摘要: At first glance, explanation-based learning (EBL) and concept formation appear to address quite different learning problems. The standard defi-nition of explanation-based learning (Mitchell, Keller, & Kedar-Cabelli, 1986; DeJong & Mooney, 1986) presents it as a process of operational-izing an existing domain theory, ie, making a theory more usable and efficient for the purpose of classification and problem solving. The stan-dard definition of concept formation (Gennari, Langley, & Fisher, 1989) presents it as an incremental form of conceptual clustering (Michalski & Stepp, 1983), ie, the formation of a conceptual hierarchy from un-classified examples.The thesis of this paper is that, despite the apparent differences, EBL can be viewed as a type of theory-driven concept formation. Standard explanation-based methods can be seen as forming a new specialized concept from a single unclassified instance. This concept represents new knowledge that is not strictly within the deductive closure of the system’s existing domain theory. Moreover, like the knowledge acquired by concept formation systems, the specialized concept learned by EBL can be used to predict missing information. The remainder of this paper elaborates the view of EBL as concept formation. Section 2 provides some background material on explanationbased learning. Section 3 describes how standard explanation-based methods can be applied to the concept formation problem. Section 4