Comparison, Categorization, and Perceptual Dimensions: A Connectionist Model of the Development of the Notion of Sameness

The process of comparison is fundamental to behaving organisms. Generalization from past to present experience involves a measure of the similarity of present perceptual input to what has been perceived before. The likelihood that we call some object a dog is a function of the perceptual similarity of that candidate object to other objects known to be dogs. But humans do far more than simply categorize objects into a single basic level. At various times and for various purposes, we compare objects in terms of their overall similarity to one another and along a wide array of perceptual dimensions. For example, we might judge a dog to be brown, to be the same color as our cat, to be large for dogs in general. Indeed, what we consider higher mental functioning|metaphor, poetry, science itself|involves pointing to and discovering novel kinds of similarity. The problem of how a child develops a system of multiple kinds of perceptual similarity together with devices for linguistically communicating about similarity is clearly of great importance to cognitive science. This is an area in which there is rich and detailed data about human development, but no current theory that adequately explains it. In this paper we describe a connectionist model of some of the basic facts of comparison along perceptual dimensions. The workings of the model are based on the idea that categorization along dimensions (what color is it? ) and comparison along dimensions (is X the same color as Y? ) make use of the same dimension representations and the same internal representations for objects. We propose that these representations develop in response to the demands of the two kinds of tasks. Our model should yield behavior which qualitatively ts the developmental data from children.