Using Preferences in Negotiations over Ontological Correspondences

The alignment of disparate ontology is of fundamental importance to support opportunistic communication within open agent systems, and as such has become an established and active research area. Traditional alignment mechanisms typically exploit the lexical and structural resemblance between the entities (concepts, properties and individuals) found within the ontologies, and as such often require agents to make their ontology available to an oracle (either one of the agents themselves or a third party). However, these approaches are used irrespectively of whether they are suitable given the intended models underlying the ontologies and hence their overlap, and usually require the disclosure of the full ontological model. This prevents the agents from strategically disclosing only part of their ontological model on the grounds of privacy or confidentiality issues. In this paper we present a dialogue based mechanism that allows two agents with limited or no prior knowledge of the other’s ontological model to determine whether it is possible to achieve some form of alignment between the two ontologies. We show how two agents, each possessing incomplete sets of private, heterogeneous (and typically ambiguous) correspondences, can derive an unambiguous alignment from a set of ambiguous, but mutually acceptable correspondences generated using an inquiry dialogue. The termination properties of the dialogue are formally proved, and the presentation and instantiation of an abstract preference-based argumentation is given. We demonstrate how ambiguity can be eliminated through the use of undercuts and rebuttals, given preference relations over the arguments.