id
stringlengths 7
12
| sentence1
stringlengths 6
1.27k
| sentence2
stringlengths 6
926
| label
stringclasses 4
values |
|---|---|---|---|
train_98900 | Then, we will propose a new feature LICENSEE for a better analysis of them. | far, we have considered the nature of the unbounded relationships in CACs. | neutral |
train_98901 | Compare this with (6a) for the analysis of ta (perfect). | in (22b), there is no such involvement of a reference point, In (20c), the perfect state is introduced straightforwardly into the main DRS, and the triggering events are not introduced into the main DRS. | neutral |
train_98902 | amedama-wa] candy-TOP tabe-tei-ru (-kedo...) eat-PROG -PRES -BUT 'It is (at least) the candy which the children are eating. | this is sufficient to identify the informational contribution of such a NP as focus. | neutral |
train_98903 | In Section 3.2.1, we have formalized the constraint on topic which if present appears in sentence-initial position. | two options are available within the principle in (16): (i) accounts for instructions with narrow focus in (25a): it requires a mother to inherit the INFO-ST value of its daughters, as represented by the structure-sharing using 2 in (30a). | neutral |
train_98904 | 7 Another advantage of using the multi-dimensional representation concerns mismatches between informational partitioning and syntactic constituency. | we conclude that topic wa and (N)RF wa function as link and focus marker, respectively. | neutral |
train_98905 | In this experiment, we used all sentences that incorporated positive instance noun-verb pairs to learn constructional and lexical preferences. | below, we outline the two methods, the first of which uses hand-generated templates to identify candidates for each role type, while the second employs a supervised learning technique to dynamically identify templates predictive of the different role types. | neutral |
train_98906 | use of case markers and SOV order). | moreover, it costs a lot to develop parsers of both the source and target language. | neutral |
train_98907 | The unknown word rates vary by datasets, CHTB has the most unknown words whereas AS has the least. | in the next row, we have shown that re-segmentaion using the disambiguation model improves the results, as those missing words (found in one context but not the other) can be corrected. | neutral |
train_98908 | A separate process for unknown word detection could help to increase the lexicon and improve on the segmentation accuracy. | first, we use the unknown word model to extract unknown word candidates from the input text and apply a pruning process to them. | neutral |
train_98909 | At first, people believed in rule based method, and tried to apply rules for word segmentation. | this is because SVM requires more computational power. | neutral |
train_98910 | Let's take the person name " " (Sun yanzi) as an example. | if the unknown word rate is low, such as AS corpus, it would be better if all the detected words are used for re-segmentation because the pruning steps eliminate too many valid unknown words (5%) relatively. | neutral |
train_98911 | Please note that the benchmark of random prediction among the three types of relations is 33.33%. | in addition, the internal relation of the VV compound has already been assigned to one of the three types: coordinate, modificational, and resultative. | neutral |
train_98912 | 2003), our work on Mandarin Chinese and the CTB to date uses a smaller feature set and a less fine-grained analysis. | a recent body of research had extended the basic paradigm of automatic PCFG acquisition from treebanks to the extraction of deep, wide-coverage, constraint-based grammars and lexical resources such as LFG (Cahill et al., 2002;Cahill et al., 2003;Cahill et al., 2004;O'Donovan et al., 2004), HPSG (Miyao et al., 2003;Miyao et al., 2004) and CCG (Hockenmaier and Steedman, 2002;Hockenmaier, 2003). | neutral |
train_98913 | The method is found to require a huge search space, consuming far more computation time than we could afford. | part of its grammar is still unsure. | neutral |
train_98914 | But if the system is designed to work at the syllable level, these linking syllables can be generated from a specific syllable in a certain context. | this process is also non-deterministic. | neutral |
train_98915 | Space is inserted as word separation, such as "prasop khwam samret" (meet -Nom. | we believe that syllabification is a necessary step for grapheme-to-phoneme conversion. | neutral |
train_98916 | In the following discussion, I will argue that shi-support is needed in order to satisfy the ECP (Empty Category Principle) requirement because of the intervening barriers 6 . | if we cannot see the difference between them, how can we further account for the asymmetry between these two different kinds of wh-phrases? | neutral |
train_98917 | Such movement is more economical, and the mysteries of this argument-adjunct asymmetry will be revealed, as shown in 7and 8 Zhangsan like someone but I not know be who ' Z h a n g s a n l i k e s someone, but I don ' t k n o w wh o ' shi exists before the wh-argument Shei ' Wh o ' is that the trace of this wh-argument cannot be antecedent-governed by its antecedent in the lower hierarchical level, and thus must be lexically-governed or head-governed by a lexical head, such as the preceding verb Zhidao ' k n o w . ' | zhangsan like Meili ' z h a n g s a n l i k e s Me i l i . ' | neutral |
train_98918 | Different from their analysis in crucial respects, however, I argued that a covert wh-movement also takes place simultaneously with the operation of copying the a n t e c e d e n t I P . | since shi-support is a language-particular phenomenon, it only functions as the last resort to rescue the ill-formed Sluicing sentences. | neutral |
train_98919 | Departing from their analysis in crucial respects, however, I argue that a covert wh-movement also takes place simultaneously with the operation of copying the antecedent IP. | we have nothing but to pursue another lexical head. | neutral |
train_98920 | *Tsukue-ni hana-ga ari-ki-tta. | cut and kiru, as main verbs, both express division sense such as in cut a cake and keeki-o kiru 'to cut a cake'. | neutral |
train_98921 | Both creation of concrete artifacts and performance of events are secondary senses derived from co-composition. | it also has its own function called type coercion, which is unpredictable from the semantics of the main verb. | neutral |
train_98922 | A target-language term should have pronunciation as close to the source-language transliterated equivalent as possible. | this algorithm establishes a conceptual foundation for term transliteration. | neutral |
train_98923 | 200 sentences were separated at random as the test set for unseen texts. | travel routes) in a choice set and "chooses" one route. | neutral |
train_98924 | Examples: (every 5th and 10th day), (door-to-door), (mimetic word; fast), gotoobi doatsuudoa byuun (access), (annoying) akusesu iratsuku A collection of complex words means the combination of nouns, verbs, particles, etc. | inui (2004) analyzed open-ended questionnaire texts about traffic infrastructure planning, focusing on the intention behind answers. | neutral |
train_98925 | Rather, " (fast)" is recognized, and " " only modifies " ." | for example, the cognition result "time" can be extracted from the sentence "Waiting takes time." | neutral |
train_98926 | Details will be reported in another paper. | when developing new products, a company must analyze customers' awareness from a bottom-up viewpoint. | neutral |
train_98927 | Statistically acquiring sufficient knowledge about a language to build a robust WSD system is extremely difficult. | generality and adaptability are, therefore, essential to a robust and portable WSD system. | neutral |
train_98928 | So far over 10 million character of Chinese texts of different types and corresponding English texts has been collected and included into the corpus. | users can know "发扬" and "民主" co-occurred 5 times in the small sub-corpus loaded by the concordance tool. | neutral |
train_98929 | Most of Chinese words do not have morphological variants. | 2) Monolingual textual structural annotation. | neutral |
train_98930 | BAI Xiaojing and ZHAN Weidong (2003) explored how English passive form could be translated into Chinese, either with Chinese preposition " 被" or not. | in a segmented corpus, both "民主" and "民主集中制 2 " are segmented as word. | neutral |
train_98931 | These recognition results are fed into the computer scoring system. | the responses to these two parts allow score uses or test administrators to listen to test-taker's spontaneous speech. | neutral |
train_98932 | "approach" and "length" were transformed into "size" and "way", respectively. | • For words with the same number of instances of "l" and "r" preceding vowels, a word PACLIC 18, December 8th-10th, 2004, Waseda University, Tokyo Table 2: Examples of transforming English sentences into sentences containing fewer instances of "l" and "r" letters preceding vowels Correct transformation We think a good approach way is to construct it using "X no Y". | neutral |
train_98933 | However, LVCs differ from idiomatic phrases in that, in most cases, their semantics (and hence their translation) is more predictable. | hence, deciding precisely which light verbs combine with which nouns to form acceptable LVCs is a difficult task. | neutral |
train_98934 | Take a lunch is a compositional combination of a verb and a noun, both contributing their literal semantics. | due to the peculiarities in the behaviour of metaphorical and idiomatic expressions, they have been mostly overlooked by researchers in computational linguistics. | neutral |
train_98935 | Example 16 (b) shows a new sentence produced by reordering words of the sentence in (a). | the more complex case in Example 9 (a) can be represented in sign language by moving the right hand shaped for 'to see' to the right-hand side of the speaker, and the left hand also shaped for 'to see' to the left-hand side of the speaker. | neutral |
train_98936 | However, expressions in sign language represent their syntactic information by varying word order rather than by varying words themselves, which is the very reason why they are classified as an isolating language. | this additional representation further clarifies the sense as shown in (c). | neutral |
train_98937 | Phrases composed of Chinese characters are considered as unknown words. | example 11 (a) 어머니께서 나에게 빨래를 하라고 하셨다. | neutral |
train_98938 | We restrict the scope of Korean sentences to be practical and concrete, avoiding the use of words beyond everyday usage, and the use of sentences with highly complex structures. | it is desirable to rephrase unknown words in an explanatory manner or to express them with the use of fingers for exact spelling. | neutral |
train_98939 | These tables show percentages according to the syntactic positions of 'market.' | chinese speakers say 'market' as above the market rather than inside the market. | neutral |
train_98940 | For example, the noun root halatra 'theft' is specified with the continuation class NTR2RWeak. | we also make use of P-type and R-type flag diacritics to model long distance dependencies between verbal and nominal affixes. | neutral |
train_98941 | For example, if we decided to implement the lexical preference for passive suffixation in -ina rather than -ana as a feature, the lexical entry for fantatra in the lexicon would be accompanied by a featureˆFpassi on the surface level as bellow. | as with any finite-state morphological transducer, our Malagasy morphological analyzer is bidirectional: it can be used in grammatical analysis to produce morphologically analyzed input to a parser, or in generation to produce a surface form from a specification of lexical properties (Beesley and Karttunen, 2003). | neutral |
train_98942 | Three-syllable roots take penultimate stress unless they end in one of the 'weak syllables' na/ny, ka, tra, in which case they usually receive antepenultimate stress and are called 'weak roots' (Keenan and Polinsky, 1998). | because they do not take effect until the interpretation phase they do not interfere with the XFST rules. | neutral |
train_98943 | They preferred the non-solids for the shape match non-solid trials, selecting it 94% of the time. | virtually no studies examined this distinction in classifier languages which supposedly lack the distinction. | neutral |
train_98944 | However, the children's pattern for shape match non-solid trials differed from adults; both groups of children were willing to pick the non-solids that matched the shape. | we tested six count classifiers: gen (rod), zhi (stick), tiao (line), zhang (sheet), pian (slice), and kuai (chunk). | neutral |
train_98945 | Virtually no studies examined this distinction in classifier languages which supposedly lack the distinction. | the closed box was correct for two of these trials. | neutral |
train_98946 | Accordingly, we propose a sequential parser that determines syntactic structure based on semantic analysis referring to the context and on relevant statistical information. | dependency structures containing them can be transformed into the form "an attribute is a value" in Figure 2. | neutral |
train_98947 | Homographs are traditionally regarded as a special case of word sense disambiguation problem, and are usually addressed by corpus based statistical methods. | examples (16) and (8) are expressed with the same predicate '싫어한다(silhehanta: dislike)', but the presence of the unconjugated adjectives for the object '밤(pam)' in example (16) helps restrict the sense of the homographic noun to 'night' with short vowel sound. | neutral |
train_98948 | No feedback was given to correct responses. | by item analysis, passives were rated less grammatical than canonical variants for simple sentences (t(23) = 3.07, p < 0.01), less grammatical than both canonical and ba variants for sentences with RCs (t(23) = 5.68, p < 0.001; t(23) = 2.27, p < 0.05), and less grammatical than canonical variants for sentences with adjunct RCs (t(23) = 4.92, p < 0.001). | neutral |
train_98949 | In (17), we observe that Skolem term dareka 'somebody' allows only the narrow scope reading. | ( -1976) "Pivot-Independent Relativization in Japanese II," Reprinted in S.-Y. | neutral |
train_98950 | In (17), we observe that Skolem term dareka 'somebody' allows only the narrow scope reading. | the transitive verb loves has the following category with its environment being the empty set: 5 (12) Application of a rule induces environment passing in the following way: if a function with environment F is applied to an argument with environment A , the environment of the argument in the resulting LF is the union of the two (F U A). | neutral |
train_98951 | Using a corpus, statistical-based MT allows flexibilty to accomodate different domains. | the evaluator used the following rating scheme: 1 -Unacceptable, 2 -Poor, 3 -Acceptable, 4 -Good, and 5 -Accurate. | neutral |
train_98952 | Viewpoints of the observer do not seem to play an important role in in and out orientations. | nei4 and wai4 are symmetric as in (1a)-(1d): guo2nei4 國內 'inside the country; domestic'/guo2wai4 國外 'outside the country; abroad'; nei4meng2 內 蒙 'Inner Mongolia'/ wai4meng2 外蒙 'Outer Mongolia'; nei4xiao1 內銷 'sell domestically'/wai4xiao1 外銷 'export', etc. | neutral |
train_98953 | This observation suggests that the choice and the distribution of referring expressions in the task-oriented dialogues depends on the way the participant collaborate to judge the most salient entity in the current discourse against their common ground. | based on Grosz, Joshi and Weinstein (1995) and Walker, Joshi and Prince (1998), I adopt this ranking in analyzing the data. | neutral |
train_98954 | (i) Miles wants to meet a policeman. | the judgment is found controversial. | neutral |
train_98955 | For example, a verbal stem '먹 mek' may simply concatenate with a tense marker to form another stem, but its syllable structure may restrict the choice of its ending, thus allowing the stem '먹었 mek.ess' only. | the end of verbalEnding entries adnoun = [ 추운 "@P.SURF. | neutral |
train_98956 | Most IE researches are concerned with producing better performance and facilitating development portability [2][5] [1]. | these common string features may be found at the string's surface form level. | neutral |
train_98957 | Given a set of n-tuple, r = {t 1 , t 2 , …, t m }, in the database initially loaded by an IE task, we will apply the set of string features to the set of n-tuple such that each value v i in the database is evaluated by six string features. | these common string features may be found at the string's surface form level. | neutral |
train_98958 | A perfect classifier will have true positive and true negative rates of 1. | the accuracy of detecting invalid data also depends on the boundary between majority and minority. | neutral |
train_98959 | In this paper, we present a string feature-based approach to textual data profiling and invalid data detection. | for most real world applications, there is always a tradeoff between TP-rate and TN-rate. | neutral |
train_98960 | Person name is one of the most frequent foreign phrases in texts, and it often causes noise as unknown words. | a Chinese name is basically made of three characters, one for family name and succeeding two for a given name. | neutral |
train_98961 | But the basic notations of the names in Latin alphabet are different in China, Hong Kong and Taiwan. | most of the researches are for documents, and though the possibility of high accuracy for shorter strings of characters, language identification for words or phrases has not been discussed much. | neutral |
train_98962 | These applications, at least theoretically, could get them automatically from the Web. | plate 3: A writing system (Thai) related to NUMERALS. | neutral |
train_98963 | Web-crawlers and IR systems are other candidates which will certainly profit from XNLRDF. | the OLAC user scenario assumes a human user looking for resources and tools, whereas XNLRDF is designed to allow applications to find resources autonomously given a text document to be processed and a task to be achieved. | neutral |
train_98964 | In terms of syntax, for example, ASL has topiccomment syntax, while English uses Subject-Object-Verb. | the word order is "what"+"name" in JSL, which is more alike spoken English. | neutral |
train_98965 | Such computer-based tests can well assess three skill areas such as reading, writing, and listening comprehension. | transcribers use Ordinate's Internet-based transcriber interface. | neutral |
train_98966 | VS sentences can be used as declarative sentences when the subject is unidentifiable as well as an answer to a question like shei2 lai2 le? | this section can be summarized as follows. | neutral |
train_98967 | A total of 400 participants, including students, tutors, and the Course Coordinator (CC), took part in the two boards. | as in P020, while the third posting was a repair to prior postings, it also became a trouble source, as there was no file attached, even though the 1 Transcription conventions in this study: data cut-off by transcriptionist (( )) commentary by transcriptionist → points out a phenomenon under scrutiny Underline highlights parts related to the phenomenon under scrutiny Re: original mark in the Web-based discussion board in front of topic line indicated the posting being in reply. | neutral |
train_98968 | Also, it has been shown that an order-M seimi-CRF is strictly more powerful than an order-M CRF. | if this measure is less than a threshold, then the result of dictionary-based n-gram method was adopted, otherwise the results of sub-word tagging were adopted. | neutral |
train_98969 | Four segmentation corpora were presented in each Bakeoff. | the basic features that we can use are characters themselves. | neutral |
train_98970 | If the current character is " ", then all active features will be " "," ", " ", " ", " ", and " ". | we see that it is not better than character-based tagging method, which is 0.974 as seen in Table 7. | neutral |
train_98971 | 1 As specified in [2] and [3], TimeML annotates for temporal relations with four TimeML tags <EVENT>, <TIMEX3>, <SIGNAL>, <MAKEINSTANCE> and also with three link tags <TLINK>, <SLINK>, and <ALINK>. | the scope of the present discussion, however, remains at the purely descriptive level. | neutral |
train_98972 | But it seems to refer both to a date and to an associated event (instance). | for inviting me to present a paper at PACLIC20, I would like to thank Maosong Sun, Tingting He, Po Hu and other Organizing, Programming, and Local Committee members, and also Chu-Ren Huang and other Steering Committee members of PACLIC. | neutral |
train_98973 | This is augmented with default syntactic and semantic specifications which reflect shallow but efficient human sentence processing. | phase 2 (300-500 ms): Lexical-semantic and morpho-syntactic information is processed in parallel without interaction, leading to the assignment of thematic roles. | neutral |
train_98974 | This is tantamount to say that the first NP is syntactically interpreted as subject and the second as object by default. | no attempts have hitherto been made to model human sentence comprehension based on the data provided by those methods from a unified formal point of view. | neutral |
train_98975 | The second step scores Bβ by examines all the wordforms ending with the string. | table 1 lists out the experiment results obtained with different dictionary scales. | neutral |
train_98976 | Too small a window will lose a lot useful information, which makes internal relation of community not tight enough, and will lose many important relations of NE pairs. | we modify e ij because of weighted network, as in equation 4; Tr(e)=∑ i e ii gives the fraction of edges in the network that connect vertices in the same community; ‖e‖indicates the sum of the ele ents of the matrix e. m where e is an edge connectin ion. | neutral |
train_98977 | The small characters in the left side of Chinese-transliterated characters represent different consonant pronunciation, called word initials. | otherwise, there are 317 items in type Cy, accounting for 32.9521% of the whole types. | neutral |
train_98978 | It is necessary to point out that the word initials and the word endings do not really stand for Mongolian sounds or syllables, in stead, they are supplementary symbols for transliterating Mongolian letters to Chinese characters. | the Secrete History of the Mongols (SHM) is a special Mongolian historical document transliterated with Chinese characters as phonetic symbols, which feature causes many multivariate analyses in later ages. | neutral |
train_98979 | We consider the advantages of case grammar [20] , transformational generative grammar [21][22] and Wordnet [23] fully, and try to form a strategy to extract the conceptual expressions, and then, process the information in a semantic way based on the conceptual expression. | one is Jelinek-Mercer Model based on Term Frequency, the other is Bayesian Model based on Term Frequency. | neutral |
train_98980 | However, it is hard to confirm the concept of each query exactly for separate queries always lack strong phraseological relationship between each other. | in this phase, the chunks in the sentence and their corresponding possible sentence category expression will be hypothesized according to the words in the sentence and their concepts recorded in the "CE" of the WCK. | neutral |
train_98981 | In this proposed method, the translated sentence in the shaded portion may not be always correct. | an over the counter component is utilized for English speech synthesis. | neutral |
train_98982 | It is different in our method that we consider This work was finished while the first author visited Microsoft Research Asia both feature template selection and tag set selection, instead of feature template focused only methods in previous work. | a feature template set we selected is shown in Table 1. | neutral |
train_98983 | Syntactic meaning is no exception. | the representation and acquisition of knowledge play a key role in studies on information processing and natural language interpretation. | neutral |
train_98984 | In conclusion, one feasible way is to make use of the preliminary result of terminology extraction without segmentation, and add these newly-found terminologies into the general dictionaries for the segmentor. | we must identify terms first before validating that they are domain specific terms, which is the second step called terminology extraction. | neutral |
train_98985 | In this paper, we have discussed the effect of segmentation errors on terminology extraction. | the criterion of judgment is very simple: The candidate string would be accepted as a term if its internal associative strength is larger than a given threshold. | neutral |
train_98986 | There are 11 types of phrase types in the corpus including BNP, BAP, BVP, BDP, BQP, BTP, BFP, BNT, BNS, BNZ, and SV (see details in [18]). | the noun phrase rules are generated from the instances which are initially extracted from corpusS, then refined by re-testing on the same corpus without BNP chunks, and finally manually checked as shown in table 1. | neutral |
train_98987 | • Applies the Rset to candidate bi-gram noun phrase collocations. | the χ 2 -test is used in this work because it does not assume normal distribution probabilities. " | neutral |
train_98988 | "generic" dictionary rules A→B A∈V C , B∈V T "generic" phrase structure rules A→α A∈V X , α∈(V C ∪V X ) ＋ "specific" dictionary rules for request ID=i A→B A∈V C i , B∈V T "specific" phrase structure rules for request ID=i A→α A∈V X i , α∈ Fig. | compared to the Web or other open domain retrieval tasks, a manual/FAQ search is a very domain-specific task, where the terms used in documents and queries are limited, and this causes high overlapping among the terms of different queries. | neutral |
train_98989 | Today Penn treebank is already seen as a gold standard of training and evaluating parser. | based on the national standard "POS tagset for Chinese information processing" (2003) and popularly used "Grammar system for middle-school teaching", we propose a set of word class with 13 types: According to the three basic elements consisting of dependency relation: governor, dependent and dependency type, the tagset of Chinese dependency type is built. | neutral |
train_98990 | This approach is based on the assumption that if two words in different languages have similar contexts, it is most likely that they are mutual translations. | bilingual Named Entity Extraction is important to some cross language information processes such as machine translation (MT), cross-lingual information retrieval (CLIR), etc. | neutral |
train_98991 | In (10b), sensei is the subject of yonda and the subject of kaita is omitted. | in this case, the type π r π for dake is inappropriate. | neutral |
train_98992 | Furthermore, Buszkowski formally proved that grammars based on free pregroups are context-free [1]. | we will prefer the alternative analysis in which we assign the new type π rs s l to the topic marker wa, as in (1c), such that the resulting sentence is of types, that is, a topicalized sentence. | neutral |
train_98993 | Their algorithm to define a term needs concept semantic knowledge and other knowledge, and segmented and tagged texts. | in Chinese texts bilingual LWPs can be classified as follows: A：complete About this type interpretation, we designed a module to process. | neutral |
train_98994 | In the second part, the CES sentence category is explained, and the third part gives out the result of chunk share in CES sentence. | moreover, we explain the data preliminarily. | neutral |
train_98995 | As the chunk with its own construction, it is necessary to think over the chunk share from another view, the whole and part. | it probes into the problem of sentence semantic chunk share based on the sentence category expression of HNC theory. | neutral |
train_98996 | Again, there is no doubt the milk is liquid food in the literal interpretation. | fourthly, we show how the derived generalizations can be applied in deep linguistic account. | neutral |
train_98997 | Kim ([11]) assumes that the first clause is a subordinate clause only when it does not bear an explicit tense marker as in (1a). | if we are going to account for the tense phenomenon in (5a) with reference to the marker -ø past , we would need to posit another null form tense marker, i.e. | neutral |
train_98998 | We saw why such ambiguity arises and figured out the differences between the two types of clauses concerned. | the first clause in sentence (1a) can be an adjunct when it does not contain any tense marker. | neutral |
train_98999 | As we can see in (11), -ko as a subordinate marker cannot combine with a verbal stem containing a tense marker. | the sentence is actually ambiguous between a past tense and a present tense reading when it does not have the adverb 2 . | 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.