Marco Schorlemmer, from the IIIA-CSIC, talks about ontology alignment.
Multi-agent interactions are based on illocutionary speech acts. But it is unfeasible: normative structures that regulate under which conditions agents can communicate and the meaning of the communication is needed. The illocutionary particles and agents roles are the interaction model, whereas the content language is the ontology. In the case of MAS, ontologies are previously engineered and agents coordinate or negotiate its use.
Share ontologies are useful for stable domains and closed communities because of the cost of maintaining them. Syntax, linguistic elements, synonyms in thesaurus, instances or hierarchies are all elements to be shared. But all that work is doe prior to interaction. But there’s equivalences that only can be detected while the communication process is being performed.
The idea is that the token of syntactic expression to be aligned is an interaction state transition where the type is the content of an illocution. So a minimal semantic agreement is needed previous to interactions and the matchings are been performed in different iterations while agents communicate. So we can transcend from a semantic web vision into a vision of a web of interactions.
Blogged with the Flock Browser