The Role of Common Ontology in Achieving Sharable, Reusable Knowledge Bases

Although AI research and commercial system development depend on bodies of formally represented knowledge that are expensive and diicult to construct, current knowledge base design does not support the accumulation or reuse of such knowledge. This paper presents a strategy for building libraries of sharable, reusable knowledge in which common ontologies play a central role as a knowledge coupling construct. Ontologies are deened as coherent sets of representational terms, together with textual and formal deenitions, that embody a set of representational design choices. Problems in the design of sharable ontologies are identiied. Research and development in AI is impeded by an inability to share and reuse knowledge bases. The expense of building serious knowledge bases and the lack of means to exchange them with colleagues makes it diicult to generate, evaluate, and build on research results which depend on \domain theories" or \background knowledge." Similarly, commercial systems development is hampered by the lack of interoperability among tools based on diierent knowledge representations, and the lack of mechanisms for accumulating libraries of reusable knowledge. Today's knowledge systems are isolated monoliths characterized by high internal coupling (e.g., among ground domain facts, procedures, terminology, axioms, and idiosyncratic ways of modelling the world) and a lack of external coupling interfaces that would enable the developer to reuse software tools and knowledge bases as modular components. The only practical way to \share" or \reuse" an existing knowledge base is to adopt the entire representation and programming environment of the existing KB.