A Flexible DL-based Architecture for Deductive Information Systems

Description Logics (DLs) are nowadays an accepted standard for decidable knowledge representation. DLs also provide the theoretical foundation for representing and reasoning with ontologies as well as for the Semantic Web. Thus, DL systems (e.g., Fact++, Pellet, RacerPro) and DL-based technology in general will play an increasingly important role for building the next generation of deductive, ontology-based information systems. The RacerPro description logic system [HM01a] implements the very expressive DL ALCQHIR+(D −), also known as SHIQ(D−) [HST99, HM01b]. RacerPro’s most prominent feature is the support for so-called ABoxes which allow for the representation of a “concrete state of the world” in terms of individuals and relationships. In the following, we assume familiarity with DLs as well as DL systems. In order to provide the information technology backbone for next generation information systems (IS), current DL systems still have a long way to go. Nowadays, building ontology-based IS with a DL system is still a non-trivial task: One the one hand, DLs somehow live in their own “realm” and are thus not really interoperable with the rest of “more conventional” IS infrastructure (e.g., existing relational databases). One the other hand, only recently issues such as persistency and powerful query languages (QLs) have been considered and incorporated into DL systems. DLs itself have their deficiencies as well and are thus not a panacea for arbitrary information representation and (ontology-based) retrieval tasks: Due to the complexity of the inference problems, scalability (in the average case) is not easy to achieve and nowadays only achievable for knowledge bases (KBs) which use simple DLs. Standard DLs are well suited for the representation of (and reasoning about) semi-structured (or even unknown/indeterminate) information, but things become more complicated if, say, n-ary relationships or special “non abstract” domains such as space are considered. Then, either non-standard DLs or complicated logical encodings are needed. For non-standard DLs, no working systems exist, and complicated logical encodings are likely to decrease the performance and complicate the information handling and maintenance. To get working systems on time today, we must thus agree upon pragmatic solutions. This also implies that existing technology (i.e., existing DL system such as RacerPro) must be exploited and possibly extended for the task of IS building. Due to the intellectual inherent complexity of the field, application and system studies on how to use DL system in real-world IS scenarios are thus of utmost importance in order to provide guidance.