On the Normativity of Failing to Recall Valid Advice

Abstract

On the Normativity of Failing to Recall Valid Advice David C. Noelle ( NOELLE @ CNBC . CMU . EDU ) Center for the Neural Basis of Cognition; Carnegie Mellon University Pittsburgh, PA 15213 USA Abstract Instructed category learning tasks involve the acquisition of a categorization skill from two sources of information: explicit rules provided by a knowledgeable teacher and experience with a collection of labeled examples. Stud- ies of human performance on such tasks have shown that practice categorizing a collection of training examples can actually interfere with the proper application of ex- plicitly provided rules to novel items. In this paper, the normativity of such exemplar-based interference is as- sessed by confronting a model of optimal memory per- formance with such a task and comparing the “rational” model’s behavior with that exhibited by human learners. When augmented with a rehearsal mechanism, this opti- mal memory model is shown to match human respond- ing, producing exemplar-based interference by relying on memories of similar training set exemplars to categorize a novel item, in favor of recalling rule instructions. Introduction Contemporary studies of human category learning have tended to focus on the acquisition of general knowledge about a new concept exclusively from exposure to a col- lection of labeled examples. In common learning envi- ronments, however, students attempting to learn a cate- gorization skill are frequently provided with more than a set of training examples. In particular, learners are often explicitly instructed in the nature of a new category be- fore being presented with instances. They are provided with definitional sentences and explicit rules (e.g., “an equilateral triangle has at least two sides of the same length” or “bugs with six legs are insects”). Direct in- struction of this kind can rapidly provide a basic under- standing of a new category, while experience with exam- ples can further shape and refine that initial understand- ing (Klahr and Simon, 1999). While it is common for the process of explicit instruc- tion following and the process of induction from exam- ples to cooperate to produce quick and robust learning, there are situations in which these two learning processes actually compete. Specifically, practice at classifying a set of training examples can cause learners to violate ex- plicitly provided categorization rules when classifying novel items. Extensive experience with examples can lead learners to categorize novel instances according to similarity to training items, rather than according to cat- egorization rules communicated through explicit instruc- tion. Thus, novel items which are highly similar to train- ing examples from another category come to be misclas- sified as a result of practice. This exemplar-based interference effect, in which ex- perience with examples interferes with proper instruction following, was investigated by Allen and Brooks (1991), as well as others (Brooks et al., 1991; Neal et al., 1995; Noelle and Cottrell, 2000). Such interference in category learning is mirrored by similar difficulties in a wide va- riety of learning contexts, such as when students come to solve math or science problems by analogy to previ- ously seen problems, rather than by application of formal principles and techniques communicated through direct instruction. Learners appear to have a tendency to disre- gard perfectly valid explicit advice in favor of knowledge induced from experiences with examples. Exemplar-based interference might be seen as the re- sult of limitations of the cognitive system, such as im- perfect working memory efficacy (Noelle and Cottrell, 2000) or difficulties recalling and applying abstract, lin- guistically encoded, rules. There is another alterna- tive, however. It is possible that human learners neglect explicit instructions in favor of experienced exemplar- similarity information because the latter form of infor- mation tends to be more reliable in a wide variety of learning contexts. Exemplar-based interference may be the result of an essentially normative process of weight- ing sources of category information according to the pre- viously established utilities of those sources. There are many aspects of common learning situations which may encourage students to rely more heavily on examples than on explicit rules. Consider, for example, how the instructions provided by teachers are frequently approximate and heuristic. Advice is often implicitly limited to a particular range of circumstances, and there are often exceptions, even within this range, to explic- itly provided rules. Also, teachers are sometimes in er- ror. In short, human learners may have strong reasons to doubt the perfect accuracy of offered categorization rules. In comparison, exemplar similarity may be seen as a highly reliable indicator of category membership. Most categories, after all, involve clusters of similar ob- jects, suggesting that similarity might be the best tool for predicting the category labels of novel instances. Even if considerations of teacher reliability are ig- nored, there are other rational reasons for a learner to