In this SSP talk I will present my thesis proposal. This proposal
addresses the problem of ontology heterogeneity in open multi-agent
systems. In open MAS, it is often an impossible requirement that
all
agents share an ontology or that all the combination of mappings
between different ontologies are foreseen. My proposal starts
from the
pragmatic assumption that understanding between agents is needed only
to perform the interactions required for reaching some common goal. I
propose that only small portions of ontologies, relevant to the context
of the interactions are mapped. The mapping framework exploits
the
experience gained in previous interactions to quickly classify the
dialogue into a context and then uses the context to filter out the
unrelated parts of the ontologies. The mapping itself is modelled as
decision-making under uncertainty.