It is easier to remember two faces than a single one

Theories of visual working memory (WM) assume that encoding produces independent “internal replicas” of the to-be-remembered items and describes retrieval as a template matching process. The present investigation questions this basic assumption. Method. Observers were asked to perform an old/new tasks with a short time interval between the memory array and the probe item. The memory array always comprised low-discriminable items generated through a morphing procedure (human faces, cars, or cat faces). The memory load was varied. Results. For car stimuli, recognition performance decreased with set size. Contrary to previous results, for face stimuli recognition performance was higher when observers were asked to remember two or three faces, rather than only one (Experiment 1). For caucasian participants, the size of this effect was larger for caucasian faces, smaller for afro-american faces, and smaller for cat faces (Experiment 2). The disadvantage of the memory load of one face with respect to two or three faces for recognition performance could not be explained by the lower discriminability of the face stimuli within the morph continua with respect to the car stimuli. In Experiment 3, we showed that the face stimuli were more easily discriminable among each other than the car stimuli. Conclusions. The present results are inconsistent with WM models assuming that items are stored discretely in different slots and suggest that different perceptual coding schemes are used by the brain for low-discriminable faces and non-faces. The novel results concerning the face stimuli are compatible with synthetic vision algorithms based on principal components analysis, such as the Eigenfaces model. These results support the hypothesis that, for low discriminable faces, information is not represented independently in WM, but rather in relation to the other items appearing in the same spatial configuration (Jiang, Olson, & Chun, 2000).