This paper is concerned with deening a formal mapping from a terminological logic to an object data model which preserves as much of the semantics of the terminological logic as possible. To this end, the paper gives concise introductions into the terminological logic FRM and the object data model COCOON. It then brieey outlines a mapping from FRM class descriptions to COCOON types and classes and shows how the terminological inference of classiication is mapped to a set of equivalent COCOON queries. Since these queries can (mostly) be submitted as a whole to the underlying database system we can take full advantage of all the results on query optimisation, on providing eecient physical access structures, as well as on parallelisation that are available in the database area to make terminological inferences more eecient. This can play a crucial role in realising knowledge base systems capable of dealing with very large knowledge bases.

We propose an architecture for large-scale knowledge base systems based on database technologies and the three levels of semantic construct-frames, objects and relations. The intermediate object level retains the structural semantics of the frame level and is therefore beneecial in bridging the semantic gap between the frame and re-lational levels and enabling the use of semantic information in query optimisation. Speciically, we outline how this approach has been adopted in the hybrid knowledge base system, HYWIBAS.