Four main issues have been raised in analyses of whether face recognition is better seen as an innate ability or as a learned skill (Nachson, 1995; Tanaka & Gauthier, 1997). They are: (a) empirical findings indicating that a preference for faces may be innate, (b) empirical evidence that there are cortical cells that respond specifically to faces, (c) findings that face recog- nition is disproportionally impaired by inversion, and (d) the existence of a face-specific neuropsychological deficit—prosopagnosia (also known as
face agnosia), a disorder that results from cerebral injury and involves the
inability to recognize familiar faces (e.g., see Bruyer, 1989). Research evi- dence on each of these points is briefly reviewed.
Infants’ Preference for Faces
It is well known that infants perceive faces as compelling stimuli, and early research (e.g., Fantz, 1961) showed that infants preferred to look at human faces more than other stimuli, such as a bull’s eye, colored circles, or newsprint. Recognition is important for the development of attachment between infants and their caregivers (Bowlby, 1969). Fantz’s findings im- plied that infants are born with some form of pattern vision, supporting the view of Zuckerman and Rock (1957), who argued that “perceptual or- ganization must occur before experience . . . can exert any influence” (p. 294). Subsequent findings have questioned this assumption, however. Studies of infant vision have shown that 1-month-old infants can obtain only limited visual information from observing a face: the outer contour as defined by the hairline, and vague darker areas in the region of the mouth and eyes (Souther & Banks, 1979). Other studies have shown an “externality effect,” a preference for attending to the boundaries of stim- uli, in the first 2 months of life (e.g., Bushnell, Gerry, & Burt, 1983). Fur- ther, Maurer and Barrera (1981) demonstrated that whereas 1-month-old infants looked with equal interest at intact and scrambled schematic faces, 2-month-olds looked longer at the face-like configuration. Other work has suggested that a preference for face-like configurations is first visible at 4 months of age (Haaf, 1977; Haaf, Smith, & Smitty, 1983). Analyzed to- gether, this body of findings suggests that infants do not respond differen- tially to faces until they are at least 2 months old, which is in contrast to the assertion that there are innate perceptual preferences for faces or face- like stimuli.
However, other work supports the existence of innate perceptual pref- erences. At a conceptual level, it has been asserted that because recogniz- ing others is of vital importance to people, “it makes sense to postulate a selective evolutionary pressure to evolve neural mechanisms specifically for the recognition of faces” (Nachson, 1995, p. 257). Goren, Sarty, and Wu (1975) found that newborn infants (median age 9 min) showed more inter- est in a moving schematic face pattern, as measured by head and eye movements, than in scrambled “faces” or a blank head outline.
Cortical Cells That Respond Specifically to Faces
The existence of cortical cells that respond specifically to faces has been supported by the discovery of neurons in the monkey’s visual cortex that respond predominantly to faces (Leonard, Rolls, Wilson, & Baylis, 1985; Rolls, 1992; Rolls, Baylis, & Leonard, 1989). Each neuron does not respond only to one face. Instead, each neuron has a different pattern of responses across a set of faces. The recordings in these studies are made primarily
with nonhuman primates because the temporal lobe, the site of this proc- essing, is more developed than in nonprimates (Rolls, 1992). It has been found that the neurons respond to human faces as well as monkey faces, and to facial photographs as well as live faces. The response of these neu- rons did not change when faces were presented sideways or inverted, or when size was changed or color was modified (Damasio, 1989). Evidence indicates that some of these neuron responses were altered by experience so that new stimuli became incorporated in the neural network (Rolls, 1992). When components of faces (e.g., mouth, eyes) are presented in iso- lation, different neurons respond to various components. It has also been found that some neurons are specialized for face recognition and for the decoding of facial expressions (Rolls, 1992).
Effects of Facial Inversion
The third set of findings that support the modularity hypothesis involve the effects of facial inversion. Generally, faces are recognized more easily than any other class of stimuli that are as similar to one another in their con- figuration. But, turn them upside down (invert them), and faces become harder to distinguish than other classes of inverted stimuli such as air- planes, stick figures, and houses. This effect was first reported in a series of studies by Yin (1969, 1970, 1978). Scapinello and Yarmey (1970) reported a similar pattern of results. Yin argued that the unique reversal of recognition accuracy for faces, from best upright to worst inverted, indicated that face recognition was the product of a system distinct from the one used for rec- ognizing other types of visual stimuli. Yin suggested that neural special- ization had evolved to support a process specific to human faces.
Subsequent research, however, has questioned Yin’s explanation of these findings. In several experiments Diamond and Carey (1986) showed that the inversion effect found for faces was not unique, but was appar- ently due to expertise with the stimulus materials. When people are expe- rienced in perceiving stimuli, such as human faces, inverting those stimuli causes a major disruption in encoding. They found that expert dog breed- ers showed the same inversion effect for dog faces (for the breed for which each was an expert) that people show for human faces. Diamond and Carey argued that experts, people who are accustomed to discriminating between highly similar objects from the same class (e.g., human faces, or dog faces from a single breed for dog breeders), rely on extracting “sec- ond-order relational features” in order to make these complex discrimina- tions. This process is most disrupted by inversion. They posited that faces are processed by utilizing a holistic strategy, rather than a piecemeal strat- egy that examines distinctive features of the stimulus object. When an ex- pert’s normal processing strategy for a class of objects (e.g., a holistic strat-
egy for faces) is disrupted by inversion, face recognition is at a particular disadvantage because the holistic strategy works poorly for inverted ob- jects.
Therefore, what first seemed to be a face-specific effect instead appears to be an “expertise-specific effect” (Cohen-Levine, 1989; Nachson, 1995). It is also true, however, that face recognition appears “special” because it in- volves more configural processing than what is necessary for the recogni- tion of other objects (Tanaka & Farah, 1993).
Prosopagnosia
The fourth set of research findings cited as supporting the modularity po- sition is derived from the study of clinical cases of prosopagnosia, the in- ability to recognize familiar faces as the result of brain injury. This disor- der is generally correlated with bilateral lesions located either in the inferior occipital region or anteriorly in the temporal region of the brain (Damasio, Tranel, & Damasio, 1990). Prosopagnosic persons may be un- able to recognize faces of friends, family members, or even their own face in a mirror, though they may be able to recognize others on the basis of gait, clothing, voice, or context (Bruyer, 1989). Because the ability to recog- nize other visual objects and words remains relatively intact, this pattern implies that face recognition constitutes a special, unique system that can be damaged in isolation.
However, in-depth study of prosopagnosia has revealed problems in recognizing other classes of objects as well (Bruyer, 1989; Damasio, Da- masio, & van Hoesen, 1982; Ellis, 1975). Damasio, Tranel, and Damasio (1990) noted that these patients have consistently demonstrated difficulty in identifying unique stimuli viewed previously from a group of objects that share certain configural properties (e.g., houses, cars, pets, articles of clothing). Hence, from the standpoint of the skill hypothesis, it might be argued that prosopagnosia involves a disruption in individuals’ ability to interpret configural information leading to recognition of unique items (Damasio et al., 1990). Studies measuring subtle “covert identification re- sponses” (either cognitive responses or psychophysiological autonomic identification responses) have found that prosopagnosics may show sub- tle recognition response patterns when observing familiar faces, even though they appear unable to recognize them (Bauer, 1984; Bruyer et al., 1983; Tranel & Damasio, 1985).
Several cases have been described, notably by Bruyer et al. (1983) and DeRenzi (1986), that appear to be instances of the “pure” syndrome, in which the only perceptual deficit was for face recognition. Ellis and Young (1989) asserted that, “for the moment we are inclined to accept that prosopagnosia can occur in a form undiluted enough to warrant the view
that it is a distinct cognitive deficit which could only arise from the exis- tence of a cognitive system containing functional components specific to face recognition” (p. 14). Addressing this same issue, Bruyer (1989) ob- served that, “All that can be said at this time is that the question is still open” (p. 455).
DEVELOPMENT OF FACE RECOGNITION ACROSS