Language Resource Addition Strategies for Raw Text Parsing

Language Resource Addition Strategies for Raw Text Parsing Atsushi Ushiku1, Tetsuro Sasada2, Shinsuke Mori3 1Graduate School of Informatics, Kyoto University 2,3Academic Center for Computing and Media Studies, Kyoto University 1,2,3Yoshidahonmachi, Sakyo-ku, Kyoto, Japan Abstract We focus on the improvement of accuracy of raw text parsing, from the viewpoint of language resource addition. In Japanese, the raw text parsing is divided into three steps: word segmentation, part-of-speech tagging, and dependency parsing. We investigate the contribution of language resource addition in each of three steps to the improvement in accuracy for two domain corpora. The experimental results show that this improvement depends on the target domain. For example, when we handle well-written texts of limited vocabulary, white paper, an effective language resource is a word-POS pair sequence corpus for the parsing accuracy. So we conclude that it is important to check out the characteristics of the target domain and to choose a suitable language resource addition strategy for the parsing accuracy improvement. Keywords: language resource addition, part-of-speech tagging, dependency parsing 1. Introduction For languages without clear word boundary natural lan- guage processing (NLP) , such as word segmentation (WS), part-of-speech (POS) tagging (PT), and dependency pars- ing (DP) is essential technology for applications such as machine translation or information extraction (Matsumoto et al., 2000; Kudo et al., 2004; Kurohashi et al., 1994; Neu- big et al., 2011; Kurohashi and Nagao, 1994; Mori et al., 2000; Kudo and Matsumoto, 2002; Flannery et al., 2012). WS and PT are now sufficiently accurate and widely used in various applications. DP is, however, not so widely used as WS and PT because of its insufficient performance. Re- cently there are many attempts at using sentence structure such as tree-based machine translation, language genera- tion, etc. Thus the accuracy of a dependency parser is get- ting more and more important. In this paper, we discuss DP from raw sentences in Japanese. Although Japanese does not have clear word boundary, we take words as the DP unit similar to NLP in other languages such as English (Buchholz and Marsi, 2006; McDonald and Pereira, 2006; McDonald et al., 2005; Nivre and Nilsson, 2005; Koo and Collins, 2010; Oflazer, 1999; McDonald et al., 2006; Sagae and Lavie, 2006; Eis- ner, 1996; Nivre and Scholz, 2004), and we adopt a three- step pipeline method consisting of a word segmenter, a POS tagger, and a dependency parser (Figure 1). Although some researchers have proposed joint approaches (Hatori et al., 2012), the pipeline method has advantages that it can utilize various types of language resources easily and it is easy to analyze the effects of each step. So we can say the pipeline method is more realistic than joint approaches. Actually for WS and PT available training data are usually larger than those for DP especially in domain adaptation situations. In this paper we assume that we want accurate DP results of raw texts in a new target domain and the method is lan- guage resource addition to each step because it is …

Content truncated to 500 words