id
stringlengths 7
12
| sentence1
stringlengths 6
1.27k
| sentence2
stringlengths 6
926
| label
stringclasses 4
values |
|---|---|---|---|
train_95000 | Determining POS of the words having the same surface form may be more ambiguous because a verb or an adjective can appear in the position of a noun as in the case of báo cáo in rows 3 and 4 of (as the case of báo cáo in rows 5 6 of Table 2). | in this section, we will discuss our method to improve the quality of annotation guidelines and to ensure correct and consistent annotation. | neutral |
train_95001 | Words that have the same surface form and pronunciation but different meanings and grammar functions occur frequently in the text. | to achieve a high agreement ratio, the annotators need to be trained carefully and to practice the annotation more on the basis of real texts so that they become familiar with annotation and analyzing the texts following the guidelines; the guidelines also need to be updated for new constructions throughout the annotation process. | neutral |
train_95002 | We call such element determined leaf. | a pattern with k nodes is called k-pattern. | neutral |
train_95003 | As the result, the method fails to correct the annotation error in (c). | t rans(τ ′ , τ t ) does not hold for any τ ′ and τ t . | neutral |
train_95004 | The constituents of the cleft construction são. | named entities), or is composed of relations between words (e.g. | neutral |
train_95005 | The comparison also shows that written Slovenian is significantly more nominal than spoken Slovenian with a higher share of nominal and prepositional phrases (the nmod and case relations) that are also more complex in terms of the number of adjectival (amod) or clausal (acl) attributives. | a detailed description of the sampled treebank, currently amounting to 3,188 utterances or 29,468 tokens, is given in Typically, spoken language annotation denotes annotation of its representation in the form of written transcription. | neutral |
train_95006 | The first version of the Spoken Slovenian Treebank is planned to be released as part of the UD release v1.3 under the CC-BY-NC-SA 4.0 licence. | the large majority of texts in the treebank include longer continuous spans of complete turns by one or more speakers, enabling posterior extension, re-segmentation or addition of other layers of linguistic annotation, such as discourse relations annotation or dialogue act annotation. | neutral |
train_95007 | This paper is organized as follows: Section 2. gives an overview of the design of the ALT corpus. | for example, the three nodes "(/PUNC'', "1/NUM" and ")/PUNC" form the phrase "(1)/NUM''. | neutral |
train_95008 | The conversion is implemented as a cascade of structural rules followed by a relation conversion procedure over the modified graph structures. | all experiments are performed on the data splits detailed in Table 6. | neutral |
train_95009 | NDT implements a somewhat broader definition of prepositions than the UD scheme and includes several cases which are counted as adverbs in the universal schemes. | the notion of control is not native to NDt, hence we approximate it by requiring the presence of an explicit subject dependent of the head verb of the clause. | neutral |
train_95010 | Without a gold standard for Norwegian UD dependencies, it is difficult to evaluate our conversion directly. | this is the case for demonstrative adverbs such as her 'here' and der 'there'. | neutral |
train_95011 | LT will bring us closer to this goal. | a truly multilingual Europe, which is supported through sophisticated Language Technologies (LT) is still far from being a reality. | neutral |
train_95012 | A certain amount of community building has indeed successfully taken place, also due to META-NET activities such as the META-FORUM conference series. | as well as data, tools and technologies distribution through open repositories. | neutral |
train_95013 | An essential step is the preparation of a convincing, realistic, yet visionary research and innovation strategy that includes solutions for the Multilingual Digital Single Market. | to non-initiated outsiders, the wider community around the Multilingual Europe topic still appears to be very much fragmented. | neutral |
train_95014 | It seems that the more a question requires self-commitment, or to give an opinion, the higher the ratio of "I do not want to answer". | finally, our proposal of a special issue of the TAL journal on Ethics and NLP has been accepted and a workshop dedicated to ethics, ETHI-CA 2 (ETHics In Corpus collection, Annotation and Application) will be held during LREC 2016. | neutral |
train_95015 | Again, out of the 200 participants, 73 people declared that they are interested in participating in a working group on ethics in NLP (31 gave their Email), which is very encouraging (see Table 1). | it seems that our question on whistle blowing ("Do you know of any way to blow the whistle (issue an alert) in case of a major ethical infringement in your domain?") | neutral |
train_95016 | Decentralised publication, however, has been shown by the linked open data community to support massive scaling in data exchange (Bizer et al 2009). | used when human intervention is required to review the Policy. | neutral |
train_95017 | An example of an accepted LR is presented in Since its launching, the ISLRN identification number has been adopted by large institutions over the world. | #NAME? | neutral |
train_95018 | The rise of speech-dependent customer service centers and mobile phone applications, as well as the growing need for better speech synthesis systems (e.g., in-car navigation tools) is reflected in industry reports as well as at conferences featuring industry solutions, like SpeechTEK. | twice each year students apply by submitting a Data Use Statement and Letter of Support from their advisor. | neutral |
train_95019 | This is unfortunate for reasons of comparison. | we present a novel corpus of tweets annotated with the MBTI personality type and gender of their author for six western European languages (Dutch, German, French, Italian, Portuguese and Spanish). | neutral |
train_95020 | It is important to mention here that all the previous large scale annotation studies known to us focused only on Level 4 (language and POS at word level). | en Trans: drkumarvishwas had said during victory celebration after anna fast that: let the janlokpal pass there will be such a celebration that the world will see. | neutral |
train_95021 | In the current study, we analyzed CS data from Twitter for pragmatic or functional aspects (Barredo, 1997;Sanchez, 1983;Maschler, 1991;Blom and Gumprez, 1972). | the nature and extent of CS depends on a number of factors, including structural, pragmatic (functional), and socio-cultural aspects. | neutral |
train_95022 | But we knew that this list was neither exhaustive nor necessarily suitable for Twitter. | an analysis of CS data requires inputs at Social, Contextual, and different linguistic and meta-linguistic levels that operate on various sub-parts of the conversation. | neutral |
train_95023 | Dependency types rely on the definition of a "clause," such as the distinction between nsubj and csubj. | for a relative clause, the head of the dependency is the noun modified by the clause; the dependent is the main predicate of the clause as shown in (8). | neutral |
train_95024 | It is possible to argue that these constructions share the same dependency type. | 7 this type is also used for fronted or postposed elements that do not fulfill the usual core grammatical relations; for example, the relation between "office" and "me" in the sentence "This is our office, me and Sam." | neutral |
train_95025 | These tools allow querying any of the UD treebanks, freely combining restrictions on the existence or absence of wordforms, lemmas, POS tags, morphological features, dependency labels, and subtrees. | figure 2 uses the french sentence Toutefois, les filles adorent les desserts au chocolat (the girls love chocolate desserts) to exemplify the different UD annotation layers, which are described in more detail in the following sections. | neutral |
train_95026 | An issue that we did not consider so far are revisions to the first stages of the pipeline, namely tokenization, sentence splitting, POS tagging and the generation of morphological features. | (iii) Can we avoid any loss of linguistic information and loss of accuracy in the process? | neutral |
train_95027 | In order to support automatic conversion into UD, the MIDT Treebank was extended with extra information (we call the enhanced version MIDT+). | mIDT in turn was obtained by merging two existing Italian treebanks, with different annotation schemes: TUT, the Turin University Treebank (Bosco et al. | neutral |
train_95028 | In such a way, we take advantage of the graph's properties, such as, the neighborhood, the edge weights, the size. | the most frequent extracted features from datasets, are frequency, mean, standard deviation, etc. | neutral |
train_95029 | We next describe our corpus collection and annotation in Section 3, and present the general idea of our system in Section 4. | instead of calling them best and mode, we thus report best, best 3 and mode 3. | neutral |
train_95030 | These papers are already provided in a clean, hand-authored XML format with a well-defined XML schema 3 . | we integrate automatic Core Scientific Concepts (CoreSC) classification into a prototype context-based citation recommendation system and investigate its usefulness to the task. | neutral |
train_95031 | We attempt to capture the possibility of Arg2 being present in that position if the connective was playing the role of a discourse connective. | begum, R., Husain, S., Dhwaj, A., Sharma, D. M., bai, L., and Sangal, R. (2008). | neutral |
train_95032 | Unlike Arg2, which is always occurs after the connective, there is no such constraint for Arg1. | a fully automated discourse analyzer would thus greatly aid discourse analysis and NLP applications such as text summarization and question answering. | neutral |
train_95033 | To keep a changing data consistent, especially during an annotation phase, a well designed annotation format does not contain redundancies. | processed with output filters to produce statistical tables. | neutral |
train_95034 | Such clauses are represented by a node with the syntactic category SINV in the PTB annotation. | to make queries in the PML-TQ simpler, we have added several redundancies to the transformed data of the PDTB-2.0, as the last step just before the transformation of the data into the database format. | neutral |
train_95035 | The tags 'Marking Aside', 'Adding to topic' and 'Managing Audience Discipline' were excluded from the final set of MD tags, due to their low occurrences in the trial study. | different instructions sets have been designed for each of the 19 MD category in the schema. | neutral |
train_95036 | Leading discourse markers such as "ok", "right" and "so" are regarded as individual utterances (unless "so" can be replaced by "hence"). | x-turns tier containing the turn boundaries for participant x x-utts tier used for segmentation and transcription of x's utterances x-en tier containing English paraphrase translation for x's utterances x-laughter tier containing the laughter and laughed speech duration intervals for participant x Segmentation: In the x-turns tiers, all continuous stretches of speech by one speaker until the other speaker takes over, modulo small overlaps, are considered one turn. | neutral |
train_95037 | Establishing reliable storyline clustering benchmarking data sets and metrics is one of the goals of the SUMMA project, as good storyline clusters are the prerequisite for downstream storyline summarization, visualization, and predictive anticipation of upcoming developments. | to enable joint multi-task learning for multilingual neural translation of morphologically rich languages we replace the attention mechanism with the sliding-window mechanism and operate the sequence to sequence neural translation model on the character-level rather than on the word-level. | neutral |
train_95038 | It is still an open issue which vectorspace projections yield the semantically best clusters (Hill, Cho & Korhonen, 2016) and further experiments are needed. | this is an order of magnitude slower than regular NEL or IF IDF bagof-words based clustering methods. | neutral |
train_95039 | The approach is informed by our previous work on machine learning (Barzdins, Paikens, Gosko, 2013), media monitoring (Barzdins et al.,2014), and character-level neural translation (Barzdins & Gosko, 2016). | for storyline (Rissen et al., 2013) clustering the signals for the stories belonging to the same storyline might be not so much the semantic similarity of the articles (they might report various developments of the storyline from differing viewpoints), but rather the matching time and location as well as same organizations and people being involved -the information typically supplied by Named Entity Linking (NEL) tools. | neutral |
train_95040 | Finally, we calculated the inter-annotator agreement for the identification of contrasting evaluations. | 3/9/2016 brat ¶ @afneil @daily_politics @TheSunNewspaper Missed off the #irony hashtag? | neutral |
train_95041 | 8 For Brazilians, these words were probably an indication that the text was not written in Brazilian Portuguese. | (2015) showed that for 52 misclassified Portuguese instances, only 22 have been labeled correctly by the annotators with low inter-annotator agreement. | neutral |
train_95042 | This is essentially NRC's first 2015 run, which performed 0.7% below the top system, trained using various training set sizes. | in addition to the full 2014 and 2015 test sets, we evaluated performance on the subset of the 2015 test set that contains the groups in the 2014 shared task, i.e. | neutral |
train_95043 | Scanned PDF files were also excluded since there is no sufficiently accurate Arabic Optical Character Recognition (OCR) system for the low resolution files we came across; although most of the OCRs available online claim high accuracy, tests reveal that their output needs editing (Abbas, Atwell and Al-Sulaiti 2016). | the 'Create Corpus' tool in SketchEngine (Kilgarriff et al 2014) enables researchers to gather texts in a range of different formats and then upload them into SketchEngine format. | neutral |
train_95044 | Based on a ceiling of 30,000 out of approximately 800,000, the duplication ratio is at most 3.375% in RCV1. | the concept of information entropy was defined by Shannon in his seminal text (1948). | neutral |
train_95045 | The paper has a substantial number of corrections which can be applied to the corpus; their revision is referred to as RCV1-v2. | the result from an experiment with an ideal 100% accurate classifier would be reported as something like 99.5% because of classification errors in the corpus -thus, it is not implausible to think that a less accurate classifier could win an evaluation. | neutral |
train_95046 | • Expenses are often referred to as an alternative to Costs. | and instead marked such a statement as FL, for 'forward-looking', under Sent_Tone. | neutral |
train_95047 | In the Accounting and Finance literature, there is an increasing body of work using basic word-list and ML approaches to examine the information content of forward-looking statements (e.g., Li (Li, 2010)) and other work has looked at the relationship of optimistic versus mild statements in the context of positive and negative financial performance (Chen et al., 2013). | we also performed conceptual anno-tation where we counted the frequency of semantic (Sem) tags A2.1 (Affect:-'Modify', 'Change') and A2.2 (Affect:-'Cause'/'Connected') those believed by the accounting and finance expert to be good indicators of attribution. | neutral |
train_95048 | We collected 500 Preliminary Earning Announcements (PEAs) released between 2010 and 2012 by firms listed on the London Stock Exchange. | another coder might well decide to make the ambiguity inherent in the statement explicit by marking this sentence as UNSURE, EXT, UNSURE. | neutral |
train_95049 | The idea is the following: the more P (c j |t) is different from P (c j ), the more important the term t i is. | recalculate the distances between new cluster and other clusters. | neutral |
train_95050 | Overall, we tried to create near realistic plagiarism settings. | a passage of size between 200 -300 words was excerpted from each source Wikipedia article. | neutral |
train_95051 | 3 Resolution of request/problem A message that aims to provide a resolution to a previously submitted request or problem report. | the idea of identifying requests and responses within online communication has been applied to email (Shrestha and McKeown, 2004;Carvalho and Cohen, 2005;Lampert et al., 2008), online forums (Ding et al., 2008;Cong et al., 2008) and online forums and issue trackers related to OSS (Korkontzelos and Ananiadou, 2014). | neutral |
train_95052 | Evaluation was per- Table 7: Evaluation results for the multi-label setting. | deciding whether to adopt an OSS project requires an assessment of its quality at the time of adoption and possibly an estimate of the perceived enthusiasm of the community towards it in the future. | neutral |
train_95053 | L2=L3 Others Percentage (%) 27.47 44.27 28.27 Table 6 Noisy Labels Distribution In this paper, we present a method that combines automatic selection and manual annotation selection method based on natural hashtag labels to construct an emotion analysis corpus. | the author feels sad because he will leave someone because of holiday and the lexicon-based method set it as sadness because of the word "miss". | neutral |
train_95054 | Hence, when combining several utterances, an utterance's matrix language and the matrix language of the previous utterance need to be represented (if they differ, that implies adding a code-alternation point between the two utterances). | it is also shown how this formal measure can be used in practice, by applying it to several code-switched corpora. | neutral |
train_95055 | 2015, and the four corpora 6 used in the shared task on word-level language detection in code-switched text (Solorio et al., 2014) organized by the workshop on Computational Approaches to Code Switching at the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP). | so does the annotation strategy chosen for the EMNLP corpora prescribe that elements such as abbreviations should be tagged with the language they belong to, while other annotations schemes treat them as language independent. | neutral |
train_95056 | For this reason, the dialectal Arabic corpus actually was the one causing most problems for the processors in the EMNLP 2014 shared task on code-switched language identification. | code-switching is most prominent in spoken language conversations and has thus traditionally mainly been studied by psycho-and sociolinguists (Auer, 1999;Muysken, 2000;Gafaranga and Torras, 2002;Bullock et al., 2014) and by speech researchers (Lyu et al., 2015), while the lack of large-scale textual corpora has made code-switching less attractive as a subject of study in computational or corpora linguistics. | neutral |
train_95057 | If an utterance x only contains language independent tokens, its code-mixing is zero; for other utterances, the level of mixing depends on the fraction of language dependent tokens that belong to the matrix language (the most frequent language in the utterance) and on N , the number of tokens in x except the language independent ones (i.e., all tokens that belong to any language This initial measure has several short-comings. | they were only interested in the relative difficulty and give no measure of the complexity as such. | neutral |
train_95058 | It is installed in our main lecture hall (Audimax) and in multiple other lecture halls. | although an archive might be useful for later studying and the preparation for the final exam, the live function allows for more situational context information. | neutral |
train_95059 | In the summer term, an active usage of the LT in all seven lectures could be observed. | it was also a possibility to get for more detailed answers. | neutral |
train_95060 | This release of the corpus consists of 300 abstracts from articles in the ACL Anthology Reference Corpus, published between 1978-2006. | we are grateful to Prof. Elke Teich and Prof. Laura Kallmeyer for supporting development and dissemination of this resource. | neutral |
train_95061 | Independently of each other, we annotated these abstracts. | this is often the case in anaphoric contexts (e.g., as in ". | neutral |
train_95062 | In recent years, machine translation (MT) became widely used by translation companies to reduce their costs and improve their speed. | the corpus includes 520 articles with an average of 192 words per article. | neutral |
train_95063 | We believe that this battery of evaluation techniques is necessary for a principled machine translation development and to understand the nature of the output. | although the results are not comparable across the language pairs, given the different nature of the languages involved, they are indicative of the amount of work to be done. | neutral |
train_95064 | The numbers of Accuracy and Fluency errors are quite similar, with slightly more Fluency than Accuracy errors. | a full MT development process often relies on a mix of automatic measures and human rating. | neutral |
train_95065 | As our measures show, self-plagiarism is a common We'd like to thank Wolfgang Hess for the ISCA archive, Douglas O'Shaughnessy, Denise Hurley, Rebecca Wollman and Casey Schwartz for the IEEE data, Nicoletta Calzolari, Helen van der Stelt and Jolanda Voogd for the LRE Journal articles, Olivier Hamon and Khalid Choukri for the LREC proceedings, Nicoletta Calzolari, Irene Russo, Riccardo Del Gratta, Khalid Choukri for the LRE Map, Min-Yen Kan for the ACL Anthology, Florian Boudin for the TALN proceedings and Ellen Voorhees for the TREC proceedings. | when we look at the five top sources (marked in the extreme right column), we see that the main sources of "inspiration" for LREC papers are ISCA and LREC itself. | neutral |
train_95066 | The majority (90%) of the documents come from conferences, the rest coming from journals. | "British National Corpus" is considered as possibly abbreviated to "BNC", as well as less regular names like "ItalWordNet" possibly abbreviated to "IWN". | neutral |
train_95067 | Document expansion, in addition to inserting text and hyperlinks, can also involve adding non textual content such as images that are topically related to document text, in order to enhance the readability (or in fact indexability) of the text. | we demonstrate that a tf-idf based term weighting for the query terms can prove useful in improving retrieval effectiveness, thus leaving open the future directions of research for effective query representation for this task. | neutral |
train_95068 | These methods are useful in order to arrive at an overview of the content of large corpora. | several technologies can detect related elements in texts, and the predicate that indicates their relation. | neutral |
train_95069 | Two rules were created to deal with this non-standard pronoun use, and anaphora resolution was limited to cases covered by these rules: An actor in the subject of a sentence's main verb (based on dependency parsing) is taken as the antecedent of a sentence-initial he/she in the following sentence. | the ENB-IPCC test-set does not come from the same corpus that the system was developed for, and serves as a way to test the system on texts from a slightly different domain, in terms of syntactic structures and lexical items involved. | neutral |
train_95070 | In the open RE task there is a great diversity of relations, which makes their classification more difficult. | the produced resources related to this paper (the subset of texts and corresponding positive relation instances manually annotated in tuples (NE1, relation descriptor, NE2) used in each experiment) are electronically available at: http://www.inf.pucrs.br/linatural/data set RE.html In future work, we plan to use open source tools; as well as performing an extension of the proposed process for other languages. | neutral |
train_95071 | Since each Web page is retrieved by AND search of the given query q j and the search engine suggest s, one or more suggests are assigned to each Web page. | those three statistics have quite high correlations. | neutral |
train_95072 | JAPAN Big Data Report 5 , search queries that have high correlation with economic boom as well as economic depression are identified. | then, from the set D N (z k , q j ), we collect Web pages which satisfy the lower bound θ lbd into the set D N (z k , q j , θ lbd ) as below: Furthermore, we collect search engine suggests assigned to each Web page d in the set D N (z k , q j , θ lbd ) into the set S(z n , q j , N, θ lbd ) as below: For each topic z n , we collect search engine suggests for a query q j (j = 1, . | neutral |
train_95073 | We finally propose how to predict the market share of a specific product genre based We examine 10 query company names "ASUS", "Lenovo", "NEC", "SONY", " " (SHARP), " " (Panasonic), " " (MITSUBISHI ELECTRIC), " " (FUJITSU), " " (HITACHI), and " " (TOSHIBA). | it has become popular recently to analyze concerns of those who search for Web pages and to compare them with people's behavior in real world. | neutral |
train_95074 | (2010) were not significant enough to influence automatic evaluation in recent work. | this leads to overly pessimistic scores since variations in the extracted keyphrases that might be judged as correct cannot be taken into account (Zesch and Gurevych, 2009). | neutral |
train_95075 | We assume that due to specialization, scientific texts exhibit greater encoding density over time, i.e. | to build such corpora, usually high manual effort is involved. | neutral |
train_95076 | (Álvarez et al., 2014). | a preliminary experiment showed that basic Levenshtein algorithm produces 1.1 % of errors, that is almost 12,000 sounds for our 30 hours of experimental speech data. | neutral |
train_95077 | The conversation was natural, the range of topics was not limited in any way. | in the read texts pauses were labelled automatically, with manual correction where necessary; additionally, when the speaker had problems with reading some words or phrases and re-read them, the defective fragment was segmented and labelled as a false-start. | neutral |
train_95078 | Furthermore, placing two strong stresses within one orthographic word was not forbidden, since it is quite possible in colloquial speech. | due to very frequent clipping in some speakers' recordings, defective fragments shorter than 100 ms were allowed. | neutral |
train_95079 | We collect tons of news from mainstream news websites and calculate the co-occurrence of Bollywood artists' names in these articles, which is a plausible profound and comprehensive way to tell the relativeness of two singers. | the same humming query is used in both figures with different scaling and the main melody. | neutral |
train_95080 | We propose a query by humming system for music retrieval that uses deep neural networks for note transcription and a note-based retrieval system for retrieving the correct song from the database. | the humming query input is generally divided into several segments and linear scaling is used on each of the segment to get the optimal distance between the query The candidate songs are then ranked by their smallest distance with the query, with the song with the lowest distance ranked first. | neutral |
train_95081 | At the output of the Speech/Non-speech Discrimination module, the specificity is high, Sp = 0.996, this parameter represents the classification rate of non-speech events classified as non-speech. | this model is unigram and contains 11018 words. | neutral |
train_95082 | Among the 1950 audio events, 277 are distress calls, 45 are colloquial sentences and 1628 are non-speech. | the speech technology, though mature is still difficult to master for non-expert and to tune for a particular usage (Vacher et al., 2015a). | neutral |
train_95083 | Taking a closer look at the random fit for each transcriber ( Figure 5) reveals that the poor performance of FS − was largely attributed to the non-expert group CROWD, who were not able to beat the ASR output with this method. | in addition, there has also been a rise in computer-assisted transcription, in which automatic transcription is used as a starting point and human intervention is only needed when the ASR produces errors (Rodriguez et al., 2007). | neutral |
train_95084 | As mentioned in §2., each participant's audio is recorded with a Microcone microphone array in a real environment. | in the offline experiments, on the other hand, the test dataset has been well defined. | neutral |
train_95085 | This may lead to disturbances in any of the components involved in the speech production, including respiratory, phonatory, resonatory, articulatory and prosodic elements. | we can point out that the spontaneous speech productions are shorter (duration and # of phones) than the read speech productions independently of pathologies. | neutral |
train_95086 | In order to account for phonemic effects such coarticulation the monophones were expanded using the full linguistic context. | without these errors the overall token-level accuracy would be 90.5%, which means that about half of the token-level errors are due to these difficult aspects of Bangla orthography and phonology. | neutral |
train_95087 | It is based on the simple hypothesis that states that during a TV newscast, when the name of a person appears on screen, that person is presently speaking. | once the speech tracks identified, the LIUM speaker diarization tool (Rouvier et al., 2013) is used to provide a speaker segmentation and clustering of the analyzed shows. | neutral |
train_95088 | For the sake of comparison, we consider that a message in chats corresponds to a breath group on the phone. | the paper is organized as follows. | neutral |
train_95089 | This can be used to identify words' variations or misspellings (Kondrak and Dorr, 2004). | we also showed that the use of post-processing patterns would not lead to a substantial increase but would be useful to balance the precision/recall ratio. | neutral |
train_95090 | The participation rate (19%) of the group members for the collaborative translation was rather high, and produced a decent number of viable translations that allow for interesting insights into the use of dialect on the Internet. | for our stop words list, we used a publicly available subcorpus of the DCK, which is called the "release corpus" 5 . | neutral |
train_95091 | Bavaria is one of 16 federal states in Germany. | most of the visits occurred during the first 4 days; the whole experiment lasted for 10 days. | neutral |
train_95092 | This is probably from the period when Wikipedia had not enforced the format of editors' signatures when they were making posts. | our corpus represents a considerably large number of interactions between the same set of people. | neutral |
train_95093 | For the identification of the NE categories Person, Organization and Place, we use the NERP-CRF output described in Section 3.2.. After, we identify the first pair of NEs in each sentence. | the authors acknowledge the financial support of CNPq (Conselho Nacional de Desenvolvimento Científico e tecnológico), CAPES (Coordenação de Aperfeiçoamento de Pessoal de Nível Superior) and FAPERGS (Fundação de Amparo à Pesquisa do Rio Grande do Sul). | neutral |
train_95094 | In (a) the noun phrases "o agrônomo " "the agronomist " and "Miguel Guerra" " Guerra" are considered coreferent, in other words, they belong to the same coreference chain. | there were 1,086 NEs identified, distributed in the ten HAREM categories. | neutral |
train_95095 | The original annotation guidelines have been created and tested for the first release of ARRAU (Poesio and Artstein, 2008). | as a part of our ongoing work, we are analyzing the outcome of our efforts, in particular, of the measures undertaken to improve the data quality. | neutral |
train_95096 | Again, it is very important to have the same corpus annotated for both bridging and coreference since these two tasks are very interrelated. | we have doubled the corpus size, expanded the selection of covered phenomena to include referentiality and genericity and designed and implemented a methodology for enforcing the consistency of the manual annotation. | neutral |
train_95097 | Presumably not, leading us to suppose that elaborations of our annotation in which distinct idioms are distinguished would be relevant. | w i represents the pool of spatial skip bigrams within the ith segment. | neutral |
train_95098 | The semantic relation we consider is relatedness of terms in thesauri for intellectual document classification. | both for Eurovoc-1 and Eurovoc-2 we only selected terms consisting of one word and we moreover required that word occurs at least once in the UKWaC. | neutral |
train_95099 | In 5 the arrival event is annotated as POS as it may still occur (conditioned on the accomplishment of the if clause). | they have annotated a corpus and have worked on the automatic detection of factuality. | neutral |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.