ISCA Archive Interspeech 2004
ISCA Archive Interspeech 2004

Integrating layer concept inform ation into n-gram modeling for spoken language understanding

Nick Jui-Chang Wang, Jia-Lin Shen, Ching-Ho Tsai

The paper presents a novel approach, integrating layer concept information into the trigram language model, to improve the understanding accuracy for spoken dialogue systems. With this approach, both the recognition accuracy and out-of-grammar problem can be largely improved. The concept error rate is therefore reduced. In the experiment using a real-world air-ticket information spoken dialogue system for Mandarin Chinese, a relative concept error rate reduction of 30% is achieved.