Both the specific similarity of test items to study items and the grammaticality of test items were found to be major determinants of performance under task conditions common in the literature. Results bearing on the issue of how item-specific effects are coordinated with knowledge pooled across items are: (a) Better item memory resulted in smaller rather than larger effects of specific similarity on judgments of grammaticality, suggesting that items can be too well differentiated to support transfer to new items, (b) Variation in the effect of specific similarity did not result in compensatory variation in grammaticality, suggesting that any scheme that tightly links the effects of the two variables is insufficient, (c) Differential reliance on the 2 knowledge resources was not under good instructional control, which poses a problem for accounts that use functional task analyses to coordinate functionally different memories.

No abstract

The effect of practice on the operative form of a rule was investigated by giving subjects an easy, perfectly predictive classification rule, followed by training in applying that rule to a set of practice items. On a subsequent transfer test, the accuracy and speed of classifying new items was strongly affected by similarity to previously seen items, suggesting that the effect of practice was not simply to automatize the rule. The effect occurred with pictorial, easily integrated stimuli, but not with lists of verbally stated features. Subjects generally did not have insight into the role of previous items in their performance. This dependence on prior episodes may be frequent in ecologically common conditions and is of special interest when the categorization rule becomes uncertain, as when a rule has only heuristic value.When dealing with a newly learned rule, one is often aware of relying on previous episodes of applying the rule. The specific materials or problems to which the rule was originally applied seem to have some privilege, particularly when the rule initially seems quite abstract. Understanding a difficult rule, or at least understanding how to apply it, seems to occur in terms of previous concrete applications. This impression is corroborated by Ross in his interesting work on remindings in problem solving (Ross, 1984(Ross, , 1987(Ross, , 1989 Ross & Kennedy, 1990). He demonstrated that performance while learning a word processing program or learning to solve simple probability problems is influenced by specific analogies with previously encountered problems.But what happens to these prior episodes of problem solving when the rule is not difficult, or when one has had sufficient practice to make it no longer seem difficult? Introspectively, the prior examples seem to disappear from active processing, a suggestion reflected in a wide variety of theories in cognitive psychology. In this article we argue that under common conditions, prior episodes retain an important role in helping apply the rule to new material, well after clear awareness of the prior episodes has disappeared. We make an ecological argument for the conditions under which episode-based knowledge has continued value. Abstractive GeneralizationAn intuitively plausible view of the changes in processing that occur during practice with a rule is what we will refer to as abstractive generalization. With increasing practice, processing comes under the control of fast, general, and automatic procedures that concentrate solely on the relevant information given in the stimulus display. With practice, a rule that was explicitly known becomes automatized, consuming a decreasThis research was carried out with the support of the Natural Sciences and Engineering Research Council of Canada, through a graduate scholarship to Scott W. Allen and a grant to Lee R. Brooks.

Non-analytic reasoning is a central component of diagnostic expertise at all levels. Clinical teaching should recognise the centrality of this process, and aim to both enhance the process through the learning of multiple examples and to supplement the process with analytical de-biasing strategies.

Preparation of this article was supported in part by operating grants to Lee R. Brooks and John R. Vokey from the Natural Sciences and Engineering Research Council of Canada. We thank Lorna Moore for her invaluable assistance in collecting the data and Kris Woeppel and Mathew Davidson for their contributions to earlier phases of the project.