Modeling Conversation Policies using Permissions and Obligations

Abstract

Both conversation specifications and policies are required to facilitate effective agent communication. Specifications provide the order in which speech acts can occur in a meaningful conversation, whereas policies restrict the specifications that can be used in a certain conversation based on the sender, receiver, messages exchanged thus far, content, and other context. We propose that positive/negative permissions and obligations be used to model conversation specifications and policies. We also propose the use of ontologies to categorize speech acts such that high level policies can be defined without going into specifics of the speech acts. This approach is independent of the syntax and semantics of the communication language and can be used for different agent communication languages. Our policy based framework can help in agent communication in three ways (i) to filter inappropriate messages, (ii) to help an agent to decide which speech act to use next, and (iii) to prevent an agent from sending an inappropriate messages. Our work differs from most existing research on communication policies because it is not tightly coupled to any domain information such as the mental states of agents or specific communicative acts. Contributions of this work include (i) an extensible framework that is applicable to varied domain knowledge and different agent communication languages, and (ii) the declarative representation of conversation specifications and policies in terms of permitted and obligated speech acts.