Integrating Knowledge Sources for the Specification of a Task-Oriented Dialogue System

We show how the specification of a dialogue system can be divided into domain-dependent and domain-independent parts. We demonstrate how comparisons of actual representations in the dialogue history can help to infer hierarchical dialogue structure. The principles guiding the inference can be expressed in domain independent rules. Using typed feature structures as the only representation formalism for domain-dependent knowledge we retain the simplicity of frame-based systems in terms of gathering necessary information to fulfil a task. On the other hand, being able to easily integrate a type hierarchy into the representations and describing the systems behavior in clauses quantifying over feature structures in the dialogue history, we not only achieve a compact specification of the system's behavior, but also a type discipline that helps to detect errors in specification before system deployment. The described implementation is a first step towards the implementation of domain-independent task-oriented dialogue processing systems