ABSTRACT
This report summarizes recent continuing work on the Copycat project, a stochastic computer model of fluid concepts, high-level perception, and analogy-making developed by Hofstadter and Mitchell (Mitchell, 1993). Copycat perceives analogies between short strings of letters, which can be thought of as representing abstract situations in an idealized microworld. An example of an analogy problem taken from this microworld might be "If abc changes to abd, how does srqp change in an analogous way?" An interesting feature of such problems is that there is no single "right" answer; rather, a range of answers is always possible for each problem. For the previous example, some possible answers might be srqo, trqp, srqd, drqp, or even abd. Of course, some answers are consistently judged by people to be better than others, for most analogy problems. Furthermore, for some problems, the answers judged to be the "best" are not at all the most "obvious" ones.
