Computer-Generated Dialog as a Resource-Bounded Activity

Research Summaries

Resource bounds abound in conversation. Conversants must work under time pressure, they have working memory limitations, and they often have incomplete or uncertain information about the state of the conversation. Fortunately, conversation can tolerate errors and misconceptions: if you do not understand something the speaker says, then you can initiate a clarification dialog, or, if you recognize an error on the speaker’s part, you can interrupt them to propose a repair. In this abstract, we discuss some past work on resource-bounds in conversation and describe our own work on modelling conversation as a dynamic constraint satisfaction problem.