Abstract
Contrary to the common belief that visual working memory can hold representations of 4 objects in which multiple features are integrated, Saiki (1999, 2000) showed that in a dynamic situation, even 3 objects' color and spatiotemporal location cannot be maintained simultaneously. Because other object features remain constant in the studies above, it is unclear whether this limitation is due to the integration failure at the level of object (feature combination) or of single features. This study investigated the maintenance and transformation of shape-color conjunctions in a dynamic situation. Stimuli were patterns of four objects behind an occluder. By the regular rotation of the pattern or of the occluder at a rate of 120 degrees per second, 360-ms visible and occluded periods alternated 10 times. During one visible period, properties of two objects, either both shape and color, shape alone, or color alone, could switch. Observers detected any type of switches, switches involving shape, and switches involving color in different sessions (Experiment 1), or identified the type of switch (Experiment 2). In general, observers' performance was much better when the pattern is stationary than when the pattern moves. If object representations are available, switch of single feature, which create new feature combinations, should be easier to detect than switch of both features. Analyses with multinomial processing tree models evaluated this prediction and revealed that shape-color conjunctions had substantial contribution to the switch detection performance only when the pattern is stationary, suggesting that integration of an object's shape and color cannot survive the 360-ms occlusion. These results suggest that previous findings of maintenance of 4 integrated perceptual objects really reflect the binding of perceptual features to static spatial locations, or that without continuous visual input, life of integrated object files is too short to survive a brief occlusion.