A topic-based framework for rational interaction
Andreas Herzig, Dominique Longin
Abstract : Cooperative dialogue is one of the most important challenges of computer science. The background of this work are the dialogue systems developed by France Telecom R& D as instantiations of its ARTIMIS (Sadek, 1999; Sadek et al., 1997; Sadek et al., 1996) generic rational agent technology. Such systems allow to manage real-time cooperative dialogues in natural language. Our framework is what we call an “intentional approach” of dialogue (Cohen & Levesque, 1990a; Sadek, 1991; Sadek, 1992; Rao & Georgeff, 1992). This approach is based on theories of Intentionality (Searle, 1983; Bratman, 1987). Within these theories, an agent is represented by its “mental state”, which is a set of informations. This set contains the different mental attitudes about the world the agent has: beliefs, goals, intentions... These theories are at the base of what is called “BDI-architectures” (for belief, desire and intention) in the literature. Intentional approaches are defined within twofaced formal theories : rational balance and rational interaction. The first theory describes, through properties of mental attitudes and action, the relationships that must be maintained as true (the relationships between the different mental attitudes of an agent firstly, and between these mental attitudes, plans and actions secondly). The second theory characterizes the inter-agent relationship within a multiagent environment (communication, cooperation, ...). Agents built on these twofaced theories are called rational agents. In this paper, we focus on the belief change process, viz. the ability to take into account the dynamics of the world.