An Empirical Semantics Approach to Reasoning About Communication Felix Fischer and Michael Rovatsos Department of Informatics Technical University of Munich 85748 Garching, Germany {rovatsos,fischerf}@cs.tum.edu Abstract. In the process of designing multiagent systems, it is often the case that some kind of specification of communication rules (in the form of protocols, ACL semantics, etc.) is available. and The question naturally arises how appropriate agents can be designed to can operate on such a specification. Moreover, if these multiagent systems are viewed as open systems, the problem is complicated even further by the fact that adherence to such a supposedly agreed specification cannot be ensured on the side of other agents. This paper presents an architecture for dealing with a very generic type of pre- specified communication patterns (which contain surface structure and logical constraint specifications) based on an empirical semantics model of communi- cation. This model allows for flexible adaptation to evolving communication se- mantics by combining existing expectations about the use of communication with empirical observation. This architecture is based on the InFFrA social reasoning framework and the con- cept of interaction frames. We show how interaction frames that represent classes of interaction situations can be used to conduct decision-theoretic reasoning about communication when interpreted using the empirical semantics approach. We introduce the abstract architecture, a formal model for its probabilistic se- mantics and present results of an experimental validation of our approach in a complex domain that illustrate its effectiveness. Keywords Agent interaction, agent communication languages, interaction protocols, multiagent learning, social reasoning 1 Introduction The process of designing agent communication languages, interaction protocols and conversation policies is primarily concerned with what goes on between rather than in- side agents. It is therefore only natural that ACL research has rarely attacked the prob- lem of how to build agents in accordance with a given set of semantic rules, protocol structures, etc. In the light of open multiagent systems, in which agents are not controlled by a central entity or pursue common goals and need not be benevolent, this design problem