Improving Rule-Based Systems Through Case-Based Reasoning

A novel architecture is presented for combining rule-based and case-based reasoning. The central idea is to apply the rules to a target problem to get a rst approximation to the answer; but if the problem is judged to be compellingly similar to a known exception of the rules in any aspect of its behavior, then that aspect is modelled after the exception rather than the rules. The architecture is implemented for the full-scale task of pronouncing surnames. Preliminary results suggest that the system performs almost as well as the best commercial systems. However, of more interest than the absolute performance of the system is the result that this performance was better than what could have been achieved with the rules alone. This illustrates the capacity of the architecture to improve on the rule-based system it starts with. The results also demonstrate a beneecial interaction in the system, in that improving the rules speeds up the case-based component.