id
stringlengths 7
12
| sentence1
stringlengths 6
1.27k
| sentence2
stringlengths 6
926
| label
stringclasses 4
values |
|---|---|---|---|
train_98800 | The integrated resource is readily usable, and we have discussed some potential use of it in WSD. | wilks, 1975;Small & Rieger, 1982;Miller et al., 1990). | neutral |
train_98801 | However, though we have these rules of thumb for our judgement, with borderline cases we have but our intuition to count on; and since it is quite difficult to distinguish such varying degrees of accuracy in practice, we will group them into a single category called Accurately Mapped (A). | hence we can exploit the structural inter-relatedness to map senses from the dictionary to the mediator and from the mediator to the thesaurus separately. | neutral |
train_98802 | This was because we observed that high level nodes were often too general to be helpful. | this may not be straightforward. | neutral |
train_98803 | Although in the current study we have focussed on the direct mapping from the mediator to the thesaurus, the evaluation criteria described below also apply to the direct mapping from the dictionary to the mediator, and most of them to the indirect mapping from the dictionary to the thesaurus. | the LDOCE definition corresponding to the sense of synset X (i.e.-Dx) is expected to be similar to the textual gloss of synset X (i.e. | neutral |
train_98804 | All of these remain in our future work. | y is a position indicator, range from 0 to 15. | neutral |
train_98805 | If the first reading in (10c) is chosen, the transition state in (10d) is CONTINUE, and if the second reading is chosen, it is ROUGH-SHIFT. | under our analysis, (5) can be represented as in (11). | neutral |
train_98806 | The second sentence describes a state. | it precedes the utterance time, because the past tense is used. | neutral |
train_98807 | Nevertheless, they need further elaborations. | they may possibly signal turn-initial positions or they may express special pragmatic functions such as surprise or hesitation. | neutral |
train_98808 | Words used to initiate turns mark the beginning position of all turns and at the same time they are also beginnings of a number of utterances. | it is very likely that discourse particles and speech disfluency are highlighted by prosodic means to emphasise the semantic and syntactic inadequacy. | neutral |
train_98809 | The extra-linguistic knowledge that the speaker is male may be indicated by describing the fact with some conjunction like where: e. where male( ) When the extra item is added as in (5), the effect is usually taken into account by restricting the domain where the semantic representation should be interpreted. | per son(i) When combined with the set of axioms (4), the axiom turns the give semantic representation into the ohter below: aru I (vehicle(x) 3.4 Points to note We ensure that the exponential, !, should not distribute under any other connectives but the multiplicative one. | neutral |
train_98810 | Generation result is: "John's sister" . | c. He was excited that he could finally buy a piano. | neutral |
train_98811 | This point is illustrated as in (14). | the focus of this phonetic experiment is on the middle stray mora, which may possibly attach either to the preceding foot (F3), or the following foot (F4). | neutral |
train_98812 | In HAPPINESS IS PLANT metaphor, HAPPINESS maps to the result of cultivating a plant. | growth is an image-schema in the source domain of plant that is preserved when mapped onto the target domain of love. | neutral |
train_98813 | Focusing some element in a sentence amounts to making a statement to the effect that some predication holds of the object that corresponds to the focused element in some other circumstances. | in (15a) is somewhat strange in isolation without "Tabe-naku-te-mo, hutoru." | neutral |
train_98814 | Contrast these with wa and mo attached to nominal phrases. | the presupposed number must be less than three. | neutral |
train_98815 | Like Cho & Chai (2000), our analysis also provides the same explanation for various scrambled sentences such as the Double Accusative Construction (DAC) in Korean.' | this difference in the scrambling possibility just follows from the fact that under our analysis, (18b) violates the RM and ANI Constraint of the Argument Constraint but (19b) does not violate any LP Constraint. | neutral |
train_98816 | (The PM value of the NP Mary is `Acc/Daf , while that of the NP ton is 'Ace) Nor does the RM ANIMACY Constraint apply to this case, since their ANIMACY is different. | when the location is focused, the adjunct NP i san can bear a (contrastive) 'focus' marker, which happens to have the same morphological form of nominative case, as shown in (9b). | neutral |
train_98817 | All features of functional categories are typically uninterpretable. | although I do not discuss the structure of abstract nouns, I assume that the analysis ensued for non-abstract nouns will carry over to them as well. | neutral |
train_98818 | However, if the noun refers to a specific object, a classifier must be present. | further, under K, both Num and CL must be lexically realized. | neutral |
train_98819 | (30) a. Tom has written a novel, but Peter never has i. b. | 4 Following Zwicky and Pullum (1983), we take n't to be an inflectional suffix. | neutral |
train_98820 | This prediction appears to be correct: It is an important semantic fact that a VP modifier never outscopes a higher verb. | based on these empirical properties of auxiliaries, this paper argues for the existence of the construction aux-headph(rase) whose subtypes include negation-ph, inversion-ph, ellipsis-ph, vp-filler-ph, and the like. | neutral |
train_98821 | That these languages do not show Complex NP and Adjunct island effects is well-known and it has been proposed that it can be explained if what is moved is the the whole island containing the WH-phrase, not the WH-phrase inside the island (Choe 1987;Nichigauchi 1986Nichigauchi , 1992Kishimoto 1992;Yoon 1999, etc.). | wH-in-situ languages like Korean, Japanese and Sinhala could be taken as languages where both the direct and indirect wH-feature checking movement are covert. | neutral |
train_98822 | However, if Based on this contrast, he argues that pied-piping is a last resort measure made possible only when it is forced by the need to ensure convergence: pied-piping the DP in (i) is grammatical since moving the WH-word alone will violate the Subject island constraint The same does not hold in (ii) since moving the WH-word out of the object DP is possible. | imbabura Quechua (9) [ima-ta Juan randi-slikato-taj pro ya-ngui? | neutral |
train_98823 | Given that the indirect WH-feature checking movement is covert in English, the prediction is that an XP can be pied-piped even when the WH-phrase is not in its Spec overtly as long as it could move into the Spec covertly. | this will be crucial in my account of pied-piping in English, as will be seen shortly. | neutral |
train_98824 | What is crucial in our discussion on the existence of covert pied-piping is that in order for an element in syntactic islands like complex NPs and adjunct clauses to be questioned in Sinhala, the second part of the WH-word, i.e., d a , must appear at the edge of the island, not next to the first part of WH-word inside the island. | in my analysis clausal pied-piping in these languages is explained in the following way: On its way to the Spec of +WH-C, the WH-phrase will pass through the Spec of intermediate CPs due to Subjacency. | neutral |
train_98825 | Given the hierarchy (12), the type nonfinite-head-subject-construction is a subtype of the type clause and also a subtype of the type head-subject-phrase, and so it inherits all the properties of the two supertypes. | it has been long recognised that verbs like begin are ambiguous: one belongs to the strict transitive type taking a pure NP complement while the other is what we might call an oblique verb as it takes an infinitival or participle VP complement obligatorily. | neutral |
train_98826 | The Subcategorization Feature Principle in Fig. | they are represented by the lack of constraints on SUBCAt feature and the non-empty value of tENSE feature in ADJCNt feature, respectively. | neutral |
train_98827 | Adjunct-head schema (b) corresponds to Head-Adjunct. | pseudo-lexical rule schema allows the higher constituent of the complex predicate to inherit the argument structure of the lower predicate, and as a result, the schema makes the two predicates form a constituent in syntax. | neutral |
train_98828 | To reflect the transition, IHRC must rather be defined in relation to the main verb. | it is characterized by the apparent syntacticsemantic discrepancy; iHRC apparently involves a clausal complement, which, however, designates not the event denoted by the clause but a certain entity participating in the event. | neutral |
train_98829 | Note that this phenomenon is separate from the development into adverbial clauses which took place 1000 years ago. | the indirect linkage via R-relation is an advantage in that it offers a more comprehensive and adequate account of IHRC. | neutral |
train_98830 | The plausibility of this kind of approach can be testified in the other well-studied cases as shown in 19). | many previous studies on gerund phrases either do not defend endocentricity, or create a new syntactic category, or restrict the analysis to some subtypes of GPs. | neutral |
train_98831 | I like (it) very much for him to apply for the position. | verbal gerunds are still a kind of verbs and their forms are determined by the usual grammatical apparatus. | neutral |
train_98832 | 3 Syntactic Analysis We use the CKY algorithm that was developed for the Chomsky normal form to parse a sentence. | we use only 1,000 sentences to find the optimal segmentation words. | neutral |
train_98833 | In addition, besides the part of speech information they used word information. | the probabilistic model is composed of two probabilities, the phrase occurrence probability and the modification probability. | neutral |
train_98834 | Therefor we constrain the position of the indices by reference to the keywords. | these are conditional probabilities on the semantic role and part-of-speech given a set of possibly contexts. | neutral |
train_98835 | The parser consists of two main modules, a training module and a parsing module. | the current standard evaluation approach which consist in testing on unseen 10% of a training corpus may serve neither of these Figure 8: The closure aims, as it does not distinguish the system quality from the representativeness of the training data for the testing data. | neutral |
train_98836 | Following Culicover andRochemont (1983, 1990) and Horvath (1986), we in this paper assume that this essentially semantic conception of 'Focus' can be characterized as a purely formal syntactic feature [+Focus] or [+F], which gets assigned to constituents at a certain appropriate level of syntactic representation, triggering such syntactic operations such as 'Movement' and 'Adjoining' under the general syntactic principles and constraints. | this generalization holds true only in Mandarin and perhaps in some other varieties of the Chinese language but certainly not in all varieties of the language. | neutral |
train_98837 | From the constraint (16), the Focusing Rule and the Focus Resolution Rule are forced to apply somewhere inside the clause. | (In this case, the whole NP with the sentential modifier is focused and associated with mo.) | neutral |
train_98838 | The meaning of this NP is specified in the CONT feature. | in (7b), the associate can either be Taroo or imooto or Taroo no imooto. | neutral |
train_98839 | In English, association with focus is more or less a clear-cut phenomenon. | in order to do so, we need the definition of the function only. | neutral |
train_98840 | But this is not the case; (6a) simply does not have such reading. | i TOP Hanako DAT Taroo ACC introduce do PAST only COP "i introduced only Taro to Hanako." | neutral |
train_98841 | in the case of NFA), this rule checks if the FOC values of the head daughter and the FP have identical values with respect to the TG and OP features. | at least according to the intuitions of mine and some native speakers I consulted, the sentence (i) does not seem to have such drastic contrast in interpretation to its counterpart in which the case marker o is missing (i.e. | neutral |
train_98842 | bread only ACC eat PAST (I) did nothing but eat bread. | the sentence would be assigned an illicit interpretation. | neutral |
train_98843 | Taroo is said to be included in the event if it is coreferential with the argument of the verb. | b. Hanako-ga tonari-no gakusei-ni piano-o asa-made hik-are-ta. | neutral |
train_98844 | assassin-Nom senator-Acc kill-past "The assassin killed the senator." | the adversative interpretation does not arise. | neutral |
train_98845 | Moreover, the V in this construction must be a transitive verb whose subject is an agent. | we have identified a potential lacuna in the organization of a clause in terms of pertinent arguments. | neutral |
train_98846 | As shown by (19a), a noun phrase in a genitive possessor induces an i-within-i violation if it is coindexed with the entire possessed noun phrase. | (15b)) b. Te-wa(,) Taro-ga naga-i (Cf. | neutral |
train_98847 | The words listed as entries in an Indonesian dictionary are usually root words; the various affixes applicable to a root word are usually listed together in the corresponding entry. | this situation is inevitable since we have to handle morphing and thus we have to consider every possible replacement of the first letter(s) of the stem. | neutral |
train_98848 | How to avoid or reduce the SVD cost is the major problem of LSI approach, especially for multi-linguistic IR. | how to represent a document on a semantic basis rather than on a pure statistical basis is one of the chief goals of our further research. | neutral |
train_98849 | Kim 1998a The key point of the KPSG treatment includes the type constraints on v-mod-w and gapintroducing rules. | for a verb stem to appear in syntax, it should be inflected at least with a mood marker (cf. | neutral |
train_98850 | The output of this stage becomes the immediate input to the syntactic analysis component, bypassing the tokenizing and morphology transductions that perform these functions for the other ParGram languages. | f-structures are meant to encode a language universal level of analysis, allowing for cross-linguistic parallelism at this level of abstraction. | neutral |
train_98851 | Topicalized noun phrases may have certain postpositions before the final ha, as in (3a). | just as with the japanese ha, the Korean topicalizer also cannot cooccur with postpositions marking the basic grammatical functions, as in (3d). | neutral |
train_98852 | We anticipate no problem for open class items with predictable subcategorization frames such as nouns, adjectives, and adverbs. | the Japanese entry for wo was ported to the Korean entry for +Acc. | neutral |
train_98853 | In this regard, Jackendoff is absolutely right, when he suggests that the mental rehearsal of sentences may be the primary means by which our own thoughts are able to become objects of further attention and reflection. | *I wish to thank Prof. Shuanfan Huang and two anonymous reviewers for their useful comments and suggestions. | neutral |
train_98854 | The proofreader can see the other examples as references to determine whether or not each tagging result is right. | for instance, in Table 1, the keyword-in-context file of the word '1111: f9: (research), which has pos-categories of verb type VE and noun type Nv, is sorted according to its left/right context. | neutral |
train_98855 | A cost-effective algorithm may not be the algorithm with the highest precision. | a better accuracy can be achieved if the tagging results are manually proofread. | neutral |
train_98856 | U (wei) 'hello' with a rising tone illustrates a greeting, particularly used to initiate a telephone conversation. | this word will bring about effects on the audience, i.e. | neutral |
train_98857 | It is shown that there is a special group of one-word small onomatepoeic speech act verbs which perform perlocutionary forces. | a Chinese character) for interjections and onomatopoeic speech act verbs in Mandarin Chinese tend to have the radical D mouth'. | neutral |
train_98858 | Since we describe the onomatopoeic speech act verb as a syntactically independent monomorphemic utterance which performs perlocutionary forces, it is feasible to say that onomatopoeic speech act verbs are a subset of interjections. | examples are ill (pei)'bah' and I"-(heng)'humph'. | neutral |
train_98859 | It is found two word forms for this particle in the Sinica Corpus IS and l ' -; . | they may appear as typical verbs as well. | neutral |
train_98860 | A: "(No, that was because) we were going to buy a mat for my father; he slipped and fell down that morning." | (Su 1998)4 In this regard, the linkingfiu acts more like an utterance-connective, even when it does not combine with such connectives as ranhou, hou-lai and shuo-yi. | neutral |
train_98861 | Results showed considerable variation and indicated that the phenomenon was relatively most common in Beijing text -up to more than 18% of verbs undergo the verb-noun transitional process to various extents. | in the last section we have seen that there is a continuum of the verb-noun shift and Table 10 shows examples with different r values. | neutral |
train_98862 | (Linebarger 1987:338) For example, NPI is grammatical in the next sentence (13), for the reason that negation does not have scope over the because-clause in LF: (11) He didn't budge an inch, because he was just a wooden statue. | (Zwarts 1993;van der Wouden 1997) Zwarts (1993,1998) classifies three types of NPIs which are "weak", "strong", and "superstrong", according to DE-ness, anti-additivity and anti-morphy respectively. | neutral |
train_98863 | The experiments by Lewis et al. | for each category we learned a binary classifier and evaluated its performance on test set by computing a value of f1 . | neutral |
train_98864 | Our results showed that the SVM and the variants achieved similar performance on Chinese with those on English for document categorisation for large categories, despite big difference between English and Chinese. | a document categorisation problem with only a few relevant documents for training may have similar situation with that in the right graph of Figure 1. | neutral |
train_98865 | Regarding that some extra errors were introduced when preprocessing the Chinese documents, we can speculate that the SVM and the variants are quite robust to the noise in feature vectors for large categories but are sensitive to the data noise for small categories. | we had to segment Chinese sentence into words. | neutral |
train_98866 | At the same time, the results of experiments are figured out and analyzed. | for example, the Construe [Church , K.W.Lisa.f.Rau 1995] is developed based on it. | neutral |
train_98867 | The results judged by machine are regarded as correct, by means of artificial discrimination. | generally, feature weight is computed and normalized. | neutral |
train_98868 | According to a study in cognitive science, people often disambiguate word sense using only a few other words in a given context (frequently only one additional word) [8] . | the blithesome collaboration with Dr. Ying Chenjin and Mr. Guo qingjian from Chinese Department is memorable for all of us. | neutral |
train_98869 | We define an left-open edge, <w,v] of a GST as the edge that contains nodes between the upper node Lemma lc: Each substring that terminates in a node between a left-open edge <w, v] has the same tf and df. | a left-open edge <w, v] of the GST has a upper node w and lower node v. The substring "aca" that terminates in <w, v] has a counts of 3;2, whose occurrences are underlined in each of the strings in the GST. | neutral |
train_98870 | The discussion in section 2.1 explain Property E can be achieved by defining domination range which classify the substrings in a GST into 0(N) classes in 0(N) time. | in addition, the time grows linearly with our corpus size. | neutral |
train_98871 | In addition, the new child state will be the suffix state of the child state. | the proof of Property F for GSt is achieved by considering the following formulae: given a substring w = xyz, where x, xy, xyz are the longest substrings in their respective classes: the above formulae can be computed in constant time, since each of the involved in the formula can be accessed from root by traversing one of the substrings: w, y, xy and yz. | neutral |
train_98872 | They are required in multigram compound discovery. | following that, we set the SID of the suffix state to j if the STD of the suffix state is less than j. | neutral |
train_98873 | A similarity between two sentences of a language is a nonempty sequence of common items (root words or morphemes) in both sentences. | the SttL and DttL are combined as the template learning algorithm (ttL). | neutral |
train_98874 | Then we use GIS algorithm to estimate parameters of all existing features in the model. | and every feature must encode a binary value that might help predicting the indexing result. | neutral |
train_98875 | We have compared the results of ME model with the method of scoring by experience (Li 2003) and Chien's(1997) The result is not very satisfying. | the first are those which have rules of composition, mainly names and foreign terms. | neutral |
train_98876 | Then we can make parameters more objective. | we can conclude that the fault doesn't lie in the ME model itself. | neutral |
train_98877 | In our method, we didn't make use of any training corpus. | the detailed results are as follows: Precision for topi-k Extracted Ms 1 2 3 4 5 6 8 9 10 K x 1000) (Fig. | neutral |
train_98878 | CLEs consist of 1,252 entries and 1,913 functional annotations for 675 words. | the Average is the average of the 200 averages for all parses for each sentence. | neutral |
train_98879 | As described in Section 1, an f-structure is meant to encode a more language universal level of analysis. | then the lexical entry for each word in the input is automatically created (Sentence Lexical Entry: SLE). | neutral |
train_98880 | The most similar word is chosen as the answer to the back-transliteration problem. | there is one further point which should not be ignored. | neutral |
train_98881 | Concluding remarks are made in Section 5. | we use the similarity-based approach to model the task of back-transliteration. | neutral |
train_98882 | Moreover, YC is derived from the extension of a chunk (hence extended chunk or EC) and is essentially a sentence (called EC sentence and represented as ErJ) in the sentence format (SF) of !31. | the English translation may or may not inherit the SC of the source sentence, despite the fact that it does, more often than not, take a different one. | neutral |
train_98883 | This candidate is also out because it violates *RED, which, according to the ranking in Cantonese, is not allowed. | useless or redundant c-structure constituents should not be allowed. | neutral |
train_98884 | However, the data of serial verb construction using manner verbs is small in our corpus. | one informant split the instructions into two parts. | neutral |
train_98885 | The HPSG valence/content analysis is that the content of a case phrase structure-shares with a part of the content of its immediately larger constituent only through the valence of its head verb. | b's response to A's utterance in (2) sounds worse than that in (1). | neutral |
train_98886 | See Kay 2002 for the meaning or function of a construction. | 3 in Sag's analysis, a phantom relation corresponding to a verb (a relation of a verb not occurring but its meaning existent) is postulated in syntax or semantics. | neutral |
train_98887 | And it may be the human nature to like and pursue the economy. | we recognize the semantic pattern of huaiyi sentences by its function, not by the meaning of huaiyi, which is not the fundamental basis of semantic recognition in speech communication." | neutral |
train_98888 | Also the training result is usually hard for human beings to understand, thus it's relatively hard to integrate expert knowledge with the machine learning method. | if there're too many unknown categories, the document might be of another domain. | neutral |
train_98889 | They employed literary texts written by foreign authors. | judging from these facts, it is very significan that the verb "fukumu (to include)" has 85 examples. | neutral |
train_98890 | and Ck has direct relations with C. Here C and R indicate concepts and semantic relations, respectively. | his method differs from ours in that we consider all semantic relations in the ontology, not taxonomy relations only. | neutral |
train_98891 | Note that in ZA detection rules, we employ the rule 1 to detect omitted cases as ZA candidates, and the rule 2 is an exceptional case which a verb appears in the initial position of an utterance but the verb is an intransitive verb tagged as VA or VH. | they are not constrained by features of any previous utterance in the discourse segment (DS), and the elements of CI(U") are partially ordered to reflect relative prominence in U,,. | neutral |
train_98892 | To investigate this, I will make a distinction between segmental and tonal neighborhood. | next, the study demonstrated that within each category, defined by whether or not the stimuli obey the postulated voicing constraint and whether or not the clusters exist in the language, there were segmental neighborhood effects. | neutral |
train_98893 | Moreover, another objective of the experiment is to obtain additional findings concerning segmental and tonal neighborhood effects on speakers' well-formedness judgments of novel words. | there exist a number of thai words that begin with the thai letter representing /t/ followed by the letter for they are not pronounced together as a cluster. | neutral |
train_98894 | As has been observed in many studies, various functional relations are expressed by particles in Japanese. | phrase structure rules and constraints for focus particles are described in the grammar rules. | neutral |
train_98895 | 1985), the dependency between which man (the filler) and loves (the head that selects it) is usually mediated through an empty category (trace) located in the filler's "canonical" position. | obviously, theories with no movement operation (GPSG, LFG, HPSG, categorial grammars, etc.) | neutral |
train_98896 | As narrated sequentially, these events took place without any overlap or inclusion. | the verb "ca" (sleep) is marked with the PAST tense marker "ss". | neutral |
train_98897 | manner adverbs) restrict the range of events referred to; so that when such adverbs are selected by a verb, they contribute a necessary part of the information that the VP supplies in defining a given situation, just as an object selected by a transitive verb. | it appears difficult for such a functional account employing the paraphrasability to explain the difference between (23a) and (23b), where an adverb is in the sentence-initial position. | neutral |
train_98898 | Just like the LICENSOR feature, it originates from a CA and acts as the syntactic "connector" of the CA and its correspondent VE. | the modifier-modified relationship can also be handled more easily once we have a syntactic mechanism connecting the two elements. | neutral |
train_98899 | The sentence would have to be grammatical in whatever order the CAs are, as far as they are under the scope of their VEs. | there are two different types of unboundedness involved: one between a concord adverbial and a verbal ending and the other between the adverbial as a modifier and a predicate. | 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.