Dialog Modeling in an Agent-based Framework

We describe an agent-based model of NL dialog, arguing that, by embedding a cooperation model in an agent architecture, the variability of dialog can be faced and linguistic phenomena can be modeled as general agent behavior. The coherence notion adopted in our model requires that the steps of a dialog contribute to the goals of the interactants; so, each turn is interpreted by matching its underlying (object level and metalevel) goals on the speakers' intentional models, to see if the turn contributes to their activities. This approach to dialog modeling allows us to interpret several apparently unrelated reactions to a turn. The architecture is based on a declarative representation of the knowledge about acting in order to satisfy a goal: metalevel operators describe recipes for plan formation and execution, while at the object level there are the domain and communicative actions.