It makes sense … and reference: A bag-of-words model for meaning grounding and sharing in multi-agent systems

摘要

In a Semantic-Web-like multi-agent environment, ontology mismatch is inevitable: we can't realistically expect agents created at different times and places by different people to commit to one unchanging universal ontology. But what happens when agents attempt to interact? Can we benefit from ontology matching techniques to prevent communication breakdown? In this paper we present the Semantic Matcher, a system created to handle semantic heterogeneity (i.e. the use of different words for similar meanings), which is one of the most common mismatches across existing web-based ontologies. We claim that independently created agents with different ontologies are bound to miscommunicate because word meanings are private and not grounded in the real world. We propose the solution of achieving reference by means of sense, a common argument within Philosophy of Language. In our work, senses are simulated by bags of words, constructed for each lexical item known to a service-requesting agent, and for selected lexical items known to a service-providing agent in a given interaction. In the end we show that adopting this notion of 'sense' for words in formal ontologies allows for successful matching and creates the groundwork for a theory in which the vast amount of data available on the web can be used to facilitate agent communication by providing meaning for mismatched terms.