Ontological mediation: Finding translations across dialects by asking questions

Even when two participants in a communication share common language, domain of discourse, and communication protocol, miscommunications can occur when one agent's ontology, or formal conceptualization of the domain, differs from the other's. One of these differences is in terminology: the two agents use different terms to denote the same domain object. We call this the "dialect problem", because it occurs often in dialects of the same natural language. For example, Americans say "elevator", Britons say "lift".; This dissertation presents a method by which a computational agent, called an Ontological Mediator (OM) can solve the following problem: Given term T used by a speaker S, but unknown to a listener L, find another term T1 used by L that means the same to L as T does to S.; Our approach is to model the two participants and the OM as three distinct intelligent agents communicating with each other. By asking questions of both speaker and listener, the OM traverses and builds its own mental representation of parts of their ontologies. The OM then uses this representation to find ontological translation candidates, and rank them according to a formula involving frequency of appearance, lexical similarity, semantic distance, and ontological skew.; The question-asking approach is domain independent and cognitively valid in that the OM (a) starts with no knowledge of the domain, and (b) is not endowed with complete access to the agent's ontologies. The OM gains all its knowledge through the asking of questions. We describe a modular architecture for the OM that permits it to communicate with agents that use a variety of communication languages.; Through empirical testing with large ontologies such as WordNet and the Unified Medical Language System (UMLS) we show that our techniques are successful across multiple ontologies in multiple domains.