id
stringlengths 7
12
| sentence1
stringlengths 6
1.27k
| sentence2
stringlengths 6
926
| label
stringclasses 4
values |
|---|---|---|---|
train_18600 | An automatic classifier is then applied to determine the character code that most likely corresponds to each character image. | such systems are still imperfect. | contrasting |
train_18601 | It can be seen that in our annotations we have tended to provide the contemporary correct form far more often than a historical form (of which there may be several, which might obfuscate results). | the results also show that the LD from a historical variant to its historical correct form is often lower than to its contemporary correct counterpart. | contrasting |
train_18602 | In order to solve this problem, TICCL will need to be equipped with a special module geared at picking up elongated stretches of very short word tokens. | the full book evaluation told us nothing about the extent of the foreign language problem in our supposedly Dutch corpus. | contrasting |
train_18603 | Our prior example of 'vossestaarten' already showed that confusion of the long 's' by 'f' is not necessarily the case. | it is the feature of digitized diachronic texts most often commented on by researchers working on digitized text (cf. | contrasting |
train_18604 | (3) Argument A2 attacks the controversial issue, which says that vaccines protect the population (with no exception). | since the number of deaths is very limited, this statement may be interpreted either as a weak attack or as a concession: Vaccine protects the population HOWEVER there are a few cases where it does not work. | contrasting |
train_18605 | On the one hand, we are able to validate the interoperable scheme in an application. | the successful application of the scheme indicates possible interoperability in existing resources. | contrasting |
train_18606 | In this pilot we measured the durations of the filled pauses that could be identified on the basis of manual transcriptions of the interviews. | these transcriptions are not always accurate. | contrasting |
train_18607 | Previous research works have proven that discourse information is beneficial to these NLP tasks. | the current discourse annotations concentrating only on monolingual languages have some insufficiency that limits its power in many multilingual NLP tasks. | contrasting |
train_18608 | In RST, the temporal reading similarly corresponds to "circumstance". | it is not at face value clear in which way während is contrastive: Does it express a mere (same-level) contrast, or is it concessive by indicating one argument as more central than the other? | contrasting |
train_18609 | The RST corpus also identifies an "antithesis" use which is a nucleus/satellite (subordinating) relation. | according to the (semantic) PDTB definition of concession (a denial of expectation), we could not find any concessive uses of während in 100 instances from German newspaper text, leaving only general "contrast". | contrasting |
train_18610 | This has resulted in the creation of discourse-annotated corpora, such as the Penn Discourse Treebank (PDTB) (Prasad et al., 2008), the Rhetorical Structure Theory (RST) Treebank (Carlson et al., 2002), and the Annodis corpus (Segmented Discourse Representation Theory, SDRT) (Afantenos et al., 2012), to name a few. | as of yet, the different frameworks are not inter-operable, nor is there a unified scheme for discourse annotation (but see the proposals by Benamara and Taboada (2015), Chiarcos (2014), Bunt et al. | contrasting |
train_18611 | With that aim in view, the labels in the different frameworks are mapped onto each other. | the mapping does not include the structures (i.e. | contrasting |
train_18612 | The authors claim that their taxonomy is robust across theoretical frameworks and can be applied to corpora from different languages. | they identify some problems for the mapping, e.g. | contrasting |
train_18613 | Other studies that have used the PDTB framework to annotate spoken data are Demirşahin and Zeyrek (2014) and Stoyanchev and Bangalore (2015). | they have not made any further changes to the PDTB framework. | contrasting |
train_18614 | This corpus is already available on line at http://sldr.org/sldr000847/fr in its first version. | many enhancements have been made in order to allow the development of automatic techniques to transcript conversations and to capture their meaning. | contrasting |
train_18615 | The goal of the DECODA project is to enable the development of these strategies and these applications while allowing the annotation effort to be less important. | the type of speech that is found in the conversations of call-centers is quite far from what usually treat ASR Systems. | contrasting |
train_18616 | It consists of two parts: On the one hand, 23 texts were written by one of the authors and have been used as examples in teaching and testing argumentation analysis with students. | 90 texts have been collected in a controlled text generation experiment, where 23 subjects wrote short texts of controlled linguistic and rhetoric complexity, discussing one of the issues they chose from a pre-defined list of controversial issues. | contrasting |
train_18617 | Argumentative SUP-PORTs cooccur with ELABORATION, EXPLANATION, and RESULT. | 48% of the supports cannot be mapped to SDRT edges, which is more than in the alignment of argumentation and RST. | contrasting |
train_18618 | fuĝ|into [fugitive]), dropped by a similar margin, from 2% to 0.8%. | most new/unknown tokens are CWs 8 , and the effect is even bigger for multiple (3+) CWs, which were 10 times as frequent among new words. | contrasting |
train_18619 | Moreover, many descriptions can be deduced from other relations because most relations are symmetrical or could be recovered transitively. | we consider this redundancy as beneficial because it eases the creation of the resource, its update, and its use. | contrasting |
train_18620 | (2014) where POS accuracy was 92.4%. | the lemma accuracy reported by Pasha et al. | contrasting |
train_18621 | The morphological part comprises deep-structure morphological analyses of German. | as it was developed in the Nineties, both encoding and spelling are outdated. | contrasting |
train_18622 | Incorporating this feedback in the development process may have influence on the success of the app. | one challenge for the developers is to deal with the overwhelming amount of reviews. | contrasting |
train_18623 | The analysis of opinions in reviews has been widely investigated within the last decade (Pang and Lee, 2008) and is 1 https://itunes.apple.com/us/genre/ios/ id36?mt=8 2 https://play.google.com/store/ 3 https://appworld.blackberry.com/ webstore/ 4 https://www.microsoft.com/en-us/windows/ apps-and-games typically referred to as sentiment analysis or opinion mining (Liu, 2012). | previous work in the area of analyzing user reviews in app stores is limited. | contrasting |
train_18624 | The number of subjective phrases within the different application categories is relatively constant. | the number of application aspects (from 149 to 287) and relations (104 to 237) varies much more. | contrasting |
train_18625 | The out-of-domain Subtask 3 had no participants. | the data developed for this subtask remains available and can be used for future research. | contrasting |
train_18626 | implies the fear emotion, brought forward because of the possible diseases which germs can transmit. | for a software system to detect that, it should be able to conclude that germs are capable of spreading diseases and people are afraid of things which can spread diseases. | contrasting |
train_18627 | The accuracy is highest for joy. | we have learnt the smallest number of property-value pairs with respect to the other emotions. | contrasting |
train_18628 | Concepts can be connected with several types of conceptual relations (Loukachevitch & Dobrov, 2014). | to SentiWordNet and other sense-based sentiment resources, we found that a lot of Russian ambiguous words have the same polarity in all available senses. | contrasting |
train_18629 | Several pro-Iraq demonstrations have taken place in the last week. | to TimeML, in the Richer Event Description 3 (Styler et al., 2016, RED) framework the event/non-event distinction is not based on morpho-syntactic information. | contrasting |
train_18630 | 12 In FactBank, the factuality value assigned to blowing up in this sentence is PR+ ("did probably happen") for both relevant sources, namely the AUTHOR and the GEN AUTHOR. | 13 the uncertaint y expressed by suspected is not so much directed towards the proposition as a whole as it is towards a specific aspect of the event: the two Libyans being the ones who did it. | contrasting |
train_18631 | There is comparable little work with respect to visualisation in lexical semantics: Kievit-Kylar and Jones (2012) present a tool that allows to identify relations between words in a graph structure, where words are visualised as nodes, and word similarities are shown as directed edges. | to our proposed tool, the actual position of a word is however not determined by its underlying feature distribution, and can even be changed using click and drag. | contrasting |
train_18632 | In previous work by Siegel and McKeown (2000), Zarcone and Lenci (2008) and Friedrich and Palmer (2014), the event type of corpus utterances is predicted on the basis of syntactic and semantic features of the main predicate and the linguistic context thereof. | in this study, we investigate lexical (inner) aspect through a French lexical resource, "Les Verbes Français" (François et al., 2007), a valence lexicon of French verbs providing a detailed morpho-syntactic and semantic description for each reading (use) of a verb. | contrasting |
train_18633 | This is justified by the fact that under reading 01, remplir is more natural with an in-adverbial than with a for-adverbial. | it is nevertheless still acceptable with for-adverbial under its 'partitive' reinterpretation (described e.g. | contrasting |
train_18634 | 7 Again, these results confirm the importance and effectiveness of the syn/sem features as well as the influence of the agentivity features. | they again do not provide further insights about the direction of this influence. | contrasting |
train_18635 | This confirms in some way our hypothesis that agentivity positively correlates with aspectual flexibility (or, in other words, that predicates are aspectually more flexible under their agentive than under their nonagentive reading). | features indicative of concreteness are characteristic both of the VARiable and TELic classes. | contrasting |
train_18636 | DSMs have been around for a while (Landauer and Dumais, 1997). | the recent enthusiasm about neural networks and word embeddings has made DSMs more accurate and faster to build using very large corpora. | contrasting |
train_18637 | The comparisons with the individual monolingual tasks confirmed that the "monolingual similarity after translation" approach works reasonably well, in the light of common insights gained in the field of cross-language information retrieval. | as expected, it is highlighted that the quality of translation would impact the end-to-end performances. | contrasting |
train_18638 | For instance, for an application involving a simple interface to a robot, quantifiers may not be needed. | there is little point in providing a single 'Simplified MRS' as an alternative to the standard ERS. | contrasting |
train_18639 | First, the subgraph S to be replaced has to be identified in the DMRS graph G. This is done via a form of exact subgraph matching, where nodes in S are associated with corresponding nodes in G and all links in this subgraph of G match the ones in S (see §4.1). | matching nodes exactly would be very inflexible and would require writing all possible concrete instantiations of a mapping pattern. | contrasting |
train_18640 | Initial experiments with DMRS mapping for paraphrase using automatically acquired rules are described by MacKinlay (2012, ch7). | when the source and target languages of the subgraphs are different, mapping becomes a method of machine translation, analogous to the use of MRS transfer rules in MT (compare e.g., Bond et al. | contrasting |
train_18641 | How-to websites like wikiHow 1 and eHow 2 help how to do things in daily life. | the quantity of wikiHow and the quality of eHow cannot compete with those of wikipedia. | contrasting |
train_18642 | Of the 30 sampled tasks, "buy movie tickets early", "buy a car", and "write a cover letter for a recruitment consultant", which achieve accuracies of 0.3000, 0.3200, and 0.2800, respectively, perform worse than the other tasks. | "download music safely" and "play experimental music", which have accuracies of 0.8200 and 0.7000, perform better. | contrasting |
train_18643 | Indeed, semantic vectors built for a single language were able to predict in an unsupervised setting typological patterns that hold across languages. | the evaluation showed that our test set is a valid semantic benchmark. | contrasting |
train_18644 | Consider the following example: RT @NewAppleDevice: Apple's smartwatch can be a games platform and here's why http://t.co/uIMGDyw08I It contains factual information that can be understood even without visiting the link. | the following tweet is not an argument, given that it does not convey an independent message while excluding the link: For all #business students discussing #Apple-Watch this morning. | contrasting |
train_18645 | The majority of works in author profiling use Supervised Learning which requires a sufficiently large corpus of clean, correctly annotated training data. | in many author profiling tasks, such data is not available. | contrasting |
train_18646 | Syntactic features are less often used; cf., e.g., (Cheng et al., 2009). | from linguistics and philology we know that syntactic idiosyncrasies are also distinctive features of the writing style of individuals and groups of individuals who share demographic, social, cultural or gender characteristics; see, e.g., (Crystal and Davy, 1969;Biber, 1989;Strunk and White, 1999;Tufte, 2006). | contrasting |
train_18647 | We believe that this is due to the composition of the feature set we use. | adding more instances in a semi-supervised fashion lets the classification further improve. | contrasting |
train_18648 | Building on this automated measurement of edit types, we use a graphical model analogous to LDA topic modeling to identify the latent roles editors occupy, much as document comprise topics. | to studies that employed either clustering analysis or principle component analysis to extract user roles (Liu and Ram, 2011), our role modeling treats an editor as comprising multiple roles at the same time. | contrasting |
train_18649 | Most written Arabic has been Classical Arabic or Modern Standard Arabic (MSA), and therefore most natural language processing (NLP) tools address these two variants of Arabic. | this situation is currently undergoing change. | contrasting |
train_18650 | dyAl has a higher degree of grammaticalization than similar forms in other dialects, such as Levantine tabaς. | to tabaς, can reduce to d and cliticize, e.g. | contrasting |
train_18651 | In other words, we expected that the less frequent a lemma is, the more regularly it is derived from the viewpoint of derivational patterns. | this was a working hypothesis and had to be proved empirically first. | contrasting |
train_18652 | The ocamorph analyzer is also implemented in OCaml. | jmorph is implemented in Java. | contrasting |
train_18653 | We evaluated the task of semantic categorization by manually counting the number of correct and incorrect words in the given category. | in order to be able to perform this validation efficiently, the result lists were clustered automatically so that these groups could be reviewed at once. | contrasting |
train_18654 | Some of these false rejections were recovered during listening by the auditor noticing and transcribing the sentence manually. | separating all false from correct rejections would yield diminishing returns. | contrasting |
train_18655 | Human auditors may comb through the many hours of data in order to verify the read speech, but this task would be quite unwieldy. | a pure automatic approach that aligns the transcript and excludes imperfect alignments may be unnecessarily conservative. | contrasting |
train_18656 | The motivation for setting up this corpus was that there is no phonetically annotated and segmented corpus for this language pair of comparable of size and coverage. | to most learner corpora, the IFCASL corpus incorporate data for a language pair in both directions, i.e. | contrasting |
train_18657 | As expected, there were more frequent pauses and more frequent disfluencies in L2, as well as longer durations of pauses filled with breath noise than those without. | the analysis also reveals that in fluent phases the vast majority of pauses contains an audible inhalation -which suggests a reinterpretation of the terms "unfilled" and "silent" pauses. | contrasting |
train_18658 | We are exploiting our manually annotated corpus to design more robust forced speech-text alignment algorithms by anticipating possible errors made by learners. | the non-native realizations are studied in order to investigate efficient acoustic feedback provided to learners, and how feedback can robustly interact with automatic segmentation provided by automatic speech recognition. | contrasting |
train_18659 | Most of the observations made for English will probably remain identical for other European languages. | some analysis elements may differ due to the specific syntactic and morphological features. | contrasting |
train_18660 | We then observe many redundancies and fuzzy or vague uses. | even with a limited number of verbs, fuzzy uses are observed, which are due to the arguments the verb is combined with. | contrasting |
train_18661 | They are full of mention-to-entity mappings from which we might be able to train a wikifier. | we decide to annotate corpora by ourselves primarily because Wikipedia is too inconsistent to be used directly as training data. | contrasting |
train_18662 | If two LUs correspond to the same frame, they are both ascribed to it. | exemplary sentences are linked to LUs appropriate for the meaning of the predicate. | contrasting |
train_18663 | The super-lexicographers can make particular portions of entries available for particular lexicographers, which en-ables a sort of specialisations (e.g., for verb, noun and adverb subdictionaries). | the entries finished in one phase constitute an input for the next one. | contrasting |
train_18664 | Statistical machine translation algorithms tend to disregard lexicons. | when no parallel corpora are available, algorithms may resort to monolingual corpora and bilingual lexicon induction (Klementiev et al., 2012). | contrasting |
train_18665 | In recent years, several datasets have been released that include images and text, giving impulse to new methods that combine natural language processing and computer vision. | there is a need for datasets of images in their natural textual context. | contrasting |
train_18666 | Decades ago Stokoe defined sign language glosses as combinations of movements, hand shapes and location (Stokoe, 1980). | many studies developed their own gloss based annotations. | contrasting |
train_18667 | On the other hand, HOME ADDRESS and SCHOOL ADDRESS are considered as phrases because the words in the phrases can be modified. | for ADDICTED, the meaning is not preserved when the words are modified. | contrasting |
train_18668 | In particular, regarding elderly living alone at home, the detection of distress situation after a fall is very important to reassure this kind of population. | many studies do not include tests in real settings, because data collection in this domain is very expensive and challenging and because of the few available data sets. | contrasting |
train_18669 | Many effective SRL systems have been developed to work with written data. | when applying popular SRL systems such as ASSERT (Pradhan et al., 2005), Lund SRL (Johansson and Nugues, 2008), SWIRL (Surdeanu and Turmo, 2005), and Illinois SRL (Punyakanok et al., 2008) on transcribed speech, which was generated by an automatic speech recognizer (ASR), many errors are made due to the specific nature of the ASR transcribed data. | contrasting |
train_18670 | These attempts at connecting language expressions to real world objects such as images are often called symbol grounding (Harnad, 1990), an exciting new area in natural language processing. | images, videos, and many other forms of media have ambiguities that make symbol grounding difficult. | contrasting |
train_18671 | For example, two clusters that contain segments with the same name are merged. | a cluster which contains segments with different labels is divided. | contrasting |
train_18672 | This seems to be partly due to the underlying being designed for sitting subjects, while in our study participants were standing so that distance reference information like knee position is not applicable. | the remaining annotations proved to be highly consistent, with percentage agreements ranging from over 80 % to about 98 %. | contrasting |
train_18673 | Those rules can then be assembled in a system that is aimed to support inhabitants of a smart home for example by suggesting a certain configuration preset to the user. | to existing home automation technology, such a system is able to infer more general rules than remembering the status of a single device at a certain time, e.g. | contrasting |
train_18674 | The number of verbs from Italian-English annotated is 515, translated is 473 out of the total 521. | annotated and translated verbs from English-Italian are 546 and 497 in number respectively out of the total 550. | contrasting |
train_18675 | But in Urdu they are semantically treated as referring to the same action verb "moving" /ghumanA/. | in the English counterpart of the sentence ID: b65d7431ghumAnA refers to two verbs: 'to turn' and 'to rotate'. | contrasting |
train_18676 | Domain ontology describes knowledge of a certain domain and has been used in many applications such as QA systems. | constructing ontologies by hand is a tedious and time-consuming work. | contrasting |
train_18677 | We therefore added a Difference measure, which shows the difference between the coverage for each list in the two corpora. | this turned out not to be enough. | contrasting |
train_18678 | While the numbers indicate that the trilingual NN produces more (reliable) triples than the German NN, this is not generally the case (e.g., for triple f-score 0.6). | our evaluation is biased towards features from the German annotation as we use the German-based MLG gold annotations for bootstrapping the annotation scheme. | contrasting |
train_18679 | 1 For other languages, several corpora of spoken language exist. | similarly to the situation for the German language, only a few of them are manually annotated with part-of-speech tags or have tagsets adapted to their needs. | contrasting |
train_18680 | This can be inferred from the high accuracy of state-of-the-art POS tagging not only for English, but also most other languages such as Arabic, which reaches 97% for Arabic and English being at 97.32% (Gadde et al., 2011). | the performance of standard POS taggers for English is severely degraded on Tweets due to their noisiness and sparseness (Ritter et al., 2011). | contrasting |
train_18681 | There is a growing interest within the NLP community to build Arabic social media corpora by harvesting the web such as (Refaee and Rieser, 2014;Abdul-Mageed et al., 2012). | none of these resources are publicly available yet. | contrasting |
train_18682 | AMIRA also breaks urls into parts since they contain punctuation marks. | mADA and Stanford do not break all Twitter-specific elements into parts since they use the space as an indicator for a new token. | contrasting |
train_18683 | Based on that, we decide to further improve the accuracy of Stanford tagger while preserving its speed by using Arabic tweets training data. | there is no labelled training data available to do so. | contrasting |
train_18684 | Another approach to utilizing state-of-the art taggers is proposed by (Eger et al., 2016) mostly based on MarMoT (Müller and Schütze, 2015) and, thus, on higherorder CRFs. | to this, we focus on 1st order CRFs by asking about the payout of such a simpler model. | contrasting |
train_18685 | If we consider the inflectional categories in isolation, TLT-CRF performs worse than several of its competitors. | if we consider joint learning of these inflectional categories, TLT-CRF performs second best again (see the line ALL in Table 2). | contrasting |
train_18686 | Two other corpora exist: MULTEXT-East, a multilingual corpus extracted from the novel 1984, sentence-aligned and with morphosyntactic annotations; a corpus of 36,150 tokens has also been annotated with dependencies in the context of the BALRIC project. | 1 these corpora can not be used for our purpose: the former does not contain information about dependencies and, as reported by (Calacean and Nivre, 2009), the latter consists in simple sentences of only 8.9 tokens in average, and no diacritics, 2 therefore not representing actual Romanian language. | contrasting |
train_18687 | This is probably to be explained by the size of the lexicon as only ∼2% of the words in the whole 500ktoken corpus are not covered by the lexicon and the fact that including suffix features already deals with tokens not covered by the lexicon. | given that in the next section we perform experiments in which we drastically decrease the size of the lexicon used, we decided to keep for the the remainder of the experiments the more complex feature of MSD hypotheses given the longest suffix. | contrasting |
train_18688 | For example, some linguists consider the expression cá f ish rô anabas {anabas} as a compound word but bệnh illness sởi measles {measles} as two words (Hoang, 1998;Diep, 2005). | these expressions have a similar construction: the combination of a categorization noun 4 and a specific noun. | contrasting |
train_18689 | For example, adjuncts indicating degree and tenses modify adjectives and verbs, respectively. | this method does not necessarily work sufficiently with real texts. | contrasting |
train_18690 | and they are expressed through the adjunct. | the adjunct is dropped frequently in the text. | contrasting |
train_18691 | By applying these rules to a treebank, the method corrects errors. | this method does not achieve wide coverage of error correction. | contrasting |
train_18692 | There are several methods of detecting annotation errors in a phrase structure treebank (Dickinson and Meurers, 2003;Ule and Simov, 2004;Dickinson and Meurers, 2005;Boyd et al., 2007;Dickinson, 2009;Przepiórkowski and Lenart, 2012;Kulick et al., 2013;Faria, 2014). | there is little work on treebank error correction. | contrasting |
train_18693 | So the treebank is constituency-based and syntactic relations are represented by means of trees. | the FTB contains several specific annotations, aiming at reducing the embedding of the trees (limitation of the projections, suppression of the VP level, affixation of clitics, etc.). | contrasting |
train_18694 | The amount of data needed to establish subtle discourse constraints is too large 5 given the costly discourse annotation process. | a solution of manipulating control 'original' texts annotated to form hypotheses-driven modified text from an intermediate size corpora seems to be a very promising option. | contrasting |
train_18695 | In the EDT version, the word-form trükikoja 'printing house-gen' is governed by the postposition eest 'of', which in turn is governed by the main verb of the clause, which appears to be a particle verb kandis hoolt 'took care'. | the word-form trükikoja 'printing house-gen' itself is modified by the main verb asus 'was located' of the relative clause, resulting in crossing arcs. | contrasting |
train_18696 | 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. | the spoken treebank contains notably more adverbial modifiers (advmod), which mostly include expressions of deixis, discourse cohesion and modality. | contrasting |
train_18697 | Loan words were segmented based on how they are segmented in the English corpus such as "ဂရိ တ် တား မန် ချက် စတာ" for "Greater Manchester", "အင် တာနက် ဖို ရမ် " for "internet forum", "ဝီ ကီ နယူ းစ် " for "Wikinews". | segmentation of some Myanmar particles and post positional markers were heuristically segmented depending on the context. | contrasting |
train_18698 | Furthermore, the building, transformation or annotation of the (sometimes huge) language resources implies to set up and drive large scale projects involving human resources, thus generating potential issues with regards to working conditions. | nLP is called for when it comes to providing tools for de-identifying documents, discovering plagiarism or performing fact checking, all applications that can be considered as ethically positive. | contrasting |
train_18699 | Here again, the questions reflected our bias in favor of a better concern for ethics, but we are glad that people did not hesitate to express some disagreements in their answers and comments. | it has to be noted that out of the 43 comments added to this poll, the large majority is surprisingly positive and pro-active (much more than the French ones), while a few criticize the form of the survey, i.e. | contrasting |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.