hide
Free keywords:
-
Abstract:
To come to grips with the complexity of the world, primates and other species have developed the ability to categorize objects. To study the mechanisms underlying categorization, we trained 3 monkeys (Macaca mulatta) to perform 2 tasks with 2 sets of parameterized stimuli (line drawings of faces and fish). In the first task the monkeys rated the pairwise similarity of the stimuli presented in triads. This data was used to derive distances between the stimuli and to determine their psychological representation using Multidimensional Scaling (MDS). In the second task, the monkeys learned 2 classes of exemplars for each stimulus set. The categories were designed to differ along 2 (out of the 4) parameterized dimensions, which were the diagnostic dimensions for the categorization. In a second phase, the monkeys assigned new exemplars to the 2 classes in an unsupervised manner. The representations and classifications of the monkeys were compared with those of 6 humans tested with the same experimental paradigms. Humans and monkeys exhibited similar categorization performance, a result suggesting that comparable neural mechanisms may underlie the process of perceptual classification. Specifically, the categorization strategy of both species could best be explained by assuming that classification of a novel stimulus was determined on the basis of its similarity to familiar exemplars, rather than to class prototypes. For both species, the 4 varying dimensions of the stimuli were not represented uniformly in the subjects' psychological space. In particular, the monkeys represented the diagnostic dimensions veridically to the original configurations of the stimuli only after the categorization training, providing evidence for perceptual sensitization induced by experience and task demands. The humans showed the same effect for the fish stimuli. In current neurophysiological experiments we are investigating the representation of stimulus-related information in the anterior inferior temporal cortex. Preliminary data suggest that neurons encode parametric information mainly about the diagnostic features of the stimuli, as well as whole exemplars with certain combinations of feature values.
