id
stringlengths 7
12
| sentence1
stringlengths 6
1.27k
| sentence2
stringlengths 6
926
| label
stringclasses 4
values |
|---|---|---|---|
train_96200 | Table 2 shows an example of the annotations for a dialogue from the Switchboard Corpus. | this new type was then considered in each subsequent utterance as a possible attribute type -along with the properties from Schema.org -if an attribute instance was found. | neutral |
train_96201 | This is the first Vietnamese dialog act corpus. | the inter-annotator agreement score is computed by Fleiss kappa measure (Fleiss and Cohen, 1973). | neutral |
train_96202 | There were several dialog act corpora available to the research community like as TRAINS (Traum, 1996), VERBMOBIL (Alexandersson et al., 1998), SWBD-DAMSL (Jurafsky et al., 1997), MRDA (Shriberg et al., 2004), AMI (McCowan et al., 2005) and so on. | in the pre-processing stage, we build a Vietnamese dialect dictionary includes 167 distinct Southern Vietnam dialect words, 55 distinct Central Vietnam dialect words and their translation to the "standard" North Vietnam dialect. | neutral |
train_96203 | suggest/AqtrH ,)اقترح/ Directives (i.e. | performative verbs express specific action belong to a purpose such as: Assertives (i.e. | neutral |
train_96204 | In this annotation, the main proposition (P1) is often not required for annotation, as it is assumed to be marked by the full stop signalling the end of the sentence. | classification experiments are carried out using the Weka 3.8 Workbench (Hall et al., 2009). | neutral |
train_96205 | Furthermore, conversations are not simply conducted while sitting around a table, which is the case in most conversation corpora. | are is a demonstrative pronoun, so it can be produced without mentioning the object it refers to when spatial-deictically used. | neutral |
train_96206 | We need to craft concise instructions. | the web pages cover a variety of domains and the first three sentences are long enough to annotate with discourse relations through crowdsourcing. | neutral |
train_96207 | We developed a computational thinking coding scheme, as shown in Table 1 Every dialogue turn is labeled with one of the codes in Table 1. | the full model architecture is shown in Figure 1. | neutral |
train_96208 | Ideally the model should be time-variant and uses time step as variable too, because the time/stage of a dialogue may have an impact on the labels. | for LSTM-CRf, our final model has 1024 LSTM units to process sequences of sentence embeddings, and the hidden layers have 1024 and 512 neurons, respectively. | neutral |
train_96209 | We detect those by inspecting annotated texts for sources of disagreement between annotators, and redefining the annotation scheme so that these disagreements are minimized. | in order to assess the reproducibility of human annotations, we used Cohen's kappa (Cohen, 1960). | neutral |
train_96210 | However, not considering fact premises as part of the annotation comes unnatural and is inadequate from a point of view of argumentation theory and of application utility. | approximately half of the words were annotated as belonging to an argument component, as can be seen in Figure 1. | neutral |
train_96211 | Claims are the central components of an argument and they either support or attack the major claim. | the number of cases where such analysis could be carried out is so small that we require a bigger corpus to obtain more significant figures and draw conclusions upon them. | neutral |
train_96212 | The vocabulary checklist application consisted of three parts. | cognitive Development, 10(2), 159-199. | neutral |
train_96213 | Collocations can be characterized by different levels of cohesion, ranging from fully fixed idioms (i.e. | the same search on the CORIS corpus (Rossini Favretti 2000) finds 1771 occurrences for nello stesso tempo and just 1908 for allo stesso tempo, so it is necessary to be careful in drawing conclusions from rough data. | neutral |
train_96214 | Revita lies at the intersection of two established areas of research: intelligent tutoring systems (ITS) and computer-assisted language learning (CALL)the project seeks intelligent solutions for language learning. | the learner needs to get more correct answers in a shorter period of time to win the competition. | neutral |
train_96215 | Weights are assigned not only to particular words, but also to their various grammatical categories; thus, if a user, e.g., sometimes makes mistakes in a certain nominal case, the system will provide more exercises for this case. | offering a fixed, limited set of exercises is in conflict with the principles of adaptability of the learning process to the profile of the particular user. | neutral |
train_96216 | This is clearly not the case (x 2 = 12.4, df = 2, p < 0.006, Yates' corrected and excluding the category 'other'). | we know very little about the prosodic realization of nouns and verbs in IDS in German. | neutral |
train_96217 | Occasionally the spelling errors may overlap with grammatical errors. | these properties have to be taken into account in error-annotation. | neutral |
train_96218 | The SGATe (SLA in Grammatically Annotated Texts) resource comprises the entire EF-Cambridge Open Language Database (EFCAMDAT) annotated with 107 pedagogically relevant grammatical structures. | the evaluation of a language learner's capacity when producing texts in a foreign language is not an easy task. | neutral |
train_96219 | In the absence of well-established written traditions and a single authority, the regional languages of France do not have unanimously recognised and accepted standards (Caubet et al., 2002;Vannin et al., 2012). | associating them to modernity (with innovative cartographic tools to present boundaries between dialectal areas), will hopefully help revitalis e our linguistic heritage, or at least give it recognition as a vehicle for creativity, for lack of being able to counteract the decline in minority language practices. | neutral |
train_96220 | We randomly split our corpus into 80% training, 10% development, and 10% testing data, based on sentences. | turning to the differences between the random and Crúbadán approaches to query generation, for each number of queries, the Crúbadán approach gives many more documents than the random approach, while the number of tokens is not drastically different between the two approaches. | neutral |
train_96221 | The consistency ratio of Alsatian, when measured this way, is equivalent to that of standard German, despite the lack of a unified orthographic norm in Alsatian, as shown in In this section, we use the pronunciation dictionaries to train a G2P system. | when there were doubts about some phonemes, specifically stops and their voice or voiceless quality, the waveforms were analyzed using PRAAT (Boersma and weenink, nd). | neutral |
train_96222 | • Output Phase Finally, the annotation is saved in a database. | in this context, ChAnot 4 is an intelligent web-based annotation tool focused in Peruvian indigenous and agglutinative languages, which allows text processing through morphological (lemma and morphemes), POS-tag, named entity and syntactic annotation (with an interface integrated with BRAT). | neutral |
train_96223 | The goal of the DLDP is to help minority language speakers' communities in the acquisition of intellectual and practical skills to create, share, and reuse online digital content in their languages. | there are a lot of necessary online and digital resources missing, and speakers often lack information about those that are available. | neutral |
train_96224 | Using textual features only, there is no significant difference between the performance of the two classifers. | for motion classification, the SVM classifier achieves accuracy of 92.1% and an f1 score of 0.921, while the MLP classifier obtains accuracy of 93.0% and an f1 score of 0.931. | neutral |
train_96225 | For the two-step model, we also propose an alternative method for determining motion sentiment that infers polarity labels from the relationship to the Government of the speakers who introduce the motions Additionally, we evaluate the use of n-gram textual features and a range of contextual features extracted from metadata related to the speakers. | considering the large overlap between the two sets of labels, for future work or to create larger datasets, manual annotation of these may not be cost-effective. | neutral |
train_96226 | 2 In this step we perform sentence segmentation, tokenization, lemmatization, morphological analysis, part-of-speech tagging and dependency parsing, following the Universal Dependencies scheme (Nivre et al., 2016). | coNLL-U files When extracting the text from the canonical HTML to pass it to UDPipe, we strip away all mark-up and discard all content marked for removal as described in Section 2.2. | neutral |
train_96227 | Each review is labeled with a manually assigned score of 1-6, as provided by the rating of the original author. | the pre-processing set-up is slightly complicated by the fact that the Norwegian language has two official written standards -Bokmål (the main variety) and Nynorsk -both of which are represented in the review corpus. | neutral |
train_96228 | In a companion publication (Rouces et al., forthcoming), we discuss applications of this resource in text mining for digital humanities research. | the intersection of the DA test set and the entries with BWS value is used for applying the rank-based measures. | neutral |
train_96229 | For each political party, we generated 3 extractive and 3 abstractive summaries, building 36 opinion In order to improve to annotation process, we established some guidelines in relation to how to build the summary, the aspects coverage and the summary length. | the primary research is centralized in some specific languages or domains (mainly English), due to the lack of summaries corpus to evaluate the process in different ones. | neutral |
train_96230 | Both steps, transcription and annotation, provide valuable information for an SLR system. | the units of interest in the assessment system in our project (cf. | neutral |
train_96231 | Our analysis showed that the distinction between permissible variants (Category 2) and incorrect productions (Category 3) of a sign was in some cases especially challenging. | participants were specifically instructed to sign the base forms of the lexical items, not modified versions based on the context evoked in the example sentences. | neutral |
train_96232 | object, but only a noticeable part of the object, as in the RSL sign CAT. | these parameters were only annotated as being iconic or non-iconic, without further analysis of iconicity. | neutral |
train_96233 | The video data used for the present analyses come mainly from the online dictionary of the Spread The Sign Project (Domfors and Fredäng, 2008) and also from LSD Visual Sign Language Dictionary (Hong Kong Sign Language: http://www.sign-aip.net/ sign-aip/en/home/index.php), Taiwan Sign Language Online Dictionary (Tsay et al. | the rest of the paper is organized as follows: Section 2 introduces some basic principles of lexicostatistics applied to SL, Section 3 describes the web application we created for phonological annotation of signs, Section 4 describes the comparative approach applied to the data, Section 5 reports the results of a case study on four Asian SLs, and Section 6 provides the analysis of handshapes of the sample of 23 SLs. | neutral |
train_96234 | List(s) of concepts/meanings which are assumed to be universally instantiated in the world's languages (e.g., blood, many, leaf, etc.) | in previous studies, Annotators' subjective perception could affect data evaluation in two steps of the procedure. | neutral |
train_96235 | The rest of the paper is organized as follows: Section 2 introduces some basic principles of lexicostatistics applied to SL, Section 3 describes the web application we created for phonological annotation of signs, Section 4 describes the comparative approach applied to the data, Section 5 reports the results of a case study on four Asian SLs, and Section 6 provides the analysis of handshapes of the sample of 23 SLs. | previous research has not attempted a systematic comparison of a large sample of SLs. | neutral |
train_96236 | Unlike translation of texts for which sentence-by-sentence progression is very likely wrong 5 , a picture-by-picture progression this time is often the result everybody expects. | for comparison, we used some of the LS-COLIN data (nonchronological tasks), which we have full access to. | neutral |
train_96237 | Contrarily to spoken languages, SLs have no written form which can be read out and captured without a trace of the preparation. | in LS-COLiN, informants were given the assignment and elicitation material in a separate room prior to standing in the studio, and could take the time to prepare their production. | neutral |
train_96238 | To enable studies of higher-level discourse operations such as rhetoric argument building, semantic sectioning and longterm contextualising, we developed a need for both prepared and long Sign Language monologues. | we argue that note support undermines the production whenever they are visible to the informant while they are performing, hence must be avoided then. | neutral |
train_96239 | In Language resource and evaluation conference (LREC), 6th workshop on the representation and processing of Sign Language: beyond the manual channel, Reykjavik, Islande. | the second argument "tourism" is a focus on the first argument "power" 2. | neutral |
train_96240 | We used the default settings of Tobii to identify the fixations and saccades, the details of which are described by Ollson (2007). | the main purpose of the corpus was to allow us to analyze the manual segmentation behavior and to identify modalities and features useful for modeling segmentation. | neutral |
train_96241 | In this paper, to detect when it is difficult for an interpreter to provide an interpretation, statistical analyses of missing translations in E-J simultaneous interpretations were described. | the cross-tabulation is shown in table 6. | neutral |
train_96242 | To confirm that the proportions of missing translations are significantly different between each part-of-speech and the overall average, a chi-squared test was applied to the proportions of missing translations of each part-of-speech and the proportions of other content words. | this also contradicts intuitive expectations because the root word is generally the main word of the sentence. | neutral |
train_96243 | In addition, the proportion of missing translations of numbers is 29.1%, which is 4% less than the average. | in addition, word translation correspondences were aligned manually for each aligned utterance unit. | neutral |
train_96244 | Note that the differences in the proportions of the missing translations between adjectives, numbers, and other content words are not significant. | names, which are proper nouns, and numbers are regarded as one of the problem triggers (Gile, 1995) and are likely to overload interpreters and be omitted in interpretations. | neutral |
train_96245 | Here, the horizontal axis is the speech rate and the vertical axis is the proportion of missing translations. | these analyses, which were based on observation, did not clarify the correlation between the identified factors and the occurrence of omissions. | neutral |
train_96246 | To better identify the pairs of labels that can be easily discriminated from those which can not, only binary models are trained thus resulting in an emotion confusion matrix. | for such speech intervals, the timbre remains the own speaker's timbre or corresponds to another character. | neutral |
train_96247 | Apart from their research value, spoken language resources are also indispensable in the development of speech recognition systems and related language technologies. | the latter two levels of annotation are provided automatically using the forcedalignment tools developed by (Koržinek et al., 2017). | neutral |
train_96248 | These efforts have been dwarfed by the vast amounts of user data that are being gathered by multinational corporate giants for the betterment of their proprietary technologies. | for the purposes of this paper, a frame width of 100ms was used throughout, giving each spectrogram a height of 64 pixels, a width of 100 pixels, and a depth of 256 shades. | neutral |
train_96249 | The corresponding audio snippets are played when the cursor hovers over a given datapoint, so listeners can hover over different regions and listen to hypothetically similar data in quick succession or simultaneously (Edlund et al., 2010) 7 . | it is less efficient if we know nothing of the data (the distribution in space will be random). | neutral |
train_96250 | The mean values of F 0 exhibit a statistically significant difference between medium and high cognitive load (p < .001). | if the length of the tolerance exceeded the limits of the original audio segment, as much silence as needed was added to fill the 200 ms at the beginning and/or the end. | neutral |
train_96251 | During this manual SAD correction and segmentation pass, annotators distinguished three categories: • Non-speech. | videos in the main corpus contain speech in English (including US, UK, and other varieties), Mandarin Chinese, and Arabic. | neutral |
train_96252 | Data scouting was conducted using a customized user interface developed for VAST, known as VScout. | if the recording contains continuous speech for 15 seconds with no noticeable pauses, annotators created a segment on the speech track whose duration is 15 seconds -even if the number of speakers or the languages spoken changed during that segment. | neutral |
train_96253 | While 6.3% and 9.3% of all sentences appear more than once in the British and French WaCs, respectively, in brWaC only 1.3% are repeated. | for Brazilian Portuguese, we describe how starting with a 1.6 billion token corpus and following a strict methodology (Section 3), the resulting corpus is expanded to encompass 2.7 billion tokens. | neutral |
train_96254 | Non-significant words are removed using Part of speech (POS) filtering and for feature selection, the mutual information method is used. | these features are integrated with word features to improve classification results. | neutral |
train_96255 | This dataset is licensed under the Attribution-NonCommercial-ShareAlike 3.0 Unported License 6 . | text documents are stored in the individual text files using UtF-8 encoding. | neutral |
train_96256 | news, Wikipedia, crawled webpages or literature (Baroni andBernardini 2004, Biemann et al. | figure 4 shows two distributions for each corpus of German. | neutral |
train_96257 | Although the choice of this theshold is quite arbitrary, it leads to a good tradeoff between the corpus size and the simplicity of sentences by selecting typically between 5 and 10 percent sentences (see Table 4 for details). | the corpora are still large enough for statistical analyses. | neutral |
train_96258 | Both initiatives have in common that rather than building comparable corpora from scratch they try to re-use existing corpora and currently mostly rely upon national and reference corpora. | 2014), within the DeutUng project. | neutral |
train_96259 | Ph, shown in green, is a monthly psychological magazine. | korAP already offers some unique features for the analysis of DeReko. | neutral |
train_96260 | The Icelandic Parsed Historical Corpus (IcePaHC) 5 is a diachronic treebank that contains about one million running words from every century between the 12th and the 21st centuries inclusive (Rögnvaldsson et al. | the downloadable corpus is divided into two parts: IGC1 and IGC2. | neutral |
train_96261 | The study of (GRABAR et al., 2014) proposes a specific lexicon in order to assess which words are potentially non-understandable and then require further explanations. | finally, the access mode as a "physician" user is an interface that offers in the translation point of view the opportunity for the physician to consult the technical term in his or her mother tongue. | neutral |
train_96262 | Moreover, despite the substantial amount of health-related information available on Internet, little is known about the quality and accessibility of that information. | this implementation is still being developed and assessed as we have done a distinction between term in popular language, by classifying it as a recurrent word mainly in oral talks between patients (for example, in clinic and hospital waiting rooms), and semi-specialist (or informative) term that has its place in informative articles and which is then extracted from written corpus. | neutral |
train_96263 | This tool is used in the indexing of articles in biomedical journals of PubMed, by the Higher Institute of Health of Rome. | most of these resources are prominently available for English and the access to terminological resources in languages other than English may not be so simple. | neutral |
train_96264 | In our work we utilize this approach of making data available based on a standardised data model, i.e. | there are approximately 8.1 million Xhosa speakers 3 , adding up to about 16% of the South African population. | neutral |
train_96265 | The equivalence of all object properties within the BantuLM vocabulary is created with the owl:equivalentProperty property. | further, it can be seen that this WordNet entry ultimately 16 leads to a sense definition of the lexeme reporter. | neutral |
train_96266 | 'These initiatives create a more mobile work environment and consequently reduce our housing footprint.' | finally, we provide some concluding remarks in sSection 7. | neutral |
train_96267 | For each morphological code in PSC-M we created a corresponding datatype property, subproperty of hasMorpholog-icalTrait, in order to relate a lexical entry with its morphological variants represented as strings. | in the rush to convert all kinds of lexicons and dictionaries into RDF it is perhaps the case that the technical limitations of this mode of publication, as well as the new potentialities which it offers, are not as thoroughly understood as they ought to be. | neutral |
train_96268 | The MetaNet repository readily provides a substantial number of such metaphors for a non-trivial number of schemas. | it turns out that these three individual instances of linguistic metaphors can be viewed as instantiations of a more general conceptual metaphor HAPPY iS UP. | neutral |
train_96269 | There is a manual alignment for verbs and nouns, while the validation step was automated. | this study aimed the development of a new lexical and synonym-based resource for the Shipibo-Konibo language. | neutral |
train_96270 | The greater the result, the relationship between words would be closer. | for instance, for the verb manéxti ("clean" in English), which was taken in the sense whose gloss is "to tie the hairs of the head or crown with other hairs of the same head", the process would be as follows: • Tie the hairs of the head or the crown with other hairs his own head to • Tying, hair, head, crown, hair, head -Only verbs and nouns • Tie, hair, head, crown -Lemmas were extracted and word repetitions were erased The similarity was expressed as a decimal number between -1 and 1 (being the result of a cosine). | neutral |
train_96271 | Their average precision for the three language pairs are: 91.8% for amod, 90.6% for nmod and 86.2% for vobj. | 1 www.dbpedia.org/page/Wiktionary In §3., we present the functions that we have developed in our API to retrieve information from the French Lexical Network in RDF/OWL format. | neutral |
train_96272 | Lexfom uses the lemon model to represent information about lexical entries and lexical senses. | we had false positives due to parsing errors. | neutral |
train_96273 | But this definition causes confusions because the modifiers do not always play geo-political entity roles (e.g., "Chinese" in "[GPE Chinese] food"). | we also attempted to project English word embeddings to Uyghur using a bilingual lexicon. | neutral |
train_96274 | We refer the reader to (Poostchi et al., 2016) for more details. | we have released all four word embeddings on GitHub 5 , which will allow easy replication of our experiments. | neutral |
train_96275 | 2 We use 0.1 for L2-regularization. | it is possible that the latter may be indicative of different parameters. | neutral |
train_96276 | The method showed advantages over the baselines. | dictionaries are not always useful. | neutral |
train_96277 | NER and NEL have been actively researched topics for decades (Finkel and Manning, 2009;Ratinov and Roth, 2009;Ritter et al., 2011;Derczynski et al., 2015; Nitish Gupta and Roth, 2017). | and MISC, we map the specific gold standard label to these four types for the evaluation. | neutral |
train_96278 | We exclude single pronouns such as I and me unless they seem extremely contextually relevant, such as in the case of It 7 , when discussing movies with the user. | the resulting entity classification is often coarse and does not encode an ontology. | neutral |
train_96279 | The lowest layers of the ANN tend to represent task-independent features, whereas the topmost layers are more task-specific. | fewer studies have been performed on transfer learning for ANN-based models in the field of natural language processing. | neutral |
train_96280 | In the upper part, there are examples of the form "#adv" (meaning either an adverb phrase or an adverb as a word) representing the time expression "when" (i.e. | to the original project of PDT, in these treebanks, the morphological and surface syntactic annotations were done automatically, and the manually annotated deep syntactic layer does not contain annotation of information structure and some other special annotations. | neutral |
train_96281 | Available treebanks are still, however, largely based on written textual resources, with a few exceptions (Dobrovoljc and Nivre, 2016;Östling et al., 2017). | a list of standardized interjections The extended annotation guidelines for the syntactic annotation of the LIA material is built on the work of Dobrovoljc and Nivre (2016), who describe the annotation of the Slovenian spoken language treebank with Universal Dependencies. | neutral |
train_96282 | It is to be noted that our benchmarking corpus SandhiKosh is expected to undergo refinements in the future as there is scope for its improvement. | for all the three tools, we considered all the results instead of just the filtered output and marked the output to be correct if any of the results matched with the expected output of Sand-hiKosh. | neutral |
train_96283 | We introduced the new version 2.0 of the treebank. | cLTT 2.0 contains 483 manually annotated relations classified into 3 categories. | neutral |
train_96284 | We divided the Chinese Penn Treebank 6 into training, development and test sets as indicated by Tse and Curran (2010). | another mistake observed in parsing subject relative clauses is caused by noun-verb confusion in Chinese. | neutral |
train_96285 | In (8), the direct object shiqing 'things' is moved to the front of the sentence. | since subject relative clauses and object relative clauses are both very frequent in the data, these two parsing possibilities are very competitive. | neutral |
train_96286 | After examining the statistics of relative clauses in Stanford dependencies of the training set, we found that when the subject is present in the relative clause while the object is missing, the relative clause is most likely to be an object relative clause; when the subject is missing but the object is present, it is most likely to be a subject relative clause; and if both subject and object are absent, it is most likely to be a subject relative clause. | we labeled an 'of-asso' dependency between the extracted topic and the subject of the relative clause. | neutral |
train_96287 | After 1200, IO > DO is relatively more frequent, with a peak around 1250. | in summary, we report the development of a shell-based enrichment pipeline for Middle High German including a CoNLL-RDF-based chunker for the analysis of Middle High German syntax and its semantic determinants. | neutral |
train_96288 | We thus provide a CoNLL-TSV-based enrichment pipeline that includes the chunker in addition to other annotators in order to assess the impact of, for example, semantic factors such as animacy on word order alternation. | while this approach is qualitatively different from conventional parsing, we adopt the terminology of classical Shift-Reduce parsing (Nivre et al., 2007, 100-104): we introduce the properties conll:SHIFT to connect (the root nodes of) adjacent partial parses, and conll:REDUCE to represent attachment within a (partial) parse. | neutral |
train_96289 | For example, the FTB contains many nominal named entities (tagged N N, e.g for persons), assuming a proper disambiguation step, those could have received a flat:name label instead. | this work was partly funded by the French ANR projects ParSiti (ANR-16-CE33-0021 and SoSweet (ANR-15-CE38-0011-01), as well as by the Program "Investissements d'avenir" ANR-10-LABX-0083 (Labex EFL). | neutral |
train_96290 | We have seen that the LISCA bins progressively gather occurrences of rarer and less prototypical relation instances, such as left-headed subjects, whose occurrence mainly concentrates in the last four bins. | the list of ordered dependencies was then subdivided into 10 groups, henceforth "bins", each corresponding to 10% of the total. | neutral |
train_96291 | In Figure 1, it is shown that in all considered languages subjects are mostly right-headed, whereas significant differences are recorded for left-headed subjects whose percentage is much higher for Italian and Spanish (i.e. | we investigated whether algorithms for measuring the plausibility of dependency relations within the output of dependency parsers could be exploited to acquire quantitative evidence from gold treebanks to shed new light in linguistic typological studies. | neutral |
train_96292 | 'Here, from each corner of the world, arrive 300 thousand patients') exemplifies a different, less common, Verb-Subject order involving a nominal subject. | it investigates whether and to what extent algorithms developed for assessing the plausibility of automatically produced syntactic annotations could contribute to shed light on key issues of the linguistic typological literature. | neutral |
train_96293 | Roller and Erk (2016) demonstrated that the prototypicality learned by CONCAT captures Hearst patterns such as "fruits such as apples." | tween the hypernym frequencies and the mean of the inner products of the trained parameter vector and combined word representations on each hypernym frequency. | neutral |
train_96294 | 2 In addition, we use $subclass for the construction of Adimen-SUMO statements in the few cases where synsets have to relate the knowledge of SUMO at the level of classes. | is not related to any of) the potential SUMO objects that are characterized by the Adimen-SUMO statement proposed above. | neutral |
train_96295 | From the word similarity point of view, the degree of similarity for 墨鏡"sunglasses" and 專利"patent" should not be high no matter which ontology is applied, for the former is a concrete object but the latter is an abstract one. | that means whenever a event "buy" occurs, that imply the event "pay (money)" happened as well. | neutral |
train_96296 | For instance, 貓 頭 鷹 'owl' is defined as {bird| 禽 :predication={SelfMove| 自 移 :duration={night| 夜},theme={~}}} and we can find several similar words defineds as the same way. | their distance in an ontology is also far from each other. | neutral |
train_96297 | This is why we use the principle of Hypernymy linkages for linking such concepts. | some of the synsets in Indian languages are too fine-grained and have a common representation in the English language. | neutral |
train_96298 | Thereby, they are created by human experts. | india is a vast country with massive language diversity. | neutral |
train_96299 | NRL is defined in a closed context. | nRL is defined in a closed context. | neutral |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.