MAIDS - a Framework for the Development of Multi-Agent Intentional Dialogue Systems

Abstract

This paper introduces a framework for programming highly sophisticated multi-agent dialogue systems. The framework is based on a multi-part agent belief base consisting of three components: (i) the main component is an extension of an agent-oriented programming belief base for representing defeasible knowledge and, in partic- ular, argumentation schemes; (ii) an ontology component where existing OWL ontologies can be instantiated; and (iii) a theory of mind component where agents keep track of mental attitudes they ascribe to other agents. The paper formalises a structured argumentation-based dialogue game where agents can “digress” from the main dialogue into subdialogues to discuss ontological or theory of mind issues. We provide an example of a dialogue with an ontological digression involving humans and agents, including a chatbot that we developed to support bed allocation in a hospital; we also comment on the initial evaluation of that chatbot carried out by domain experts. That example is also used to show that our framework supports all features of recent desiderata for future dialogue systems.