment for 10-folds cross validation is conducted 10 times on 10 sets of corpora. | it is defined as concatenation of elementary syntactic functions tagged by Survey Parser. | neutral |
train_99315 | AMIRA (Diab et al., 2004) and its successor AMIRA 2.0 (Diab, 2009) are collections of tools for processing Arabic language based on supervised learning.. AMIRA 2.0 is a toolkit for processing Arabic text. | of omitting diacritics, complex morphology and peculiar orthographical rules, processing Arabic text is a difficult task whether performed using statistical or rule-based methods. | neutral |
train_99316 | The threshold set is that words with frequency greater than or equal to 10 and appears in more than 1 document are chosen as the lexical entries. | in view of these criteria, only 1, 386 words out of 9,965 or 13.91% of the lexical candidates were chosen as the final lexical entries, 45 of which are proper nouns. | neutral |
train_99317 | The process of building the corpus includes query generation, document sampling via google API and document classifier. | this study implemented an automated system that builds a language specific lexicon by using an open corpus or the World Wide Web. | neutral |
train_99318 | Table 3 shows the relationship between the characteristics of feature in knowledge node and the number of knowledge nodes which are relevant to the sentence. | in order to solve these problems, it is necessary to enhance the use of sentence features while dealing with the similarity judgment with knowledge nodes, the methods are as follows: Only to extract the keywords in the input sentence, that is, just find out the words exists in F, the set of knowledge feature, and remove the words which is not feature as stop words. | neutral |
train_99319 | In this experiment, in total 42 terms and 49 terms are registered on the Japanese-and Danish education system ontologies, respectively. | one single term unambigously designates one single concept. | neutral |
train_99320 | PrefixSpan (http://prefixspan-rel.sourceforge.jp/) is a method of mining sequential patterns efficiently. | furthermore, we examined the effects on the accuracy of user identification when we relaxed the criteria of selecting learning examples. | neutral |
train_99321 | When the user with the highest probability differs from the user submitted a series of input messages, our method determines that these users have similar styles of writing. | some users abuse the anonymity and disrupt communications in a community site. | neutral |
train_99322 | In the following sections, we present how we detect nasty comments on a BBS with many posts using an n-gram model. | the classification result of the SVM is shown in Figure 3 and table 4. | neutral |
train_99323 | Our proposed method can detect nasty comments based on nasty phrases and over-segmented words. | detection is difficult because we need to consider the neighboring words. | neutral |
train_99324 | We repeat this 10,000 times and obtain 10,000 data sets. | table 4 indicates the ratios when the method eliminating only one feature and using all the remaining features obtained a higher F-measure than the method using all the features and the ratios when the method using all the features obtained a higher F-measure than the method eliminating only one feature and using all the remaining features. | neutral |
train_99325 | 5 In all cases, the relative clause does not describe a temporary state because it refers to neither an in-progressive state nor a target state, and hence all the sentences in (12) will be judged unacceptable. | this paper has proposed that kes-RCs are stages, which instantiates at the time of the main clause's event. | neutral |
train_99326 | (10) [Cip-i olyastongan pyeyka-lo pangchi-toy-ø-n] kes-ul pucilenhan [house for a long time deserted house-as neglect-Pass-Prf-Rel] kes-Acc diligent senwen hana-ka soncilhya-ess-ta sailor one-Acc repair-Pst-Decl 'The house had been abandoned for a long time and a diligent sailor repaired the house.' | what can be deduced from (8)-(11) is that if the predicate within the relative clause describes an active event, only telic-perfect predicate is allowed, whereas such a constraint does not apply for perfect predicates describing a stative event. | neutral |
train_99327 | Number inflections depend on the gender of noun and ending vowels/diacritics. | noun inflection FSTs discussed here are not only useful in inflected word formation but also in morphological analysis/parsing of Sindhi nouns. | neutral |
train_99328 | Progressive marking proceeds from activity to accomplishments or achievements. | moreover, this may suggests that Japanese learners analogize from their L1 -te i-and expand the prototypical meaning of -ko iss-to the resultant state by the influence of L1 transfer. | neutral |
train_99329 | In the interpretation task, the participants were asked to select the best matching picture for the sentence from three choices that were supplied. | there were two times as many learners who over-generalized -ko iss-as there were learners who over-generalized -a iss-. | neutral |
train_99330 | As can be seen, the progressive -ko iss-was more frequent than the other forms, within all levels. | regarding the acquisition of Korean aspect markers -ko iss-and -a iss-by Japanese learners, we found that the acquisition order was the progressive -ko iss-→ the resultative -ko iss-→ the resultative -a iss-. | neutral |
train_99331 | This action-inprogress meaning is obtained when the progressive marker is attached to activity verbs and accomplishment verbs. | this study tests whether this arguably universal developmental pattern (the progressive is acquired earlier than the resultative) holds true for the L2 acquisition of Korean by Japanese learners, and in addition, this study surveys the influence of instructional order and acquisition order on the L2 acquisition of Korean imperfective aspect. | neutral |
train_99332 | In this experiment, three annotators (A 1 , A 2 , A 3 ) annotated the target reviews. | figure 1 shows the outline of our method. | neutral |
train_99333 | However, a review basically consists of sentiments with various aspects (i.e., "image quality" and "usability" of a camera). | (2007) has estimated the valid number of clusters by statistically evaluating the clustering result. | neutral |
train_99334 | These similar opinions have possibilities to be redundantly extracted as a summary. | we identify representative words of each cluster by using this feature. | neutral |
train_99335 | Then, we define the tf -idf value of a word i in target reviews j as below. | it is difficult to know the optimum number of divided clusters beforehand. | neutral |
train_99336 | In such a case, it is probably that the driver only saw a certain facial part of him, not the whole body of him; thus the compositional operation here is accommodation. | distributional information based on distributional analysis is widely used in many natural language processing tasks. | neutral |
train_99337 | An argument like " " (screen) is just a physical object, but it is clear that in the following sentence " " (screen) refers to the information on the internet or of a computer. | in Table 2, under the sense "visual activity", the operation of "kàn" is much richer (namely: dot exploitation, artifactual exploitation, dot introduction and qualia introduction). | neutral |
train_99338 | Since the parser was developed for news, science and teaching texts, and is currently used and optimized for the translation of Wikipedia articles, the most likely explanation for these performance variations is the difference between the formal and professionally transcribed political jargon of Europarl on the one hand and the creative and hastily written chat texts on the other. | at least in terms of 3rd person pronouns this can also be said of the Europarl corpus, which consists of isolated monologues, and employs a more abstract and elaborative style, which is also compounded by a high attribute/noun ration, and the high incidence of (pronoun-compensating) demonstratives. | neutral |
train_99339 | Given the clear inter-corpus differences we documented, and the high chat corpus error rate in particular, it would make sense to perform a detailed error analysis of our annotation, which in turn would allow us to genre-adjust the analysis lexicon, and permit the introduction of genre-specific rules into the parser in order to improve its performance. | the Europarl data boasts the longest words and longest sentences 6 , and scores highest for subordination and infinitive subclauses, as well as the rare past participle subclauses, all of which considerably complicate syntactic trees. | neutral |
train_99340 | This means that, at the present time, sufexbased CSEs for corpora with hundreds of millions of tokens require a relatively high, but not unrealistic, hardware investment. | for example, given a set S of words or patterns of interest, submitting a query g L sg R for each s ∈ S, where g L and g R are "gap-like" wildcard expressions for the left and right context, and having sufex generate frequency lists of the content found for g L and g R , makes it possible to iteratively refine a set of contexts, which in turn can be used to find further elements belonging to S. While such bootstrapping has already been used in many areas of NLP research, sufex improves it in two ways: (a) Contexts can be defined as complex expressions, including annotation symbols, rather than mere windows of N words; (b) Sufex is an index and provides answers within milliseconds (the more specific the queries become in the process, the faster the retrieval, due to the least-frequency property). | neutral |
train_99341 | In this respect it allows us to match a phrase structure: It may be nested, even together with the within operator, enabling truly complex queries which are still computed efficiently. | we present extensions to the CQL 'Corpus Query Language' for intuitive creation of syntactically rich queries, and demonstrate that they can be computed quickly within our tool even on multi-billion word corpora. | neutral |
train_99342 | Many linguists are interested in dialectology and the possibility of studying languages that differ only marginally from each other. | the latter is based on a query language of regular expressions. | neutral |
train_99343 | This is an imperative in the attempt to make different databases interoperable with one another. | data will be added in the near future on dialects of English (FREd) and Slovene (Slovene dialect database on doubling). | neutral |
train_99344 | McCarthy, 1991), it is undeniable that a focus on sentence-level form tends to be emphasized in many foreign language teaching contexts. | we have attempted to characterize each sub-corpus in terms of its (nominal) entity coherence within the framework of centering. | neutral |
train_99345 | Several analyses have been done on the developed sentiment lexicons to understand the sentimental behavior of people depending upon location, age, sex, profession and etc. | the word "blue" get tagged by different players around the world. | neutral |
train_99346 | The probability of the walker to be at a particular vertex is given by the Markov chain on the graph. | recently, Ferre-i-Cancho (Ferrer-i-Cancho and Sole, 2001) and Sigman (Sigman and Cecchi, 2002) who work on graph theory highlighted the "hierarchical small world" (HSW) structure used in language studies. | neutral |
train_99347 | Then DicoSyn.Verb has been given symmetric and reflexive. | this Markov chain is defined by the transition Matrix If ∆ 0 is the vector in V of the initial distribution probability of the walker (that is the probability to be at any vertex at the time t 0 = 0), the vector in V of the probability distribution of this walker after t steps becomes: to measure, in G = (V, E), the proximity between a vertex s ∈ V and the initial state ∆ 0 of the walker, we define prox(G, ∆ 0 , s, t) by the t steps random walk as: For a graph G = (V, E) and an initial state ∆ 0 , we can now define the flexemity at the time t of a vertex s ∈ V : the s f lex(emity) is the s prox(imity) weighted by the inverse of s degree. | neutral |
train_99348 | Such usage should be rather considered in terms of semantic approximations, an attempt to utter with the conventional verb due to the semantic cognitive flexibility. | those verbs used by adults are almost verbs with a specific meaning in the study. | neutral |
train_99349 | However, we want to perform a graph analysis of the experimental data as well. | several issues have been raised however about the cross-linguistic validation. | neutral |
train_99350 | Tuna cannot be used as an animal and food at the same time frame 6 . | the relation between constructive type theory and tIL's approach to natural language analysis needs further investigation. | neutral |
train_99351 | The word tuna can however occur in seemingly co-predicative context, such as example (8), but only when the two stages referred to by the word tuna are separated by different time frames. | dependent sort is not a set theoretic object. | neutral |
train_99352 | Many efforts have been paid to find out the crucial semantic elements that influence children"s extension of verbs. | given the above concerns, the present study aims to explore the research questions as the followings: (1) Can 5-and 7-year-old Mandarin-speaking children properly paraphrase and act out familiar verbs with novel instrument extensions? | neutral |
train_99353 | The result of the production (paraphrase) task suggests that not all of them can properly paraphrase the novel instrument extensions of familiar verbs. | the block-made instruments were in no way look like the names they were labeled, children can only access the knowledge of the instrument by the provided labels (i.e., names), not by their appearance. | neutral |
train_99354 | Twenty-two native Mandarin-speaking adults were asked to do a free recall on the instruments that could be used in these verbs. | five-and seven-yearolds in this study should be acquainted with these labels and have no difficulty recognize them. | neutral |
train_99355 | The subject of the complete sentence, Bob, is shared by the following gapping parts, with # marking the gaps. | from the software side, the mechanism of multiple visits is easily programmed. | neutral |
train_99356 | Typologically, in verb-framed languages (Talmy 2000) such as K/J, Hindi, Romance, Semitics, Bantu and Polynesian languages, an existential/stative location postposition/preposition such as -ey (K), -ni (J), à (French), -meN (Hindi) is further used only for basic directed motion verbs but not for manner verbs, e.g. | we can possibly represent [load the truck with hay] (pile/spray ---) as a change-of-state lcp with FORMAL = (globally) affected (e2, <2>), AGENTIVE = (e1, move_act (e1, <1>, <3>, <2>). | neutral |
train_99357 | John is more rich than Mary is beautiful There is in (8a), (8b) and (9) an implied notional comparison of the extent or magnitude of two distinctly different but related attributes: wealth vs beauty, which is possible in English sometimes, but not in other languages, such as Chinese, where only the magnitude of the same attribute may be compared. | this equated attribute allows for comparison and pragmatic manipulation by means of logical reasoning uniquely incorporated into the human communication chain. | neutral |
train_99358 | Lift or interestingness takes the correlation between and , it tests on two hypotheses, first is independent of , so they do not co-occur in the same webpage. | and it is computed as follow (Shi 2010). | neutral |
train_99359 | The translation candidate extraction is based on the idea that most OOV terms have their correspondent human translation nearby (Cheng, Teng et al. | sVM works well with OOV terms that have similar properties. | neutral |
train_99360 | In order to take the context into consideration, instead of segment pair unigrams, we also tested combined Maximum Matching with contextual n-gram probabilities for ranking. | the longest acceptable consonant sequences on the left of the vowels were then marked as onset (O), and finally all remaining consonants were marked as coda (C). | neutral |
train_99361 | Different from other tasks, language model is incorporated under a log-linear framework in SMT. | the language model could thus play a more important role in the translating procedure, which further reveals a large value relatively. | neutral |
train_99362 | Let D be the development dataset and T be the test dataset. | the cross entropy of test dataset increases as LM weight decreases, as shown in figure 2, in which we use the same dataset pairs as in section 3. | neutral |
train_99363 | This difference suggests that semantic representations play significant roles since the syntactic structure of (1) and (2) are all the same, 1 and that the quantificational structure introduced by the universal quantifier makes (2) easy to interpret. | a syntactic approach with the minimal attachment cannot account for the fact that the universal quantifier reduces the GP effect. | neutral |
train_99364 | For instance, the word "concept" is defined as "an abstract or general idea inferred or derived from specific instances" which scored 6 despite the word "abstract" explicitly appears in the definition. | the genus is also checked if it is possibly an abstract or mass noun, which is currently approximated by the absence of any indefinite articles and plural markers. | neutral |
train_99365 | The distinction between concrete and abstract concepts is psychologically valid but so far it can hardly be quantified in any objective way, which prevents it from being further studied in computational linguistics. | very common concrete concepts may not need so much detailed explanation as others in a dictionary. | neutral |
train_99366 | 1994;Talmy 1988, Sweetser 1991.The extant knowledge has also shown that the negation of modal sentences can affect the semantic manifestation of modality. | the negative marker bu (不) is used to substitute de (得) and turns the affirmative potential construction into a negative form [Verb-bu(不)-C-O] as exemplified in (3) and (4). | neutral |
train_99367 | Furthermore, while m5-sii2 (毋使) indicates '¬□p (not necessary)', the linguistic form of m5-zo3 (毋做) does not have the modal function. | (19) 垃圾使毋得儘採亂丟。 la2sab4 sii2-m5-ded4 cin2cai2 lon7-diu1 trash can't random litter 'Don't litter at random.' | neutral |
train_99368 | The negative operator m5 (毋) is infixed in the modal verb zo3-ded4 (做得). | the role to denote the idea of "not necessarily" falls on m5-sii2 (毋使) exclusively. | neutral |
train_99369 | In line with the concept of lexicalization, this study analyzes the intricate syntactic and semantic features exhibited by modal verbs in Taiwanese Hakka. | type A construction is contracted by two subclauses : the position of O is not concerned with any movement. | neutral |
train_99370 | In these two cases, joek6 "MEDICINE" is put in the first argument position in the action tier because it is the instigator of the result "dead" The third type of resultative construction is a marked construction in Mandarin Chinese. | one major difference is that certain causing verbs or main verbs are not adjacent to the means verbs or resulting predicates. | neutral |
train_99371 | We set the similarity threshold as 0.4, and the statistics of the graphs are shown in Table 2. | from the table we can see that about 36.3% of sentence pairs in the bilingual graph are isolated sentence pairs. | neutral |
train_99372 | If the node v does not connected to any other node, then we call the node v as an isolated sentence pair. | their experiments showed the improvement in the translation quality. | neutral |
train_99373 | Taking the above two steps, it avoids updating and sorting the importance of each sentence pair that the selected sentence pair can reach through a connected path. | the relationship between sentence pairs, and the importance will be updated dynamically during the selection. | neutral |
train_99374 | We can even obtain better results than the whole corpus when selecting 80% of the corpus, which suggests that the corpus may contain noisy data and decrease the quality of the translation. | in the beginning, φ = S , and the quantity of information for each sentence pair will be the largest value. | neutral |
train_99375 | This is more pertinent with respect to translation from English to Indian languages, Hindi in particular. | this will prove helpful in the translation process and improve the quality of the translation in a machine translation system. | neutral |
train_99376 | 2) Laughter relates to "laugh" which is translated as "hansnaa" and so on. | the major problem therefore is to identify whether the first or the second kind of semantic representation is appropriate? | neutral |
train_99377 | This has been shown in Table 5. | the solution scheme needs to find answers to the following questions: -whether the verb + preposition combination is a true VPC, as opposed to the preposition being used to take care of an intransitive verb; -how to identify the VPC even when the preposition is not adjacent to the underlying verb; -how to identify the exact sense of the VPC, given that the same VPC may have multiple senses, as per WordNet; -how to choose the proper translation in the target language. | neutral |
train_99378 | Finally, new kinds of data have to be created in the perspective of validating our model against spoken language corpora. | (Gibson, 2000), (Vasishth, 2003) or (Lewis and Vasishth, 2005) for example). | neutral |
train_99379 | Longitudinal interaction data from three Korean boys was collected and analyzed. | generally speaking, the meaning of -te i-is dependent on the inherent aspectual value of the verbs to which -te i-is attached. | neutral |
train_99380 | Two of them associated imperfective aspect marking with achievement verbs early on, but one child associated imperfective aspect marking with activity verbs. | further research is needed to verify the Aspect Hypothesis completely, including the acquisition of the progressive tense. | neutral |
train_99381 | The token frequency is in parentheses. | whereas -ko iss-is compatible with all transitive verbs, -a iss-can only be used with intransitive verbs in describing a persisting state resulting from a completed action, as in (3) (Lee, 1991). | neutral |
train_99382 | The coefficients of correlation in every combination were very high ( Table 3). | the average score of two evaluators for each sample sentence was used in the comparisons and analyses. | neutral |
train_99383 | However, all the evaluation methods mentioned that include human subjective evaluation provide less information for error analysis because they give only one score for each document or sentence. | it represents synonyms and phrases that should exist (or not exist) using regular expressions. | neutral |
train_99384 | We verified that the results of our automatic evaluation method are consistent with that of human evaluation for each question, and clarified that our method can be used for error analysis of grammatical items. | our method has high correlation with subjective evaluations by humans, even when only our method is used (which is comparable to conventional automatic evaluation methods and question-based evaluation by humans). | neutral |
train_99385 | • Baseline 5-gram: baseline system using 5-gram language model. | • The extraction of SCFG rules is merely based on the word alignment information. | neutral |
train_99386 | Moreover, offline methods require syntax analysis which does not always produce convincing result on certain languages. | the result turned out a decline on BLEU score. | neutral |
train_99387 | A certain degree of unavoidable over-generation of ambiguity can be handled by an ordinary disambiguation CG -the same contextual rules working for true ambiguity (e.g. | but while only generation will ensure the inclusion of all forms of a lemma, many of these forms may be very rare, and a reasonable fullform list may be built simply by extracting all unique word forms (and their different analyses) from an annotated corpus. | neutral |
train_99388 | At the disambiguational level, early work by Chanod & Tapanainen (1995) showed that even a small set of CG rules can compete with an off-the-shelve statistical tagger. | frequencies close to the hapax level, do not contribute. | neutral |
train_99389 | To maintain tokenization compatibility, and thus matchability with the gold standard annotation, we retained the system's own preprocessor, but for a new language a working version could be produced with a few lines of code, basically line-breaking on space and punctuation, with a few exceptions for abbreviations and -depending on the languagecontractions, enclitics and possibly names. | we reduced the original morphological analyzer to a mere look-up program with access only to a mini-lexicon of irregular and closed-class words, inactivating the real, verbo-nominal lexicon. | neutral |
train_99390 | #spr labels a reentrancy between the values of two The parsing and generation algorithms require that the type hierarchy be a bounded complete partial order (BCPO), i.e., for any two types that share subtypes, there must be one unique most general such type, called the 'greatest lower bound (glb)'. | the LKB uses numbers to assign each added glb-type a unique name. | neutral |
train_99391 | Because we expect the candidates returned by our method to be manually examined by a lexicographer, we are particularly interested in whether the top-ranked items are correct. | taking inspiration from evaluation approaches in word sense disambiguation that make use of artificiallyambiguous words (e.g., Schütze, 1992;Gale et al., 1992), we propose evaluation methods that use synthetic examples of semantic change. | neutral |
train_99392 | If correct, this would refute the universality of the noun-verb distinction in Linguistics, with significant implications for Linguistic Typology and Theoretical Syntax. | the first component of the EM algorithm (the Expectation Step) is the same as the process by which k-means assigns a data point to a cluster, however in the second step of EM (the Maximisation Step), the cluster centroids are recomputed such that the likelihood of the parameters of the distributions are maximised. | neutral |
train_99393 | The lexical processing of each word component in compound words is similar to single word. | furthermore, the clinical staffs will be involved in the evaluation process to have a better estimation of system performance. | neutral |
train_99394 | We can conclude that when applying an automated proof reading process, the clinical note is more informative as more medical concepts and instances are revealed. | the proofed corpus becomes much more informative for automatic information extraction as well as human comprehension and interpretation. | neutral |
train_99395 | A review of past papers using the TOEFL synonym test as a benchmark 1 , suggests that the corpus used, preprocessing of documents and resulting vocabulary size may impact the performance achieved (Stone et al., 2008). | the main contribution of this paper is to present a novel, efficient approach to using Kronecker products to encode word order information within a semantic space. | neutral |
train_99396 | It was created by the Knowledge and Data Engineering research group at the University of Kassel to provide a useful service to the research community and to collect data for their social bookmarking research. | next, we turn to a qualitative analysis of polysemous lexemes found in the both generated disambiguated graphs from Section 2.3. | neutral |
train_99397 | However, this is not the case. | figure 2 shows a sentence with dependency arcs (below) and its equivalent dependency tree structure(top). | neutral |
train_99398 | The quality of the parallel text alignments was intrinsically evaluated by computing the F-score of the hunalign outputs against manually aligned data. | a second round of cleaning was carried out to 1 STB hosts a series of tourism related websites in particular websites with parallel texts, viz. | neutral |
train_99399 | • Single primary document: the amount of text associated with a given exhibit area varies, but is generally of the order of slightly over 1000 words, in a single primary document authored by museum curators; this is complemented with a secondary document from Wikipedia which is anywhere from 600 to slightly over 3000 words in length. | it is worth noting that there are no statistically significant (p ≤ 0.05) factor combinations for "Themed" and "Unthemed" outside the top-15 factor combinations for "All" presented in the table. | neutral |
Subsets and Splits
No saved queries yet
Save your SQL queries to embed, download, and access them later. Queries will appear here once saved.