Inconsistencies in Hybrid Knowledge Bases

The development of intelligent systems based on structured representation of knowledge that are capable of exhibiting human-like reasoning is a key goal of artificial intelligence in general and the field of knowledge representation research in particular. Advances in the latter have given rise to various formalisms motivated by different application domains. The most prominent families of such formalisms are ontologies in Description Logics (DLs) and nonmonotonic logic rules. While ontologies are well-suited for terminological modelling especially in the Semantic Web context, rules are widely accepted for common-sense reasoning and solving difficult search problems using declarative means. Many multi-purpose and task-specific applications, however, require a combination of ontologies and rules. The need for such a combination has led to the development of different approaches within Hybrid Knowledge Bases (KBs). Among several others, loose coupling has emerged as a prominent approach, in which an ontology and rules are treated separately, although, a level of interaction between them is permitted via a well-defined interface. Description Logic (DL)-programs are a representative of loosely coupled hybrid KBs. The bidirectional information exchange between rules and ontology realized in DL-programs makes them powerful systems that can be effectively used for solving advanced reasoning tasks on top of ontologies. However, this sophisticated information flow can also be a reason for inconsistencies which, as practice shows, occur in DL-programs. An inconsistent system does not yield any useful information, and as such is viewed as broken and in need of repair. Due to a possibly complex interaction between the rules and ontology, the repair problem is far from trivial and represents a significant challenge. Unfortunately, currently available engines for evaluation of DL-programs (e.g. dlvhex, DReW) suffer from the inability to handle inconsistencies with ease. This forms a major obstacle to wider acceptance of such systems. Therefore, the aim of this thesis is the development of a sophisticated framework for handling inconsistencies in DL-programs. The main results of the effort are briefly summarized below: • We offer a novel general repair semantics for DL-programs, and facilitate its practical applicability by applying formal methods in computer science to analyze computational complexity.