Knowledge Modeling and Reusability in ExClaim

This paper presents E x Claim, a hybrid language for knowledge representation and reasoning. Originally developed as an operationalization language for the KADS knowledge based systems (KBS) development methodology, E x Claim has a meta-level architecture: it structures the knowledge on three levels, namely the domain, inference and task level. An extension of a description logic is used for implementing the domain level. The inference and task levels are general logic programs integrated with the domain level by means of upward and downward reflection rules which describe the automatic domain operations performed whenever arguments of inferences or tasks are accessed. Inferences and tasks support non-deterministic reasoning, which in turn requires a non-monotonic domain level. Description logics offer a set of inference services (some not available in other knowledge representation languages) which are extremely useful in knowledge modeling. Such inference services include domain-level deduction, semantic consistency verification and automatic classification of concepts. We argue that such validation and verification facilities are important in assisting a knowledge engineer in developing models. These models are reusable due to the layered architecture as well as to the possibility of writing generic inferences using a reified membership relation. 1 Introduction Knowledge based systems (KBS) are typically large and complex software systems aiming at solving difficult problems in knowledge-intensive domains. Knowledge engineering in general and KBS development in particular are notoriously difficult not only because of the sheer size of the problem description, but also because they typically involve complex ontologies, which are usually not easily representable in a single knowledge representation formalism. In order to assist the knowledge engineer in developing KBSs, a large number of KBS development tools have been built since the eighties. Two main tendencies were followed in the early years. On one hand, a great number of expert system "shells" were put forward. Systems like KEE, ART, Knowledge Craft, Nexpert Object etc. were successfully used in building a large number of expert systems. These "shells", however, had an important drawback: they used a given symbol-level representation (for instance a frame-based system augmented with rules, daemons, message passing, etc.), which is usually not appropriate for describing reusable knowledge-level models. An alternative approach to building KBS development tools was inspired by the traditional software engineering (SE) tools. SE tools are nevertheless inappropriate as KBS tools since the domain knowledge (the ontology) is much more complex in the case of a KBS than in …