The Role of Cognitive Modeling in Communicative Intentions

A discourse planner for (task-oriented) dialogue must be able to make choices about whether relevant, but optional information (for example, the "satellites" in an RST-based planner) Should be communicated. We claim that effective text planners must explicitly model aspects of the Hearer's cognitive state, such as what the hearer is attending to and what inferences the hearer can draw, in ord@r to make these choices. We argue that a mere representation of the Hearer's knowledge is inadequate. We support this claim by (1) an analysis of naturally occurring dialogue, and (2) by simulating the generation of discourses in a situation in which we can vary the cognitive parameters of the hearer. Our results show that modeling cognitive state can lead to more effective discourses (measured with respect to a simple task). 1 Introduction Text planning is the task for a speaker (S) of deciding what information to :communicate to a hearer (H) and how and when to communicate it. Over the last few years a consensus has emerged that the text planning task should be formulated in terms of communicative goals or intentions [19, 25, 23, 16]. Consider, for example, the RST-based planners developed at ISI [13, 21, 14]. These planners use the discourse relations proposed by Rhetorical S'tructure Theory (RST) [18] as plan operators, by interpreting the requirements on the related segments as preconditions, and the resultant effect of the discourse relation as a postcondition in a traditional AI planning architecture. Two types (at least) of discourse relations have been identified in the literature. A SUBJECT-MATTER relation [18] or SEMANTIC relation [14] simply reflects a relation that exists independently in the world, such as causation. Each subject-matter relation can be seen as a rhetorical strategy for the linguistic realization of a