Flexible and Personalizable Mixed-Initiative Dialogue Systems

This paper describes our vision for a future time when end users of mixed-initiative spoken dialogue systems will be able to dynamically configure the system to suit their personalized goals. We argue that spoken dialogue systems will only become a common utility in society once they can be reconfigured, essentially instantaneously, to support a new working vocabulary within a new domain or subdomain. For example, if a user is interested in restaurants in Seattle, the system would go off-line to gather information from resources such as the Web, and would infer from that knowledge an appropriate working vocabulary, language models, and dialogue control mechanism for a subsequent spoken conversation on this topic. In addition to painting this vision, the paper also discusses our recent research efforts directed towards the technology development necessary to realize this larger goal.