Automatic Support for Ontology Evaluation Review of Entailed Statements and Assertional Effects for OWL Ontologies

The typical goal of representing knowledge formally is to make knowledge accessible to machines. Hence, knowledge representation formalisms are chosen to accommodate well to the strengths of machines, often at the expense of being less easy to understand for humans. At the same time, formal representations are most often created or maintained by humans. This discrepancy provides the motivation for the central idea of this thesis, namely to facilitate evaluating the conceptual correctness of ontologies by means of reasoning services. For instance, when a logical axiom is added to an ontology, this addition may lead to logically consistent but conceptually wrong inferences. In the context of this thesis, specifically description logics with a view on those underlying the W3C recommended web ontology language OWL are considered as knowledge representation language in which ontologies are described. First, it has been investigated both by analysis and in practice how the review of automatically inferred statements can serve the purpose of evaluating an existing ontology. In a second venture, the usage of instance data to give ontology engineers concrete examples of the implications of general statements, i.e. terminological and role axioms in description logic languages, has been studied. Indeed, the effects of terminological and role axioms on instance data have not been studied in literature so far, although from a mathematical point of view the question of conservative extensions, i.e. the logical difference between two sets of terminological and role axioms, is very close. Thus it was necessary to provide a formal definition for effects that capture the intension of “knowledge lost or gained about data”, as well as find and prove a decision procedure. Both research ventures have led to concrete implementations of ontology evaluation functionalities within MoKi, a wiki-based ontology engineering environment. As a consequence, MoKi is now, to the best of the author’s knowledge, the only state-of-the art ontology engineering environment that explicitly supports conceptual ontology evaluation rather than ontology debugging in the sense of finding logical contradictions.