Looking Beyond Looks

Abstract

How do people acquire knowledge about the world? Do the sources of information that underlie knowledge acquisition in young children differ from those in adults? These fundamental questions have permeated scientific inquiry since the time of Socrates and Aristotle. Following in this tradition, a recent article by Sloutsky, Kloos, and Fisher (SKF; 2007) is ambitious—indeed, classic. Their goal was to uncover the contributions of conceptual and perceptual information in children’s categorization and induction about natural kinds. But experimental evidence is only as good as the theory and logic upon which it rests. Unfortunately, SKF’s approach to each of the key constructs—concepts, perceptual information, categorization, and induction—misses its mark. To quickly review: SKF taught children two novel categories of buglike entities, stipulating that the categories could be distinguished by the ratio of buttons to fingers (members of one category had more buttons than fingers, members of the other had more fingers than buttons). SKF introduced novel words (ziblet and flurp) for these categories. Children successfully extended these names to new instances. SKF then told children that one particular individual had a certain property (e.g., ‘‘thick blood’’). The children extended that property to items of the same overall appearance, ignoring the ziblet/flurp distinction. SKF interpreted this result as showing that ‘‘looks are everything’’ in children’s inductive inferences. We illustrate our concerns with an analogy. Suppose we aim to study the role of conceptual versus perceptual similarity in reasoning about natural kinds. We teach children two novel categories, stipulating that they are evensies (dogs with an even number of whiskers) and oddsies (dogs with an odd number of whiskers). We find that children learn these words without difficulty. We then tell the children that one particular dog (e.g., a collie evensy) has a certain kind of blood inside. We find that they extend that property on the basis of appearance, generalizing to perceptually similar dogs, disregarding the evensy/ oddsy distinction (e.g., they extend the property to a collie oddsy, but not a Chihuahua evensy). From this evidence, can we conclude that children use perceptual similarity rather than natural-kind membership in their inductive inferences? We think not. The experiment lacks construct validity. The novel categories created for this experiment are not natural kinds, and naming them (evensy or oddsy) does not make them so. The appearance of the items is a better guide to natural-kind membership than are the labels. Our concerns about this hypothetical example apply equally to SKF’s study.