Axiom Pinpointing Using an Assumption-Based Truth Maintenance System

Abstract

The problem of axiom pinpointing [1, 22], that is, finding the minimal set of axioms responsible for an unwanted consequence, is an important problem in ontology debugging. One approach to identifying the axioms responsible for an unwanted consequence is to trace dependencies between inferences leading to the consequence. Several authors have proposed truth maintenance systems as a means of keeping track of dependencies or inferences in ontologies, e.g., [21, 5, 7]. In this paper we show that truth maintenance systems can also be used for axiom pinpointing. More specifically, we present a system which returns all minimal sets of axioms responsible for the derivation of inconsistency in an unfoldable ALC ontology. Following Sirin et al [24], we refer to these sets of axioms as explanations. Our approach involves using a modified Assumption-Based Truth Maintenance System (ATMS) [11] to trace inferential dependencies between formulas and compute the minimal sets of ontology axioms responsible for a contradiction. The main technical contribution of the paper is extending the ATMS to deal with disjunctions. We generalise the notion of an ATMS environment (a set of axioms from which a formula is derivable) to include the non-deterministic choices required for the derivation of the formula. We show that this extended ATMS (which we call the D-ATMS), combined with a tableau reasoner, produces correct, complete and minimal explanations for a contradiction in an unfoldableALC ontology. We have developed a prototype implementation of our approach which we call AOD. Preliminary results of experiments comparing AOD, MUPSter and the Pellet explanation service are encouraging, and suggest that AOD can outperform MUPSter and Pellet on both synthetic and real-world ontologies.

Extracted Key Phrases

3 Figures and Tables

Cite this paper

@inproceedings{Nguyen2012AxiomPU, title={Axiom Pinpointing Using an Assumption-Based Truth Maintenance System}, author={Hai H. Nguyen and Natasha Alechina and Brian Logan}, booktitle={Description Logics}, year={2012} }