Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

Fine Keyword Clustering Using a Thesaurus and Example Sentences for Speech Translation

Yumi Wakita (1,3), Kenji Matsui (1), Yoshinori Sagisaka (2,3)

(1) Advanced Technology Research Laboratories, Matsushita Electric Industrial Co., Ltd., Kyoto, Japan
(2) ATR Spoken Language Translation Research Laboratories, Kyoto, Japan
(2) ATR Spoken Language Translation Research Laboratories, Kyoto, Japan
(3) Graduate School of Science and Technology, Kobe University, Kobe, Japan

For robust speech translation, we propose a new language translation method in which speech recognition results are mapped to example sentences using keywords. In this method, the keyword clustering is used to cope with recognition errors and the wide variety of words that do not appear in the training corpus. Initial classes defined using only thesaurus are redefined by using the dependency between the keywords in limited number of example sentences. The effectiveness of our keyword clustering method is confirmed through example sentence search experiments. These experiments were done using keyword sets of (a) different sentences including keywords not in the example sentences and (b) recognition results those sentences in which recognition errors were obtained. Compared with the search method which uses keyword sets defined by using only a thesaurus, our proposed method offered improved search error rates.

Full Paper

Bibliographic reference.  Wakita, Yumi / Matsui, Kenji / Sagisaka, Yoshinori (2000): "Fine keyword clustering using a thesaurus and example sentences for speech translation", In ICSLP-2000, vol.3, 390-393.