An approach to the automated acquisition of production rules

The major thrust of this research effort has been the development and unification of the necessary underlying theoretical foundations for an adequate approach to knowledge elicitation from repertory grid data. To achieve this aim, the research entailed developing a new approach to the interactive derivation of hypothesis from repertory grid data provided by a domain expert. These hypothesis are submitted to a quantitative logic of confirmation which provides a basis for the automatic generation of production rules for expert systems. The proposed logic of quantitative confirmation is consistent with personal construct psychology in that it produces epistemic probabilities by the measurement of overlap of a person's constructs. A theory of non-quantitative confirmation is offered as a necessary condition for the elaboration of an adequate probabilistic logic of confirmation. This approach incorporates the basic tenants of personal construct psychology directly into the logic as a basis for the determination of relevance. The concept of an $\alpha$-plane is introduced as a binary decomposition of repertory grid data that furnishes the necessary realization of construct extensions (or ranges of convenience) needed to determine the range of relevance of a particular generalization or hypothesis, thus providing the uniquely determined string of incidences required for Bundy's truth functional incidence calculus. An application of this work to the elicitation of expertise in the domain of radiology is given.