id
stringlengths 7
12
| sentence1
stringlengths 6
1.27k
| sentence2
stringlengths 6
926
| label
stringclasses 4
values |
|---|---|---|---|
train_95400 | An interesting symptom of this is that the T E X engine is conventionally described as a living organism (Knuth, 1984, chapter 7) with eyes, mouth and stomach, rather than in terms of compiler theory. | many commonly used commands crucially depend on these functionalities. | neutral |
train_95401 | • Many representation schemes have been designed with a particular language (or language family) in mind, which hinders its adoption for languages with different morphosyntactic structures. | application developers face many challenges when combining NLP modules, which are often tailored to the particular needs of the task at hand and, thus, difficult to integrate. | neutral |
train_95402 | It was found that all the repetitions in them were reduced into a single letter. | if there are some remaining tokens without tags at the end of the pipeline, they get assigned the final encoding of the output text. | neutral |
train_95403 | For every character found in a word classification in the I(0)BES scheme will be collected and frequencies will be counted for I,B,E and S respectively. | previous research on automated digitization focused on producing 3D scans of tablets (Sect. | neutral |
train_95404 | As a logosyllabic writing system, it shares important structural characteristics with Chinese and Japanese (Ikeda, 2007), so that we evaluate word segmentation methods successfully applied to these languages. | non-restorable characters were ignored and thus are not represented in the resulting texts. | neutral |
train_95405 | Additional information can be found such as intensity marker (e.g., severe back pain) or frequency marker (e.g., chronic back pain). | human annotators can focus on unseen annotations. | neutral |
train_95406 | This is a typical annotated segment based on MAP, where only the lexical-entry segments are explicitly annotated with semantic structures and the semantic structures of larger segments are implicitly derived by the above rules. | speaking, an eq link represents the coreference between the head nodes of the two linked segments. | neutral |
train_95407 | We address this question by generating artificial data with the methods described in Section 3. | in Dalli and Wilks (2006) researchers train a classifier to predict the publication date of texts within a time span of nine years. | neutral |
train_95408 | A gender-specific first-person expression is a firstperson expression which is specific to men or women due to grammatical constraints or requirements (usually related to the grammatical gender). | the task of automatic gender attribution has been considered in computational linguistics as well. | neutral |
train_95409 | A regexp (matching time stamps and greetings) was handcrafted to isolate posts and only posts containing gender-specific first-person expressions were kept. | erotic or pornographic fiction was marked in this category. | neutral |
train_95410 | Table 4: Percentage of annotations assigned per category in the Multi-CoreSC CRA corpus. | for this purpose, we used the trained Conditional Random field (CRf) model (see (Liakata and others, 2012)) to automatically assign CoreSC concepts to the CRA corpus and evaluated the results on the gold standard. | neutral |
train_95411 | We note here that this procedure has three shortcomings: that we potentially bias the gold-standard towards (i) certain CoreSC labels through the use of our prioritisation table, (ii) the most reliable annotator and (iii) that the measure is conservative and favours singlelabel annotations. | in the case of three distinct labels the top three most popular labels are assigned scores of 0.6, 0.3 and 0.1 respectively. | neutral |
train_95412 | For this purpose, we used the trained Conditional Random Field (CRF) model (see (Liakata and others, 2012)) to automatically assign CoreSC concepts to the CRA corpus and evaluated the results on the gold standard. | this was not followed strictly by all annotators leading to some variation in the number of annotated sentences across the different annotators. | neutral |
train_95413 | In this case, the perceiver is considering "could the agent have been forced to carry out an action" or "was the agent tricked into carrying out an action". | to the best of our knowledge, this is the first piece of work exploring an automated approach for blame/praise detection from text. | neutral |
train_95414 | Wikipedia has become one of the most popular resources in natural language processing and it is used in quantities of applications. | this is translated in SQL as: sqlContext.sql(""" SELECt uri FROM wikidata WHERE predicate = 'wd:P31' AND value1 = 'urn:wikidata:Q5' """).cache().registertemptable ("persons") and results in a table called persons. | neutral |
train_95415 | Colloquial creole languages such as Spanglish (a mix of Spanish and American English) or Singlish (English-based creole from Singapore) or even official languages such as the Haitian creole (which merges Portuguese, Spanish, Taíno, and West African languages), are some of the bestknown situations of spoken code-switching. | table 4 shows the accuracy obtained on table 5: Micro-averaged F1 (%) on the code-switching set the code-switching collection by: (1) the English monolingual model (en) , (2) the Spanish monolingual model (es) and the multilingual model (en-es). | neutral |
train_95416 | The average agreement between the annotators is fair and the average agreement between the annotators and the curator is moderate. | wordNet) or a flat hierarchy of topical groupings (cf. | neutral |
train_95417 | For example, in the context of calculating available seats in the room, loveseat and sofa are not synonymous, as they by usual definition have a different number of seats. | we further assume that it most representatively shows the gaps in this kind of knowledge base. | neutral |
train_95418 | We want to know the probability p(x, y) that x is the translation of y. where Improvements of the method were later proposed. | it is a widely used corpus in cross-language text analysis and machine translation. | neutral |
train_95419 | Papers in PDF format corresponding to the most relevant search results were downloaded. | alignment of our dataset at both sentence-and chunk-level was also needed in order to evaluate the performance of different methods on different types of texts but also on different sizes of texts. | neutral |
train_95420 | The annotation is still ongoing and the resource continues to grow. | we do achieve a rate of 73.9% of text units with code-switching (see below), which contrasts with code-switching ratios of around 20% in data sets in previous literature (Solorio et al., 2014). | neutral |
train_95421 | (2013), such that all the evaluation scripts used in the SemEval 2013 CL-WSD task can also be used on this data. | there is an expression like "melting pot", which is not in the dictionaries. | neutral |
train_95422 | The enrichment is supported via a query template mechanism. | resource: /e-entity/{ner-engine}/documents Input: Berlin is the capital of Germany. | neutral |
train_95423 | We use a weighted arithmetic score combination for LSA, PCA or ICA as they are based on the same similarity measure (Normalized Euclidean distance) and a weighted arithmetic ranks combination while using SA in the combination process (SA is based on Jaccard similarity). | we notice that SA performs better than the three other models for the breast cancer corpus. | neutral |
train_95424 | We provide the results and code for a number of comparisons and benchmarks that test the usability of this toolkit against other toolkits in the literature. | c(king) − c(man) ≈ c(queen) − c(woman), corresponding to some sort of royalty concept. | neutral |
train_95425 | [2003]'s neural language model, with a number of tricks to boost performance. | for each update in word2vec, bivec performs 4 updates: source to source, source to target, target to target and target to source. | neutral |
train_95426 | A follow up project, TaaS 2 , expanded this idea to all EU languages. | note we ensure that the time difference between those two bins is no greater than 7 days. | neutral |
train_95427 | Corpora of original and translated language are essential for empirical investigation of theoretically-motivated hypotheses from the field of translation studies (Baker, 1996). | in each subcorpus, each paragraph is annotated with meta-information; in particular, the original language in which the paragraph was uttered. | neutral |
train_95428 | Do note that since we added beginning markers ('%') and end markers ('# ') to words, our rules allow to explicitly distinguish between initial and final positions within the words, which lends itself well to the inflectional schemes of the Slavic languages under consideration. | polish uses ks instead of x: CS-pL maximum -maksymum, export -eksport ((x:ks) is identified in iteration 190). | neutral |
train_95429 | This explains the variation in the amount of words in Table 1 The words in the lists belong to different parts of speech. | although the Czech internationalisms use the letter combinations ti and di, t and d are not palatalized by the i as they CRITIC parts in characters with diacritics, with the hypothesis that differences in BaSE would be stronger than differences in DIa-CRITIC. | neutral |
train_95430 | In both cases, one deals with certain ensembles of units (Akhmanova, 1971). | in the CS-PL cognate pair večer -wieczór 'evening', the model suggests the correspondences (r:r), (e:e), (e:ie), While the model does not replicate all of the diachronicallybased rules, not all of them are necessarily truly applicable. | neutral |
train_95431 | Furthermore, there are no exceptions for internationalisms in Polish orthography, in contrast to Czech. | we assume that Russian verbs of the first or the second regular conjugations may correspond to Bulgarian verbs of the first, the second or the third regular conjugations. | neutral |
train_95432 | Like us, they provide metaphors covering a variety of syntactic relationships for English, Spanish, Russian, and Farsi. | metaphors with higher metaphoricity are expected to exhibit a higher degree of the features above, whereas metaphors with lower metaphoricity are expected to exhibit lower sensory perceptibility and to be less expressive, more common, and less jarring when read and interpreted literally. | neutral |
train_95433 | These data falls under one of the following categories: 2) Annotator-Units assignment: including information about each unit (post, tweet) assigned to the annotators such as post-id, genre-id, assignment-id, path of the assigned file; Finally 3) Language-Unit assignment: which includes information about which unit belongs to which dialect/language. | the administrator manages only one central database. | neutral |
train_95434 | Multilingual corpora, combined with alignment and search tools, are today acknowledged for their theoretical as well as practical importance in cross-linguistic studies and applications: they provide a rich basis of language correspondences in context that can serve as testbeds for linguistic theories and hypotheses, but they are also essential for applications in the fields of lexicography, natural language processing, automatic or machine-assisted translation and language teaching (Altenberg and Granger, 2002;Johansson, 2007). | an ID Gloss and a [begin,end] time interval) to the corresponding part of the French translation fragment where the meaning of this sign is given in context. | neutral |
train_95435 | A subeffect test showed significant simple main effect of language in Grounding act "init" (F(1, 99) = 4.53, p < .05), "ack-init" (F(1, 99) = 8.91, p < .01) "cont" (F(1, 99) = 50.84, p < .001). | three sets of NAC EMR-9 head-mounted eye trackers and headsets with microphones recorded their eye gazes and voices. | neutral |
train_95436 | Additionally, unexposed broadcast data from prior LDC collection efforts contributed additional recordings for Indian English, Mandarin, Modern Standard Arabic and US English. | to create the CtS portion of the corpus, we recruited native speakers following the same approach used for LRE11, in which recruited callers (known as "claques") were required to make single calls to multiple unique individuals within their established social networks (friends, family, acquaintances). | neutral |
train_95437 | A massive, but expected, drop compared to the 96.3% on the newswire WSJ corpus. | in our FlexTag architecture (see Figure 1d), we thus give the user flexible control over which feature extraction and machine learning should be used. | neutral |
train_95438 | Training FlexTag Most other trainable taggers only support one input format and users are supposed to transform their data in the required format. | in order to make this simplicity possible, FlexTag relies on DKPro Core 1 (Eckart de Castilho and Gurevych, 2014) for preprocessing and model loading, and DKPro TC (Daxenberger et al., 2014) for feature extraction and classification. | neutral |
train_95439 | In 2015 we set the bar to 500 thousand tokens, mostly because of our results on morphosyntactic annotation of Slovene (Ljubešić and Erjavec, 2016) which showed corpus supervision to be of much greater importance than lexicon supervision. | croatian lijep, Serbian lep) which was already encoded during the croatian annotation. | neutral |
train_95440 | One training epoch on a tagged corpora of average size (15-50 thousand sentences) runs for 15-90 minutes on a regular 2-core CPU, and it usually took 15-20 epochs until converged. | table 1 shows our results on this dataset, compared to those reported in (Collobert et al., 2011). | neutral |
train_95441 | ], NN [=Noun], VB [=Verb] or PR [=Pronoun], and additional features, e. g. Number (Sg. | this phenomenon makes checking for congruence difficult without a morphological component. | neutral |
train_95442 | Here are some examples: In the sentence Libri u botua., engl. | to words like assesi, engl. | neutral |
train_95443 | In February 2016, Amazigh language was made an official language by the Algerian government. | finally, in Section 6 we draw some conclusions and describe the work to be done in the near future. | neutral |
train_95444 | These unsupervised or semisupervised approaches make use of distributional semantics (Turian et al., 2010). | data deviating from the norm requires tools adjusted to this data. | neutral |
train_95445 | To make sure that the batch size of 200 sentences is not too big, we experiment with 100, 50 and 1. | word embeddings have been used by Lin et al. | neutral |
train_95446 | We perform an initial automated POS tagging on the corpus. | each occurrence was marked with its sense tag. | neutral |
train_95447 | This issue is somewhat similar to that of compound words in Swedish (Heppin and Gronostaj, 2012). | instead of adding an extra annotation layer 5 , like in the SweFN project, we decided to followed Petruck's recommendation; borrowing from the Spanish FrameNet project, we annotate such roles as "externally constructionally null instantiated" (ECNI) (Subirats, 2009). | neutral |
train_95448 | This is a specific feat of digital newspaper collections, where users often search for news items relating either to their own family history or a location that they live(d) in or have some other personal relation to, and which are of particular interest to genealogists. | this can be due to the annotators accidentally missing an entity (only 1 run was made per page) or, some of the existing annotations having been scraped during post-processing. | neutral |
train_95449 | For each occurrence of a T C, the expert should answer if: (1) it is syntactically well-formed, (2) it belongs to the scientific lexicon (3) it belongs to the domain lexicon (here linguistics), (4) it is a T O. | among these association rules, let us give a focus on : -NTO] means that when Sa gives a N T O diagnosis when HB and LS cannot take decision, then it is not always a good diagnosis (confidence is only 0.7). | neutral |
train_95450 | Potential terms are n-grams, mainly unigrams, occurring in Swedish patent texts. | n T Os are more equally Figure 1: Decision Tree computed for occurrences of the T C Aspect: each node is an XML tag and each edge exhibits the normalized distance between on occurrence and the closest tag of this type, for each decision (T O or n T O), the proportion of True Positives is given distributed within the document. | neutral |
train_95451 | In Spanish, poetic lines are measured in syllables. | the system consists of four main modules (see Figure 1): 1. | neutral |
train_95452 | Let the prominence estimate P r(r i ) be the relative frequency with which the resource r i appears linked on Wikipedia compared to the frequency of all other resources in R. Formally: We estimate entity prominence through PageRank (Page et al., 1999). | we hope that this work may foster metric-aware and less biased benchmark datasets to be created, as well as it may stimulate a more sensitive discussion of results produced using those benchmarks. | neutral |
train_95453 | First, the number of additional atoms that need to be added to the network can be large depending on the domain size of the predicate's arguments. | the system configurations underlying the three rows in this table are the same as those in table 2. | neutral |
train_95454 | A few other points deserve mention. | the ability to learn from data is one of the advantages of MLNs over ILP: unlike MLNs, ILP is a pure inference framework that does not have the ability to learn. | neutral |
train_95455 | In this subsection, we provide a brief overview of two joint inference frameworks, ILP and MLNs. | an opinion candidate can be related to source and target arguments if it is not an argument-implicit opinion. | neutral |
train_95456 | To verify the usefulness of our corpora we built machine-learning models using an exhaustive list of features, and while utilizing the annotated corpora for training and testing. | we collected at least 4 annotations from different annotators for each blog. | neutral |
train_95457 | squared error loss (f (x) − y) 2 . | one of the assumptions is that domain experts are knowledgeable about the characteristics of the phenomena that they would like to be captured by the corpus. | neutral |
train_95458 | The point is not to find all coreferences in advance or we would not need to make the tool, but to insure enough variance in the documents to produce a subnetwork representing the coreferences adequately. | other optimisation methods could be easily incorporated. | neutral |
train_95459 | The first three verbs were selected because they are often referred to as typical reported speech verbs, while the other three were typical cases of describing other speech properties, as shown in the examples below, lightly adapted from COMPARA: • A Jean contouà Betty e a Betty contou-me. | drury and colleagues compiled a large quotation corpus in the financial domain (drury et al., 2011;drury and Almeida, 2012) and used it to identify trends in that domain. | neutral |
train_95460 | In a time of extremely short-lived fame, "who said that" is mostly relevant if it is modified by, or served as "has just said so", and one needs automatic reporting boots. | in their study of human tagging in English, (Bruce and Wiebe, 1999) already showed that attribution was hard, and in (Wiebe et al., 2003) the problem of automated opinion mining is further discussed. | neutral |
train_95461 | New contexts for the exploration of reported speech have recently been provided by research in Digital Humanities, see (Mambrini et al., 2012). | 3 it should be stressed that our interest is also linguistic, in the sense that we wish to define and study a lexical field, that of language-talk-speech, that contains the words related to this (central) property of mankind. | neutral |
train_95462 | The First CLIN Dutch Shared Task at CLIN26 was based on the Dutch section, while the EVALITA 2016 FactA (Event Factuality Annotation) shared task, based on the Italian section, is currently being organized. | previous work is based on the use of corpora aligned at the token level, whereas our method envisages an alignment at the markable level, where each annotated element is aligned to English on a semantic rather than syntactic basis. | neutral |
train_95463 | The focus of visual attention is one such behavior and we wish to gain deeper understanding of how it is effected and explore patterns it exhibits under different interaction conditions. | spatial alignment is done on several levels as well. | neutral |
train_95464 | The symbols used are outlined in Table. | humans listening to speech can miss or imagine the existence of objectively measured silences of short duration, especially when there is elongation of previous or following syllables (Martin, 1970), and are known to have difficulty recalling disfluencies from audio they have heard (Deese, 1980). | neutral |
train_95465 | Secondly, given that an average user is not even aware of this processing taking place, it is practically impossible for him to exercise rights that are granted to him by the Directive, such as the right of access (art. | the first case -processing one's own data -does not seem to raise any particular concerns as far as the lawfulness is concerned. | neutral |
train_95466 | a person who processes data on behalf of the controller). | shannon's work on Information Theory to treat translation as a code-breaking problem (Hutchins, 1999). | neutral |
train_95467 | People have already developed some initial applications; but they have to develop everything by themselves. | intra-Grid Executor as well as inter-Grid Executor are played the important role in our framework. | neutral |
train_95468 | The use of mobile devices to assist people in overcoming language barriers is becoming more and more popular such as translation via an image of unknown word, the spoken word, and gestures. | our proposed framework includes appropriate modules to deal with those issues. | neutral |
train_95469 | When the user clicks on a category (e.g. | b) The set of tags adopted should be interpretable across linguistic theories and valid across time in order to last as long as possible as old linguistic atlases did. | neutral |
train_95470 | 1 For these reasons, the methodological and technological boundaries existing in each linguistic project need to be overcome in order to find common grounds where linguistic material can be shared and re-used over a long period of time. | the FLaReNet Strategic Agenda highlighted the most pressing needs for research areas, such as Natural Language Processing, and presented a set of recommendations for the development and progress of LRs in Europe (Soria et al., 2014). | neutral |
train_95471 | 10 See for example the overview of the different NIST evaluations on the Linguistic Data Consortium (LDC) website https://www.ldc.upenn.edu/collaborations/ evaluations/nist. | the development of a language 23 See https://bitbucket.org/dcavar/ elan2split for more details. | neutral |
train_95472 | By the time of this writing, they have been implemented and tested on small data sets but not on the actual large data sets. | as a workaround, corpora are sometimes distributed as sentence shuffles under the assumption that single sentences never reach the required threshold of creativity, for example COW (Schäfer and Bildhauer, 2012;Schäfer, 2015b) or the Leipzig Corpora Collection (Biemann et al., 2007). | neutral |
train_95473 | A potential difference of the reviewed approaches lies within the flexibility that is offered to the users with respect to the level of customisation of the annotation schemas. | the graphical interface of the annotation page is divided into main three parts, as shown in Figure 4. | neutral |
train_95474 | The main difference between these annotation tools, is that WebAnno separates the role "Curator" as the user who reviews, merges the annotations of the different users and produces the final result. | the proposed annotation tool exploits the power of cloud computing in order to allow users to perform annotation tasks without the need of any kind of software. | neutral |
train_95475 | In batch processing all the cluster nodes are busy processing a large set of documents, whereas in a streaming processing scenario resources may be idle, waiting for documents to arrive. | depending on the requirements and quantity of the data this can be done in different ways. | neutral |
train_95476 | For example, in finding target names of prominent Australians in the data we obviously see many instances of different individuals with the same name. | alveo also aims to provide access to the data in a way that facilitates automatic processing of the text rather than the document-by-document interface provided by the Trove web aPI. | neutral |
train_95477 | The corpus contains a wide range of document types from the wide range of newspapers included in the Trove archive. | doing any computational work on the entire Trove corpus or even on significant subsets is not efficient using the Alveo (or Trove) web APIs. | neutral |
train_95478 | The term less commonly taught seems to have been borrowed in the HLT community from the second language teaching community where it refers to instruction within a specific target market, be it the United States, the Western Hemisphere or perhaps outside of the region where the language is official. | notwithstanding the number of native speakers, if most of those also speak another language with an even greater population or prominence, that could reduce the language's importance to projects whose goal is to develop news understanding technologies. | neutral |
train_95479 | Data are typically segments of broadcast and telephone conversations audited for the linguistic variety spoken, speaker number and sex, and sound quality. | after some turn-over in project management, TIDES focused the bulk of its attention on English, (Mandarin) Chinese and (Modern Standard) arabic but planned for Surprise Language Exercises. | neutral |
train_95480 | Similarly the nature of the morphology affects HLT development, not only whether the language tends toward an analytic or synthetic morphology but also such factors as the number of morphological classes and the degree of irregularity and syncretism present. | differences in the terminology, available information and goals of low resource language efforts lead to variability and some obscurity in the language selection process. | neutral |
train_95481 | These textual productions are updated by new entries more often than static content. | the non-controlled use of terms, the absence of (standard) orthographic conventions, loose grammatical constraints, and high dependence on the individuals' writing style, give rise to data sparsity issues affecting the performance of SMt systems, which present a poorer performance with informal genres (van der Wees et al., 2015a). | neutral |
train_95482 | APSyn was tested on the ESL and TOEFL questions, outperforming the Vector Cosine and the co-occurrence, plus several lexicon-based and hybrid models. | the former rely on lexicons or semantic networks, such as WordNet (Fellbaum, 1998), measuring the distance between the nodes in the network. | neutral |
train_95483 | 83.33% = 0.833) and considered for our task only sentences for which the difference between L1 and L2 was less than 0.30 5 . | independently of the different conceptual framework adopted, the majority of the systems in literature model metaphor as a discrete property, ignoring the fact that several degrees of metaphoricity are possible. | neutral |
train_95484 | In order to include only prototypical pairs in EVALution-MAN, we filtered the CWN pairs and then further assessed them through manual annotation through both rating and tagging methods. | the relatas'frequency was calculated in a combined corpus of Sinica and Chinese Gigaword. | neutral |
train_95485 | zalupati 'start banging' to lupiti 'hit') • We separate suppletive forms and do not explicitly relate them to each other (e.g. | the two verbs are morphologically related. | neutral |
train_95486 | VoxML treats objects and events in terms of a dynamic event semantics, Dynamic Interval Temporal Logic (DITL) (Pustejovsky and Moszkowicz, 2011). | when we impose more constraints on the values of an attribute, we arrive at more structured domains. | neutral |
train_95487 | To evaluate this approach, test sentences used in the previous study were prepared and the results from the proposed method were compared with those of the previous study. | in the 90 test sentences, there were 108 nouns the proposed method and baselines determined as 'Metonymic' or 'Literal'. | neutral |
train_95488 | In other words, each is an accuracy of the first correct word in the top N. As shown on the table, the proposed method expressed higher accuracies than the baselines. | table 9 lists two rates of correctness for each metonymic concept. | neutral |
train_95489 | It uses synsets to obtain node distance and the rest to do the same as CF-GT. | the proposed method exhibited higher accuracy (0.85) than the two baselines. | neutral |
train_95490 | We evaluated the quality of the common-sense knowledge obtained by the framework presented in this paper through manual review of randomly sampled relations encoded in the knowledge graph, as well as implicit relations inferred in the embedding space. | perhaps the most popular of them is the Unified Medical Language System (UMLS) (Bodenreider, 2004) which designed by the National Library of Medicine in order to provide a standard set of concepts allowing for commonalities amongst various medical terminologies to be interlinked, for example, by mapping ICD-9 diagnostic codes (Cimino et al., 1993) to Medical Subject Headings (MeSH) (Lowe and Barnett, 1994). | neutral |
train_95491 | The reports include the patient's clinical history and medications, a description of the EEG setting and configuration, as well as the physician's impressions and clinical findings. | similarly, medical knowledge is learned from two large sets of electronic health records. | neutral |
train_95492 | Because we considered lexical relations between concepts, rather than individual words, the resulting knowledge graph was quite large, containing 32,522,807 vertices and 52,209,411 edges. | this demonstrates that the embedding procedure is able to account for the affects of individual words on the semantics of the discovered lexical relationships. | neutral |
train_95493 | Attempts to address these limitations was tackled in our previous work (Goodwin and Harabagiu, 2013) where a qualified medical knowledge graph (QMKG) was presented. | h(•), uses the bag-of-words vector representations expressing the lexicalized relation to generate a vector representation of the relation. | neutral |
train_95494 | We preserve the structures of infoboxes and tables in our extraction. | as the pages of the websites we used for the knowledge database are divided into sections, we believe that this problem can be at least partially solved by using the names of these sections to solve some further co-references and lacks of context. | neutral |
train_95495 | We use a Davidsonian style representation for events; for example, an entity x 1 dropping an item x 2 is represented as drop(e), dropper(e, x 1 ), dropped(e, x 2 ). | our goal is to build a QA system that can translate a question asked by Non-factoid questions: What is the best way to spawn the two different types of Golem? | neutral |
train_95496 | Coverage is already very high with the first model but deviations from annotated phone boundaries in terms of RMSE continue to decrease with iterations. | we most certainly not yet maxed out the alignment coverage. | neutral |
train_95497 | Table 6 shows accuracies of ASR with models adapted by the adaptation technique proposed in our previous work (Hirayama et al., 2015). | we evaluated the collected data in automatic speech recognition (ASR) and Kana Kanji conversion (KKC) (Mori et al., 1999;Takahashi and Mori, 2015) systems. | neutral |
train_95498 | Tireless efforts of data collection of speech data and their transcriptions from the early stage of NLP and SLP researches drastically improved accuracies of a variety of NLP and SLP tasks. | dialect speakers convert the selected 100 sentences into their own dialect sentences. | neutral |
train_95499 | The originality of the corpus is to allow a comparison between controlled and less controlled speech in the two populations. | in each class, phones are characterized by a set of features considered as relevant for the discrimination task (see Laaridh et al. | neutral |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.