Effects of class complexity, class frequency, and preexperimental bias on rule learning.

Abstract

The difficulty of four binary conceptual rules (conjunctive, disjunctive, conditional, and biconditional) was assessed using the rule-learning paradigm in a variety of stimulus populations. Decreasing the number of unique stimuli representing each truth-table class had a facilitative effect on learning the conditional and biconditional rules. Manipulating the relative frequency of truth-table classes changed the difficulty of particular truth-table classes but had no effect on overall measures of performance. It was suggested that either increasing the occurrence of a difficult stimulus or simplifying that stimulus facilitates learning. Furthermore, changes in relative frequency affect the stability of preexperimental biases, and changes in complexity affect choice of strategy. Based on the finding that rule learning was a function of preexperimental bias, a multibiased modification of the Sawyer-Johnson rule-difficulty model was developed and tested. An additional finding was that preexperimental bias may be affected by the specific relevant attributes used to assess it.A binary conceptual rule is a plan for partitioning a population of stimuli into two discrete groups based on the presence or absence of two particular attributes. If there are two attributes and each can either be present or absent, then four situations are possible: Both attributes can be present (truth-table class TT), just one present (TF), just the other present (FT), or neither present (FF). A binary rule is a system whereby, based on its truth-table class, a stimulus is assigned one of two pos-

Rule-specific dimensional interaction effects in concept learning

“…As expected, integral stimuli promote no significant facilitation. Since dimensional structure is important in disjunction, we might have expected separable stimuli to facilitate acquisition; however, as Reznick and Richman (1976) observed, many people naturally take a disjunctive approach to all rules. For these people disjunction is learned without their necessari1y learning the dimensional structure ; hence, separable stimuli would have a less facilitative effect.…”

“…Past research has shown that, for these rules, certain truth-table classes are particularly difficult to learn. In the conditional, TF and FF are hardest; in the biconditional, FF is hardest (Bourne & Guy , 1968;Reznick & Richman, 1976) . For these rules, the data of this experiment show that separable stimuli, probably by heightening dimensional salience, facilitate acquisition of these difficult responses.…”

“…When subjects are told the relevant features and must acquire the rule defining the concept, conjunctive learning is more rapid than disjunctive, disjunctive more rapid than conditional, and conditional more rapid than biconditional (Bourne, 1970;Bourne & Guy, 1968;Salatas & Bourne, 1974). In some experiments two clusters emerge,with the conjunctive-disjunctive cluster learned more rapidly than the conditionalbiconditional cluster (Bourne, 1967;Bourne & O'Banion, 1971;Reznick & Richman, 1976).…”

“…Recently, our understanding of this problem has been improved by focusing on rule difficulty as an interaction between initial response tendencies of the person learning the rule and the rule to be learned. For example, Reznick and Richman (1976), Salatas and Bourne (1974), and Sawyer and Johnson (Note 1) have developed models in which rule difficulty results from differences between the response assignments a person expects and those that are required by the rule.…”

Thinking

“…Ciborowski (21) shows that a modified biconditional rule in which ITs are "+," TFs and FTs are "-," and FFs are "other" is considerably easier than a standard biconditional problem, lending support to one of the Salatas & Bourne assumptions. But re: cent data (34,164) suggest that a majority of subjects may not enter a rule learning problem with a conjunctive bias -instead disjunctive and one-dimensional rule:; may be fa vore d. A modified model that takes account of the subject's initial bias does a better job of accounting for rule difficulty. And Dominowski & Wetherick (34) �nd that intra truth-table-class trans fer (from stimulus to stimulus) depends on the particular rule, which is difficult to account fo r within a simple bias model.…”