Using knowledge on word-islands to improve the performance of spoken dialogue systems

摘要

This paper proposes a technique to improve the performance of spoken dialogue systems that not only consider knowledge about the semantic frames used by systems to understand the spoken language but also employ knowledge about the words in the system application domain that are used to complete frame slots. Using both knowledge sources, the technique considers specific word sequences to form what we call word-islands and which are then employed to create the language models and dictionary used by the system’s speech recogniser. Word-islands are easier to recognise than the words comprising the islands, which leads to improved spoken language understanding and system performance. Experiments have been conducted using two spoken dialogue systems which had previously been developed in our lab: one to provide fast food information and the other to provide bus travel information. Results show that the proposed technique improves the performance of both systems by improving speech recognition and spoken language understanding of sentence types that are difficult to process.