Abstract
Objects are often composed of two or more parts whose relative spatial arrangement can vary while the elements within parts remain invariant under a variety of 2- and 3-D transformations. What determines which configuration of individual elements gets encoded as parts in the absence of low-level mechanisms, such as the similarity or common motion of elements? Previously, we showed that human observers can learn the spatial configuration of shape-pairs or -triplets embedded within multiple exemplars of complex scenes. Passive observation of several dozen exemplars was sufficient for learning that some shape-pairs or -triplets that occurred consistently in the scenes comprised a set of base elements, whereas other shape-pairs or -triplets that occurred inconsistently comprised ‘accidental coincidences’. In the present study we extended these findings by showing that shape-pairs are encoded differently depending on whether or not they are embedded in a larger part. Subjects viewed a series of 120 displays consisting of 6 elements in apparently random configuration, but constructed from one of two 4-element shape-quads and one of two shape-pairs (from an inventory of 12 simple shapes). Each scene was passively viewed for 2 sec. A 2AFC post-exposure test revealed that subjects could easily discriminate shape-pairs and shape-quads which were not embedded in shape-quads from the exposure set from novel configurations of shape-pairs and shape-quads [t(19)=3.56, t(19)=4.16, both p<.001]. However, discrimination was at chance for shape-pairs embedded in familiar shape-quads when compared to novel configurations of shape-pairs [t(19)=0.78, n.s.], despite the fact that they appeared the same number of times as the non-embedded pairs. This pattern of results suggests that instead of the successive build-up of larger parts from smaller parts, subjects are biased to extract the largest configuration of consistently co-occurring elements in multi-element scenes (‘parts’), and that within an extracted larger configuration (‘whole’) the independent accessibility of embedded substructures is diminished.