8th European Conference on Speech Communication and Technology

Geneva, Switzerland
September 1-4, 2003


Cross Domain Chinese Speech Understanding and Answering Based on Named-Entity Extraction

Yun-Tien Lee, Shun-Chuan Chen, Lin-shan Lee

National Taiwan University, Taiwan

Chinese language is not alphabetic, with flexible wording structure and large number of domain-specific terms generated every day for each domain. In this paper, a new approach for cross-domain Chinese speech understanding and answering is proposed based on named-entity extraction. This approach includes two parts: a speech query recognition (SQR) part and a speech understanding and answering (SUA) part. The huge quantities of news documents retrieved from the Web are used to construct domain-specific lexicons and language models for SQR. The named-entity extraction is used to construct a domain-specific named-entity database for SUA. It is found that by combining domain classifiers and named-entity extraction, we can not only understand cross-domain queries, but also find answers in a specific domain.

Full Paper

Bibliographic reference.  Lee, Yun-Tien / Chen, Shun-Chuan / Lee, Lin-shan (2003): "Cross domain Chinese speech understanding and answering based on named-entity extraction", In EUROSPEECH-2003, 2821-2824.