Making Role Limiting Shells More Flexible

Although expert system shells based on role-limiting methods are very powerful in supporting expert system development by model-guided and graphical knowledge acquisition, it is a legitimate and often mentioned criticism that such shells usually are hard-wired and therefore hardly flexible if a given problem does not totally match the predefined method. In this paper we analyse the inner structure of role-limiting methods of two shells implemented within our group, and break them down into smaller mechanisms in order to enable new configurations of role-limiting methods and corresponding shells. Method configuration is supported both by offering a library of problem-solving specific mechanisms of how a subtask can be solved, and by allowing the introduction of new mechanisms and subtasks and their combination with the existing ones within the predefined framework. We demonstrate our approach both with assignment problems and with classification tasks. The gained flexibility substantially increases the applicability of role-limiting methods and — by allowing the reuse of mechanisms and user-interface — drastically reduces the costs of new method development.