Conversation as Action Under Uncertainty

Tim Paek
Stanford University
Stanford, CA 94305

Eric Horvitz
Microsoft Research
Redmond, Washington 98052

Author email:,


Conversations abound with uncertainties of various kinds. Treating conversation as inference and decision making under uncertainty, we propose a task independent, multimodal architecture for supporting robust continuous spoken dialog called Quartet. We introduce four interdependent levels of analysis, and describe representations, inference procedures, and decision strategies for managing uncertainties within and between the levels. We highlight the approach by reviewing interactions between a user and two spoken dialog systems developed using the Quartet architecture: Presenter, a prototype system for navigating Microsoft PowerPoint presentations, and the Bayesian Receptionist, a prototype system for dealing with tasks typically handled by front desk receptionists at the Microsoft corporate campus.

Click here to access postscript or pdf.

Keywords: Bayesian user modeling, common ground, joint activity, conversational systems, dialog systems, computational linguistics.

In: Proceedings of the Sixteenth Conference on Uncertainty in Artificial Intelligence, Stanford, CA, June 2000. San Francisco: Morgan Kaufmann.

Related Papers

E. Horvitz and T. Paek, A Computational Architecture for Conversation, Proceedings of the Seventh International Conference on User Modeling, Banff, Canada, June 1999. New York: Springer Wien, pp. 201-210.

E. Horvitz. Uncertainty, Action, and Interaction: In Pursuit of Mixed-Initiative Computing, Intelligent Systems, Sept./ October Issue, IEEE Computer Society.

T. Paek and E. Horvitz, Uncertainty, Utility, and Misunderstanding: A Decision-Theoretic Perspective on Grounding in Conversational Systems, AAAI Fall Symposium on Psychological Models of Communication in Collaborative Systems, Cape Cod, MA. November 5-7, 1999.

T. Paek, E. Horvitz, E. Ringger, Continuous Listening for Unconstrained Spoken Dialog, 6th International Conference on Spoken Language Processing (ICSLP 2000), Beijing, November 2000.

E. Horvitz and T. Paek, DeepListener: Harnessing Expected Utility to Guide Clarification Dialog in Spoken Language Systems, 6th International Conference on Spoken Language Processing (ICSLP 2000), Beijing, November 2000.