Third International Conference on Spoken Language Processing (ICSLP 94)

Yokohama, Japan
September 18-22, 1994

An Example-Based Approach to Semantic Information Extraction from Japanese Spontaneous Speech

Otoya Shirotsuka, Ken'Ya Murakami

Laboratory for Information Technology, NTT DATA Communications Systems Corporation, Kawasaki, Kanagawa, Japan

Dealing with the linguistic phenomena of spontaneous speech by the existing rule-based approach requires the preparation of complex analysis rules, which takes a great deal of effort. This paper describes a new method of extracting semantic information extraction from Japanese spontaneous speech by an example-based approach (EBA). Compared to the rule-based approach, EBA is robust and requires little effort for knowledge acquisition and its formation. In experimental evaluations of a semantic information extractor based on EBA, transcriptions of one hundred spontaneous dialogues are used as an example corpus and a testing corpus. The best performances of the extractor are 81.6% for precision rate and 62.2% for coverage rate in semantic feature extraction. The results suggest that our method is robust against unknown words and ill-formed sentences, and the extractor proved that EBA can be used as an effective tool for extracting semantic information from spontaneous speech.

Full Paper

Bibliographic reference.  Shirotsuka, Otoya / Murakami, Ken'Ya (1994): "An example-based approach to semantic information extraction from Japanese spontaneous speech", In ICSLP-1994, 91-94.