Abstract

Existing dialogue systems typically use only one dialogue strategy: finite-state based, frame-based or agent-based. This paper describes research done on a dialogue manager (DM) that uses all three dialogue strategies. This DM can determine the most suitable dialogue strategy to be used according to circumstances, and change to a different strategy whenever required. The research puts emphasis on how such a multistrategy DM improves the naturalness of human-computer interaction. Determining best use of different dialogue strategies, rather than tailoring dialogue management to a particular business domain, motivates the research.