Domain-Related Focus-Shifting Constraints in Dialogues with Knowledge Based Systems

Dialogues with knowledge based systems often are unsatisfactory to users, because they usually are exclusively oriented at the problem solving process. In contrast to that, naturally occuring domain-specific dialogues are structured in respect to a domain model. We extracted some heuristics which are a base for determining the discourse structure from a dialogue between an expert and novices. This dialogue occurs in a problem solving situation which can be characterised by the problem solving method hypothesis-and-test. The heuristics ensure a thematic coherence by adding information about the domain structure when shifting the focus of dialogue. We show, how the KADS modeling approach for knowledge-based systems could be extended to provide discourse plans that ensure the generation of coherent and hence understandable natural language dialogues. Starting out from a distribution of tasks between system and user we apply several rules which formalise the heuristics in a way that they might be used as a part of the knowledge engineering process.