Chinese Dependency Parsing with Large Scale Automatically Constructed Case Structures

Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 1049–1056 Manchester, August 2008 Chinese Dependency Parsing with Large Scale Automatically Constructed Case Structures Kun Yu Graduate School of Infor- matics, Kyoto University, Japan kunyu@nlp.kuee.kyo to-u.ac.jp Daisuke Kawahara National Institute of Informa- tion and Communications Technology, Japan dk@nict.go.jp Sadao Kurohashi Graduate School of Infor- matics, Kyoto University, Japan kuro@i.kyoto- u.ac.jp Abstract This paper proposes an approach using large scale case structures, which are automatically constructed from both a small tagged corpus and a large raw cor- pus, to improve Chinese dependency parsing. The case structure proposed in this paper has two characteristics: (1) it relaxes the predicate of a case structure to be all types of words which behaves as a head; (2) it is not categorized by semantic roles but marked by the neighboring modifiers attached to a head. Experimen- tal results based on Penn Chinese Tree- bank show the proposed approach achieved 87.26% on unlabeled attach- ment score, which significantly outper- formed the baseline parser without using case structures. 1 Introduction Case structures (i.e. predicate-argument struc- tures) represent what arguments can be attached to a predicate, which are very useful to recognize the meaning of natural language text. Research- ers have applied case structures to Japanese syn- tactic analysis and improved parsing accuracy successfully (Kawahara and Kurohashi, 2006(a); Abekawa and Okumura, 2006). However, few works focused on using case structures in Chi- nese parsing. Wu (2003) proposed an approach to learn the relations between verbs and nouns and applied these relations to a Chinese parser. Han et al. (2004) presented a method to acquire © 2008. Licensed under the Creative Commons Attri- bution-Noncommercial-Share Alike 3.0 Unported license (http://creativecommons.org/licenses/by-nc- sa/3.0/). Some rights reserved. the sub-categorization of Chinese verbs and used them in a PCFG parser. Normally, case structures are categorized by semantic roles for verbs. For example, Kawahara and Kurohashi (2006(b)) constructed Japanese case structures which were marked by post posi- tions. Wu (2003) classified the Chinese verb- noun relations as ‘verb-object’ and ‘modifier- head’. In this paper, we propose a new type of Chinese case structure, which is different from those presented in previous work (Wu, 2003; Han et al., 2004; Kawahara and Kurohashi, 2006(a); Abekawa and Okumura, 2006) in two aspects: (1) It relaxes the predicate of a case structure to be all types of words which behaves as a head; (2) It is not categorized by semantic roles but marked by the neighboring modifiers at- tached to a head. The sibling modification infor- mation remembers the parsing history of a head node, which is useful to correct the parsing error such as a verb 看 (see) is modified by two nouns 电影 (film) and 简介 (introduction) as objects (see Figure 1). Figure 1. Dependency trees of an example sen- tence (I see the introduction of a film). We automatically construct large scale case structures from both a small tagged corpus and a large raw corpus. Then, we apply the large scale case structures to …

Content truncated to 500 words