Integrating multiple layers of concept information into n-gram modeling for spoken language understanding

The paper presents a novel approach, integrating multi-layer concept information into a 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. In an experiment using a real-world air-ticket information spoken dialogue system for Mandarin Chinese, a relative concept error rate reduction of 33% is achieved.