Integration of Dependency Analysis with Semantic Analysis Referring to the Context

Proceedings of PACLIC 19, the 19th Asia-Pacific Conference on Language, Information and Computation. Integration of Dependency Analysis with Semantic Analysis Referring to the Context Yuki Ikegaya* Yasuhiro Noguchi* Satoru Kogure† Tatsuhiro Konishi† Makoto Kondo† Asoh Hideki‡ Akira Takagi§ Yukihiro Itoh† *Graduate School of Science and Engineering, Shizuoka University {cs7005@cs.,noguchi@}inf.shizuoka.ac.jp †Faculty of Informatics, Shizuoka University {kogure@cs,mkondo@ia,konishi@cs,itoh@ cs}.inf.shizuoka.ac.jp ‡Information Technology Research Institute, AIST h.asoh@aist.go.jp §CSK Corporation/ Information Technology Research Institute, AIST akira-takagi@aist.go.jp Abstract This paper describes how to perform syntactic parsing and semantic analysis using the contextual information as a language understanding component in a dialog system. Although syntactic parsing and semantic analysis are often conducted independently of each other, correct parsing of a sentence often requires the semantic information on the input and/or the contextual information prior to the input. We therefore merge syntactic parsing with semantic analysis, which enables syntactic parsing to take advantage of the semantic content of an input and its contextual information. To use contextual information, the semantic representation of an input should have a comparable form to the semantic content of the preceding context. Accordingly, we employ a framework for semantic representations that achieves such comparison. We take dialogs of hotel search and reservation for example, and demonstrate the effectiveness of the proposed method. The experimental results confirm that the proposed system achieves high accuracy in parsing and generation of semantic representations. 1. Introduction When we have a talk, we presuppose the situation and particular context.1 The situation/context enables us to understand utterances without talking in great detail. The situation/context also makes it possible to convey a wide variety of contents with a certain linguistic expression. This means that a language understanding component in a dialog system should not interpret a given sentence alone; rather, it should interpret a sentence using the context information. In our previous work, we have developed a Japanese dialog system for hotel search and reservation (Noguchi et al. 2002). The system accepts free input from the keyboard. We are planning to build an audio input module into the system. Audio input raises the importance of the context information because the effective use of the context information may circumvent possible recognition errors. In this paper, we focus on how a language understanding component can perform syntactic parsing and semantic analysis using the contextual information. It may be possible to integrate syntactic parsing with semantic analysis by taking advantage of LFG (Kaplan et al. 1982) and HPSG (Pollard et al. 1994). However, they do not consider the use of the context information. In this paper, we propose the method to integrate dependency analysis with semantic analysis referring to the context. Recently, many researches have adopted a statistical approach toward parsing, and it has been shown that the statistical approach is effective in obtaining correct dependency structure. There are 1 The term “context” primarily means association of utterances. In particular, it refers to logical relationship among words and/or sentences. In this paper, the term is used to indicate the interpretive results of input/output sentences …

Content truncated to 500 words