Time-Bounded Persistent Goals and their Role in Discourse

Cohen and Levesque have developed a theory of rational action that can be used to guide the actions of an intelligent agent. A fundamental building block of Cohen and Levesque's theory is their notion of a persistent goal. In this paper, we extend persistent goals by recasting them in terms of add/drop conditions, and introducing time intervals. These two extensions are then used to deene time-bounded persistent goals, which are useful for modelling the kinds of goals that arise in discourse. In particular, we illustrate how time-bounded persistent goals can be used to initiate discourse interruptions motivated by inconsistencies and plan ambiguities. We use examples to demonstrate the forms of dialogs which can be accounted for in terms of these goals. Finally, we discuss the potential for employing time-bounded persistent goals to manage the turn taking activity of intelligent agents, comparing with other work on turn taking in goal-oriented dialogs. Abstract Cohen and Levesque have developed a theory of rational action that can be used to guide the actions of an intelligent agent. A fundamental building block of Cohen and Levesque's theory is their notion of a persistent goal. In this paper, we extend persistent goals by recasting them in terms of add/drop conditions, and introducing time intervals. These two extensions are then used to deene time-bounded persistent goals, which are useful for modelling the kinds of goals that arise in discourse. In particular, we illustrate how time-bounded persistent goals can be used to initiate discourse interruptions motivated by inconsistencies and plan ambiguities. We use examples to demonstrate the forms of dialogs which can be accounted for in terms of these goals. Finally, we discuss the potential for employing time-bounded persistent goals to manage the turn taking activity of intelligent agents, comparing with other work on turn taking in goal-oriented dialogs.