Propose-and-revise modeled in Karl

This paper reports an evaluation study for the specification of an average sized expert system for configuring elevator systems using the language KARL (Knowledge Acquisition and Representation Language). Two results have been gained in this study: (i) a formal model of the used problem-solving method (PSM) Propose-and-Revise has been developed and (ii) the adequacy of the language KARL for specifying such systems has been evaluated. KARL is based on a strong conceptual model: the KARL model of expertise, which represents different aspects of the model at different layers. It clearly separates domain specific knowledge fim probleni-solving-specific knowledge which allows to reuse both parts independenty from the other. KARL provides language primitives on a high level of abstraction, independent of implementation issues. KARL is a formal language which allows to represent knowledge unambiguously. KARL is an executable language which allows to validate the resulting model by testing and debugging. It turned out that KARL is well-suited for such specification issues. It also turned out that due to a flexible connection between domain knowledge and problem solving knowledge provided by KARL both different kinds of knowledge may be specified nearly independenty from the other which supports their reuse. This study gave us various insights into the adequacy of the language KARL for representing the knowledge on an abstract level. In spite of the encouraging results we gained this study also revealed some deficiencies of the language KARL which are currently eliminated for a future version of KARL.