The social dimension of interactions in multiagent systems

Abstract

Communication is a fundamental issue when developping multi-agent systems (MAS). Most MAS frameworks adopted a simplified version of speech act theory in which agents’ speech acts arc interpreted in terms of communicative intentions and expressed using mental states. These formalizations are too simplistic when considering real conversations. First, they consider that agents perform speech acts in a direct way and that an agent readily accepts other agents’ speech acts, since the speech act is supposed to immediately transform her mental model. However, human locutors most often perform indirect speech acts and negotiate about the information conveyed by other agents' speech acts, often asking for explanations or justifications or even refusing it. Most MAS frameworks do not take into account the social relationships existing between locutors. In this paper we show that the notion of organizational role and social power is useful not only to reason on agent's commitments, but that it must be part of the formalization of speech acts. Assuming that locutors are always aware of the roles they play in the organizational setting in which the conversation takes place, we introduce the notions of role, decision power, social network and communicative conventions. We show that in order to explain how locutors understand their interlocutor's communicative intentions, even if they are expressed indirectly, it is necessary to model the implicit information conveyed by their utterances. We show how agents' roles and the socio-organizational context in which the conversation takes place influence the way speech acts are interpreted and implicit information determined. This framework enables us to introduce the social dimension in the specification of agents' speech acts.

This research is supported by the Natural Sciences and Engineering Research Council of Canada (grant OGP 05518) and FCAR.