SBIR-STTR Award

W propose a conceptual framework and a collection of tools to develop a general knowledge repository. This repository is designed to provide facilities to encode a large collection of existing and to-be-developed knowledge bases, and to support general retrieval across these knowledge bases as required to inform the development of new decision support (or other knowledge based) systems. The fundamental technological commitments are as follows. We employ an architecture that provides very strong support for the encapsulation of knowledge bases as well as any number of tools to operate on those knowledge bases. We employ a base representation language such as loom to serve as the target language for reformulation of existing knowledge bases. Within this base language, all prior knowledge will be classified along problem-solving dimensions the goals, constraints, and contextual factors associated with using thatKnowledge to solve a given problem. We identify a critical kind of knowledge, called problem solving skill knowledge, which defines when and how data is used to solve problems. This knowledge is plan-like and is critical to the processes that support knowledge reuse. Problem-solving skill knowledge is classified as mentioned, and intelligent browsing facilities are used to match this knowledge to user requirements thereby defining a retrieval paradigm . Knowledge repositories are significantly stronger than knowledge bases, since they encode problem-solving skill knowledge, not just a collection of facts. They will be useful in all applications of knowledge-based technology.

Knowledge Bases (KBs) and thus their otologies are rarely general encodings of a domain; rather, specialized by the tasks that the KB will serve. Knowledge is selected for inclusion and systematized to make the ontology efficient for the tasks. This fact has very strong implications for KB reuse. Reuse depends on the relationship between the original task and the one calling for reuse; it is not solely a matter of superficial matches of ontological elements. During Phase I we created the Task Modeling theory, which conceptualizes ontology design as a decision-making activity. An ontology is the product of objectives selecting over modeling alternatives. Objectives derive from the task, the organization calling for the KB, and pure design criteria. A record of this design decision-raking information is needed for effective reuse. During Phase II we propose to formalize the Task Modeling theory normatively, using the language of multi-attribute utility theory and its associated decision theory. We also propose to develop a theory of the process of reuse, based on a linguistic theory of language evolution and a cognitive theory of adaptation. We also propose to design and implement tools to test our theory experimentally. Anticipated Benefits/Potential Applications - Existing research and development in support of reuse does not address task-specificity of otologies or the process of KB reuse. Our research fills this gap and thus contributes in an important way to the development of a successful infrastructure for KB maintenance and reuse, both by the government and the commercial sector. It will be particularly relevant when the KB supports a design activity.