Turn-Taking in Discourse and Its Application to the Design of Intelligent Agents

Abstract

ly, the turn-taking problem is about when to take a turn in an interaction with another intelligent agent; a turn-taker must weigh two sets of knowledge: 1) its own BDI's, and 2) the actions that indicate the relevant BDI's of the other conversant. A uniform way to handle these two sources of knowledge is to treat taking a turn as solving a single constraint satisfaction problem (CSP). The variables could be divided into two sets: the goal variables V 1 : : : Vn , representing the n turn-taking goals the agent can consider, and the pragmatic variables such as V volume , V intonation , V pause , V focus , etc. For V 1 : : : Vn , constraints can be based on the relative importance of the propositional content of the goal (e.g. how bad would it be if the turn was never taken?), logical consistency, linguistically motivated coherence constraints, such as following standard focus/topic pattern progressions. For example, in utterance (ii), two turn-taking goals are triggered, with the cons...