Abstract
The capacity of visual working memory for faces is extremely limited, but direct evidence for the cognitive locus of these limitations is currently lacking. Some accounts of working memory for complex objects such as faces assume that a fixed number of three or four items are encoded regardless of object complexity. According to fixed-capacity accounts, the apparent severe capacity limitations for complex objects arise during the sample-test comparison process rather than reflecting the actual number of items retained. We provide direct electrophysiological evidence (using the N2pc and CDA components as markers of attentional selection and working memory maintenance) that memory capacity limits for faces arise from the limited capacity of focal spatial attention during memory encoding. Participants performed a match to sample task with lateralised memory displays that contained one face (Load One) or two faces (Load Two). We show that only a single face is focally attended on most trials, and that the allocation of spatial attention predicts whether a particular face will be retained in memory. ERPs also reveal that on a minority of trials, two faces are simultaneously encoded. However, this type of divided attention delays face matching responses, and impairs sample-test comparison processes, as reflected by attenuated N250r components relative to trials with focal attention on the stored matching face. These results demonstrate a trade-off between the number of items stored in memory and the fidelity of visual working memory representations. We suggest that the capacity and distribution of selective spatial attention is a dynamic resource that constrains both working memory capacity and resolution. Because visual working memory representations are maintained through the allocation of selective spatial attention, they are position-dependent. This hypothesis is supported by an additional ERP experiment, which demonstrates that interference between visual memory representations occurs in a position-specific manner.
Meeting abstract presented at VSS 2015