id
stringlengths 32
33
| x
stringlengths 41
1.75k
| y
stringlengths 4
39
|
|---|---|---|
d9c0e641f8ceb61e5d6e416bfc6492_0 | It has been done for constituency parsing for example by Collins (1999) but also for dependency parsing for example by <cite>Nilsson et al. (2007)</cite> . | background |
d9c0e641f8ceb61e5d6e416bfc6492_1 | <cite>Nilsson et al. (2007)</cite> modified the representation of several constructions in several languages and obtained a consistent improvement in parsing accuracy. | background |
d9c0e641f8ceb61e5d6e416bfc6492_2 | In this paper, we will investigate the case of the verb group construction and attempt to reproduce the study by <cite>Nilsson et al. (2007)</cite> on UD treebanks to find out whether or not the alternative representation is useful for parsing with UD. | uses |
d9c0e641f8ceb61e5d6e416bfc6492_3 | <cite>Nilsson et al. (2007)</cite> have shown that these same modifications as well as the modification of nonprojective structures helps parsing in four languages. | background |
d9c0e641f8ceb61e5d6e416bfc6492_5 | <cite>Nilsson et al. (2007)</cite> show that making the auxiliary the head of the dependency as in Figure 2 is useful for parsing Czech and Slovenian. | background |
d9c0e641f8ceb61e5d6e416bfc6492_6 | We will follow the methodology from <cite>Nilsson et al. (2007)</cite> , that is, to transform, parse and then detransform the data so as to compare the original and the transformed model on the original gold standard. | uses |
d9c0e641f8ceb61e5d6e416bfc6492_7 | For comparability with the study in <cite>Nilsson et al. (2007)</cite> , and because we used a slightly modified version of their algorithm, we also tested the approach on the versions of the Czech and Slovenian treebanks that they worked on, respectively version 1.0 of the PDT (Hajic et al., 2001 ) and the 2006 version of SDT (Deroski et al., 2006) . | uses differences |
d9c0e641f8ceb61e5d6e416bfc6492_8 | In this paper, we have attempted to reproduce a study by <cite>Nilsson et al. (2007)</cite> that has shown that making auxiliaries heads in verb groups improves parsing but failed to show that those results port to parsing with Universal Dependencies. | uses background |
da2429450c8d1f1f3e72383c86ec73_0 | Our approach was to build up on the system of the last year's winning approach by NRC Canada 2013 <cite>(Mohammad et al., 2013)</cite> , with some modifications and additions of features, and additional sentiment lexicons. Furthermore, we used a sparse ( 1 -regularized) SVM, instead of the more commonly used 2 -regularization, resulting in a very sparse linear classifier. | extends |
da2429450c8d1f1f3e72383c86ec73_1 | Our approach was to build up on the system of the last year's winning approach by NRC Canada 2013 <cite>(Mohammad et al., 2013)</cite> , with some modifications and additions of features, and additional sentiment lexicons. | extends uses |
da2429450c8d1f1f3e72383c86ec73_2 | Compared to the previous NRC Canada 2013 approach <cite>(Mohammad et al., 2013)</cite> , our main changes are the following three: First we use sparse linear classifiers instead of classical dense ones. | extends |
da2429450c8d1f1f3e72383c86ec73_3 | We tried to reproduce the same classifier as in <cite>(Mohammad et al., 2013)</cite> as a baseline for comparison. | uses |
da2429450c8d1f1f3e72383c86ec73_4 | Unfortunately our replica system of<cite> Mohammad et al. (2013)</cite> only achieved an F1-score of 63.25 on the Twitter-2013 test set, while their score in the 2013 competition on the same test set was 69.02, nearly 6 points higher in F1. | differences |
da2429450c8d1f1f3e72383c86ec73_5 | For each lexicon, the 4 scores were the same as in <cite>(Mohammad et al., 2013)</cite> , i.e. per tweet, we use the number of tokens appearing in the lexicon, the sum and the max of the scores, and the last non-zero score. | similarities |
da2429450c8d1f1f3e72383c86ec73_6 | All text was transformed to lowercase (except for those features in <cite>(Mohammad et al., 2013)</cite> which use case information). | uses |
da2429450c8d1f1f3e72383c86ec73_7 | We used the same set of lexicons as in <cite>(Mohammad et al., 2013)</cite> , with one addition: | extends uses |
da2429450c8d1f1f3e72383c86ec73_8 | To construct the lexicon, we extracted the POS n-grams (as we described in Section 3.1.1 above) from all texts. In comparison,<cite> Mohammad et al. (2013)</cite> used noncontiguous n-grams (unigram-unigram, unigrambigram, and bigram-bigram pairs) . We only used POS n-grams with 2 tokens kept original, and the remaining ones replaced by their POS tag, with n ranging from 3 to 6. | differences |
da2429450c8d1f1f3e72383c86ec73_9 | While in <cite>(Mohammad et al., 2013)</cite> , the score for each n-gram was computed using point-wise mutual information (PMI) with the labels, we trained a linear classifier on the same labels instead. | differences |
da2429450c8d1f1f3e72383c86ec73_10 | We used the same 3 existing sentiment lexicons as in <cite>(Mohammad et al., 2013)</cite> . | similarities uses |
da2429450c8d1f1f3e72383c86ec73_11 | The NRC hashtag sentiment lexicon was generated automatically from a set of 775k tweets containing a hashtag of a small predefined list of positive and negative hashtags <cite>(Mohammad et al., 2013)</cite> . | uses |
da2429450c8d1f1f3e72383c86ec73_12 | Our system is built up on the approach of NRC Canada <cite>(Mohammad et al., 2013)</cite> , with several modifications and extensions (e.g. sparse linear classifiers, | extends uses |
db1fd6f10a3ee22e22093d50395217_0 | The other attempt of same 6 way PIBOSO classification on the same dataset is presented by <cite>(Verbeke et al., 2012)</cite> . | background |
db1fd6f10a3ee22e22093d50395217_1 | The other attempt of same 6 way PIBOSO classification on the same dataset is presented by <cite>(Verbeke et al., 2012)</cite> . Unlike us and Kim et al. (2011) they have used SVM-HMM 2 for learning. | differences |
db1fd6f10a3ee22e22093d50395217_2 | Please note that the way we categorised an abstract as structured or unstructured might be a bit different from previous approaches by Kim et al. (2011) and<cite> Verbeke et al. 2012</cite> . | differences |
db1fd6f10a3ee22e22093d50395217_3 | Using sentence ordering labels for unstructured abstracts is the main difference compared to earlier methods (Kim et al., 2011;<cite> Verbeke et al., 2012)</cite> . | differences |
db1fd6f10a3ee22e22093d50395217_4 | However, we compare our results with (Kim et al., 2011) and <cite>(Verbeke et al., 2012)</cite> using the microaveraged F-scores as in Table 3 . | uses |
db1fd6f10a3ee22e22093d50395217_5 | However, we compare our results with (Kim et al., 2011) and <cite>(Verbeke et al., 2012)</cite> using the microaveraged F-scores as in Table 3 . Our system outperformed previous works in unstructured abstracts (22% higher than state-of-the-art). | differences |
db1fd6f10a3ee22e22093d50395217_6 | Our system outperformed earlier existing state-of-art systems (Kim et al., 2011;<cite> Verbeke et al., 2012)</cite> . | differences |
db6794da83b12336ab946e5777346d_0 | The title of our talk-an implicit reference to the English cliché like a spider weaving her webintends to attract one's attention to the metaphor that can be drawn between the dance of a spider weaving her web and a new lexicographic gesture that is gradually emerging from the work on Net-like lexical resources (Fellbaum, 1998; Baker et al., 2003;<cite> Gader et al., 2012)</cite> . | uses |
db6794da83b12336ab946e5777346d_1 | The title of our talk-an implicit reference to the English cliché like a spider weaving her webintends to attract one's attention to the metaphor that can be drawn between the dance of a spider weaving her web and a new lexicographic gesture that is gradually emerging from the work on Net-like lexical resources (Fellbaum, 1998; Baker et al., 2003;<cite> Gader et al., 2012)</cite> . | uses |
db6794da83b12336ab946e5777346d_2 | Work performed on the French Lexical Network <cite>(Gader et al., 2012</cite> ) will serve to demonstrate how the lexicographic process can be made closer to actual navigation through lexical knowledge by the speaker. | future_work |
db6794da83b12336ab946e5777346d_3 | Computational aspects of the work on the French Lexical Network are dealt with in<cite> (Gader et al., 2012)</cite> . | background |
dc6d4eb1870ed5b0bbcbbf6686e5be_0 | Understanding the temporal information in natural language text is an important NLP task (Verhagen et al., 2007 (Verhagen et al., , 2010 UzZaman et al., 2013; Minard et al., 2015; Bethard et al., 2016 Bethard et al., , 2017 . A crucial component is temporal relation (TempRel; e.g., before or after) extraction (Mani et al., 2006; Bethard et al., 2007; Do et al., 2012; Mirza and Tonelli, 2016; <cite>Ning et al., 2017</cite> Ning et al., , 2018a . | background |
dc6d4eb1870ed5b0bbcbbf6686e5be_1 | Annotators in this setup usually focus only on salient relations but overlook some others. It has been reported that many event pairs in TimeBank should have been annotated with a specific TempRel but the annotators failed to look at them (Chambers, 2013;<cite> Ning et al., 2017)</cite> . | motivation |
dc6d4eb1870ed5b0bbcbbf6686e5be_2 | It has been reported that many event pairs in TimeBank should have been annotated with a specific TempRel but the annotators failed to look at them (Chambers, 2013;<cite> Ning et al., 2017)</cite> . | background |
dc6d4eb1870ed5b0bbcbbf6686e5be_3 | Two recent TempRel extraction systems (Mirza and Tonelli, 2016; <cite>Ning et al., 2017</cite> ) also reported their performances on TB-Dense (F) and on TempEval-3 (P) separately. However, there are no existing systems that jointly train on both. | motivation |
dc6d4eb1870ed5b0bbcbbf6686e5be_4 | Two recent TempRel extraction systems (Mirza and Tonelli, 2016; <cite>Ning et al., 2017</cite> ) also reported their performances on TB-Dense (F) and on TempEval-3 (P) separately. | background |
dc6d4eb1870ed5b0bbcbbf6686e5be_5 | Note that Algorithm 1 is only for the learning step of TempRel extraction; as for the inference step of this task, we consistently adopt the standard method by solving Eq. (1), as was done by (Bramsen et al., 2006; Chambers and Jurafsky, 2008; Denis and Muller, 2011; Do et al., 2012;<cite> Ning et al., 2017)</cite> . | uses |
dc6d4eb1870ed5b0bbcbbf6686e5be_6 | Then the ILP objective is formulated aŝ where {r m 3 } is selected based on the general transitivity proposed in<cite> (Ning et al., 2017)</cite> . | uses |
dc6d4eb1870ed5b0bbcbbf6686e5be_7 | A standard way to perform global inference is to formulate it as an Integer Linear Programming (ILP) problem (Roth and Yih, 2004 ) and enforce transitivity rules as constraints. Then the ILP objective is formulated aŝ where {r m 3 } is selected based on the general transitivity proposed in<cite> (Ning et al., 2017)</cite> . | background |
dc6d4eb1870ed5b0bbcbbf6686e5be_8 | We believe that global inference makes better use of the information provided by P; in fact, as we show in Sec. 4, it does perform better than local inference. A standard way to perform global inference is to formulate it as an Integer Linear Programming (ILP) problem (Roth and Yih, 2004 ) and enforce transitivity rules as constraints. Then the ILP objective is formulated aŝ where {r m 3 } is selected based on the general transitivity proposed in<cite> (Ning et al., 2017)</cite> . | uses |
dc6d4eb1870ed5b0bbcbbf6686e5be_9 | Results are shown in Table 2 , where all systems were compared in terms of their performances on "same sentence" edges (both nodes are from the same sentence), "nearby sentence" edges, all edges, and the temporal awareness metric used by the TempEval3 workshop. The first part of Table 2 (Systems 1-5) refers to the baseline method proposed at the beginning of Sec. 3, i.e., simply treating P as F and training on their union. The second part (Systems 6-7) serves as an ablation study showing the effect of bootstrapping only. While System 7 can be regarded as a reproduction of<cite> Ning et al. (2017)</cite> , the original paper of<cite> Ning et al. (2017)</cite> achieved an overall score of P=43.0, R=46.4, F=44.7 and an awareness score of P=42.6, R=44.0, and F=43.3, and the proposed System 9 is also better than<cite> Ning et al. (2017)</cite> on all metrics. | differences |
dc6d4eb1870ed5b0bbcbbf6686e5be_10 | While incorporating transitivity constraints in inference is widely used,<cite> Ning et al. (2017)</cite> proposed to incorporate these constraints in the learning phase as well. | background |
dc6d4eb1870ed5b0bbcbbf6686e5be_11 | One of the algorithms proposed in<cite> Ning et al. (2017)</cite> is based on Chang et al. (2012) 's constraint-driven learning (CoDL), which is the same as our intermediate System 7 in Table 2 ; the fact that System 7 is better than System 1 can thus be considered as a reproduction of<cite> Ning et al. (2017)</cite> . | uses |
dc6d4eb1870ed5b0bbcbbf6686e5be_12 | Despite the technical similarity, this work is motivated differently and is set to achieve a different goal:<cite> Ning et al. (2017)</cite> tried to enforce the transitivity structure, while the current work attempts to use imperfect signals (e.g., partially annotated) taken from additional data, and learn in the incidental supervision framework. | similarities differences |
dc6d4eb1870ed5b0bbcbbf6686e5be_13 | System 7 can also be considered as a reproduction of<cite> Ning et al. (2017)</cite> (see the discussion in Sec. 5 for details). | uses |
dcc866dcfb5f9233170d633d052e8b_0 | The use of various synchronous grammar based formalisms has been a trend for statistical machine translation (SMT) (Wu, 1997; Eisner, 2003; Galley et al., 2006; <cite>Chiang, 2007</cite>; Zhang et al., 2008) . | background |
dcc866dcfb5f9233170d633d052e8b_1 | For instance, in our investigations for SMT (Section 3.1), the Formally SCFG based hierarchical phrase-based model (hereinafter FSCFG)<cite> (Chiang, 2007)</cite> has a better generalization capability than a Linguistically motivated STSSG based model (hereinafter LSTSSG) (Zhang et al., 2008) , with 5% rules of the former matched by NIST05 test set while only 3.5% rules of the latter matched by the same test set. | uses background |
dcc866dcfb5f9233170d633d052e8b_2 | The rule extraction in current implementation can be considered as a combination of the ones in<cite> (Chiang, 2007)</cite> and (Zhang et al., 2008) . | uses |
dcc866dcfb5f9233170d633d052e8b_3 | For example,<cite> (Chiang, 2007)</cite> adopts a CKY style span-based decoding while (Liu et al., 2006 ) applies a linguistically syntax node based bottom-up decoding, which are difficult to integrate. | background |
dcc866dcfb5f9233170d633d052e8b_4 | FSCFG An in-house implementation of purely formally SCFG based model similar to<cite> (Chiang, 2007)</cite> . | similarities |
ddd23a034c366b62b53d15128edd45_0 | Our contributions are summarized as follows: (1) we extend a probablistic model used in the <cite>previous work</cite> which concurrently performs word reordering and dependency parsing; (2) we conducted an evaluation experiment using our semi-automatically constructed evaluation data so that sentences in the data are more likely to be spontaneously written by natives than the automatically constructed evaluation data in the <cite>previous work</cite>. | extends |
ddd23a034c366b62b53d15128edd45_1 | To solve the problem, we previously proposed a method for concurrently performing word reordering and dependency parsing and confirmed the effectiveness of their proposed method using evaluation data created by randomly changing the word order in newspaper article sentences <cite>(Yoshida et al., 2014)</cite> . | background |
ddd23a034c366b62b53d15128edd45_2 | This paper proposes a new method on Japanese word reordering based on concurrent execution with dependency parsing by extending the probablistic model proposed by <cite>Yoshida et al. (2014)</cite> , and describes an evaluation experiment using our 1 Bunsetsu is a linguistic unit in Japanese that roughly corresponds to a basic phrase in English. | extends |
ddd23a034c366b62b53d15128edd45_3 | We use the same search algorithm as one proposed by <cite>Yoshida et al. (2014)</cite> , which can efficiently find the approximate solution from a huge number of candidates of the pattern by extending CYK algorithm used in conventional dependency parsing. | uses |
ddd23a034c366b62b53d15128edd45_4 | In this paper, we refine the probabilistic model proposed by <cite>Yoshida et al. (2014)</cite> to improve the accuracy. | extends |
ddd23a034c366b62b53d15128edd45_5 | The structure S is defined as a tuple S = ⟨O, D⟩ where In the probablistic model proposed by <cite>Yoshida et al. (2014)</cite> , P (S|B) was calculated as follows: | uses |
ddd23a034c366b62b53d15128edd45_6 | We extend <cite>the above model</cite> and calculate P (S|B) as follows: | extends |
ddd23a034c366b62b53d15128edd45_7 | Therefore, we mix Formulas (3) and (4) by adjusting the weight α depending on the adequacy of word order in an input sentence, instead of using the constant 0.5 in the previous model proposed by <cite>Yoshida et al. (2014)</cite> . | differences |
ddd23a034c366b62b53d15128edd45_8 | Each factor in Formula (2) is estimated by the maximum entropy method in the same approximation procedure as that of <cite>Yoshida et al. (2014)</cite> . | uses |
ddd23a034c366b62b53d15128edd45_9 | Therefore, <cite>our previous work</cite> (<cite>Yoshida et al., 2014</cite>) artificially generated sentences which were not easy to read, by just automatically changing the word order of newspaper article sentences in Kyoto Text Corpus 3 based on the dependency structure. | background |
ddd23a034c366b62b53d15128edd45_10 | Therefore, <cite>our previous work</cite> (<cite>Yoshida et al., 2014</cite>) artificially generated sentences which were not easy to read, by just automatically changing the word order of newspaper article sentences in Kyoto Text Corpus 3 based on the dependency structure. However, just automatically changing the word order may create sentences which are unlikely to be written by a native. | motivation |
ddd23a034c366b62b53d15128edd45_11 | That is, if a subject judges that a sentence generated by automatically changing the word order in the same way as <cite>the previous work</cite> (<cite>Yoshida et al., 2014</cite> ) may have spontaneously written by a native. | background |
ddd23a034c366b62b53d15128edd45_12 | We compared our method to <cite>Yoshida</cite>'s method (<cite>Yoshida et al., 2014</cite>) and two conventional sequential methods. | uses |
ddd23a034c366b62b53d15128edd45_13 | All of the methods used the same training features as those described in <cite>Yoshida et al. (2014)</cite> . | uses |
ddd23a034c366b62b53d15128edd45_14 | The dependency accuracy of our method was significantly lower than that of the two sequential methods, and was higher than that of <cite>Yoshida's method</cite> although there was no significant difference. | differences |
ddd23a034c366b62b53d15128edd45_15 | On the other hand, the sentence accuracy of our method was highest among <cite>all the methods</cite> although there were no significant differences in them. | differences |
ddd23a034c366b62b53d15128edd45_16 | As a result of analysis, especially, our method and <cite>Yoshida's method</cite> tended to improve the sentence accuracy very well in case of short sentences. | similarities |
ddd23a034c366b62b53d15128edd45_17 | Especially, we extended the probablistic model proposed by <cite>Yoshida et al. (2014)</cite> to deal with sentences spontaneously written by a native. | extends |
debdaa202ebd856991e09e5e00a12b_0 | This architecture has been empirically shown to perform well at Named Entity Recognition (NER) tasks <cite>(Lample et al., 2016)</cite> . | background |
debdaa202ebd856991e09e5e00a12b_1 | In the Named Entity Recognition task, we utilized a deep learning approach, given the demonstrated effectiveness of such an architecture in this domain <cite>(Lample et al., 2016)</cite> . | uses motivation |
e0b72115e1905226d22876e72aa304_0 | The basic structure of our CKB completion model is similar to that of<cite> Li et al. (2016b)</cite> . | similarities |
e0b72115e1905226d22876e72aa304_1 | The basic structure of our CKB completion model is similar to that of<cite> Li et al. (2016b)</cite> . The main difference between ours and theirs is that our method learns the CKB completion and generation tasks jointly. | differences |
e0b72115e1905226d22876e72aa304_2 | Previous model<cite> Li et al. (2016b)</cite> defined a CKB completion model that estimates a confidence score of an arbitrary triple ⟨t 1 , r, t 2 ⟩. They used a simple neural network model to formulate score(t 1 , r, t 2 ) ∈ R. | background |
e0b72115e1905226d22876e72aa304_3 | Our model Our CKB completion model is based on Li et al.'s (2016b) . | extends |
e0b72115e1905226d22876e72aa304_4 | Li et al. (2016b) formulate the phrase embedding by using attention pooling of LSTM and a bilinear function. | background |
e0b72115e1905226d22876e72aa304_5 | For the experiments with English, we used the ConceptNet 100K data released by<cite> Li et al. (2016b)</cite> 1 . | uses |
e0b72115e1905226d22876e72aa304_7 | CKB completion As baselines, we used the DNN AVG and DNN LSTM models<cite> (Li et al., 2016b</cite> ) that were described in Section 3.1. | uses |
e0b72115e1905226d22876e72aa304_8 | The threshold was determined by using the validation1 data to maximize the accuracy of binary classification for each method, as in<cite> (Li et al., 2016b)</cite> . | uses similarities |
e0b72115e1905226d22876e72aa304_9 | The bottom two lines show the best performances reported in<cite> (Li et al., 2016b)</cite> . The results indicate that our method improved the accuracy of CKB completion compared with the previous method. | differences |
e0b72115e1905226d22876e72aa304_11 | In Wiki gen, we used triples extracted by using the POS tag sequence pattern for each relation according to<cite> Li et al. (2016b)</cite> and scored each triple with CKB completion scores. | uses |
e0b72115e1905226d22876e72aa304_12 | This tendency is similar to the results reported in Li et al.<cite> (Li et al., 2016b)</cite> . | similarities |
e0b72115e1905226d22876e72aa304_13 | In particular,<cite> Li et al. (2016b)</cite> and Socher et al. (2013) proposed a simple KBC model for CKB. | background |
e0b72115e1905226d22876e72aa304_14 | The formulations of CKB completion in the two studies are the same, and we evaluated<cite> Li et al. (2016b)</cite> 's method as a baseline. | uses |
e0e21b4e473ad6fde28378b2dc4f34_0 | Methods to learn sparse word-based translation correspondences from supervised ranking signals have been presented by Bai et al. (2010) and<cite> Sokolov et al. (2013)</cite> . | background |
e0e21b4e473ad6fde28378b2dc4f34_1 | Our approach extends the work of<cite> Sokolov et al. (2013)</cite> by presenting an alternative learningto-rank approach that can be used for supervised model combination to integrate dense and sparse features, and by evaluating both approaches on cross-lingual retrieval for patents and Wikipedia. | extends differences |
e0e21b4e473ad6fde28378b2dc4f34_2 | The algorithm of<cite> Sokolov et al. (2013)</cite> combines batch boosting with bagging over a number of independently drawn bootstrap data samples from R. In each step, the single word pair feature is selected that provides the largest decrease of L exp . | background |
e0e21b4e473ad6fde28378b2dc4f34_3 | The baseline consensus-based voting Borda Count procedure endows each voter with a fixed amount of voting points which he is free to distribute among the scored documents (Aslam and Montague, 2001; <cite>Sokolov et al., 2013)</cite> . | background |
e0e21b4e473ad6fde28378b2dc4f34_4 | We use BoostCLIR 1 , a Japanese-English (JP-EN) corpus of patent abstracts from the MAREC and NTCIR data<cite> (Sokolov et al., 2013)</cite> . | uses |
e0e21b4e473ad6fde28378b2dc4f34_5 | A JP-EN system was trained on data described and preprocessed by<cite> Sokolov et al. (2013)</cite> , consisting of 1.8M parallel sentences from the NTCIR-7 JP-EN PatentMT subtask (Fujii et al., 2008) and 2k parallel sentences for parameter development from the NTCIR-8 test collection. | uses |
e0e21b4e473ad6fde28378b2dc4f34_6 | PSQ on patents reuses settings found by<cite> Sokolov et al. (2013)</cite> ; settings for Wikipedia were adjusted on its dev set (n=1000, λ=0.4, L=0, C=1). | uses |
e177758a227506bbf9de48f8f35715_0 | The second step is to perform dictionary induction by learning a linear projection, in the form of a matrix, between language vector spaces <cite>(Mikolov et al., 2013b</cite>; Lazaridou et al., 2015) . | background |
e177758a227506bbf9de48f8f35715_1 | It is one of the most competitive methods for generating word vector representations, as demonstrated by results on a various semantic tasks (Baroni et al., 2014;<cite> Mikolov et al., 2013b)</cite> . | background |
e177758a227506bbf9de48f8f35715_2 | To induce a bilingual dictionary for a pair of languages, we use the projection matrix approach <cite>(Mikolov et al., 2013b</cite>; Lazaridou et al., 2015) . | uses |
e264c45391853fb008c838aa7ccca8_0 | This paper proposes an extension of Sumida and Torisawa's method of acquiring hyponymy relations from hierachical layouts in Wikipedia<cite> (Sumida and Torisawa, 2008)</cite> . | extends |
e264c45391853fb008c838aa7ccca8_1 | We extract hyponymy relation candidates (HRCs) from the hierachical layouts in Wikipedia by regarding all subordinate items of an item x in the hierachical layouts as x's hyponym candidates, while <cite>Sumida and Torisawa (2008)</cite> extracted only direct subordinate items of an item x as x's hyponym candidates. | differences |
e264c45391853fb008c838aa7ccca8_2 | Many NLP researchers have attempted to automatically acquire hyponymy relations from texts (Hearst, 1992; Caraballo, 1999; Mann, 2002; Fleischman et al., 2003; Morin and Jacquemin, 2004; Shinzato and Torisawa, 2004; Etzioni et al., 2005; Pantel and Pennacchiotti, 2006; Sumida et al., 2006;<cite> Sumida and Torisawa, 2008)</cite> . | background |
e264c45391853fb008c838aa7ccca8_3 | On the other hand, <cite>Sumida and Torisawa (2008)</cite> have shown that you could easily obtain numerous hyponymy relations from Wikipedia; in particular, they have acquired more than 0.63 million hyponymy relations only from hierarchical layouts in the 2.2GB Japanese version of Wikipedia (e.g., Figure 1 shows a hierarchical structure of a Wikipedia article shown in Figure 2) . | background |
e264c45391853fb008c838aa7ccca8_4 | Although the above studies extracted hyponymy relations from the English version of Wikipedia, <cite>Sumida and Torisawa (2008)</cite> extracted hyponymy relations from definition sentences, category labels, and hierarchical structures in Wikipedia articles. | background |
e264c45391853fb008c838aa7ccca8_5 | Although the above studies extracted hyponymy relations from the English version of Wikipedia, <cite>Sumida and Torisawa (2008)</cite> extracted hyponymy relations from definition sentences, category labels, and hierarchical structures in Wikipedia articles. We thus focus on the hierarchical structures to acquire more hyponymy relations. | uses |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.