Domain Specificity in Face Perception Nancy Kanwisher

Domain specificity in face perception Nancy Kanwisher

Is face perception carried out by modules specialized only for processing faces? Or are faces perceived by domain-general mechanisms that can also operate on non-face stimuli? Considerable evidence supports the domain-specific view.

A primary goal in scientific research is to object recognition. Perhaps the strongest the ‘basic level’ (for example, chair, coat, ‘carve nature at its joints.’ This is emphat- evidence that distinct mechanisms may be pen), without determining the specific ically true in cognitive science, where involved in the recognition of faces comes exemplar of the category (my chair, Joe’s philosophers and psychologists have from the neuropsychological literature. In coat), for faces we usually proceed beyond sought for centuries to discover the fun- the syndrome of prosopagnosia8, patients the general category ‘face’ to determine damental components of the human are unable to recognize previously famil- the identity of the particular individual. mind. In his influential book Modularity iar faces, despite a largely preserved ability Second, we probably look at and discrim- of Mind, Fodor1 contrasts two competing to recognize objects. Even more striking inate between faces more than any other .com views on the functional organization of are patients with the opposite syndrome. class of visual stimuli (with the possible the mind. One view holds that the mind Patient CK is severely impaired at reading exception of words, discussed later). We is divided according to the content of the and object recognition, yet completely are all face experts. Is it possible that the information processed. Proponents of this normal at face recognition9. Importantly, mechanisms we use to process faces are osci.nature approach seek distinct mechanisms for CK’s face-recognition abilities are much not specialized for face processing per se, functions like voice recognition, spatial more disrupted than those of normal sub- but rather for making fine-grained dis- navigation and perception of visual jects when faces are inverted or fractured criminations between visually similar motion. In the alternative view, the mind into pieces, consistent with the idea that exemplars of any category? Or might the http://neur ¥ is organized instead around the kinds of face-specific mechanisms are holistic and putative face-specific mechanisms actual- processes it carries out. Accordingly, inversion-sensitive. Thus the neuropsy- ly be specialized for making any discrim- mechanisms might exist for functions like chology literature contains evidence for a inations for which we have gained judgment, volition and categorization. double dissociation between face and substantial expertise? That is, might a The current discussion of the nature of object recognition. domain-general account of face percep- face perception is a microcosm of this long- Functional brain imaging investigations tion be possible? standing debate2. Is face perception carried of the normal human brain have comple- A seminal study of the behavioral con- out by domain-specific mechanisms, that mented the evidence from neuropsychol- sequences of visual expertise23 found that is, by modules specialized for processing ogy. Many studies show that a region in the expert dog judges show inversion costs 2000 Nature America Inc.

faces in particular? Or are faces handled by fusiform gyrus is not only activated when when recognizing dogs, much like the © domain-general mechanisms that can subjects view faces10,11, but activated at least inversion costs all subjects show in rec- operate on nonface visual stimuli as well? twice as strongly for faces as for a wide ognizing faces. On the basis of these variety of nonface stimuli, including letter results, the authors inferred that it is Evidence for domain-specificity strings12, assorted objects13,14, animals with- unlikely that a neural substrate dedicat- Many behavioral studies suggest that out heads15 (but see also ref. 16) and the ed to face encoding exists. Instead, they domain-specific mechanisms are involved backs of human heads17. Similarly, selec- hypothesized that the mechanisms in processing faces. For example, face tive responses to faces are reported using involved in face recognition are also recognition is more disrupted by inversion scalp ERPs18 and MEG19, as well as direct engaged when subjects make discrimina- (turning the stimulus upside down) than is electrical recordings from the surface of the tions between structurally similar exem- object recognition3. Accuracy at discrim- human brain20–22. As this brief overview of plars of a category for which they have inating individual face parts (such as the the literature shows, considerable evidence gained substantial visual expertise. The nose) is higher when the entire face is pre- supports the domain-specific view: face authors of this study23 and another24 pro- sented than when the parts are presented perception seems to involve different cog- pose that face recognition and expert in isolation, whereas the same ‘holistic’ nitive and neural mechanisms from those exemplar discrimination both involve advantage is not found for parts of houses involved in the recognition of nonface extracting ‘second-order relational fea- or inverted faces4. These and other find- objects. Note, however, that the question tures,’ that is, computing the distinctive ings5–7 suggest that face recognition of the domain specificity of face-process- features of the presently seen exemplar involves special mechanisms that are more ing mechanisms is independent from the compared to the average of the category. inversion-sensitive and more holisitic (that question of the innateness of such mecha- Note however, that these important is, less part-based) than those involved in nisms, which is not the focus of the present results23 are consistent either with a uni- discussion. tary mechanism for the processing of Nancy Kanwisher is at the Department of Brain both faces and dogs in dog judges, or with and Cognitive Science, Massachusetts Institute Evidence for domain-generality two distinct mechanisms (one for faces of Technology, Cambridge, Massachusetts Faces differ from other objects in several and one for dogs) that function in similar 02139, USA. important respects. Whereas it is often ways, at least with respect to inversion. e-mail: [email protected] sufficient to identify common objects at Although it is difficult to choose between

these two hypotheses based on behavioral however that when the fusiform face area Concerning the role of visual expertise, data alone, brain-based evidence can pro- was defined using the standard criteria alphanumeric characters are the only visu- vide clues. adopted in my lab, face areas could be al stimuli for which visual expertise is on a Indeed, prosopagnosia is often accom- found in the right hemisphere in only 5 par with expertise for faces (at least for aca- panied by deficits in discriminating of the 19 subjects in this study; of these, demics!). Yet the response of face-selective between similar exemplars of other non- the magnitudes of the expertise and cate- regions in the fusiform gyrus is extremely face categories25–27, consistent with the gorization-level effects are quite small, and low during viewing of letter strings12. One hypothesis that a single common mecha- the response to faces remain at least twice might also argue that another category of nism may underlie the discrimination of as strong as that to the expert category. stimuli for which we do expert discrimi- faces and of different exemplars of non- Further studies will be necessary to resolve nation is places or scenes, yet the FFA face categories. On the other hand, this this apparent dependence of effect size on responds only very weakly to photographs finding is also consistent with the possi- the precise criteria used to localize the of indoor and outdoor scenes33. bility that face recognition and recogni- FFA. In any event, these findings suggest Thus considerable evidence shows that tion of other expert categories go on in that both within-category discrimination the FFA does not consistently produce a distinct but nearby cortical regions that and visual expertise may account for part strong response during either discrimina- are frequently damaged together. After all, of the apparent face selectivity of regions tion between similar exemplars of a non- the chance that a stroke or head trauma in the fusiform gyrus. face category or during viewing of to visual cortex will obliterate all of the nonface stimuli for which the subject has hypothesized face-processing region of Problems for domain-generality gained visual expertise. Instead, numer- .com cortex without affecting nearby cortical How far can domain-general accounts go? ous findings support the domain-specif- areas is similar to the chance that an aster- Can all the behavioral and neural evidence ic view: the presence of faces per se seems oid hitting New England would obliterate for the ‘specialness’ of faces be accounted to be an important determinant of the all of the state of Rhode Island without for in terms of the fine-grained nature of FFA response. osci.nature affecting Massachusetts or Connecticut. face discrimination and the expertise we Although it is not clear how the bird However, brain-imaging studies of nor- have with faces? Although a resolution of and car expertise results30 can be recon- mal subjects can contribute importantly this question will require considerable fur- ciled with these earlier studies, one possi- here because they have the potential to ther research, reported findings already bility is that the expertise effects in this http://neur ¥ distinguish between the activation of pose substantial challenges for the study may in part reflect the greater inter- nearby but distinct cortical areas. domain-general view. est and attentional engagement the sub- Two neuroimaging studies28,29 show Brain imaging studies provide evi- jects may have had in the category for increased activation of face-sensitive dence against a variant of the domain- which they were expert, compared to the regions of the fusiform gyrus as subjects general hypothesis23, according to which category for which they were not expert. become experts at discriminating novel distinguishing between objects within a This hypothesis is consistent with prior categories. However, the ‘greeble’ stimuli category (as opposed to making demonstrations that the fusiform face area used in these studies are facelike in several between-category discriminations) can can be strongly modulated by visual atten- important respects. First, they look like engage purported face-specific mecha- tion34,35 and that in this study expertise 31 2000 Nature America Inc.

animate figures. Second, subjects are nisms even without expertise . The produced activations that extended well © trained to identify them using proper response of the fusiform face area (or beyond the FFA30 even into one area (the names, which is likely to encourage an ani- FFA) during discrimination of individ- parahippocampal place area, or PPA) that mate and possibly human interpretation. ual faces is over four times as high as that is completely unresponsive to faces33,36. Third, their key distinguishing features measured during discrimination of indi- (The complete failure of the PPA to show consist of two horizontally displaced parts vidual hands14. Further, the response in any activation for faces above that for fix- arranged symmetrically above two verti- the FFA when subjects discriminate ating on a point argues against a similar cally displaced parts in a facelike configu- between hands is no higher (measured in attentional account of the FFA activation ration. Thus, activation of face-selective percent signal change, or as a ratio of the for faces.) Note, by the way, that this acti- mechanisms by greeble expertise is con- response to faces) than when such fine- vation of the PPA for expert within-cate- sistent with the conclusion that face-spe- grained discrimination is not required gory discrimination of nonfaces, but not cific mechanisms may only be recruited (that is, when the same subjects passive- for faces, shows that the two processes for expert subordinate-level discrimina- ly viewed either hands or assorted produce different activations, and a sin- tion of stimuli that share numerous prop- objects). Similar results are obtained gle domain-general mechanism cannot erties with faces in the first place. when discrimination between individual account for both. These authors also present stronger faces is compared to discrimination Thus, conflicting neuroimaging data evidence30 for the domain-general between individual houses17. These data raise questions for the claimed roles of hypothesis: in bird experts and car experts show that within-category discrimina- both expertise and within-category dis- scanned with fMRI while viewing birds, tion is not sufficient to strongly engage crimination in the activation of face-spe- cars, faces and objects, the activity in a the FFA. Note further that there is sub- cific regions of cortex. Further, even if the face-selective region of the fusiform gyrus stantial evidence that the FFA may not activation of the FFA by expertise and/or is weakest during viewing of assorted be involved (exclusively) in face recogni- within-category discrimination is found objects, next strongest for the nonexpert tion, but may instead (or in addition) be more consistently in the future, such imag- category (cars for bird experts and vice involved in face detection17,32; if these ing data leave two important questions versa), stronger yet for the expert catego- arguments are right, then they also refute unanswered. First, activation of the same ry (cars for car experts, and birds for bird the (exclusive) role of this area in expert area in an fMRI study by faces and by experts) and strongest for faces. Note within-category discrimination. stimuli for which the subject has gained

expertise is consistent (as the authors of First, although several prosopagnosic studies shows that even if face-specific ref. 30 note) with the possibility that the patients are impaired at both face recog- mechanisms are sometimes activated by response of this area reflects two distinct nition and at discriminating between visu- nonface stimuli, such activations are but physically interleaved neural popula- ally similar exemplars of nonface apparently neither necessary nor sufficient tions (for example, one for faces and one categories (see above), the deficit in other for the expert discrimination of exemplars for cars in car experts). Second, activations prosopagnosic patients is more specific. of those categories. of a given cortical area in any imaging For example, although patient MT42 is study need not reflect neural computations prosopagnosic, he is nonetheless normal Varieties of domain-generality necessary for the task. Next I consider evi- at discriminating between brands and Despite the substantial evidence against a dence from neurophysiology bearing on models of cars, as well as between differ- domain-general account of face recogni- the first question, and evidence from neu- ent fruits and vegetables. Further, the tion, it is always possible that future work ropsychology bearing on the second. severely prosopagnosic patient WJ43, who will prove this hypothesis to be right. It is In two recent studies, monkeys were became a sheep farmer after his stroke, therefore worth stepping back to consider intensively trained over several months to learned to recognize and name many of more closely what might be entailed in the discriminate between visually similar stim- his sheep, and performs much better on domain-general account. uli, and neurons selectively responsive to tests of sheep recognition than on compa- If a common mechanism existed that stimuli from the trained set were found in rable tests with human face stimuli. was necessary for both face recognition inferotemporal cortex after training37, in Another patient is profoundly prosopag- and for expert discrimination of visual- greater proportion than those found nosic despite showing an excellent ability ly similar exemplars of a nonface catego- .com before training38. A critical question is to discriminate between different models ry (but not for ‘basic-level’ categorization whether these neurons tuned to the newly of the same make of car, a category for of objects), what would that mechanism trained stimuli were the same or different which he had previously gained visual do? One hypothesis is that attaining as the face-selective neurons reported pre- expertise44. Cases such as these suggest that expertise in discriminating faces, cars, or osci.nature viously39,40. Although these two studies prosopagnosia is not simply a general loss other categories entails the discovery and were not designed to systematically test of expert subordinate-level discrimination. increased use of the visual features that this question, responses both to the trained Most of these patient studies report distinguish different exemplars of that stimuli and to faces were collected in both only accuracy measures, so it is possible category. However, a key problem here is http://neur ¥ studies. Neither study found a tendency that very slow reaction times occurred in that the visual features that are diagnos- for cells responsive to the trained stimuli some conditions, suggesting a subtle deficit tic in discriminating between cars (for to also be particularly responsive to faces41 that eluded detection when only accuracy example) are bound to be different from (K. Tanaka, personal communication), measures were used45. Indeed, it will be those that are diagnostic in discriminat- and anatomical information from one important for future patient studies report ing between faces. If the job of the pro- study41 suggests that the two populations both accuracy and reaction time data to posed domain-general mechanism is to are even physically segregated. avoid such ambiguities. Nonetheless, pre- encode distinguishing features of the Indeed, the possibility that individual liminary evidence from developmental exemplars of each category, then in what neurons might be tuned to two very dif- prosopagnosic subject BC indicates that sense would the mechanisms involved in 2000 Nature America Inc.

ferent stimulus types (for example, faces despite his severely impaired performance doing this for two different expert with- © and stimuli from the trained set) seems in the recognition of faces46, his d’ and in-category discriminations be any more implausible given the known response reaction time are no different from con- closely related to each other than either properties of inferotemporal neurons40. It trols in an analogous task in which the is to the mechanisms involved in basic- seems more likely that a distinct neural stimuli were shoes, (B. Duchaine, person- level recognition? population would be tuned to the features al communication). It will be of particu- One of the classic claims about the way of each visual category for which the lar interest to test the ability of patients like that face recognition differs from object monkey or person had gained expertise. BC to make discriminations on categories recognition is that face recognition is more If, as this preliminary evidence from mon- for which they have gained expertise. holistic or configural4. What exactly is meant keys suggests, distinct neural populations The converse case is even more strik- by configural or holistic processing? One and cortical loci are involved in face recog- ing. Patient CK, who has selectively pre- possibility is that holistic processing of faces nition and in making expert discrimina- served face recognition, is unable results from the tendency of face-selective tions within a (nonface) category, then it distinguish between individual airplanes cells to respond to the whole face, rather would be difficult to argue for a single or tin soldiers, despite being an expert than to parts of the face. Indeed, work using domain-general mechanism for all expert at both before his accident9,47. Thus the stimulus-reduction method40 shows that within-category discrimination. mechanisms that are sufficient for face the image of a face cannot be simplified Observing neural activations with recognition are evidently not sufficient without losing maximal activation of face- fMRI or single-neuron physiology can for expert subordinate-level identifica- selective cells. (It is also true, however, that provide information about the neural tion of nonfaces. some face-selective cells respond well to a populations that respond in a particular In sum, double dissociations between particular face feature or set of features.) In task, but these techniques cannot deter- face recognition and expertise and/or contrast, the preferred stimuli for other non- mine which of these activations reflect within-category discrimination of nonface-selective cells in inferotemporal cortex computations that are necessary for the faces in the neuroimaging, neurophysiol- are typically simpler, often corresponding task. Neuropsychological studies are par- ogy and neuropsychology literature to features rather than whole complex ticularly helpful here, and indeed they provide strong evidence that distinct objects48. Thus, the selectivities of cells in raise the greatest challenge to domain- neural substrates underlie these process- inferotemporal cortex are consistent with general accounts of face processing. es. Further, the evidence from patient the idea that faces are processed in a more

4. Tanaka, J. W. & Farah, M. J. Parts and wholes configural or holistic fashion than are non- criterion is generally considered a weak in face recognition. Q. J. Exp. Psychol. A 46, face stimuli. Is the same also true of nonface indicator of domain specificity. Sperber52 225–245 (1993). stimuli for which the person or monkey has argues that human cognition is likely to be 5. Tanaka, J. W. & Sengco, J. A. Features and their obtained expertise? It is argued49 that characterized by the flexible use of even configuration in face recognition. Mem. although the selectivity of trained cells’ domain-specific mechanisms. Pinker53 Cognit. 25, 583–589 (1997). responses might be reducible to less-com- argues against the actual use criterion most 6. Young, A. W., Hellawell, D. & Hay, D. C. Configurational information in face plex feature constellations, these data are pointedly: “The only thing that can be spe- perception. Perception 16, 747–759 (1987). more consistent with the hypothesis that cial about a perception module is the kind 7. Moscovitch, M. &. Moscovitch, D. A. these cells respond to the whole object (as of geometry it pays attention to, such as the Superface inversion effects for isolated internal seen from a particular point of view). These distance between symmetrical blobs…If or external features, and for fractured faces. preliminary data are consistent with the idea objects other than faces (animals, facial Cogn. Neuropsychol. 17, 201–219 (2000). that expertise in discriminating exemplars expressions, or even cars) have some of 8. Bodamer, J. Die prosopagnosie. Archiv für Psychiatrie und Nervenkrankheiten 179, 6–53 of a category leads to the formation of neu- these geometric features, the module will (1947). rons responding holistically to exemplars of have no choice but to analyze them, even 9. Moscovitch, M., Winocur, G. & Behrmann, M. that category. Nonetheless, it would be sur- if they are most useful for faces. To call a What is special about face recognition? prising if individual neurons coded for both module a face-recognizer is not to say it can nineteen experiments on a person with visual whole faces and, for example, whole cars (in handle only faces, it is to say that it is opti- object agnosia and dyslexia but normal face recognition. J. Cognit. Neurosci. 9, 555–604 car experts). Assuming they did not, it mized for the geometric features that dis- (1997). would be strained to argue that the set of all tinguish faces because the organism was .com 10. Haxby, J. V. et al. The functional organization cells, each one tuned to one of two very dif- selected in its evolutionary history for an of human extrastriate cortex: A PET-fCBF ferent kinds of whole objects for categories ability to recognize them” (pp 273–274). study of selective attention to faces and on which the subject had gained expertise, Carving nature at its joints has rarely locations. J. Neurosci. 14, 6336–6353 (1994). constitutes a domain-general mechanism been straightforward in any scientific field, 11. Sergent, J., Ohta, S. & MacDonald, B. osci.nature Functional neuroanatomy of face and object for expert within-category discrimination. and the effort to discover and characterize processing: a positron emission tomography Another version of the domain-gener- the fundamental components of the mind study. Brain 115, 15–36 (1992). 23 1 al story proposes that what is common is no exception. Although Fodor argued 12. Puce, A., Allison, T., Asgari, M., Gore, J. C. & to face recognition and expert within-cat- that modular organization was restricted McCarthy, G. Differential sensitivity of human http://neur

¥ visual cortex to faces, letter strings, and textures: egory discrimination of nonfaces is that to perceptual and language systems, inten- a functional magnetic resonance imaging study. each is done by the extraction of second- sive research is now striving to discover the J. Neurosci. 16, 5205–5215 (1996). order relational features, that is, deviations organization of higher-level cognitive func- 13. McCarthy, G., Puce, A., Gore, J. & Allison, T. from the average member of the catego- tions. A vigorous debate is now addressing Face-specific processing in the human fusiform ry. The challenge for this view is to pro- the question of whether discrete function- gyrus. J. Cognit. Neurosci. 9, 605–610 (1996). vide an account of how the existence of a al components exist in the frontal lobes, 14. Kanwisher, N., McDermott, J. & Chun, M. common mechanism for the extraction of and—to the extent that they do—whether The fusiform face area: a module in human extrastriate cortex specialized for face second-order relational features is not they are organized along process-specific perception. J. Neurosci. 17, 4302–4311 (1997). 54 ruled out by the data reviewed above. or domain-specific lines, or both . It seems 15. Kanwisher, N. Stanley, D. & Harris, A. The 2000 Nature America Inc.

unlikely that all of cognition will be sub- fusiform face area is selective for faces not © CONCLUSIONS served by discrete modular mechanisms, animals. Neuroreport 10, 183–187 (1999). Ultimately, the answer to the question of and also unlikely that all modules that exist 16. Chao, L. L., Martin, A. & Haxby, J. V. Are face- whether face processing is domain specific will be domain specific. A more reasonable responsive regions selective only for faces? Neuroreport 10, 2945–2950 (1999). will require us to decide which of the many hypothesis is that the degree of modulari- 17. Tong, F., Nakayama, K., Moscovitch, M., possible criteria of domain specificity are ty and the degree of domain specificity Weinrib, O. & Kanwisher, N. Response central: first, the most common use of the within modules will vary across the brain properties of the human fusiform face area. module; second, the possible use the mod- and across aspects of cognition. Cognit. Neuropsychol. 17, 257–279 (2000). ule can be put to under some circumstances 18. Bentin, S., Allison, T., Puce, A., Perez, E. & (even if these are rare); third, the functions McCarthy, G. Electrophysiological studies of ACKNOWLEDGEMENTS face perceptions in humans. J. Cognit. for which the module is necessary (rather This commentary was supported by NIH grant 56037 Neurosci. 8, 551–565 (1996). than simply involved); fourth, the origins to NK and has benefited greatly from conversations 19. Liu, J., Higuchi, M., Marantz, A. & Kanwisher, of the module in the development of the with and comments by numerous colleagues, includ- N. The selectivity of the occipitotemporal 50,51 individual or the evolution of the species ing M. Behrmann, S. Carey, B. Duchaine, A. Harris, M170 for faces. Neuroreport 11, 337–341 (2000). or both. Face processing is likely to be N. Logothetis, M. Moscovitch, K. Nakayama, M. 20. Allison, T., Puce, A., Spencer, D. D. & domain-specific according to the criteria of Potter, P. Sinha, K. Tanaka, F. Tong and A. Wagner. McCarthy, G. Electrophysiological studies of common use and origins, because for human face perception. I: Potentials generated almost all humans (and probably also for in occipitotemporal cortex by face and non- RECEIVED APRIL 4; ACCEPTED JUNE 21, 2000 their primate ancestors), faces are the pri- face stimuli. Cereb. Cortex, 9, 415–430 (1999). mary if not the only stimulus category for 21. McCarthy, G., Puce, A., Belger, A. & Allison, T. which we are experts at making within-cat- 1. Fodor, J. Modularity of Mind (MIT Press, Electrophysiological studies of human face Cambridge, Massachusetts, 1983). perception. II: Response properties of face- egory discriminations. As argued above, specific potentials generated in occipitotemporal existing neuropsychology data argue for 2. Hirschfeld, L. A. & Gelman, S. Mapping the cortex. Cereb. Cortex 9, 431–444 (1999). Mind: Domain Specificity in Cognition and domain-specificity of face recognition Culture (Cambridge Univ. Press, Cambridge, 22. Puce, A., Allison, T. & McCarthy, G. according to the criterion of necessity. UK, 1994). Electrophysiological studies of human face 30 perception. III: Effects of top-down The car and bird expertise study tests 3. Yin, R. K. Looking at upside-down faces. processing on face-specific potentials. Cereb. the possible use of a module. However, this J. Exp. Psychol. 81, 141–145 (1969). Cortex 9, 445–458 (1999).

23. Diamond, R. & Carey, S. Why faces are and are 33. Epstein, R., Harris, A., Stanley, D. & 44. Sergent J. & Signoret, J. L. Varieties of not special: An effect of expertise. J. Exp. Kanwisher, N. The parahippocampal place functional deficits in prosopagnosia. Cereb. Psychol. Gen. 115, 107–117 (1986). area: recognition, Navigation, or Encoding? Cortex 2, 375–388 (1992). 24. Rhodes, G. & McLean, I. G. Distinctiveness Neuron 23, 115–125 (1999). 45. Gauthier, I., Behrmann, M. & Tarr, M. Can and expertise effects with homogeneous 34. Wojciulik, E., Kanwisher, N. & Driver, J. face recognition really be dissociated from stimuli: towards a model of configural coding. Modulation of activity in the fusiform face object recognition? J. Cogn. Neurosci. 11, Perception 19, 773–794 (1990). area by covert attention: an fMRI study. 349–371 (1999). 25. Damasio, A. R., Damasio, H. & VanHoesen, G. W. J. Neurophysiol. 79, 1574–1579 (1998). 46. Duchaine, B. Developmental prosopagnosia Prosopagnosia: Anatomic basis and behavioral 35. O’Craven, K. M., Downing, P. E. & Kanwisher, with normal configural processing. mechanisms. Neurology 32, 331–341 (1982). N. fMRI evidence for objects as the units of Neuroreport 17, 70–83 (2000). 26. De Renzi, E. in Aspects of Face Processing (eds. attentional selection. Nature 401, 584–587 47. Assal, G., Favre, C. & Anderes, J. P. Non- Ellis, H. D., Jeeves, M. A., Newcombe, F. & (1999). reconnaissance d’animaux familiers chez un Young, A. W.) 243–252 (Martinus Nijhoff, 36. Epstein, R. & Kanwisher, N. A cortical paysan. Zoo-agnosie ou prosopagnosie pour Dordrecht, 1986). representation of the local visual les animaux. Rev. Neurol. (Paris) 140, 580–584 27. Faust, C. Die zerebralen Herstorungen bei environment. Nature 392, 598–601 (1998). (1984). Hinterpauptsverletzungen und ihre 37. Logothetis, N. K. & Pauls, J. Psychophysical 48. Tanaka, K. Mechanisms of visual object Beurteilung (Thieme, Stuttgart, 1955). and physiological evidence for viewer- recognition: monkey and human studies. 28. Gauthier, I., Tarr, M. J., Anderson, A. W., centered object representations in the Curr. Opin. Neurobiol. 7, 523–529 (1997). primate. Cereb. Cortex 3, 270–288 (1995). Skudlarski, P. & Gore, J. C. Activation of the 49. Logothetis, N. K., Pauls, J. & Poggio, T. Shape middle fusiform ‘face area’ increases with 38. Kobatake, E., Wang, G. & Tanaka, K. Effects of representation in the inferior temporal cortex expertise in recognizing novel objects. Nat. shape-discrimination training on the of monkeys. Curr. Biol. 5, 552–563 (1995). Neurosci. 2, 568–573 (1999). selectivity of inferotemporal cells in adult monkeys. J. Neurophysiol. 80, 324–330 (1998). 50. Jacobs, R. A. Computational studies of the .com 29. Gauthier, I., Tarr, M. J., Anderson, A. W., development of functionally specialized neural Skudlarski, P. & Gore, J. Does visual 39. Gross, C. G., Roche-Miranda, G. E. & Bender, modules. Trends Cogn. Sci. 3, 31–38 (1999). subordinate-level categorization engage the D. B. Visual properties of neurons in the functionally defined fusiform face area? inferotemporal cortex of the macaque. 51. Polk, T. A. & Farah, M. The neural Cognit. Neuropsychol. 17, 143–164 (2000). J. Neurophysiol. 35, 96–111 (1972). development and organization of letter recognition: evidence from functional osci.nature 30. Gauthier, I., Skularski, P., Gore, J. C. & 40. Tanaka, K. Inferotemporal cortex and object neuroimaging, computational modeling, and Anderson, A. W. Expertise for cars and birds vision. Annu. Rev. Neurosci. 19, 109–139 behavioral studies. Proc. Natl. Acad. Sci. USA (1996). recruits brain areas involved in face 95, 847–852 (1998). recognition. Nat. Neurosci. 3, 191–197 (2000). 41. Gauthier, I. & Logothetis, N. Is face 52. Sperber, D. in Mapping the Mind: Domain 31. Gauthier, I., Anderson, A. W., Tarr, M. J., recognition not so unique after all? Cognit. http://neur Specificity in Cognition and Culture (eds. ¥ Neuropsychol. 17, 143–164 (2000). Skudlarski, P. & Gore, J. C. Levels of Hirschfeld, L. A. & Gelman, S.) 37–67 categorization in visual recognition studied 42. Henke, K., Schweinberger, S. R., Grigo, A., (Cambridge Univ. Press, Cambridge, UK, using functional magnetic resonance imaging. Klos, T. & Sommer, W. Specificity of face 1994). Curr. Biol. 7, 645–651 (1997). recognition: recognition of exemplars of nonface objects in prosopagnosia. Cortex 34, 53. Pinker, S. How the Mind Works (Norton, New 32. Kanwisher, N., Downing, P., Epstein, R. & York, 1997). Kourtzi, Z. in The Handbook on Functional 289–296 (1998). Neuroimaging (eds. Kingstone, A. & Cabeza, 43. McNeil, J. E. & Warrington, E. K. 54. Wagner, A. D. Working memory R.) (MIT Press, Cambridge, Massachusetts, in Prosopagnosia: a face-specific disorder. contributions to human learning and press). Q. J. Exp. Psychol. A 46, 1–10 (1993). remembering. Neuron 22, 19–22 (1999). 2000 Nature America Inc.

nature neuroscience • volume 3 no 8 • august 2000 763