So far, much effort in agent research concerning agent interaction has focused on the semantic and pragmatic foundations of different agent communication languages (ACLs) based on speech act theory[23, 24, 11]. Researchers in the ARPA Knowledge Sharing Effort have proposed agent communication languages (ACLs) as the means to allow the exchange of knowledge among software agents in order to make easier their interoperation . Generally, an ACL is composed of three main elements: an open-ended vocabulary appropriate to a common application area, an inner language (KIF-Knowledge Interchange Format) to encode the information content communicated among agents, and an outer language (KQML-Knowledge Query and Manipulation Language) to express the intentions of agents. Apart from KQML, there are other ACLs in use such as FIPA ACL, KQML Lite, etc. However this research is currently broadening from the specification of individual utterances to include the characterization of goal-directed conversations for which agents will use ACLs. Thus new works in speech act research, exemplified by efforts such as KAoS, Dooley Graphs, COOL and MAGMA, attempt at representing and reasoning about the relationships within and among conversations, or groups of utterances. A number of formalisms have been proposed for the modeling of conversations: FSMs[4, 32], Dooley graphs, Petri Nets, etc. Our approach proposes a new model based on PDTs that allows to store the context of ongoing conversations, and, in contrast with other approaches, that provides a mapping from specification to implementation. Moreover, as a distinctive feature from other approaches, we provide our model with a detailed analysis that studies the properties that conversation protocols must exhibit in order to ensure protocol compatibility, and therefore the soundness of agent conversations.
Although there is a large number of software tools for developing agents, not many of them happen to provide support for the specification of conversation protocols. AgentTalk, COOL, JAFMAS , Agentis, Jackal and InfoSleuth, do offer conversation constructs. JAFMAS, for instance, provides a generic methodology for developing speech-act based multi-agent systems using coordination constructs similar to COOL. In addition to this, as far as our knowledge goes, none of them offers dynamically and incrementally specifiable conversation protocols except for InfoSleuth. We attempt to make headway in this matter with respect to other agent building tools by introducing interagents, autonomous software agents that permit both the dynamic and incremental definition of conversation protocols by both the agent engineer and the owner.
We have chosen the conceptualization of interagents as autonomous software agents instead of an agent's built-in conversation layer as proposed in other agent architectures because of the need to separate the agents' logics from the agents' interactions --such separation has proven to be valuable in the development of a particular type of agent-based systems, namely electronic institutions such as FM. Interagents --like KQML facilitators-- are inspired by the efficient secretary metaphor already introduced in the Actors model of concurrent computation . Nonetheless, interagents --unlike KQML facilitators-- offer the conversational level required by agents to cooperate in non-trivial ways.