Coordination Disambiguation without Any Similarities

Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 425–432 Manchester, August 2008 Coordination Disambiguation without Any Similarities Daisuke Kawahara National Institute of Information and Communications Technology, 3-5 Hikaridai Seika-cho, Soraku-gun, Kyoto, 619-0289, Japan dk@nict.go.jp Sadao Kurohashi Graduate School of Informatics, Kyoto University, Yoshida-Honmachi, Sakyo-ku, Kyoto, 606-8501, Japan kuro@i.kyoto-u.ac.jp Abstract The use of similarities has been one of the main approaches to resolve the ambigui- ties of coordinate structures. In this pa- per, we present an alternative method for coordination disambiguation, which does not use similarities. Our hypothesis is that coordinate structures are supported by surrounding dependency relations, and that such dependency relations rather yield similarity between conjuncts, which hu- mans feel. Based on this hypothesis, we built a Japanese fully-lexicalized genera- tive parser that includes coordination dis- ambiguation. Experimental results on web sentences indicated the effectiveness of our approach, and endorsed our hypothesis. 1 Introduction The interpretation of coordinate structures directly affects the meaning of the text. Addressing co- ordination ambiguities is fundamental to natu- ral language understanding. Previous studies on coordination disambiguation suggested that con- juncts in coordinate structures have syntactic or semantic similarities, and dealt with coordination ambiguities using (sub-)string matching, part-of- speech matching, semantic similarities, and so forth (Agarwal and Boggess, 1992). Semantic sim- ilarities are acquired from thesauri (Kurohashi and Nagao, 1994; Resnik, 1999) or distributional simi- larity (Chantree et al., 2005). c⃝ 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported li- cense (http://creativecommons.org/licenses/by-nc-sa/3.0/). Some rights reserved. For instance, consider the following example: (1) eat Caesar salad and Italian pasta The above methods detect the similarity between salad and pasta using a thesaurus or distributional similarity, and identify the coordinate structure that conjoins salad and pasta. They do not use the information of the word eat. On the other hand, this coordinate structure can be analyzed by using selectional preference of eat. Since eat is likely to have salad and pasta as its ob- jects, it is plausible that salad and pasta are coor- dinated. Such selectional preferences are thought to support the construction of coordinate structures and to yield similarity between conjuncts on the contrary. We present a method of coordination disam- biguation without using similarities. Coordinate structures are supported by their surrounding de- pendency relations that provide selectional prefer- ences. These relations implicitly work as similari- ties, and thus it is not necessary to use similarities explicitly. In this paper, we focus on Japanese. Coor- dination disambiguation is integrated in a fully- lexicalized generative dependency parser (Kawa- hara and Kurohashi, 2007). For the selectional preferences, we use lexical knowledge, such as case frames, which is extracted from a large raw corpus. The remainder of this paper is organized as fol- lows. Section 2 summarizes previous work related to coordination disambiguation and its integration into parsing. Section 3 briefly describes the back- ground of this study. Section 4 overviews our idea, and section 5 describes our model in detail. Sec- tion 6 is devoted to our experiments. Finally, sec- …

Content truncated to 500 words