REVIEWS simulation of its.The fusifor us,the cted system of these The boeanatomicalinvestigatiosaecomplt sing EVENT-R (ERPs)an of face features.such a at is related to face-spec 6aepol ed s ategoriz ind d th of fa rith al id full-body interp process c study of f g order v s.as pes I 94 ing The sults point to a or the fus re rel tors of per al lob he s al gv an are such a nd motion,suc as gaze shifts' In line the role ofthe uperior tempor nd ventral vis mot more abstract n TURE REVIEWSNATURE REVIEWS | NEUROSCIENCE VOLUME 4 | MARCH 2003 | 167 REVIEWS simulation of it15. The fusiform gyrus, the superior tem￾poral gyrus and other less well specified regions of occipitotemporal cortex could therefore be thought of as an interconnected system of regions that construct a spatially distributed perceptual representation of different aspects of faces. There is good evidence that activation in all of these regions can be modulated by attention16 and by the context in which the visual social signal appears17,18. The above anatomical investigations are comple￾mented by data on the timing of face processing. Studies using EVENT-RELATED POTENTIALS (ERPs) and MAGNETOENCEPHALOGRAPHY (MEG) show that some coarse CATEGORIZATION of face features, such as gender and emo￾tion, can occur at latencies as short as 100 ms (REFS 19–22). Peak activity that is related to face-specific processing near the fusiform gyrus is seen around 170–200 ms (REF. 23). Whereas the construction of a detailed struc￾tural representation of the face therefore seems to require about 170 ms, some rapid, coarse categoriza￾tion can occur with substantially shorter latencies, pre￾sumably indicating coarse perceptual routes that are parallel to a full structural encoding of the stimulus. A recent study24 investigated these different levels of categorization in detail and corroborated the idea of a fast, superordinate categorization of faces at a relatively short latency (around 100 ms). This categorization was followed by a subordinate categorization with a longer latency (around 170 ms), which was sufficient to dis￾criminate individual identity. Similar evidence for the extraction of information at subordinate levels with increasing processing time has been provided by sin￾gle-unit studies of face-selective cells in the monkey inferotemporal cortex25. At least three non-exclusive mechanisms could implement such computations: initial feed-forward processing followed by top–down modulation from higher regions, progressive process￾ing within a region, or iterative cycles of processing between a region and others (either ‘lower’ or ‘higher’ in a processing hierarchy). A growing body of work has used visual stimuli that signal biological motion (FIG. 2) to study social cogni￾tion. Social psychologists first showed our propensity to make social inferences from visual motion of abstract shapes in the 1940s (REFS 26,27), and recent studies indicate that specific movement cues might generate attributions of ANIMACY, intentionality and AGENCY28,29. Visual motion stimuli elicit attributions of intentionality and animacy in infants, and robustly elicit intentional, emotional and personality attribu￾tions in adults, even when only static depictions of their trajectories are shown. More specific information about the movements of a human body are offered by POINT-LIGHT DISPLAYS30, which generate exceptionally robust shape-from-motion cues that allow the recogni￾tion of identity31, gender32, emotions33 and personality traits34. In line with the role of the superior temporal cortex in processing dynamic aspects of faces, this region is also activated by viewing biological motion in whole bodies35 or their point-light displays36,37, and by more abstract movements of geometric shapes38,39. full-body interpersonal interactions8 . Not only are we exceedingly sensitive to the social signals themselves, but we are also sensitive to the details of the context in which they occur. Regions of non-primary sensory cortices might be relatively specialized to process certain socially relevant stimulus attributes. The best evidence comes from the study of faces, for which higher-order visual cortices can be regarded as an assembly of MODULES that process dis￾tinct attributes, as borne out by various lesion studies, scalp and intracranial recordings, and functional imag￾ing data. The results point to a role for the fusiform gyrus in processing the structural, static properties of faces, which are reliable indicators of personal identity. By contrast, regions more anterior and dorsal in the temporal lobe (such as the superior temporal gyrus and sulcus) are involved in processing information about the changeable configurations of faces, such as facial expressions, and eye and mouth movements9,10 (FIG. 3). Activation along the superior temporal sulcus and gyrus has been found when subjects view stimuli depicting biological motion, such as gaze shifts11,12 and mouth movements13. Processing in this region might draw on both dorsal and ventral visual streams in integrating shape and motion information14, and it might reflect a comparison of the observed action with the viewer’s ANIMACY The subjective impression that a stimulus is alive. AGENCY The subjective impression of a willful, goal-directed action. POINT-LIGHT DISPLAYS Visual motion stimuli created by attaching small lights to an actor’s body joints and filming the person moving in an otherwise dark room. Although they seem random when static, the biological motion of the lights immediately generates the compelling perception of a person moving about the room. Detailed perceptual processing • Fusiform gyrus • Superior temporal gyrus Emotional response in body • Visceral, autonomic, endocrine changes Social reasoning • Prefrontal cortex Coarse perceptual processing • Superior colliculus Self￾regulation Reappraisal Modulation of cognition (memory, attention) • Cingulate cortex • Hippocampus • Basal forebrain Representation of emotional response • Somatosensory-related cortices Representation of perceived action • Left frontal operculum • Superior temporal gyrus Motivational evaluation • Amygdala • Orbitofrontal cortex • Ventral striatum Figure 1 | Processes and brain structures that are involved in social cognition. It is possible to assign sets of neural structures to various stages of information processing, as I argue in this review. However, the flow of social information defies any simple scheme for at least two reasons: it is multidirectional and it is recursive. A single process is implemented by a flexible set of structures, and a single structure participates in several processes, often during distinct windows of time. Processing routes differ in terms of their automaticity, cognitive penetrability, detail of the representations they involve and processing speed. The structures outlined in this figure share some core features of a social information processing system: selectivity (they make distinctions between different kinds of information), categorization and generalization, and the incorporation of past experience. Several of the components of social cognition (inside the grey shaded region) contribute to social knowledge. Reappraisal and self-regulation are particular modes of feedback modulation whereby evaluation and emotional response to social stimuli can be volitionally influenced