Extending WordNet with Hypernyms and Siblings Acquired from Wikipedia

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 874–882, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP Extending WordNet with Hypernyms and Siblings Acquired from Wikipedia Ichiro Yamada†‡ Jong-Hoon Oh† Chikara Hashimoto† Kentaro Torisawa† Jun’ichi Kazama† Stijn De Saeger† Takuya Kawada† †Information Analysis Laboratory, National Institute of Information and Communications Technology, 619-0289 Kyoto, Japan {rovellia,ch,torisawa,kazama,stijn,tkawada}@nict.go.jp ‡NHK Science & Technology Research Laboratories, 157-8510 Tokyo, Japan yamada.i-hy@nhk.or.jp Abstract This paper proposes a method for extend- ing WordNet with terms in Wikipedia. Our method identifies a WordNet synset by in- tegrating evidence derived from the struc- ture of an article in Wikipedia and distri- butional similarity of terms. Unlike previ- ous methods, utilizing the hypernym and siblings of the target term acquired from Wikipedia, the proposed method can deal with terms other than Wikipedia article titles and can work well even when re- liable distributional similarity of a target term is unavailable. Experiments show that the proposed method can identify synsets for 2,039,417 inputs at precision rate of 84%. Furthermore, it is estimated from the experimental results that there should be 328,572 terms among all the inputs whose synset our method can cor- rectly identify, while previous methods re- lying only on distributional similarity and lexico-syntactic patterns cannot. 1 Introduction As a comprehensive repository of word senses, WordNet (Fellbaum, 1998) has played an im- portant role in many natural-language-processing (NLP) tasks. Hand-crafted semantic knowledge, however, has low coverage of named entities and domain-specific knowledge. To address this is- sue, many researchers have proposed methods for extending WordNet by mapping new terms to a WordNet “synset.”1 In this paper, we propose a novel method that extends WordNet by integrating the Wikipedia ar- ticle structure and distributional similarity. With this method, an appropriate synset for a term in Wikipedia is identified by using the distributional 1A synset is a set of cognitive synonyms, each expressing a distinct concept. similarity of the term and that of the term’s hy- pernym and siblings, which are automatically ac- quired from Wikipedia. (Hereafter, trg is used for a target term for which we identify the appropriate synset, hyper is used for the hypernym of trg and sib is used for the sibling of trg.) The reason for using hyper and sib can be ex- plained as follows. In general, when an unknown term is encountered, its context helps in interpret- ing the term. Especially, if the unknown term’s hy- pernym and/or its semantically similar terms (its sibs) were somehow learned as context, it would often be possible to successfully guess its mean- ing. In WordNet expansion, trg may correspond to terms for which reliable distributional similarity is unavailable. In such cases, the distributional sim- ilarity of hyper and sib can help. Even when the distributional similarity of trg is available, that of hyper and sib can boost the performance of synset identification by providing additional sources of information. In this study, trg, hyper and sib are derived from hyponymy relation instances acquired from …

Content truncated to 500 words