The Influence of Prior Theories on the Ease of Concept Acquisition

The finding that conjunctive concepts are easier for human subjects to learn than disjunctive concepts is reported in most introductory books on cognitive psychology. In this paper, we report some conditions under which this finding may not be true. In particular, we demonstrate that the prior causal knowledge of subjects can influence the rate of concept learning. We report on an experiment that indicates that disjunctive concepts which are consistent with prior knowledge take fewer trials to learn than conjunctive concepts which arc not consistent with prior knowledge. We present a computer model of this learning task. Introduction In concept identification tasks, it has been found that conjunctive concepts require fewer trials to learn than disjunctive concepts (Bruner, Goodnow, & Austin, 1956). More recently, it has beeti suggested (e.g., Murphy & Medin, 1985; Schank, Collins, & Hunter, 1986; Pazzani, in press) that a person’s prior knowledge influences the speed or accuracy of learning. These claims are in part responsible for interest in explanation-based approaches to learning (DeJong & Mooney, 1986). More recently,~ a number of experiments have shown that with proper background knowledge people arc capable of the single-instance generalization predicted by explanationbased learning (Ahn, Mooney, Brewer, & DeJong, 1987). In this paper, we explore the interaction between the prior knowledge and the logical form of concepts. We first present an experiment in which these factors interact. Then, we present a computer model of this learning task. Ease of Concept Acquisition: An Experiment The purpose of this experiment was to investigate the interaction between prior knowledge and the acquisition of conjunctive and disjunctive concepts. Subjects were divided into two groups. The Influte group had to perform a prediction task. This group observed photographs of a person and a balloon and had to learn to predict under which conditions a balloon could successfully be inflated. The second group, Alpha, used the same materials, but had a concept identification task that required learning which photographs belonged to a category called “alpha.” These groups were then divided into conjunctive and disjunctive groups. The conjunction to be learned was that a balloon whose size is small&whose color is yellow was an alpha (or could be inflated). The disjunction to be learned was that a person whose age is adult aa person who is stretching a balloon is an alpha (or could inflate the balloon). Note that for the prediction task, the conjunctive concept is not consistent with prior knowledge while the disjunctive concept is. It is also important to stress that the prior background knowledge’ (e.g., adults are stronger than children, stretching a balloon makes it easier to inflate) is not sufficient for subjects to deduce the 1. In a prior experiment (Pazzani, 19871, we asked subjects a series of True-False question about which balloons are easier IO inflate. Almost all subjects indicated that adults could inflaw balloons more easily than children and that a balloon that had been stretched was easier IO inflate. Subjects also indicated that the color of a balloon, or dipping a balloon in water did not affect the ease of inflation. Some subjects also responded thar long skinny balloons were harder IO inflate than round balloons.