Learning by Failing to Explain

Explanation-based Generalization depends on having an explanation on which to base generalization. Thus, a system with an incomplete or intractable explanatory mechanism will not be able to generalize some examples. It is not necessary, in those cases, to give up and resort to purely empirical generalization methods, because the system may already know almost everything it needs to explain the precedent. Learning by Failing to Explain is a method which exploits current knowledge to prune complex precedents and rules, isolating their mysterious parts. This paper describes two techniques for Learning by Failing to Explain: Precedent Analysis, partial analysis of a precedent or rule to isolate the mysterious new technique(s) it embodies; and Rule Re-analysis, re-analyzing old rules in terms of new rules to obtain a more general set.