Detecting Inconsistencies in Large First-Order Knowledge Bases

Large formalizations carry the risk of inconsistency, and hence may lead to instances of spurious reasoning. This paper describes a new approach and tool that automatically probes large first-order axiomatizations for inconsistency, by selecting subsets of the axioms centered on certain function and predicate symbols, and handling the subsets to a first-order theorem prover to test for unsatisfiability. The tool has been applied to several large axiomatizations, inconsistencies have been found, inconsistent cores extracted, and semi-automatic analysis of the inconsistent cores has helped to pinpoint the axioms that appear to be the underlying cause of inconsistency.