Dialogue Systems

One of our major research efforts is the development of a system that can engage in a collaborative dialogue with other agents. This requires that the system be able to recognize and reason about the other agents' plans and goals (see plan recognition research, model the agents' beliefs, preferences, etc. (see user modeling research, perform robust interpretation of new utterances, and generate effective responses. In our earlier work, we developed computational strategies for handling ill-formed input and for generating responses that are tailored to the user's beliefs, plans, and cognitive capability.

More recently, we have been investigating the modeling of collaborative expert consultation dialogues within a plan-based framework, with emphasis on understanding negotiation subdialogues and generating appropriate collaborative responses. Current models of discourse cannot handle negotiation subdialogues, since these systems are unable to recognize when new beliefs are adopted and when communicated propositions are questioned or rejected. We have investigated how to recognize when an agent is expressing doubt at a proposition by contending that a second proposition is true. In research with Lynn Lambert, we developed a plan-based strategy for recognizing complex discourse acts that takes into account 1)contextual knowledge which suggests expectations based on the structure of the preceding dialogue, 2)world knowledge which provides evidence for specific discourse acts, and 3)linguistic knowledge which suggests certain generic discourse acts, a speaker's beliefs, and the strength of those beliefs. This algorithm was used to identify the structure of negotiation subdialogues, including recognizing both implicit acceptance of communicated propositions and negotiation subdialogues embedded within other negotiation subdialogues. We then began exploring response generation in a collaborative environment. In research conducted with Jennifer Chu-Carroll, we developed strategies for engaging in negotiation subdialogues when the system disagrees with a proposal from the user and for initiating information-sharing subdialogues when the system has insufficient information to intelligently evaluate a user proposal.

Our work has also examined other dialogue problems. Responses to Yes-No questions often take the form of indirect answers. This research, conducted with Nancy Green, produced a computational model for interpreting and generating indirect answers. Its main features include a plan-based approach to implicature captured in a reversible architecture for generation and interpretation. To interpret an indirect answer, the system first derives a set of candidate discourse plans plausibly underlying R's response and then evaluates the relative plausibility of each. Generation of a discourse plan consists of content planning during which stimulus conditions are used to trigger speaker goals to include appropriate extra information, and plan pruning during which parts of the plan are identified that do not need to be stated explicitly. If the direct answer can be pruned, an indirect response results. Our approach overcomes the limitations of sentence-at-a-time processing for recognizing implicatures and can capture a variety of implicatures not handled before.

Relevant Publications (since 1990)

 Sandra Carberry, Lynn Lambert, and Leah Schroeder.  Toward Recognizing and Conveying an Attitude of Doubt Via Natural Language.  Applied Artificial Intelligence,  16(7), pp. 495-517, 2002.   

Schroeder, Leah and Sandra Carberry. Realizing Expressions of Doubt in Collaborative Dialogue. Proceedings of the 18th International Conference on Computational Linguistics, pp. 740-746, 2000.

Chu-Carroll, Jennifer and Sandra Carberry. Conflict Resolution in Collaborative Planning Dialogues. International Journal of Human-Computer Studies, 53(6) pp.~969-1015, 2000.

Carberry, Sandra and Lynn Lambert. A Process Model for Recognizing Communicative Acts and Modeling Negotiation Subdialogues. Computational Linguistics, 25(1), pp. 1-53, 1999.  postscript version

Green, Nancy and Sandra Carberry. A Computational Mechanism for Initiative in Answer Generation. User Modeling and User-Adapted Interaction, 9(1-2), pp. 93-132, 1999.

Green, Nancy and Sandra Carberry. Interpreting and Generating Indirect Answers. Computational Linguistics, 25(3), pp. 389-435, 1999.

Chu-Carroll, Jennifer and Sandra Carberry. Collaborative Response Generation in Planning Dialogues. Computational Linguistics 24(3), pp. 355-400, 1998.

Carberry, Sandra and Jennifer Chu-Carroll and Lynn Lambert. Modeling Intention: Issues for Spoken Language Dialogue Systems. Proceedings of the International Symposium on Spoken Dialogue, pp. 13-24, 1996.

Chu-Carroll, Jennifer and Sandra Carberry. Conflict Detection and Resolution in Collaborative Planning. Proceedings of the IJCAI-95 Workshop on Agent Theories, Architectures, and Languages, pp.~67-79, Aug. 1995.

Chu-Carroll, Jennifer and Sandra Carberry. Generating Information-Sharing Subdialogues in Expert-User Consultation. Proceedings of the 14th International Joint Conference on Artificial Intelligence, pp.~1243-1250, Aug., 1995.

Chu- Carroll, Jennifer and Sandra Carberry. Response Generation in Collaborative Negotiation. Proceedings of the 33rd Annual Meeting of the Association for Computational Linguistics, pp.~136-143, June 1995.

Chu-Carroll, Jennifer and Sandra Carberry. Communication for Conflict Resolution in Multi-Agent Collaborative Planning. Proceedings of the International Conference on Multi-Agent Systems, pp.~49-56, June 1995.

Chu-Carroll, Jennifer and Sandra Carberry. A Plan-Based Model for Response Generation in Collaborative Task-Oriented Dialogues. Proceedings of the Twelfth National Conference on Artificial Intelligence, pp.~799-805, 1994.

Green, Nancy and Sandra Carberry. A Hybrid Reasoning Model for Indirect Answers. Proceedings of the 32nd Annual Meeting of the Association for Computational Linguistics, pp.~58-65, 1994.

Green, Nancy and Sandra Carberry. Generating Indirect Answers to Yes-No Questions. Proceedings of the Seventh International Workshop on Natural Language Generation, pp.~189-198, 1994.

Lambert, Lynn and Sandra Carberry. Using Linguistic, Contextual, and World Knowledge in a Plan Recognition Model of Dialogue. Proceedings of the 14th International Conference on Computational Linguistics, pp. 310-316, 1992.

Lambert, Lynn and Sandra Carberry. Modeling Negotiation Subdialogues. Proceedings of the 30th Annual Meeting of the Association for Computational Linguistics, pp. 193-200, Newark, Delaware, 1992.

Green, Nancy and Sandra Carberry. Conversational Implicatures in Indirect Replies. Proceedings of the 30th Annual meeting of the Association for Computational Linguistics, ACL-92, pp. 64-71, 1992.

Mooney, David and Sandra Carberry and Kathleen McCoy. Capturing High-Level Structure of Naturally-Occurring Extended Explanations Using Bottom-up Strategies. Computational Intelligence Journal, pp. 334-356, 1991.

Lambett, Lynn and Sandra Carberry. A Tripartite Plan-Based Model of Dialogue. Proceedings of the 29th Annual Meeting of the ACL, pp. 47-54, 1991.

Mooney, David and Sandra Carberry and Kathleen McCoy. The Generation of High-Level Structure for Extended Explanations. Proceedings of the 13th International Conference on Computational Linguistics, pp. 276-281, 1990.

Sarner, Margaret and Sandra Carberry. Tailoring Definitions Using a Multifaceted User Model. Proceedings of the Eighth Canadian Conference on Artificial Intelligence, pp. 106-113, 1990.

Sandra Carberry, carberry@cis.udel.edu
Back to my homepage.