Integrating derivational analogy into a general problem solving architecture∗

True expertise requires knowledge-intensive reasoning as a primary mental activity and a fallback to general problem solving in order to bridge gaps in the domain knowledge or to handle otherwise unexpected situations. This paper unites derivational analogy, a form of case-based reconstructive reasoning driven by rich episodic knowledge, with a general problem solving and planning architecture. The primary learning mechanism consists of extracting and organizing derivational traces of search-intensive problem solving episodes augmented with (and indexed by) explicit justification paths of success and failure conditions. This store of new episodic case knowledge drives the derivational analogy machinery in subsequent reasoning about problems that share substantial parts of the justification structure. The method is being implemented in the PRODIGY problem solving system and demonstrated in the domain of linear algebra. ∗This research was supported in part by ONR grants N00014-79-C-0661 and N0014-82-C-50767, DARPA contract number F33615-84-K-1520, and NASA contract NCC 2-463. The views and conclusions contained in this document are those of the authors alone and should not be interpreted as representing the official policies, expressed or implied, of the U.S. Government, the Defense Advanced Research Projects Agency or NASA.