Classification-Based Programming: A Deep Integration of Frames and Rules

摘要

Rules and frames are two knowledge representation schemes whose strengths and weaknesses are complementary to each other. Although several systems have attempted to integrate the two, few efforts have been made to incorporate the classification reasoning of the frame representations into the rule-based systems. To achieve a deep integration of the two schemes, we have developed and implemented a classification-based programming paradigm where the rules' LHS and the functionalities of their RHS are represented in the terminological space. The major processes of the architecture consist of a pattern matcher that is driven by semantic representations rather than by structural patterns, a rule base organizer that classifies rules and rule classes into a taxonomy, a conflict set manager that filters rules, and a rule interpreter that selects and executes rules. The architecture facilitates the representation of control knowledge by inferring the specificity of rules and by inheriting the control strategies from rule classes. The paradigm not only enhances the reasoning capabilities of rule-based systems, but also encourages explicit representation of various kinds of knowledge implicit in rules so that they can be shared by different domains and used for multiple purposes. (KT)