Generating Discourse Across Several User Models: Maximizing Belief While Avoiding Boredom and Overload

Abstract

In this paper we present a content planning system which takes into consideration a user's boredom and cognitive overload. Our system applies a constraint-based optimization mechanism which maximizes a probabilistic function of a user's beliefs, and uses a representation of boredom and overload as constraints that affect the possible values of this function. Further, we discuss two orthogonal policies for relaxing the parameters of the communication process when these constraints are violated: conveying less information or breaking up the material into smaller chunks. 1 Introduction It is generally accepted that to generate competent discourse, a speaker must take into consideration the beliefs and inferences of the addressee. In fact, several discourse planning systems rely on some sort of user model to generate appropriate descriptions, e.g., [ Paris, 1988; Cawsey, 1990; Zukerman and McConachy, 1994 ] . However, little attention has been paid to the possibility that the user model b...