Abstract
Recent models of visual object recognition (Hochstein & Ahissar, 2002; Bar, 2004) suggest that rapidly deployed, high-level processes may dictate or aid low-level, fine-scale visual analysis. To examine the effects of high-level knowledge on basic visual processes, we developed a training paradigm in which subjects were trained to categorize novel objects based on their constituent features while learning the functional properties of those features. Subjects studied two prototypical objects from separate categories (tools). During study, subjects were provided with a story regarding the practical use of each object and learned the specific function of each feature. For each category, half of the features were assigned functions that were essential to object usage (critical features), and half were assigned secondary functions (secondary features). Functional significance for a given feature was counterbalanced across subjects. Following study, subjects were tested for their recall of category names and the function of individual features. After this testing, subjects were briefly (100ms) presented with variants and rotated versions of the prototype objects. In some trials, a number of critical or secondary features (1, 2, or 3) had been removed. Subjects were asked to rapidly judge whether a test object belonged to one of the trained categories. Results (n=6, trials>3,000) indicate that subjects were significantly (p<0.032) less likely to categorize objects into one of the trained categories when critical features were missing, than when only secondary features had been removed. These findings suggest that high-level knowledge of object functionality may influence the representation of novel object classes.