Construction of an Idiom Corpus and its Application to Idiom Identification based on WSD Incorporating Idiom-Specific Features

Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing, pages 992–1001, Honolulu, October 2008. c⃝2008 Association for Computational Linguistics Construction of an Idiom Corpus and its Application to Idiom Identification based on WSD incorporating Idiom-Specific Features Chikara Hashimoto Graduate School of Science and Engineering Yamagata University Yonezawa, Yamagata, 992-8510, JAPAN ch@yz.yamagata-u.ac.jp Daisuke Kawahara National Institute of Information and Communications Technology Sorakugun, Kyoto, 619-0289, JAPAN dk@nict.go.jp Abstract Some phrases can be interpreted either id- iomatically (figuratively) or literally in con- text, and the precise identification of idioms is indispensable for full-fledged natural lan- guage processing (NLP). To this end, we have constructed an idiom corpus for Japanese. This paper reports on the corpus and the re- sults of an idiom identification experiment us- ing the corpus. The corpus targets 146 am- biguous idioms, and consists of 102,846 sen- tences, each of which is annotated with a lit- eral/idiom label. For idiom identification, we targeted 90 out of the 146 idioms and adopted a word sense disambiguation (WSD) method using both common WSD features and idiom- specific features. The corpus and the experi- ment are the largest of their kind, as far as we know. As a result, we found that a standard supervised WSD method works well for the idiom identification and achieved an accuracy of 89.25% and 88.86% with/without idiom- specific features and that the most effective idiom-specific feature is the one involving the adjacency of idiom constituents. 1 Introduction Some phrases like kick the bucket are ambiguous with regard to whether they carry literal or idiomatic meaning in a certain context. This ambiguity needs to be resolved in the same manner as ambiguous words that have been dealt with in the WSD liter- ature. We term the resolution of the literal/idiomatic ambiguity as idiom identification, hereafter. Idiom identification is classified into two kinds; one is for idiom types and the other is for idiom to- kens. With the former, phrases that can be inter- preted as idioms are found in text corpora, typically for compiling idiom dictionaries. On the other hand, the latter helps identify a phrase in context as a true idiom or a phrase that should be interpreted literally (a literal phrase, henceforth). In this paper, we deal with the latter, i.e., idiom token identification. Despite the recent enthusiasm for multiword ex- pressions (MWEs) (Gr´egoire et al., 2007; Gr´egoire et al., 2008), the idiom token identification is in an early phase of its development. Given that many NLP tasks like machine translation or parsing have been developed as a result of the availability of lan- guage resources, idiom token identification should also be developed when adequate idiom resources are provided. To this end, we have constructed a Japanese idiom corpus. We have also conducted an idiom identification experiment using the corpus that we hope will be a good reference point for fu- ture studies on the task. We drew on a standard WSD framework with machine learning exploiting both features commonly used in the …

Content truncated to 500 words