id
stringlengths 7
12
| sentence1
stringlengths 6
1.27k
| sentence2
stringlengths 6
926
| label
stringclasses 4
values |
|---|---|---|---|
train_96000 | South Africa has 11 official languages and since the majority of South Africans are multilingual, code-switching occurs commonly and spontaneously. | the distribution of code-switch examples is sparse, which could make modelling of unseen code-switch events challenging. | neutral |
train_96001 | This means that our existing ASR already recognises such segments correctly. | edited transcript is a written document, thus it contains only final, grammatically correct and re-formulated sentences, which makes it not 100% accurate. | neutral |
train_96002 | Later we will add these segments directly to the training data, so each segment should be reasonably short and contain speech only from one speaker. | the improvement from adding these sequences to the training data will be limited because existing ASR already recognised them correctly; they already match the acoustic model quite well. | neutral |
train_96003 | So, we wanted to create a web-based application going forward that could potentially be accessed by anyone in the country. | in the case of IE, we face the issue of not knowing which native language(s) should be chosen for adaptation. | neutral |
train_96004 | Metadata for manual annotations includes amongst other information annotator ids 7 , details of manual curation as well as applied annotation guidelines. | in an adjudication step all three annotators then decided on the annotation, and remaining hard cases were discussed in the project context and documented separately. | neutral |
train_96005 | Eckart and Gärtner (2016) for a motivation and additional details of this additional layer): Based on the audio signal, some features of orality were re-introduced. | the interviews furthermore contain coreference chains and bridging links. | neutral |
train_96006 | Cognition The identification of Cognition is charactized by the importance of features such as the ratio of questions. | we focus on elements such as pauses, social noise, disfluencies, but also on the vocabulary used. | neutral |
train_96007 | Results revealed that the best performance is obtained when English is used for both training and testing. | results are presented in terms of percentage Equal Error rate (EEr), calculated using the standard NIST software (Safavi et al., 2012). | neutral |
train_96008 | This study investigates the effect of different languages on the performance of speaker verification systems. | in the cases which the users are only capable of speaking their mother-tongue, it is best to use that language for both training and testing. | neutral |
train_96009 | For this evaluation, we conducted 10 training epochs. | base, emotional change information of the event sentence readers is also associated with the sentences. | neutral |
train_96010 | In recent years, the importance of dialogue understanding systems has been increasing because interactive interfaces handling a natural language such as smart speakers have become popular. | we think that these relationships are also important for understanding speakers' motivations in conversations (especially in emotional conversations) in addition to the relationships used in Hasegawa et al. | neutral |
train_96011 | After the task, for each trigger utterance, overlapping replies and extremely low-quality replies (such as copied and pasted trigger utterances and empty replies) were discarded. | the example shown in table 1 presents typical acquired data. | neutral |
train_96012 | There is also a corpus based on a collection of logs extracted from Ubuntu-related chat rooms that is mainly composed of technical support conversations (Lowe et al., 2015;Lowe et al., 2016). | judgements on the reasonableness of the 23,196 dialogues were acquired. | neutral |
train_96013 | Following Jeffersonian transcribing system (Jefferson, 2004), we transcribe video-recordings of physiciancaregiver conversation to capture both what is said and how it is said. | intervention measures such as providing physicians with trainings of communication skills to resist caregiver pressure are likely to produce desirable outcomes. | neutral |
train_96014 | We also took bust (chest, shoulders, and head) shots of each participant (recorded at 30 Hz). | in an experiment, we found that our estimation model using dialogue acts outperformed those using morpheme information alone. | neutral |
train_96015 | They represent moreover the most difficult part of the system to be developed. | this platform, as described in Section 4, is semiautonomous: some modules of the architecture are simulated by an experimenter. | neutral |
train_96016 | The way doctors deliver bad news related to damage associated with care has a significant impact on the therapeutic process: disease evolution, adherence with treatment recommendations, litigation possibilities (Andrade et al., 2010). | different gestures of both doctors and patients have been annotated: head movements, posture changes, gaze direction, eyebrow expressions, hand gestures, and smiles. | neutral |
train_96017 | This offers an optimal sensorial immersion of the user. | our objective is not to learn a model (in a machine learning point of view) but to extract automatically and manually information to model the virtual patient's behavior (Porhet et al., 2017) that, then, will be validated through perceptive studies. | neutral |
train_96018 | We extracted the lexical features based on Google's Word2Vec (Mikolov et al., 2013). | the difference occurs because utterances in a tV drama form a conversation that is affected by the story's settings, characters, and language style. | neutral |
train_96019 | In Figure 4, we displayed an example of complex queries, in which the right order of operators "per date" and "highest" must be inferred. | quest has been evaluated against a complex corpus, which includes both SLqs and NLqs. | neutral |
train_96020 | Of these, 335 are removed in both the manually-coded text and by the constraints. | as above, constraints are specified for animacy, number, and obviation combinations, as well as whether the verb upon which a nominal is dependent is to its left or right. | neutral |
train_96021 | BPEmb and code to replicate our experiments is available at https://github.com/bheinzerling/bpemb. | recently, based on the assumption that a word's meaning can be reconstructed from its parts, several subwordbased methods have been proposed to deal with the unknown word problem: character-based recurrent neural networks (rNN) (Luong and Manning, 2016), character-based convolutional neural networks (CNN) (Chiu and Nichols, 2016), word embeddings enriched with subword information (FastText) (Bojanowski et al., 2017), and byte-pair encoding (BPE) (Sennrich et al., 2016), among others. | neutral |
train_96022 | FastText performance shows the lowest variance, i.e., it robustly yields good results across many different hyperparameter settings. | averaging a mention's associated embeddings is the worst architecture choice. | neutral |
train_96023 | This is expected for character-based models, but somewhat surprising for tokenbased models, given the fact that averaging is a common method for representing mentions in tasks such as entity typing (Shimaoka et al., 2017) or coreference resolution (Clark and Manning, 2016). | bPEmb and characterbased models show higher variance, i.e., they require more careful hyper-parameter tuning to achieve good results. | neutral |
train_96024 | For a review of the main challenges in the field, see (Krahmer and van Deemter, 2012) and (van Deemter, 2016). | monologue and dialogue situations of communication may not be entirely comparable or, at the very least, may elicit different referring expressions. | neutral |
train_96025 | However, studies of this kind will often rely on large collections of pre-recorded linguistic examples produced by every single speaker of interest, and on every domain under consideration, to obtain meaning-to-text mappings from which the lexicalisation model is built. | for simplicity, in what follows we shall focus on the choice of words that realise input properties, and we will leave aside issues of linearisation, agreement and others. | neutral |
train_96026 | The inclusion of the context identifier id is intended to reflect the practical observation that a concept may not have exactly the same meaning (and therefore not necessarily the same wording) in different contexts. | a post hoc analysis suggested that the use of personality information is particularly helpful in the lexicalisation of affective information (e.g., properties conveying attributes such as smile, emotion etc. | neutral |
train_96027 | REG involves two distinct issues: content determination (or deciding what to say) and surface realisation (or deciding how to say it in a given language). | a set of 368 descriptions of interest was obtained. | neutral |
train_96028 | The attention model tends to produce more specific descriptions and therefore is likely to be given lower scores. | we performed a quantitative analysis of results based on four evaluation metrics, including BLEU (Papineni et al., 2002), a precision-based evaluation metric used in machine translation. | neutral |
train_96029 | In less than two months, they successfully adapted the system to generate syntactic trees for wedding ring descriptions, speech pathologist reports, weather forecasts and flight details in French, soccer game descriptions in English, and lexicographic illustrative sentences for the lexical field of emotions in Mandarin, all with very satisfying results. | one must be able to control what the theme/topic of the sentence is. | neutral |
train_96030 | It outperforms both systems (SBMT and Dress-LS) by the number of sentences which undergone at least one change (97.1% instead of 82.9% and 67.1%, respectively) and by the simplicity gain (+0.28 as opposed to +0.03 and +0.14, respectively), while achieving similar grammaticality of the output (4.34 as opposed to 4.28 and 4.27, respectively). | and lastly, we note the importance of keeping a reduced size of the vocabulary -no more than 50,000 words. | neutral |
train_96031 | 6c NTS-SARI-cross As a result, many mosques will not enforce violations, both men and women when attending a mosque must follow these guidelines. | more explicitly, we model the hypothesis number as a hyper-parameter that we select after the model finished training. | neutral |
train_96032 | That is probably the reason for the much lower number of NE errors by the NTS systems trained and tested on the Wikipedia dataset. | 5b NTS-SARI-cross Graham attended Wheaton College from 1939 to 1943. | neutral |
train_96033 | Statistics regarding the size of the Wikipedia datasets are rendered in Table 3. | those with more experience were given more challenging training. | neutral |
train_96034 | 270 mentions), refer to generic uses of shooting and killings. | this is not the case for all expressions in the GVC. | neutral |
train_96035 | For example, the predicate dbo:knownFor, which Portuguese label is "conhecido por", has the first word identified as an inflection of the verb "conhecer" (en:know). | if this sentence was translated into English, we would have indicated to whom the dictionary belonged, her or his. | neutral |
train_96036 | Moreover, RDF has demonstrated a promising ability to support the creation of NLG benchmarks (Gardent et al., 2017). | rDF (rDF Working Group, 25 February 2014) statements are based on graph data models for representing knowledge. | neutral |
train_96037 | Early work applying traditional semantic role labeling (SRL) techniques to iSRL was met with deflating results. | we observed that using both the head word and the closest coreferent non-pronoun head word is better than using the head word only on our development set. | neutral |
train_96038 | This applies even more strongly to ON5V. | 1 the words and phrases that fill each role are very diverse, as illustrated by the examples in Table 2. | neutral |
train_96039 | We make the hypothesis that there are webpages in which it is probable to find definitions for highly specialized domains, because some feature a higher density of specialized vocabulary than others and some are explicitly characterized as explanatory. | the Digital Dictionary of the German Language (Digitales Wörterbuch der deutschen Sprache, DWDS 1 ) is a long term project of the Berlin-Brandenburg Academy of Sciences (BBAW). | neutral |
train_96040 | To not come up with either an incorrect or too ambiguous annotation, we consider the use of this relation to help the system. | of this work, we make available the first public online repository containing manually annotated Spanish AMRs. | neutral |
train_96041 | We also make the most of the richness of the encoding, while implementing some graphical choices to make their interpretation as clear as possible for end users. | it is assumed that from the point of view of users a graph is a more suitable tool to visualize and navigate through the terminological structure of a domain especially since some of its features can be exploited by designers to highlight different properties of relations and the nodes that these relations link. | neutral |
train_96042 | In the next section, we refer to our published results for which we used the heads discovered by our head-finding algorithm; this extrinsic evaluation of our heuristics proved to be highly successful. | in the next section, we refer to our published results for which we used the heads discovered by our head-finding algorithm; this extrinsic evaluation of our heuristics proved to be highly successful. | neutral |
train_96043 | Our automated phase generates theatre presentation as a candidate noun compound, with the label {CAUSE TO PERCEIVE, ENTITY}, implying that the presentation is by the theatre. | are called frame elements (FEs). | neutral |
train_96044 | However, this seems like a corner case, as a presentation is more likely to be at the theatre. | one can also use hierarchy of frame elements (defined along with frame relations) to generalize the semantic relations. | neutral |
train_96045 | Firstly, the approaches to creating some of the relation inventories and datasets are statistically motivated, rather than being linguistically motivated. | we do not include such examples in our dataset. | neutral |
train_96046 | We propose a novel annotation approach to address this dilemma. | it's worth noting that more than twice as much statives in news have a stative parent (57%) than the ones having an eventive parent (24%), contributing to deep stative branches, while in narratives a much higher percentage of statives directly depend on an eventive (49%), contributing to a large number of short stative branches. | neutral |
train_96047 | Other recent work in semantic parsing has also focused on programming languages, including regular expressions (Locascio et al., 2016), IFTTT scripts (Quirk et al., 2015), and SQL queries (Kwiatkowski et al., 2013;Iyer et al., 2017;Zhong et al., 2017). | other recent work in semantic parsing has also focused on programming languages, including regular expressions (Locascio et al., 2016), IFTTT scripts (Quirk et al., 2015), and SQL queries (Kwiatkowski et al., 2013;Iyer et al., 2017;Zhong et al., 2017). | neutral |
train_96048 | Our experiments confirmed the hypothesis that some forms of world knowledge can improve existing AMR parsers. | our analyses of gold standard NER features also revealed some of the limitations of using world knowledge with existing AMR parsers. | neutral |
train_96049 | The audio signal is first cut into segments by an audio segmentation algorithm. | furthermore, a search-engine enables the user to find all media files in which a keyword or phrase was spoken by searching the transcripts of spoken words provided by the automatic speech recognition. | neutral |
train_96050 | Although retrieval as well as analysis is based on the transcription to date only half of the interviews are transcribed and saved as text files. | the interview is not structured by questions but is open for the course of memories coming into the interviewee's mind when telling his entire life story from birth and childhood into the present. | neutral |
train_96051 | However, to achieve satisfactory results the speech recognition performance has to be improved further. | in this work we focus on acoustic modeling only and use our default language model described in subsection 2.2.4.. | neutral |
train_96052 | We also present the challenges caused by Oral History interviews for ASR-systems. | for evaluation we mostly used the clean subset of the DiSCo corpus in our previous work. | neutral |
train_96053 | In the last step of analysis keywords are extracted from the ASR-generated transcript using a tf-idf (term frequencyinverse document frequency) approach. | we recently trained language models for specific tasks with more than one million words in the lexicon. | neutral |
train_96054 | According to equation (2) we used a different room-impulse-response of the same room, than the one applied to the speech-signal, if one was available. | due to the computational time needed for training sophisticated neural networks we decided to use a 128 h subset of the GerTV1000h corpus for the experiments in this work. | neutral |
train_96055 | Experiments with different content of the acoustic training data indicate that careful selection of phonotactical data could be advantageous when developing a speech corpus of limited size. | the Hjal project was successful as a pilot project, but has not been maintained. | neutral |
train_96056 | The text was tokenized, thus separating punctuation from words, and lowercased. | due to the extra effort needed to collect such data and the possible loss in recording quality due to reading mistakes, a clear positive impact of special phonotactical data should be evident. | neutral |
train_96057 | Details of the training set and processed test sets B1 and B2 of subtask 1 are shown in Table 6. | it is suggested that our augmentation strategy is effective and texts from forums are more suitable than news texts for the DSL task. | neutral |
train_96058 | As we can see, with each feature removed, the performance of our system decreases to different extent except for the word unigrams. | identifying languages from very little data, from multilanguages input or discriminating between extremely similar languages are bottlenecks of this field (Ljubešic and Kranjcic, 2014;Zampieri et al., 2015b). | neutral |
train_96059 | We approach this problem by successfully adapting a high-performance standard German speech Table 6: WER [%] results of directly trained Swiss German speech recognitions systems using different types of pronunciation lexicons; standard German G2P (GG2P), combined data-driven Swiss German and standard German G2P (SGG2P) or grapheme sequences (Graph.) | in some use cases these resources are not available. | neutral |
train_96060 | Over the last 15 years, MaryTTS (Schröder and Trouvain, 2001) has become one of the reference systems for open source text-to-speech synthesis (TTS). | this includes, at the very least, the sequence of phonemes, i.e., the pronunciation, but typically also other features related to 1 Examples of such dependencies in MaryttS include thirdparty libraries for text tokenization (Jtok), number expansion (ICU4J), and part-of-speech (POS) tagging (OpenNLP publishes voice-somename-xy the voice project resolves the processed data from the cloud (e.g., downloading from GitHub), runs all steps required to build a voice, and is published to Bintray. | neutral |
train_96061 | ROUGE only analyzes the overlapping between the candidate compression and the references. | a Tesla Motors anunciou acordo para comprar a SolarCity por US$ 2,6 bilhões . | neutral |
train_96062 | In Section 4 we analyze the results achieved by state-of-theart methods using our dataset. | they differ in many details of their grammar and lexicon. | neutral |
train_96063 | On investigating, the feedback from the language expert was that the summaries are "good enough". | the monolingual extractive summaries were provided by Veooz 7 and Google translate was used for automatic translations. | neutral |
train_96064 | The delay can be just a few hours, days or sometimes the news does not appear in the regional languages at all. | • The pluggable and generic architecture provides possibility of using the workbench for almost any language pair, and with any set of external tools to plug into the pipeline. | neutral |
train_96065 | Moreover, there was almost no difference between a method using only the utterance position feature and others. | by dividing the utterances into topic units by topic segmentation, it is possible to take into account of topics of the discussion in conversation summarization. | neutral |
train_96066 | Conversation summarization is useful to understand the content of conversations for both participants and nonparticipants. | other tags were appended to the end of each utterance for sentence-level identification. | neutral |
train_96067 | The beginning part of the conversation usually contains a trigger of the discussion. | the accuracy was not enough. | neutral |
train_96068 | Gigaword corpora are also available in Arabic, Chinese, French, German, and Spanish. | over the past few years, artificial neural networks have shown promising results for abstractive summarization. | neutral |
train_96069 | They use the Wikipedia featured articles as source for summary texts, since they are (i) well-written, (ii) comprehensive, (iii) well-researched, (iv) neutral, and (v) stable according to the Wikipedia featured article criteria 2 . | we were not able to retrieve results for a significant number of sentences with the search engine in particular for German sentences (English: 10,102 (14.20%), German: 9,325 (41.81%) sentences with an empty search results). | neutral |
train_96070 | English and German lead sections have an average length of 13.87 and 8.99 sentences. | the evaluation results might be inaccurate if the summarization models are only evaluated on a few topic. | neutral |
train_96071 | Finally, the model has low accuracy for the Non-Entailment cases. | the participant HOLE is omitted, but mentioned as mandatory in the pattern, i.e. | neutral |
train_96072 | Target: influence (B2) Candidates: determine (B1), impress (A2), change A1In 2, where the target is the verb "influence" (B2), "determine" (B1) passes the grammatical reformation stage but fails at the definition stage because it does not cover the semantics of "influence" in the meaning of affecting the way someone thinks or behaves, though it might do so in another context where determination causes a change in someone's way of thinking. | this dataset has 137K pairs of simplified and unsimplified sentences that can be used for creating simplification rules. | neutral |
train_96073 | In order to serve the research community, the adherence to the principles of Open Science not only demands clarity about the data services offered, but also about what is to be expected from the analysis tools offered. | language also plays a role as the reflection of scientific and societal knowledge, as an instrument for human communication, as one of the central components of the identity of individuals, groups, cultures or nations, as an instrument for human cognition and expression, as a formal system, and as historical records in need of documentation and preservation. | neutral |
train_96074 | As in prior years, LDC has provided data and assessment for all six TAC KBP tracks including the endto-end Cold Start task, which builds a knowledge base from scratch along with evaluations with EDL, Slot Filling Events and Belief/Sentiment. | we computed fundamental frequency and log-scale pitch range controlling for individual and sex differences and correlating to neuropsychiatric tests and measures of gray matter atrophy. | neutral |
train_96075 | • Implementation: There are two aspects of implementation of multi-language services: implementation of service composition, and embedding of composite services into application systems. | '); Figure 2: Example of invoking the atomic translation service KyotoUJServer ;; Create the atomic language service client ;; Only URL is different with that of previous example $client = ClientFactory::createTranslationClient ('http://langrid.org/service_manager/wsdl/ kyoto1.langrid:GoogleTranslate'); ;; Specify the service authentication information $client->setUserId('someUserId'); $client->setPassword('somePassword'); ;; Set invocation parameters and get the result $result = $client->translate( Language::get('en'), Language::get('ja'), 'Have a nice day! | neutral |
train_96076 | This lack becomes most evident in Machine Translation applications for smaller European languages as well as other standard NLP tools and systems (according to approx. | the survey created a total of 634 responses and, considering the number of questions, a surprisingly high completion rate of 27%. | neutral |
train_96077 | Further, participants were asked to define the research fields, areas and sub-areas, methods and applications they work on. | when asked about the biggest challenge the European Language Technology field is facing at the moment (Q14) around 16% of all provided survey answers stress that the neglect of smaller languages is a severe threat, which is leading to a fragmented rather than a united and multilingual Europe. | neutral |
train_96078 | 2003) topics to represent story structure and measured similarity between trajectories. | the disruption leads to high tension and the restoration of equilibrium lowers the tension. | neutral |
train_96079 | The studies reviewed above exemplify a rich research tradition using statistical analysis of wine review corpora. | for the terminological features, we therefore considered the 15,357 most domain-specific terms, sorted by termhood. | neutral |
train_96080 | Variability caused by the selection of the actual texts for the corpora might contribute to the observed differences. | the performance of both variants stabilizes for text lengths greater than 7,000 words. | neutral |
train_96081 | For all methods, the central 50% of samples lie in a fairly narrow range of ±5 percent points around the median (the colored boxes). | we also observed considerable performance differences of roughly ten percent points between English, French and German. | neutral |
train_96082 | Nevertheless, assuming the existence of an appropriate dataset, architectures for generating face descriptions are likely to share many of the challenges in the more familiar image description task. | by including country of origin, age and gender among the demographic details we request from participants, we hope to be able to address these questions in a more informed manner. | neutral |
train_96083 | (Ordonez et al., 2011) and (Kuznetsova et al., 2012) are examples of retrieval in visual space; other approaches rely on retrieval in multimodal space (Hodosh et al., 2013, Socher et al., 2014. | this raises the possibility of more in-depth investigation of the use of language in a specific domain, including the identification and description of salient or distinguishing features (say, eye colour, the shape of a face, emotion or ethnicity) and the investigation of the conditions under which they tend to feature in human descriptions. | neutral |
train_96084 | While the majority include physical features, there are also emotional descriptors, as well as analogical descriptions (as when a person is described as resembling a criminal; see Figure 1a) and inferred characteristics (such as the inferred nationality of the man in Figure 1a). | the system saved session variables, meaning that there was a limited time period during which participants could revisit the online interface and resume the description exercise from where they had left off. | neutral |
train_96085 | The residuals of the model (i.e., respDev ∼ 1 + adj + σ p + µ p + σ p : µ p + (1|turker)) showed that the data was heteroskedastic. | in order to build technology that has the ability to answer questions relevant to national and global security, e.g., on food insecurity in certain parts of the world, one has to implement machine reading technology that extracts causal mechanisms from texts. | neutral |
train_96086 | The resulting resource we provide 1 consists of a linear model for each adjective that takes as input the typical distribution of the item being modified and in turn provides the predicted size of the change. | responses generated using MTurk can be quite noisy, so to reduce the amount of noise in our data we excluded data based on several criteria. | neutral |
train_96087 | Intuitively, by using word embeddings trained over a large corpus, we already know some of the underlying semantics of the unseen adjectives. | when describing events, authors often make use of gradable adjectives, i.e., adjectives (such as small) that can take a range of magnitudes or degrees (e.g., something can be a little small, very small, extraordinarily small, etc.) | neutral |
train_96088 | To ensure that this difference between story-wise cross-validation and the default test set reflects the challenge of encountering new storyspecific content and isn't simply an artefact of the test set, we performed random cross-validation with held-out sets of the size of the average narrative. | by reducing the data requirements we make automatic transcription technology more feasible in a language documentation setting. | neutral |
train_96089 | (2014) (pitch), and a combination of both (fbank+pitch). | we assess the performance of the systems as training data scales from 10 minutes to 224 minutes of spontaneous speech of a single Na speaker, and between 12 and 50 minutes for a single speaker of Chatino. | neutral |
train_96090 | Since large amounts of text are used for language model training, such systems often do not incorporate pitch information for speech recog-nition of tonal languages (Metze et al., 2013), as they can instead rely on contextual information for tonal disambiguation via the language model (Le and Besacier, 2009;Feng et al., 2012) even though there is no computational burden in additionally using pitch features. | as for the sister, she stayed at home. | neutral |
train_96091 | This paper presents a general use corpus for the Native American indigenous language Choctaw. | the data set comprises text, audio, and video. | neutral |
train_96092 | This paper introduces a general use corpus for Choctaw, an American indigenous language. | of this history, the Choctaw language has few published works and little text representation online. | neutral |
train_96093 | Currently only North Sami has a Wikipedia, but during the data collection sessions, interest was also sparkled within Inari Sami community to start a Wikipedia of their own. | the question also concerns how to define the appropriate ways to use the data, who can use it and who has the right to access the data. | neutral |
train_96094 | Since the culture and traditions are carried by the language, there is always an inherent element of cultural knowledge in the topics and the use of the language, and although corpus collection can aim at a neutral goal of documenting a less-resourced language, in practice it may not be possible to collect a culturally "neutral" corpus. | (2018) use an end-to-end dialect recognition system based on the deep learning method and discuss its use as enabling technology for building interactive applications. | neutral |
train_96095 | For example, the French word ordinateur (computer) is commonly used in Tunisian Arabic, so it might be transliterated as Awr-dynAtwr. | • Add new words that are not on the AUTO list. | neutral |
train_96096 | For the individual task, the native speakers can do the task anywhere before the deadline as shown in Figure 11. | • T 2(L ID , L A ): From an estimated duration of 10 seconds per translation and a daily wage of JPY5,000/8 hours, the estimated total translation per day is 1 × 6 × 60 × 8 = 2, 880 and the estimated cost is JPY1.74 per correct translation. | neutral |
train_96097 | All body part nouns are traditionally considered to be relational (Laczkó et al., 2009). | this includes NomBank (Meyers et al., 2004), and other work on nominal semantic role labelling (Padó et al., 2008;Gerber and Chai, 2010). | neutral |
train_96098 | To generate multilingual cognate tables, we employ an automatic method of clustering words from our lexical resources. | the character transition probabilities are obtained from alignment using GIZA++. | neutral |
train_96099 | In this paper, we describe a new system to extract, index, search, and visualize entities in Wikipedia. | this system could be improved in many ways. | neutral |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.