Abstract
Description Logic-based systems include extensive, complex reasoning components that may produce results that surprise users, yet these systems typically provide little or no explanation support. In this paper, we explore the explanation of subsumption reasoning in Description Logics from one perspective|explanation for knowledge engineers. We adopt a deductive framework for presenting the variety of inferences supported in these systems, and thereby achieve a simple, uniform specification of the notion of explanation. We then address the problem of overly-long and overly-detailed explanations by introducing the notions of atomic descriptions and atomic explanations. We also provide an implementation perspective by discussing the design space and presenting some desiderata for explanation modules. We have implemented our approach in the Classic knowledge representation system.