Prioritized Base Debugging in Description Logics

The problem investigated is the identification within an input knowledge base of axioms which should be preferably discarded (or amended) in order to restore consistency, coherence, or get rid of undesired consequences. Most existing strategies for this task in Description Logics rely on conflicts, either computing all minimal conflicts beforehand, or generating conflicts on demand, using diagnosis. The article studies how prioritized base revision can be effectively applied in the former case. The first main contribution is the observation that for each axiom appearing in a minimal conflict, two bases can be obtained for a negligible cost, representing what part of the input knowledge must be preserved if this axiom is discarded or retained respectively, and which may serve as a basis to obtain a semantically motivated preference relation over these axioms. The second main contributions is an algorithm which, assuming this preference relation is known, selects some of the maximal consistent/coherent subsets of the input knowledge base accordingly, without the need to compute all of of them.