Explanation-based learning with incomplete theories: a three-step approach

Abstract A weakness of EBL is its inability to explain when the theory is incomplete. This paper presents a three-step approach to deal with incomplete theories based on abduction, analogical reasoning and case-based reasoning. Abduction allows us to explain an example in the context of an incomplete theory, without modifying the theory. The simple variant of analogical reasoning used here not only provides an explanation, but also extends the domain theory. We show that the overhead imposed by our method on EBL is acceptable (at most polynomial). The approach presented in this paper was implemented in a system called LISE (Learning in Software Engineering). LISE is a system which translates informal and nonoperational user requirements into formal and operational software specifications.