Abstract
We present new experiments that test the hypothesis that human vision uses two geometrical rules to carve shapes into parts: the minima rule and the short-cut rule. These rules use the intrinsic geometry of a shape, whether 2D or 3D, to decompose the shape into parts. The minima rule states that human vision divides 2D shapes into parts at negative minima of curvature of their bounding contours; it divides 3D shapes into parts at negative minima of the principal curvatures. The short-cut rule states that human vision uses the shortest possible cuts, in conjunction with the minima rule, to make complete part cuts. The new experiments indicate that the parts defined by these two rules, the “minima parts”, are among the representational units passed on by the visual system to higher cognitive processes. In particular, there is a “minima-part bias” in language learning: These minima parts are the units of shape that adults recognize and attach names to in certain tasks of ostensive definition, and they are the units that adults use to generalize the application of part names across changes in scale and location. This minima-part bias is a constraint on higher cognitive processes, one that arises from the endogenous computational processes of the visual system itself. The minima-part bias interacts with other well-known biases in word learning, such as the whole-object bias, the mutual exclusivity principle, and the syntax of count nouns. Together these biases and principles allow the word learner to sift through the countless possible meanings of a novel word, and to fix on the correct interpretation.