ABSTRACT
Perception of multidimensional stimuli composed of separable attributes requires some process of integration to combine the attributes into the correct compounds. We hypothesize that when more than one such stimulus is presented, subjects avoid perceiving illusory conjunctions composed of wrongly paired attribute values by focusing attention on one location at a time and thus processing the stimuli serially. Any of three conditions should then allow parallel processing with divided attention: (1) the stimuli vary along one attribute only (and acuity does not limit performance); (2) the stimuli are multidimensional but the correct conjunctions are irrelevant; (3) the stimuli are multidimensional but share no attributes which could be wrongly recombined. We report two experiments designed as first attempts to explore the implications of this hypothesis for perception in a visual search task and for memory in a successive matching task. Latencies to detect targets in arrays of varying size suggested that parallel processing could occur when the target differed from nontargets either in color alone or in a shape feature (curvature) alone, but that when a particular conjunction of color and shape defined the target, subjects were forced to scan the stimuli serially. In the successive matching task, two multidimensional targets (colored letters in one experiment and schematic faces in another) were compared with two multidimensional test stimuli. When a test stimulus wrongly combined two attribute values from the target stimuli, errors were frequent and response latencies increased by a constant duration. This suggests that: (1) attribute values are poorly integrated in memory when the physical stimulus is no longer present, and (2) a separate stage of processing is required to check whether two matching values which are spatially combined in the test stimulus were also combined in a memorized target.
