A model of intentional communication: AIRBUS (Asymmetric Intention Recognition with Bayesian Updating of Signals)

Abstract

The rapid and fluent nature of human communicative interactions strongly suggests the existence of an online mechanism for intention recognition. In this paper, we discuss the requirements of such a system, drawing upon philosophical and empirical considerations, and propose a new mathematical model that addresses these requirements. Our model provides a way of integrating knowledge about both directions of the relationship between linguistic expressions and communicative intentions, through a rapid process of Bayesian update. It enables us to frame predictions about the process of intention recognition. Moreover, by equipping this model with a measure of entropy, we can also draw predictions about the circumstances under which utterance planning can start before the preceding utterance is completed, and about when explicit other-repair mechanisms must be invoked to resolve misalignment. We sketch an example of this model in operation and discuss its implications for a broader theory of communication.