Tailoring inference-rich descriptions through making compromises between conflicting cooperation principles

Acting cooperatively in a communicative situation requires the computer agents to behave collaboratively. This activity is complicated by many sorts of conflicts that may arise in complex interaction environments. In order to capture essential forces driving dialogues in these environments, a number of formal models have been proposed, which focus on choosing among action and reaction types in comparably simple sequences. In this paper, we propose a method to enhance these models by the capability to produce longer dialogue contributions, which is done through tailoring variations in richly structured system actions and choosing among them, guided by quantitative and partially conflicting cooperativity principles based on the Gricean conversation maxims. We have elaborated this model in the context of an interactive, formal proof presentation system. Acting cooperatively for that system means collaborating with the human user by considerably manipulating a proof as produced by a theorem prover to obtain a human-adequate form, and to resolve conflicts between completeness of detail, length of the descriptions given and degree of explicitness required. Our approach complements models of cooperativity and collaboration that are grounded in basic principles of rationality by mediating between partially conflicting conversation goals in producing more complex dialogue contributions.