Abstract
A key assumption of structural description accounts of object representation is that the representation of the relations among parts are distinguished from the representation of the shapes of parts themselves, e.g., by different units for the different kinds of information. Models that do not represent relations explicitly leave this distinction to some undefined, later cognitive process, an assumption somewhat at odds with recent results from selective attention tasks showing that parts are perceptually distinguishable (or analyzable) from relations (Hayworth & Biederman, 2004). Subjects passively viewed four kinds of “flip movies” (produced by rapidly cycling between two images) as well as an LO and an MT localizer. In Part Shape movies, the shape of a part of a two-part object cycled between two geons, say, a cylinder and a wedge, both on top of a brick for 3 s, with each frame being 250 msec. In Relation movies, the relation between the two parts changed with, for example, the cylinder shuffling back and forth over the surface of the brick. A control (Separate Parts) for the Relation condition separated the cylinder from the brick to assess whether activation produced by the Relation condition was reflecting within object-relations rather than activity from a large translation of a part. In View movies, the whole object either translated, changed size, or rotated in depth. The magnitude of motion energy was equated for the four conditions. All four types of movies activated MT equally, confirming the motion energy calculations. Part changes produced greater activation in area LO than any of the other conditions whereas the Relation changes produced the greatest activation in a region in the parietal cortex.
NSF 04207994 and 0426415; Human Frontiers Science Program 99-53