Reasoning in Inconsistent Stratified Knowledge Bases

This paper proposes a discussion of inconsistency-tolerant consequence relations in prioritized knowledge bases. These inference techniques extend methods for reasoning from inconsistent, non-stratified, knowledge bases to the case where priorities between formulas are available. Priorities between formulas are handled in the framework of possibility theory and allow for the use of pieces of information having various levels of confidence. A comparative analysis of several approaches is carried out, namely, the possibilistic inference and its extensions, three inference methods based on a selection of maximal consistent subsets of formulas, and two inference methods based on arguments.