Adaptation learning provides knowledge to enable another type of learning, learning to refine similarity criteria. A central role of similarity judgments in case-based reasoning is to determine which cases to apply to a new situation and how to adapt them to fit new circumstances. As Smyth & Keane [1995] observe, CBR systems often base similarity judgments on semantic similarity, but the real goal of ``similarity assessment'' in CBR is to determine adaptability: how easily an old case can be adapted to fit the requirements of a new situation. If new adaptation strategies are learned, static similarity criteria do not keep pace with new capabilities for performing adaptations. When adaptation learning makes it easier to apply particular cases, those cases should be judged more relevant to a new situation. Thus similarity assessment criteria should change as new adaptation knowledge is acquired.
DIAL improves on the traditional indexing techniques for retrieval by comparing the adaptability of response plans. DIAL uses learned adaptation cases to make judgments about the ease or difficulty of fixing problems in a new situation. The candidate case with the least expected adaptation effort is chosen. Thus similarity assessment can provide the adaptation component with a case requiring less work, improving the system's overall performance. We hypothesize a synergistic effect between similarity and adaptation learning, with learned improvement of one reducing effort and increasing quality of solution in the other. Further experiments are planned to support these hypotheses.