Explanation-based reasoning

In artificial intelligence there has also been some effort to produce theories of novel concept understanding, although these approaches are not as language-related as the previously discussed work on conceptual combination. Much of the work focuses on the use of explanation-based reasoning.       For example, DeJong and Waltz (novel:dejong-waltz-1983) explained how a variety of novel language use could be understood by appealing to known concepts and explaining the differences between the known and the novel.     Similarly, Ram (read:ram1) produced the AQUA theory--this model was designed to be a more competent reader than previous systems. In particular, AQUA made extensive use of explanations; when confronted with an unfamiliar situation in a story, AQUA would attempt to perform explanation-based learning in order to continue the comprehension process. The process was guided by structures known as explanation patterns which described generic explanations which could be made. This ability allowed AQUA to do a simple form of what I am calling creative understanding. AQUA could also make use of this to skim material which it judged to be uninteresting. ISAAC's method of creative understanding is more well-defined, as well as making use of a new technique, base-constructive analogy.

  In general, the creative understanding theory within ISAAC is similar to these past approaches in that the theory proposes ways in which novel concepts may be understood in an explanation-driven fashion. The specifics are different, particularly in that I propose a complete cycle of creative understanding. I demonstrate how memory, analogy, and problem reformulation work together to produce the behavior which can, at times, be called creative. The DeJong and Waltz approach was more failure-driven in the sense that the understanding of novel concepts occurred only when normal understanding processes failed. The ISAAC approach is more integrative in that there is just one understanding process; at times it appears creative, at other times it appears mundane. Regarding the AQUA approach, the ISAAC theory proposes a different mechanism by which novel concepts are understood. And, there is a different control mechanism at work in the ISAAC approach; again, it is the conceptual level heuristics which guide the process.