Abstract
A key assumption of structural description accounts of object representation is that the relations among parts are distinguished from the parts themselves, e.g., by different units for the different kinds of information. Models that do not represent relations explicitly, such as templates or theories that posit a projection of a constellation of features (such as Hmax) leave this distinction to some undefined, later cognitive process. Are parts perceptually distinguishable (or analyzable) from relations? Subjects performed a same-different judgment on whether the members of a sequentially presented pair of novel objects, composed of two parts, were identical or different (metrically). The shapes of the parts composing the objects varied in random-appearing fashion from trial to trial, as did the location of the second object relative to the first. On a given block of trials, the pair of objects on each trial could differ in either the length of a part or its position (relation) relative to the other part or both. On some blocks of trials only one kind of information changed on a “different” trial; on other blocks, both part and position changes could occur. Subjects were instructed to respond “different” (by key press), only if there was change in a part (on some blocks) or only if there was a change in a relation (on other blocks). Consistent with an explicit distinction between parts and relations, one or the other kind of information could be processed with little or no effect of whether the other kind of information was constant or varied. This pattern of independence is apparent for “privileged” combinations (Stankiewicz, 2002) of part attributes or different types of relations, such as the width and axis curvature of a part but not for arbitrary combinations of attributes or attributes and relations.
Supported by Human Frontiers Science Program Organization RG0035/2000B, MURI ARO DAAG55-98-1-0293, and James McDonnell Foundation 99–53.