Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

Automatic Lexicon Generation and Dialogue Modeling for Spontaneous Speech

Hiroki Mori, Hideki Kasuya

Faculty of Engineering, Utsunomiya University, Japan

This paper describes novel framework for dialogue modeling based on a superword model, a superset of word n-gram. This has a remarkable advantage, because only transcribed text is needed to obtain the model, and no word dictionary is needed. In this paper, it is shown that the expressions specific to dialogue speech are extracted automatically from the transcriptions of spoken dialogue corpora by applying the acquisition method of the superword model. From experimental results based on a Japanese spoken dialogue database which consists of 42 sessions from 6 different tasks, it has been found that the proposed language modeling method has an ability to acquire task-independent lexical entry characteristic of dialogue speech, and many lexical entries are found to be relevant to discourse structures.


Full Paper

Bibliographic reference.  Mori, Hiroki / Kasuya, Hideki (2000): "Automatic lexicon generation and dialogue modeling for spontaneous speech", In ICSLP-2000, vol.1, 577-580.