Knowledge engineering for SALES : a limited legal expert system

Expert systems are computer programs which try to model the thinking process of an expert. This thesis explores how a knowledge engineer organizes a loosely ordered knowledge base so that it can be used for expert system conclusions and explanations.

SALES is a limited legal expert system based on Article 2 of the Uniform Commercial Code. It is concerned with black letter law or statute law. It does not deal with interpretation, case law or analogy. It is intended to provide a user with a preliminary understanding of his legal rights regarding Article 2. It is not a final authority.

SALES is a knowledge base for DECIDE, an inference engine or processor, which is written in Prolog and was improved to handle SALES. SALES uses the decision expression methodology of DECIDE to order the knowledge and the interaction with the user. Hierarchical identities are used to replace general questions/decisions with detailed expressions. These identities are the mechanism in the DECIDE system whereby knowledge is clustered. Each knowledge cluster maybe arranged as a separate file.

The decision expression structure makes SALES deterministic. All its paths must be explicitly coded when the system is built. But this does provide for an explanation hierarchy and concluding "leaps." Concluding "leaps" are made by avoiding the expansion of a question with a detailed decision expression. The system provides an explanation of why the current question is being asked and also provides a summary of the interaction so far. Explanations are given in terms of the expert system's main question or the question which invoked the detailed decision expression.

SALES is a simple system which relies heavily on the knowledge engineer to order its knowledge and its file movement. The integrity of the expert system is dependent on the knowledge engineer.