Tzong-Han Tsai, National Taiwan University
Dialogue models fall into two categories. Structural approaches do not emphasize the contextual nature of communication. Plan-based approaches, on the other hand, attempt to recognize users’ goals and plans, and to produce corresponding effects. These two approaches are effective as long as users follow the preplanned scripts closely. However, when digression occurs, it is difficult for the structural approach to handle due to the rigid nature of finite state machines and grammar rules. And the plan-based approach wouldn’t incur the high cost of frequent re-planning and discourse context switching. In this paper, we shall deal with these issues using an ontology-based approach. We use a knowledge representation framework, InfoMap, as our ontology to identify events in utterances. By identifying events in the user’s high level query, InfoMap can create a dialogue strategy that guides the user to an appropriate deeper event, where the system can provide more concrete solutions. When digression occurs, the system either ignores it (if the digressive level is too high in the ontology) or moves to a related topic.