Erscheinungsort / Ort: Karlsruhe
Referent(en): Rudi Studer, Peter H. Schmitt, Pascal Hitzler
Formal models of domain-specific knowledge abound in science and technology. It is desirable that such models can be managed, exchanged, and interpreted in computer systems, and the term “ontology” was coined to refer to the respective modelling artefacts. A prominent application field for ontologies is the Semantic Web where the Web Ontology Language OWL is the predominant modelling language. The formal semantics of OWL is largely based on the description logic (DL) family of knowledge representation formalisms that are well-suited for terminological modelling. Rule-based knowledge representation languages, in contrast, have a stronger focus on modelling relationships between instances. Both perspectives are relevant in applications but the combination of rules and DLs turns out to be difficult, since vital computational properties such as decidability are lost easily. The subject of this work is to advance the development of hybrid DL rule languages based on first-order Horn rules. Reasoning for SWRL – the combination of DLs with (first-order) datalog – is known to be undecidable, and we identify DL Rules as a novel class of decidable SWRL fragments that is closely related to DLs. New decidability results for DLs with role constructors allow us to include simple role conjunction and concept products into DL Rules. DL Rules are further extended with DL-safe variables to arrive at DL+safe rules. The latter properly generalise DL Rules and the known approaches of DL-safe rules and role-safe recursive CARIN. This leads to expressive DL rule languages with high computational complexities, motivating the study of more restricted languages. We introduce Horn DLs to generalise the known DL Horn-SHIQ, and show that many of these DLs exhibit high reasoning complexities in spite of their low data complexity. DLP has been proposed as a logic in the “expressive intersection” of DLs and datalog. We question the meaning of this description, and develop formal design criteria for DLP that let us specify the largest datalog-expressible fragment of description logics. Combining these insights, we arrive at a new tractable DL rule language ELP which extends both DLP and the light-weight DL EL++, although the union of these languages is intractable. ELP incorporates DL Rules and a certain form of DL+safe rules, and we present a reasoning procedure based on a direct reduction to datalog that preserves the structure of rules. This also lets us derive a new datalogbased inferencing procedure for the DL SROEL(⊓s, ×) which extends EL++. This work advances the understanding of the relationship of rules and description logics, leading to concrete new knowledge representation formalisms of practical relevance. DL+safe rules constitute one of the broadest classes of decidable SWRL fragments known today. ELP provides a tractable DL rule language that generalises the novel light-weight ontology languages OWL RL and OWL EL as standardised by W3C, and that has been adopted as the basis for the WSML-DL v2.0 dialect of the Web Service Modeling Language. Our work also suggests new rule-based implementation methods for supporting these languages based on a single inferencing algorithm.