Clusters Precede Shapes in Perceptual Organization

Does perceptual grouping require attention? Recent controversy on this question may be caused by a conflation of two aspects of grouping: element clustering (determining which elements belong together) and shape formation (determining cluster boundaries). In Experiment 1, observers enumerated diamonds that were drawn with either lines or dots. These two types of stimuli were subitized (enumerated rapidly and accurately in the range from one to three items) equally well, suggesting that clustering dots into countable entities did not detnand attention. In contrast, when target diamonds were enumerated among distractor squares in Experiment 2. only line-drawn items could be subitized. We propose that clustering and shape formation not only involve different percepttial processes, but play different functional roles in The classic illustrations of the Gestalt laws of grouping are elegant in their simphcity (Koffka. 1935; Wertheimer, 1923/1950). For example, one detnonstration of the law of proximity begins with a square grid of dots. Simply changing the relative distances between dots can dramatically alter the percept, so that the grid is seen as a set of columns, rows, or even diagonals (see Kubovy & Wagemans. 1995). The simplicity of this demonstration belies a complexity of processes. First, to see the dots as rows instead of columns or diagonals, one must link together some dots but not others; we refer to this step as element clustering. Second, shape formation must occur; that is, the principle orientation of the cluster of dots must be determined. Koffka (1935. pp. 125-127) first distinguished between these two operations, but recent theories of grouping appear to have ignored the distinction, arguing instead about whether shapes are determined before or after figureground relations have been established (e.g.. Baylis & Driver, 1995; Kellman & Shipley, 1991; Palmer & Rock. 1994; Peterson. 1994). Both of these operations presuppose that some elements have been clustered together. The neglect of clustering may be because it is really inseparable from shape formation, but it is also possible that researchers have been misled by the immediacy of their own perceptions. We believe that the conflation of clustering and shape formation has contributed to a confusion in the literature. For example, some researchers have argued that grouping is performed by preattentive mechanisms (Julesz, 1984; Treisman, 1982), whereas others have claimed that it is performed only much later (Palmer & Rock, 1994) and with the benefit of attention (BenAv, Sagi, & Braun, 1992; Mack, Tang, Tuma, Kahn, & Rock, Address correspondence to Lana Trick, Department of Psychology, University of British Columbia. 2136 West Mall. Vancouver. B.C.. Canada V6T 1Z4; e-mail: [email protected]. 1992). Interestingly, although the former studies involved search or texture segregation tasks in which cluster presence or absence was manipulated, the latter studies required observers to report on cluster shape (e.g.. Are the clusters arrayed horizontally or vertically? Do the clusters have a dominant orientation?). This study demonstrates that clustering and shape formation are separable operations that have different attentional demands. We used a visual enumeration task to make this point for two reasons. First, enumeration requires observers to process each member of a specified set of visual items (targets). If an item as a whole is to be enumerated, then its component elements must be linked; otherwise enumeration would be based erroneously on the constituent elements. When only targets are displayed, clustering, but not shape formation, is relevant to the task. However, if targets are displayed among distractors differing only in shape, then shape formation becomes important. The separability of clustering and shape formation can therefore be demonstrated if an experimental factor influences one operation differently than the other. Second, the enumeration of small numbers of items (up to three or four) is very sensitive to attentional demands. Generally, if items can be detected preattentively, then their enumeration is very rapid (40 to 100 ms per item) and accurate (Trick & Pylyshyn, 1993,1994). This pattern of results is called subitizing. However, if target items are too similar to distractors, or are defined by their spatial relations to other items, then subitizing does not occur and response time (RT) slopes for small numbers of items are indistingui.shable from those for larger quantities. The slower (250 to 350 ms per item) and more error-prone enumeration process observed in this case is called coimting. Our strategy was to test the importance of element connectedness in two enumeration tasks; one that required only clustering (Experiment 1) and one that also required shape formation (Experiment 2). Element connectedness is ideal for this test because it is widely recognized as critical to perceptual grouping (Elder & Zucker. 1993; Enns & Rensink, 1991; Palmer & Rock, 1994; Rensink & Enns, 1995). If clustering and shape formation are really inseparable, then this variable would be expected to have the same effect in both experiments. Moreover, if attention is required either for clustering or for forming a shape from disconnected elements, then subitizing would be expected for the connected forms, but not the disconnected forms, in the corresponding experiment.