Transformation Rules for Knowledge-Based Pattern Matching

Many AI tasks require determining whether two knowledge representations encode the same knowledge. For example, rule-based classification requires matching rule antecedents with working memory; information retrieval requires matching queries with documents; and some knowledge-acquisition tasks require matching new information with already encoded knowledge to expand upon and debug both of them. Solving this matching problem is hard because representations may encode the same content but differ substantially in form. Previous approaches to this problem have used either syntactic measures, such as graph edit distance, or semantic knowledge to determine the “distance” between two representations. Although semantic approaches outperform syntactic ones, previous research has focused primarily on the use of taxonomic knowledge. As a result, mismatches between representations go largely unaddressed. In this paper, we investigate whether semantic approaches can be augmented with additional non-taxonomic knowledge to further improve matching. To test this hypothesis, we built a matcher that uses both taxonomic and non-taxonomic knowledge in the form of transformation rules and applied it to the task of critiquing military Courses of Action. We compared our matcher’s performance to both a syntactic and a semantic matcher applied to the same task. From this study, we found the results show that using additional non-taxonomic knowledge further improves matching.