KQML+: An Extension of KQML in Order to Deal with Implicit Information and Social Relationships

Abstract 2 On one hand agent-based applications require a well defined language in order to implement agent interaction protocols - KQML or FIPA are proposed for that purpose. On the other hand, industry is actively working on "conversational interfaces" which will integrate an increasing number of features found in human conversations. We show that current agent communication languages cannot be used to develop applications involving such interfaces. The languages should be extended in a way which allows an agent to manipulate indirect speech acts and to use the various kinds of speech acts found in human conversations. For that purpose, and after a careful study of speech act theory, we propose an extension of the KQML language. Introduction Agent technology is used more and more frequently to help users achieve various tasks in diverse domains such as network management (Skarmeas and Clark 96), air-traffic control (Chu-Carrol and Carberry 95), telecommunication (Busuioc 96), or electronic commerce (Chavez and Maes 96). In most of the systems developed in those domains we can distinguish two levels of communication: agent/user interaction and interagent interactions. Agents interact with users in order to characterize their needs and to provide them with answers or solutions. Agents interact with each other in order to exchange various kinds of information. In order to successfully interact together, agents need to share a common language.To deal with inter-agent communication, industry tries to find a standard communication language such as KQML (Finin et al. 95) or FIPA (Fipa). When it comes to agent/user communication the current technology is based on Graphical User Interfaces (GUI). However, the industry is seriously considering applications involving "conversational interfaces": For example, “by providing