Reflecting the looser sense of modeling in earlier versions of ACT-R there have not been detailed models of how ACT-R 2.0 analogy was involved in the moment-by-moment execution of experiments. One of the lessons we have learned is the need to compare proposed mechanisms closely with data. This production compilation mechanism, being new and still somewhat tentative, does not have an abundance of correspondences to data. However, researchers are starting to develop models that involve production compilation in the detailed accounting of empirical results. Chapter 10, on learning by analogy to examples, will contain a fairly elaborate application. This subsection will describe a simpler application by Niels Taatgen (see Taatgen, 1997, for an earlier model) to a classic experimental paradigm in psychology, which is the contrast between reversal-shift and extra-dimensional-shift problems (Kendler & Kendler, 1959).
In these experiments subjects have to classify stimuli that vary on two dimensions such as size (large, small) and color (red, green). Initially, one dimension is relevant -- for instance, all red objects might be positive and all green objects might be negative. After the subject has reached the criterion of 10 consecutive correct classifications, the reinforcement scheme changes. In the reversal-shift condition, the values switch on the same dimension -- for instance, green will now be positive and red negative. In the extra-dimensional shift the other dimension now becomes relevant -- for instance, large objects might be positive and small objects negative. Young children typically find extra-dimensional shifts easier and older children and adults find reversal shifts easier.
This model was written by Niels Taatgen