Integrating intention and convention to organize problem-solving dialogues
Elise H. Turner, Richard E. Cullingford · UPT. Syiah Kuala University Library (Syiah Kuala University) · 1989
The structure of problem solving dialogues reflects both the conventions of language and the intentions, or goals, of the conversants. For a computer system to be an effective participant in such a conversation, it must account for both of these influences in language processing. We present a representation for conversation that combines intention and convention, and a mechanism to control that representation which provides a planner for intentionally motivated conversation. The representation, the conversation MOP, has been suggested as the cognitive structure which captures discourse in humans (Kellermann, et al., 1989). We show how these structures can represent discourse structure in computer systems and how they can be used as plans for conversation. The control mechanism uses the conventions captured by the conversation MOPs to create a plan for the conversation that will achieve the system's goals in a way that seems natural to the user. The planner is flexible enough to react to the changing goals of the system as well as to many types of unanticipated user utterances. The system achieves flexibility because it dynamically alters the plan as new goals arrive and because it combines the effects of intention and convention to make decisions about plan execution while participating in the conversation. Because goals are fit into the conventional framework of the accepted conversation MOPs, a natural language interface can organize even an incoherent stream of goals from a problem solver to form a coherent dialogue. Because the system allows a goal's priority to influence the actual execution of the plan, it can override convention to respond to important goals in a timely manner. Because intention and convention are combined, the resulting dialogue is both easy to understand and responsive to the needs of the problem solvers. We present an implemented conversation controller, JUDIS, which embodies the ideas presented in this thesis.