Hybrid Reasoning using Universal Attachment

Abstract

Hybrid representation frameworks provide a powerful basis for constructing intelligent systems. Universal attachment is a domain-independent mechanism for integrating diverse representation and reasoning methods into hybrid frameworks that contain a subsystem based on deduction over logical formulas. Although based on the same principles as previous attachment methods, universal attachment provides a much broader range of connections between general-purpose deduction and specialized representation and reasoning techniques. This paper defines a formal inference rule of universal attachment and discusses the properties of soundness, completeness and correctness for this rule. The relationship between universal attachment and other integration techniques is explored. Finally, policies based on experimentation with an implemented universal attachment system are presented that lend guidance in exploiting the expanded representational and inferential capabilities that hybrid systems provide.