The project's long-range goal is to create a unified architecture for interactive discourse, incorporating both interpretation and generation. The immediate goal is to develop an environment for simulating and testing dialog in cooperative planning environments.
The project is premised on the use of simulated discourse to collect data about discourse, and to test specific hypotheses about discourse interpretation and generation. Note that a discourse simulator can be tested (a) by generating entirely simulated dialogues in which discourse strategies and conditions are varied, or (b) by testing the simulations' interactions with human subjects. And (c) the simulation architecture can also be used as an environment for gathering data concerning conversations between human subjects in the task domain. The project will use all these forms of testing.
The basis of our approach to interpretation is the abductive model of Jerry Hobbs and his associates at SRI. (Jerry Hobbs, Mark Stickel, Douglas Appelt, and Paul Martin, "Interpretation as abduction," Artificial Intelligence 63, 1993, pp. 69-142.) The basis of our approach to generation is the plan-based approach of Johanna Moore. (See Johanna D. Moore, Participating in Explanatory Dialogues, MIT Press, 1995.) The theoretical work of the project consists largely of unifying these approaches. The discourse simulation will be based on this unification.
Developed an "open-domain system" approach to finite-state natural language processing for use in applying the basic research of the project.