A deterministic dependency parser for Japanese

A Deterministic Dependency Parser for Japanese Ulrich Germann USC Information Sciences Institute, Marina del Rey, CA 90292 germann@isi.edu in: Proceedings of the MT Summit VII: MT in the Great Translation Era. Asia-Pacific Association for Machine Translation. Singapore, 1999, pp. 547-555; also available from http://www.isi.edu/˜germann Note: This is a slightly revised version of the original paper. Some typographical errors have been corrected. This version was produced on 5th November 1999. Abstract We present a rule-based, deterministic depen- dency parser for Japanese. It was implemented in C++, using object classes that reflect linguis- tic concepts and thus facilitate the transfer of linguistic intuitions into code. The parser first chunks morphemes into one-word phrases and then parses from the right to the left. The av- erage parsing accuracy is 83.6%. 1 Introduction Dependency grammars have recently received increased attention and interest from computational linguists. While there has always been active research in dependency gram- mars and related theories since the publicationof Tesni`ere’s ´El´ements de syntaxe structurale (1959), dependency-based approaches now also seem to become more popular for practical projects and applications such as annotating cor- pora (Kurohashi and Nagao, 1997; Hajic, 1998) and build- ing parsers (e.g., Kurohashi and Nagao, 1994; Tapanainen and J¨arvinen, 1997; Arnola, 1998; Haruno et al., 1998; Oflazer, 1999; Sekine et al., 1999; Uchimoto et al., 1999). This interest has also manifested itself in a recent work- shop on the processing of dependency-based grammars (Kahane and Polgu`ere, 1998). The objective of this paper is to present the results of a rule-based implementation of a deterministic parsing algo- rithm for Japanese. Although our parser does not perform quite as accurately as recent statistical parsers or Kuro- hashi and Nagao’s knowledge-based parser (see section 7), we believe that our experiment provides some interesting insights. On the one hand, our results offer support for some observations made by Sekine et al. (1999) regarding the feasibility of deterministic parsing. On the other hand, we would also like see them as a counterargument to the claim made by Uchimoto et al. (1999) and Sekine et al. (1999) that rule-based systems are practically not feasible. In the remainder of the paper, we first briefly present some core ideas of dependency grammars, and a summary of specific characteristics of Japanese that are relevant to the discussion. We then discuss and argue for parsing Ja- panese “backwards”, i.e., from the last element of the sen- tence to the first. Finally, we present and discuss the results of a rule-based implementation of this algorithm. 2 Dependency Grammars Dependency grammars assume direct, binary and asym- metric dependency relations between the elementary units (nuclei) of a sentence. For the sake of simplicity, these nu- clei can be assumed to more or less correspond to words in English, although J¨arvinen and Tapanainen (1998:4) point out that “there is no one-to-one correspondence between nuclei and orthographic words, but the nucleus consists of one or more, possibly discontinuous, words or parts of words. The segmentation belongs to the linearisation, which obeys language-specific rules.” …

Content truncated to 500 words