Dimension-based attention in the recognition of facial identity and facial expression

Although the human visual system is constantly flooded with sensory information, the brain is remarkable in inferring structures from the massive inputs and selectively attending to behaviorally relevant information. However, how the two processes interact remains largely unknown. Can top-down attention efficiently select the task-relevant dimension (e.g. gender) during face recognition to override interference in the task-irrelevant dimension (e.g. expression)? To address this issue, participants were asked to classify real face images according to gender or expression, which were preceded by other faces (masked priming task) or words (face-word Stroop task). Results show that face classification was 1) affected by the task-relevant but not the task-irrelevant dimension of the preceding faces, and 2) modulated by words depicting the task-relevant but not the task-irrelevant dimension of the face. These results suggest that high level dimensions such as facial expression and facial identity can serve as units of attentional selection, possibly due to the late binding of the two dimensions