Decision lists for determining adjective dependency in Japanese

Decision Lists for Determining Adjective Dependency in Japanese Taiichi Hashimoto†, Kosuke Nishidate†, Kiyoaki Shirai‡, Takenobu Tokunaga† and Hozumi Tanaka† †Department of Computer Science Tokyo Institute of Technology ‡School of Information Science, Japan Advanced Institute of Science and Technology †2-12-1 Ookayama, Meguro-ku, Tokyo 152, Japan ‡l-1 Asahidai, Tatsunokuchi, Ishikawa 923-1292, Japan taiichi@cl.cs.titech.ac.jp, nishidy@cl.cs.titech.ac.jp, kshirai@jaist.ac.jp, take@cl.cs.titech.ac.jp, tanaka@cl.cs.titech.ac.jp Abstract In Japanese constructions of the form [N1 no Adj N2], the adjective Adj modifies either N1 or N2. Determing the semantic dependencies of adjective in such phrase is an important task for machine translation. This paper describes a method for determining the adjective dependency in such constructions using decision lists, and inducing decision lists from training contexts with correct semantic dependencies and without. Based on evaluation, our method is able to determine adjective dependency with an precision of about 94%. We further analyze rules in the induced decision lists and examine effective features to determine the semantic dependencies of adjectives. Keywords Determining Adjective Dependency, Japanse, Decision List, Lexical Preferences, Exploration of Effective Feature 1 Introduction 1.1 Background In Japanese constructions of the form [N1 no Adj N2], Adj can modify either N1 or N2, where N1 and N2 are nouns, Adj is an adjective and “no” is an adnominal function word. The following are examples of this construction type: a) zo no nagai hana (elephant) (long) (trunk) “an elephant’s long trunk” b) kami no nagai shojo (hair) (long) (girl) “a girl with long hair” The adjective “nagai (long)” modifies the pro- ceeding noun “hana (trunk)” in noun phrase a), while the adjective “nagai (long)” modifies the preceding noun “kami (hair)” in noun phrase b). In terms of syntactic dependency, both ad- jectives modify the proceeding noun, because all words syntactically modify the succeeding word in Japanese. In terms of semantic dependency, however, adjectives can modify the preceding noun as in b). Obviously, determining the se- mantic dependencies of adjective is an impor- tant task for machine translation. Notice that the syntactic composition of a) and b) are the same in Japanese, but different in English. As the part-of-speech (POS hereafter) con- text of the two example noun phrases is the same, the determination of whether Adj seman- tically modifies N1 or N2 is strongly influenced by lexical items in the noun phrase. There- fore, lexical preferences are necessary to analyse semantic dependencies between adjectives and nouns. 1.2 Related Work Tanaka and Ogino (1980) observed that the se- mantic dependency of the adjective is highly correlated with the semantic dependency be- tween the two nouns in constructions of the form [N1 no Adj N2], and proposed a set of principles to determine the modifiee of the ad- jective (“Noun-Noun Dependency Principles”) as follows: In Japanese constructions of the form [N1 no Adj N2], Principle 1 If [N1 no N2] is felicitous,1 then Adj modi- fies N2. Principle 2 If [N2 no N1] is felicitous, then Adj modi- fies N1. Principle 3 If both [N1 no N2] and [N2 no N1] are fe- licitous, then Adj …

Content truncated to 500 words