Using Task-Based Dialogues for Multi-agent Collaborations

Abstract

New collaboration models for software agents can be made more flexible than pre-defined protocols. In pursuit of this, we have used the analysis of human interactions from Dialogue Structure Theory (DST) and devised an agent interaction approach which is based on Task-Based Dialogues (TBDs). This paper describes the following basic constructs of TBDs: a shared task hierarchy to represent agent intentions in the system, a dialogue model that enables agents to construct a meaningful sequence of messages, and a repository of dialogue states. The approach integrates these fundamental ingredients in software agents, and allows agents to implement different strategies in order to achieve their tasks. An example of TBDs shows how a sequence of messages can evolve that is beyond the capability of a single protocol, when the dialogues are driven by shared task specifications and not by a sequence of message types or performatives.