Acquiring and adapting probabilistic models of agent conversation

Abstract

Communication in multiagent systems (MASs) is usually governed by agent communication languages (ACLs) and communication protocols carrying a clear cut semantics. With an increasing degree of openness, however, the need arises for more flexible models of communication that can handle the uncertainty associated with the fact that adherence to a supposedly agreed specification of possible conversations cannot be ensured on the side of other agents.As one example for such a model, interaction frames follow an empirical semantics view of communication, where meaning is defined in terms of expected consequences, and allow for a combination of existing expectations with empirical observation of how communication is used in practice.In this paper, we use methods from the fields of case-based reasoning, inductive logic programming and cluster analysis to devise a formal scheme for the acquisition and adaptation of interaction frames from actual conversations, enabling agents to autonomously (i.e. independent of users and system designers) create and maintain a concise model of the different classes of conversation in a MAS on the basis of an initial set of ACL and protocol specifications. This resembles the first rigorous attempt to solve this problem that is decisive for building truly autonomous agents.

Adaptiveness in Agent Communication: Application and Adaptation of Conversation Patterns

“…As we haven shown in [6], d * is indeed a metric on the set M * c of finite-length message sequences.…”

“…Since messages as defined above are essentially first-order objects, we could simply use a general purpose first-order distance like the one proposed in [24]. In [6], we have instead introduced a family of mappings on messages that are parametrised on two functions d s and D s and allow us to add a "semantic" flavour in the form of domain-specific knowledge. The most basic (and domain-independent) instance of this family is in fact a metric on messages (in particular, it satisfies the triangle inequality), and can easily be extended to message sequences.…”

Practical Strategic Reasoning and Adaptation in Rational Argument-Based Negotiation

“…As a result, A cannot attack the alternative means Y for achieving G, independent of the fact if she really holds G as a goal. 8 To allow for the exchange of multiple arguments, [19] further defines six frames for iterative IBN, corresponding to a successful proposal, challenged proposal, and rejected proposal (i.e. edges leading out of the proposal node of figure 2) and to successful challenge, successful justification, and successful attack (i.e.…”

“…Since agents may deviate from pre-defined frames, frame adaptation cannot be restricted to a mere update/extension of counter values and condition/substitution sets. To allow for the adaptation of frames from actual interaction experience, [8] extends the aforementioned distance metric on message sequences to frames and interprets frames as clusters in the space of possible conversations. Cluster validation techniques are then used to decide whether (and how) new observations should be merged into existing frames or whether they should be used to create a new frame.…”

“…In other words, we require agents to obey the overall communicative regime, which will in this case force them to concede to anything that follows from their assumed goal structure and beliefs. As a consequence, 8 For reasons of simplicity, we suffice with this, somewhat naive, approach for handling the complex notion of commitment in dialogue [25]. An elaborate account of commitment management is beyond the scope of this paper.…”