Abstract
Behavioral studies and single cell recordings in monkey IT have shown that there is greater sensitivity to nonaccidental property (e.g., straight versus curved), than metric property (e.g., curved versus more curved) changes of simple shapes (Biederman et al., 2009; Kayaert et al., 2003). Are we similarly more sensitive to nonaccidental (NAP) than metric (MP) differences in the relations between objects? We generated a set of stimulus triplets that could, from a base stimulus (e.g., a cylinder separated from a brick), undergo a NAP relational change (e.g., the cylinder attached to the brick) or a MP relational change (e.g., the cylinder further separated from the brick). Critically both relational variations were matched in physical dissimilarity using the Gabor-jet system (Lades et al., 1993), a model of V1 similarity. In an adaptive staircase match-to-sample paradigm, a target stimulus was presented with two samples, one of which was an identical match and the other, a distractor that varied in either a NAP or MP change. To achieve 75% accuracy, subjects required more than double the amount of presentation duration for detecting differences for MP (795 ms) than for NAP (330ms) relational changes. The results could not be explained by differences in local or global features. Given recent findings that the lateral occipital complex (LOC) is a critical region for the coding of relations between objects (Kim & Biederman, in press) and within parts of an object (Behrmann et al., 2006), a prediction is that the differential sensitivity of NAP vs. MP relational changes likely arises in LOC as well.
NSF BCS 04-20794, 05-31177, 06-17699 to I.B.