Can We Lose of ? Content Specificity and Awareness in a Prosopagnosic

Nancy L. Etcoff Department of and Cognitive Sciences Massachusetts Institute of Technology Neuropsychology Laboratory Massachusetts General Hospital Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/3/1/25/1755723/jocn.1991.3.1.25.pdf by guest on 18 May 2021 Roy Freeman Division of New England Deaconess Hospital Beth Israel Hospital Harvard Medical School Kyle R. Cave Department of Psychology University of California, San Diego

Abstract H Prosopagnosia is a neurological in which patients nonfacial channels. The only other categories of shapes that he cannot recognize faces. Kecently it has been shown that some has marked trouble recognizing are animals and emotional prosopagnosics give evidence of “covert” recognition: they expressions, though even these impairments were not as severe show greater autonomic responses to familiar faces than to as the one for faces. Three measures (sympathetic skin re- unfamiliar ones, and respond differently to familiar faces in sponse, pupil dilation, and learning correct and incorrect learning and interference tasks. Although some patients do not names of faces) failed to show any signs of covert recog- show covert recognition, this has usually been attributed to an nition in LH, though the measures were sensitive enough to “apperceptive” deficit that impairs perceptual analysis of the reflect autonomic reactions in LH to stimuli other than faces, input. The implication is that prosopagnosia is a deficit in access and face familiarity in normal controls. Thus prosopagnosia to, or awareness of, memories of faces: the inducing brain cannot always be attributed to a mere absence of awareness does not destroy the memories themselves. We present (i.e., preserved information about faces whose output is dis- a case study that challenges this view. LH suffers from proso- connected from conscious cognitive processing), to an apper- pagnosia as the result of a closed . He cannot rec- ceptive deficit (i.e., preserved information about faces that ognize familiar faces or report that they are familiar, nor answer cannot be accessed due to improperly analyzed perceptual questions about the faces from , though he can (1) input), or to an inability to recognize complex or subtly varying recognize common objects and subtly varying shapes, (2) match shapes (i.e., loss or degradation of shape memory in general). faces while ignoring irrelevant information such as emotional We conclude that it is possible for brain injury to eliminate the expression or angle of view, (3) recognize sex, age, and like- storage of information about familiar faces and certain related ability from faces, and (4) recognize people by a number of shapes. H

Prosopagnosia, a term introduced by Bodamer in 1947, patient continues to recognize familiar voices, gaits, and is a disorder of visual recognition of human faces. In the so on. It is generally believed that bilateral damage to usual case, the patient cannot recognize the faces of the medial occipitotemporal regions is necessary for this individuals known prior to the onset of illness, nor those deficit to occur and all patients whose have been met since. The overt deficit is absolute in that the face examined postmortem have been found to have bilateral does not elicit any sense of familiarity. Recognition of damage (Damasio, Damasio, & Van Hoesen, 1982; Da- people proceeds undisturbed via other channels, and the masio, Damasio, & Tranel, 1986). However, as with all

0 I991 Masrachusetts Institute of Technology Journal of Cognitive Neuroscience Volume 3, Number 1

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.1991.3.1.25 by guest on 28 September 2021 other aspects of prosopagnosia, the neuroanatomy is the cortex of monkeys have shown small populations of neu- subject of controversy: prosopagnosics with lesions only rons sensitive to faces (e.g., Perrett, Smith, Potter, Mislin, in the right hemisphere as revealed by CT scan have Head, Milner, & Jeeves, 1984) and thus there exists the been reported (DeRenzi, 1986; Landis, Cummings, Chris- possibility that faces have a distinct and separate repre- ten, Bogen, & Imhof, 1986; Michel, Perenin, & Sieroff, sentation in the brain. 1986). The second question is one of more recent origin. In this investigation we focus on two other unresolved Within the past 5 years, several laboratories have shown questions: the extent to which prosopagnosia is a dis- that prosopagnosics give evidence of selective sensitivity order limited to the recognition of faces as opposed to to familiar faces if the tasks do not require explicit rec- a more generalized deficit in recognizing objects or ob- ognition. For example, Tranel and Damasio (1985) found jects within a category, and the extent to which it rep- that galvanic skin responses were greater to familiar than resents a disorder of awareness of recognition rather to unfamiliar faces in the prosopagnosics they tested (see than of recognition itself. also Bauer, 1984, 1986; Tranel & Damasio, 1988). Covert Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/3/1/25/1755723/jocn.1991.3.1.25.pdf by guest on 18 May 2021 The first is a basic question, for if prosopagnosics are recognition has also been demonstrated using behavioral not selectively disturbed at face recognition, why suggest measures of face matching, interference, and priming a separate disorder at all, or at least why call it pro- (e.g., Bruyer, Laterre, Seron, Feyereisen, Strypstein, Pier- sop(face)? One might argue (as some have) that rard, & Rectem, 1983; De Haan, Young, & Newcombe, faces pose a difficult perceptual problem; they are harder 1987;Young, Hellawell, & De Haan, 1988), where familiar to discriminate than most other objects in the visual and unfamiliar faces have different effects. Covert rec- world. Perhaps prosopagnosia is a very mild form of ognition has also been reported via the measurement of , one which disrupts only the most difficult P300 amplitude of brain potentials in a prosopaganosic visual discriminations. If this was the case, one would (Renault, Signoret, Debruille, Breton, & Bolger, 1989). predict that patients might exist for whom face recogni- These results raise several questions. Will most pro- tion was selectively disturbed but that patients should sopagnosics demonstrate these effects, suggesting that not exist for whom object recognition is disturbed in the prosopagnosia is mainly a disorder of access to conscious presence of preserved face recognition. This “mild form awareness? (See Schacter, McAndrews, & Moscovitch, of visual agnosia” argument can be refuted by the pres- 1988 for a description of various disorders of this type.) ence of case reports where object do exist in If so, how can we explain prosopagnosics who do not patients with preserved face recognition (Hecaen, Gold- demonstrate covert effects? A few such cases have been blum, Masure, & Ramier, 1974; Ferro & Santos, 1984) reported (Bauer, 1986; Newcombe, Young, & De Haan, and, to a lesser extent, by studies where face recognition 1989; Young & Ellis, 1989; Sergent & Villemure, 1989) is not fully preserved but object recognition shows far and they all have been interpreted in a similar way: that greater impairment (Albert, Reches, & Silverberg, 1975; these patients have perceptual impairments that disrupt McCarthy & Warrington, 1986; see review in Young, their ability to construct representations of faces, so that 1988). incoming information about the physical properties of On the other hand, virtually all prosopagnosics have the face are not specific or accurate enough to activate been reported to have difficulties making at least some stored face representations. In fact, the presence or ab- other within-category visual discriminations (the one sence of covert recognition in prosopagnosics has been notable exception being the patient described by De- suggested as an “index” (Newcombe et al., 1989) of Renzi, 1986). The categories posing difficulties tend to which type of prosopagnosia a patient might have: one be “natural-kind’’ categories, such as animals or foods, where the problem is mainly perceptual in nature, or but there are also reports of difficulties recognizing cars, one where the problem is mainly one of memory for coins, or other sets of similar-appearing objects (e.g., face information. Bornstein, 1963; Damasio et al., 1982). Perhaps, then, The distinction between perceptual and mnestic forms prosopagnosia reflects a more generalized disorder of of agnosia dates back to Lissauer (1890) who proposed individuating members of a visually similar class of ob- a division of the agnosias into “apperceptive” and “as- jects. Furthermore, faces vary continuously along multi- sociative” forms. In “apperceptive” agnosia there is an ple quantitative dimensions. A more refined version of impairment in visual beyond the level of the within-class hypothesis would be that prosopagnosia primary sensory functions (such as visual field cuts). is not specific to facelike shapes but is a difficulty in “Associative” agnosics perceive adequately but do not distinguishing objects that share the same global shape recognize what they perceive-as Teuber (1968) re- but differ quantitatively along continuous dimensions marked, there is “a normal percept stripped of meaning.” (Levine & Calvanio, 1989). Finally, of course, there is the By the classic definition of these two types of agnosias, possibility that prosopagnosia is correctly defined: that the apperceptives are those that cannot copy or match there is a disorder of visual recognition confined to faces, visual forms, and the associatives can copy and match a class of objects with great biological significance. In- forms but cannot associate them with knowledge that vestigations of the activity of neurons in the temporal would allow them to name or define the category they

26 Journal of Cognitive Neuroscience volume 3, Number I

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.1991.3.1.25 by guest on 28 September 2021 fall into. More recently, Warrington and Taylor (1978) proposed a division between patients with “perceptual categorization” impairments and those with “semantic categorization” impairments. They define “perceptual categorization’’ as a “postsensory but presemantic” pro- cess whereby two stimulus inputs are given equivalent geometric descriptions. Such patients are impaired at recognizing and matching objects photographed at un- conventional views, and are believed therefore to have trouble extracting viewpoint-invariant shape representa- tions from two-dimensional (2-D) objects (see Marr, 1982;Pinker, 1984). Patients impaired at the second stage, that of “semantic categorization,” are poor at linking semantic knowledge to the representation. Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/3/1/25/1755723/jocn.1991.3.1.25.pdf by guest on 18 May 2021 Although different researchers have used different cri- Figure 1. Right sagittal TI weighted MRI image teria for categorizing a patient as an “apperceptive” ver- sus an “associative” prosopagnosic, all have agreed that patients who do not show covert recognition are in the former category whether it be defined by the presence of a more general object agnosia, by an inability to copy or match objects, by problems in recognizing noncanon- ical views of objects, or, more specifically, by problems recognizing faces transformed in some way (either in angle of view or expression or in the way the faces are illuminated). If this association between apperceptive deficits and lack of covert recognition turns out to be true of all cases, it carries with it an interesting consequence. It implies that prosopagnosia is inherently a deficit of awareness, with preserved knowledge of faces; the only reason that someone can lack covert recognition is that access to a presumably intact store of face memories is disturbed by imperfectly analyzed perceptual input. Fur- Figure 2. sagittal T1 weighted MRI image. ther, the implication would be that no lesion can actually Left eliminate the stored information specific to faces, which would in turn suggest that such information is widely tex, and dilation of the temporal horn of the lateral distributed through the brain. Localized lesions could ventricle (Fig. 2). The T2 weighted axial image at the only knock out the perceptual input to face memories level of the demonstrates extensive loss of (apperceptive prosopagnosia) or the output from face brain tissue in the right anterior including memories to the cognitive systems underlying awareness partial injury to the and hippocampus. There (prosopagnosia with covert recognition). In this paper, is relative sparing of the medial temporal lobe. The left we study a patient who provides a critical test of these temporal horn is dilated and there is white matter injury ideas. in the paratrigonal region. A shunt artifact is present (Fig. 3). An axial proton density weighted image at the level of the bodies of the lateral ventricle shows abnormal CASE DESCRIPTION signal in the paraventricular white matter in the right LH, a 40-year-old ordained minister and social worker, frontal and parietooccipital region and to a lesser extent suffered a severe closed head injury in an automobile in the left parietooccipital region (Fig. 4). In sum, LH has accident at the age of 18. The accident and the surgical bilateral lesions affecting visual association cortices and procedures it necessitated (a right anterior temporal lo- the subjacent white matter. These sites include the right bectomy and insertion of a ventriculoatrial shunt) re- temporal lobe, the left subcortical occipitotemporal sulted in the pattern of illustrated in white matter, and bilateral parietooccipital regions. Figures 1-4. These images were taken from an MRI scan LH’s visual acuity is 20/50 in the left eye and 20/70 in done in 1989. The right sagittal T1 weighted image shows the right. He has a left homonymous superior quadran- extensive atrophy and brain loss in the anterior temporal tanopia, a partial right inferior quadrantanopia, and some and’frontal cortex (Fig. 1). The left sagittal T1 weighted diffuse shrinkage of the remaining peripheral fields. image shows atrophy, preservation of the temporal cor- Color vision is abnormal as is color imagery (see Levine,

Etcog et al. 27

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.1991.3.1.25 by guest on 28 September 2021 et al., 1980, 1985) revealed a remarkably intact intellect. His WAIS Verbal IQ of 132 was in the 2nd-3rd percentile for adults his age, and his Performance IQ of 93 was in the average range. He achieved a Memory Quotient of 121 on the Wechsler Memory scale indicating superior performance. He has no discernible language deficits, writes normally, and can read, albeit slowly. He draws excellent copies of objects and complex drawings. His only striking deficit is his inability to recognize a face. Although LH can recognize most pictures of objects and most objects encountered in daily life, he is unable to recognize the faces of his wife, children, friends, or mem- bers of his family of origin. Recognition of individuals via other channels such as their voices remains intact, as Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/3/1/25/1755723/jocn.1991.3.1.25.pdf by guest on 18 May 2021 does his retention of biographical information about in- dividuals he does not recognize visually. The following report is divided into two sections. In the first, we describe a series of studies investigating LH’s ability to derive information from faces, and his ability to perform seemingly related perceptual and memory tasks. The aim is to determine the extent to which LH’s Figure 3. T2 weighted axial MRI image at the level of the mid-brain. face agnosia is embedded in a more general perceptual or . Second, we describe three tests of covert face recognition designed to test. the idea that prosopagnosia is a disorder of visual awareness, a con- dition that can be unmasked if recognition is tested im- plicitly rather than explicitly.

TESTS OF FACE, PERSON, AND OBJECT RECOGNITION Visual Face Recognition Recognizing Familiar Faces LH likens the experience of looking at a face to attempt- ing to read illegible handwriting: you know that it is handwriting, you know where the words and letters stop and start, but you have no clue as to what they signify. LH is not able to identify photographs of individuals known before his accident (his mother, his closest high school friends) or since (his wife, children, and co-work- ers). Photographs of these individuals were mixed in with a set of photographs of strangers of similar age, and LH was asked to rate the relative familiarity of the faces Figure 4. Axial proton density weighted MRI image at the level of on a 1-5 scale, with 5 being familiar. The familiar and the bodies of the lateral ventricle. unfamiliar faces were given the same mean rating of 1.3. The face that seemed most familiar to him (he gave it a 3) was that of a stranger whom he thought might be a Calvanio, & Wolf, 1980; Levine, Warach, & Farah, 1985 friend of his wife’s. for details of visual testing). He has no nystagmus, and LH showed a similar profound impairment when asked pursuit and saccadic extraocular eye movements are nor- to identify pictures of famous individuals. Shown 79 pho- mal.. His pupils are 4 mm, equal in size and reactive to tographs of famous people taken from newspapers from light. the 1920s to the 1970s (Marslen-Wilson & Teuber, 1975) LH responded in heroic fashion to his . He LH identified only 5, for an accuracy rate of 7% (exclud- . completed his studies at an Ivy League university (he was ing ones he showed no knowledge of when given their a freshman at the time of his accident) and attained two names). Two of these had distinctive hairstyles (the Bea- Master’s degrees. Testing done in 1975 and 1976 (Levine tles and Einstein) which he called “dead giveaways.” For

28 Journal of Cognitive Neuroscience Volume 3, Number I

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.1991.3.1.25 by guest on 28 September 2021 the others he used nonface cues; for example, when he whether the photograph depicted a possible human face identified Hitler he said-“I see the Third Reich insignia, or not. the swastika, so it must be Hitler . . . you wouldn’t be This provides a stringent test of recognition of faces mean enough to show me an underling like Goebbels.” as a category, since information about the number of Similarly, he recognized Mickey Mantle and Roger Maris features and their position above or below one another because they wore Yankee uniforms, while misidentify- is not enough for recognition. LH was able to distinguish ing Babe Ruth as “Mantle or Maris” because he, too, wore a true from an impossible face with 97% accuracy on the the Yankee uniform. Interestingly, LH was able to rec- upright and 94% accuracy on the inverted faces. He was ognize many clothing and background cues and there- always able to recognize a normal face and to recognize fore guess the likely occupation of the person, yet this that a face with scrambled features was not a face. His did not improve his performance except in the few cases only errors occurred on the faces with inverted features, above. In contrast, he was able to give precise biograph- where he made one error when faces were upright (88% ical information about 73 of these 79 figures when told accuracy) and two errors when faces were inverted (75% their name, although they continued to look unfamiliar accuracy). LH’s performance on this task is comparable Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/3/1/25/1755723/jocn.1991.3.1.25.pdf by guest on 18 May 2021 to him. to that of normal adults in the faces-upright condition, Finally, LH’s face agnosia did not improve when we and actually better than that of normal adults in the faces- gave him a test directly tapping his memory represen- inverted condition. Normal adults usually perform at tations of faces and bypassing his ability to perceptually chance when features are inverted and the entire faces analyze pictures of faces. We asked him 20 face imagery are shown upside down (Carey & Diamond, unpublished questions about famous individuals. Each of the ques- data). tions involved a forced choice, for example, which in- LH can truly recognize faces as faces, and is sensitive dividual had thicker lips or curlier hair. Some of the not just to gross information such as number of features questions involved an individual feature; others involved and relative placement, but to subtler relational infor- facial structure (such as which figure had a narrower face mation about feature orientation. However, the fact that or closer set eyes). After giving the test, we presented he is less subject to the usual illusion created by inver- the name of each celebrity to LH to be sure that he was sion does suggest that he may not be perceiving the face familiar with the individual; one trial needed to be elim- as a gestalt, a process which leads to an immediate per- inated for this reason. LH performed at chance on the ception of grotesqueness when the face is upright but imagery task: 47% correct. Two age-matched normal con- masks it when the faces are inverted (Thompson, 1980; trols performed this task at 85% and 90% accuracy. The see also Carey, 1982). This would also accord with his result suggests that LH’s difficulties recognizing faces behavior. LH slowly studied each face in both conditions, cannot be reduced entirely to problems of perceptual and was not struck by the disparity so remarkable to analysis of pictures, as the difficulty persists when long- normal perceivers. term representations are tapped through imagery. Matching Pictures of Faces Recognizing the Face Categoy LH scored 36/54 on the Benton-Van Allen face matching It is commonly accepted that prosopagnosics are always task (Benton & Van Allen, 1968), which puts him into the aware that a face is a face but have trouble recognizing category the authors call “Moderate Impairment,” based its specific identity. In an experimental test of this idea, on their test norms. He was 100% accurate when the Bruyer et al. (1983) showed that a prosopagnosic patient target and test faces were identical, 71% correct when was able to discriminate a human face from other com- the test and target differed in angle of view, and clearly plex visual stimuli such as animal faces, car fronts, and impaired (54% correct) when the test and target differed house fronts. in lighting and appeared fragmented and silhouetted. We investigated LH’s ability to recognize human faces Further evidence that LH is unimpaired at matching iden- in a task that required specific knowledge about facial tical views of faces was provided when we presented him structure. We presented him with 32 faces, 16 of which with our own face-matching task. LH was shown a black were normal, 8 of which contained duplicate features or and white photograph taken from a college yearbook, features in the wrong place such as the nose above the and two faces, one of which was the same and the other eyes, and 8 of which had either the eyes or the mouth a foil. The foils were from the same yearbook and (4 pictures) or both the eyes and the mouth (4 pictures) matched for pose, hairstyle, and hair color. LH matched inverted (see Thompson, 1980). Normal faces also had these faces at 90% accuracy. the eye -and mouth areas cut out and then replaced We then tested LH’s ability to attend selectively to the normally oriented, in order that the faint lines resulting identity of a face despite changes in its appearance due from the cutting would be present in all faces. Thirty- to facial expression, and his ability to attend selectively two Faces were presented upright, and then 32 new faces to the expression of the face despite differences in the were presented upside down. The task was to indicate exact rendering of that expression across individuals. We

Etco& et al. 29

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.1991.3.1.25 by guest on 28 September 2021 used the Garner sorting procedure (Garner, 1974), de- Although LH is able to perceive these faces quite well, scribed in Etcoff (1984): LH was first shown pictures of he cannot retain that knowledge for even 5 sec. two women displaying two different facial expressions (happy and sad) and asked how many people there were Recognizing Other Kinds of Information from a and how many expressions there were. Unlike many of Face: Age, Sex, Likeabilig, Expression the patients with right hemisphere damage that Etcoff (1984) studied previously, LH found this task easy, im- LH can recognize age and sex from facial appearance. mediately seeing them as two individuals, as women, and We presented him with color photographs of two sets of as conveying a happy and a negative expression (he was faces of individuals ranging from their teens to their not sure whether the expression was angry or sad). He fifties and asked him to order them from youngest to was then asked to sort sets of 32 photographs into two oldest. Each set contained one face from each decade. piles. On different trials, the two piles corresponded None of the individuals had grey hair or was balding. LH either to the two expressions or the two women. Each performed this task without error. To test discrimination of these types of sorts was done equally often with decks of sex, we showed LH a set of 30 black and white pho- Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/3/1/25/1755723/jocn.1991.3.1.25.pdf by guest on 18 May 2021 of cards whose dimensions were correlated (woman A tographs of males and females taken from a college year with expression A, woman B with expression B), orthog- book. The faces were chosen so that hairlength and style onal (woman A with expression A, woman A with ex- would not provide a reliable cue as to sex (many of the pression B, woman B with expression A, and woman B males had long hair; many of the females had short hair) with expression B), or with one of the dimensions con- and the pictures were cropped just below the chin so stant (woman A with expression A or B; woman A or B that the neck and clothing were not visible. Three male with expression A). LH did this task slowly, but with high control subjects ranged from 70 to 91% correct on this accuracy, making errors on less than 1% of trials. Inter- task. LH got 70% correct, indicating that he is at the low estingly, he was not integral: he did not show “orthog- end of the normal range. onal interference” for identity or emotion. That is, the LH also appears to be able to derive information about means for the correlated, constant, and orthogonal trials a face’s pleasantness and likeability. We showed LH and for identity were 84.7, 86.3, and 83.3 sec, respectively. a control subject matched for age and sex a set of 76 For emotion, the means were 62.3, 57.7, and 59.3 sec. faces taken from a college yearbook. Fifty-two of these Thus, LH was able to disentangle these two sorts of facial faces had a neutral expression and 24 were smiling. We information: he was highly accurate and was not slowed asked LH and the control to rate these faces on a 1 to 7 down when he had to sort faces differing in expressions scale for how “appealing or likeable” the person ap- or when he had to sort emotions rendered across dif- peared to be. We then did a multiple correlation analysis ferent faces. This stands in contrast to the group of pa- on this data with presence/absence of smile and the tients with unilateral right hemisphere damage studied control subject’s data as the independent variables, and by Etcoff (1984), who had difficulty selectively attending LH’s data as the dependent variable. We found a signifi- to these two sorts of information. Although LH was slow, cant effect of smiling (p C.007);both LH and the control he was actually a bit faster than that group of FWpatients saw smiling faces as more pleasant. However, even after when sorting identity (their mean was 91.1 sec) and far this was accounted for, there remained a marginally sig- faster when sorting emotions (their mean was 92.3 sec). nificant predictive effect of the control subject’s ratings Most interesting, his pattern of results most resembled (p c.06). That is, like the control subject, LH was very those of the control groups in that study. influenced by facial expression, but, independent of that, whatever cues the control subject used to rate these faces were correlated with the criteria used by LH. His behav- ior outside the laboratory is consistent with these data: Remembering Pictures of Faces LH speaks of people’s faces in evaluative terms, for ex- We gave LH a memory version of the face-matching task ample, describing a co-worker as an “attractive person described above. Twenty black and white photographs all the way around; very appealing smile” and one of his of faces taken from college yearbooks were presented to physicians as “distinguished looking.” LH one at a time, and he had to choose the correct face LH was poor, however, at recognizing facial expres- from two photographs presented 5 sec later. LH per- sions of emotion. He was shown three sets of seven facial formed at 65% accuracy, which is not a statistically above expressions (happiness, surprise, anger, sadness, fear, chance performance, and far below his performance on disgust, and neutral) taken from Ekman and Friesen’s the face-matching version. Since we used different faces Pictures of Facial Meet (1975). He only recognized hap- in this test than in the face-matching task, we had to rule piness reliably. When the photographs of one model out the possibility that these faces were simply more were placed before him and he was asked to place the . difficult for LH to discriminate. To do this, we repeated seven expressions of the other two models with the the test with all three faces presented simultaneously. LH corresponding emotional expressions of this model, he achieved 90% accuracy, as he did on the earlier version. made only two errors; this shows that he is able to

30 Journal of Cognitive Neuroscience Volume 3, Number I

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.1991.3.1.25 by guest on 28 September 2021 discriminate among the different expressions and rec- glass, bowl, bottle; pitcher, cup, frying pan, pot; pipe, ognize them across different individuals. His problem spoon; picket fence, comb; desk, dresser; watering can, was not just one of finding a verbal label (an anomia for kettle; vase, wine glass; and toothbrush, hairbrush, facial expressions) as he also performed very poorly on broom. He was 75% accurate in recognizing the mem- a task of matching facial expressions with corresponding bers of the group peach, orange, pear, cherry, lemon, emotional tones of voice. Since he is very accurate at apple, tomato, potato; and he was 80% accurate in nam- recognizing emotion through tone of voice and since ing the members of the group cigar, cigarette, chisel, bat, this task does not require verbal labeling, one must screwdriver, nail, needle, asparagus, banana, knife, pen, assume that LH’s problems recognizing facial expressions pencil, rolling pin, nail file. In contrast, he misidentified are in making the conscious link between a visual display 15 of 30 four-footed animals, 5 of 8 birds, and 4 of 8 and his knowledge about specific emotions (a detailed insects, with animal misidentifications comprising 51% discussion of LH’s performance on this and other emo- of total errors. LH could not name pictures of animals tion tasks will be presented in a forthcoming paper). even with unique parts (kangaroo pouch, rabbit ears). The results both suggest that LH’s deficit is not exclusive Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/3/1/25/1755723/jocn.1991.3.1.25.pdf by guest on 18 May 2021 to human faces (it extends in a milder form to animals) Recognition of People Other Channels via but that it cannot be characterized as a general deficit in LH’s ability to recognize people via routes other than the making visual distinctions within a category (see Etcoff, visual is intact. He says that he recognizes people’s voices, Tarr, & Carey, 1991). and formal testing confirmed this: he was correct 90% of the time on a test of Famous Voice Recognition (Van Dkcriminating Real porn Unreal Objects Lancker & Kreiman, 1987). He also performed within the and Animals normal range on a task we devised of imagery for famous voices, analogous to the imagery for famous faces task Further evidence of a marked difference between LH’s described earlier. LH was presented with 20 questions ability to recognize objects versus animals was provided about the voices of celebrities, in which he had to choose by his performance on an object decision task (supplied which of two individuals had more or less of a particular by Dan Bub). LH was shown line drawings of 45 animals, voice quality (e.g., which individual had a more nasal and then 24 objects. His task was to indicate whether the voice, or breathier voice, or spoke more quickly). LH got pictures depicted a real animal, in the first task, or a real 85% of these questions correct. Three age-matched male object, in the second. Twenty drawings of real animals controls were 80-95% accurate on this task. Although LH and 25 drawings of nonexistent animals were presented reports relying mainly on individuals’ voices to achieve in part one. The nonexistent animals were constructed recognition, he also uses other cues. He once recognized by placing the head of one animal together with the body the first author by the sound of her gait, and claims to of another (e.g., a rabbit head with a mouse body). The recognize individuals at work by the smell of their per- heads and bodies were taken from the real animals hme. When forced to use the visual route, LH will at- shown in the set. The nonexistent objects were con- tempt to identify distinctive hairstyles or clothing. structed in the same way. LH performed at chance on the animal decision task (judging 70% of the animals as Recognition of Objects and Animals real and 68% of the chimeras as real, for an overall accuracy rate of 49%), and, strikingly, only twice recog- Recognizing Real Objects and Animals nized that the bodies and heads were mismatched. This LH says that he has some difficulty identifying animals, a occurred when isolable discrete features occurred in fact he discovered when he could not identify his chil- impossible combinations (e.g., a camel’s hump with an dren’s animal crackers. Our tests of visual recognition antlered head). After he completed this task, we asked outside the faces domain uphold this observation: LH is LH to label the animals he judged as real, and to say why able to recognize most common objects, even discrimi- he rejected the others. We did not find any tendency to nating objects within a category quite well, but has no- favor the body over the “face” or head on his false alarms. ticeabk difficulty recognizing animals. For example, Indeed, for several of his false alarms he created whole when shown the 260 black and white line drawings of new animals (e.g., a pig’s head with a rabbit’s body was animals, foods, and common objects published by Snod- called a “fat dog”). He also made many errors on true grass and Vanderwart (1980), LH named 82% correctly. animals, rejecting 7 of the 20, and correctly labeling only Within all the categories but animals (fruits, vegetables, 10 of 20. Interestingly, Farah, Hammond, Levine, and musical instruments, vehicles, and so on), he was 82- Calvanio (1988) found LH to be deficient in an animal 100% accurate. Furthermore, we created 10 subsets of imagery test (deciding whether or not a given animal objects with similar global shapes and composed of sim- had a long tail), but normal on a task tapping nonvisual ilar parts and found LH to be relatively unimpaired in knowledge about these animals. recognizing objects within these categories. LH was 100% In contrast, LH was 85% accurate on the object deci- accurate in naming the members of the following groups: sion task, judging 87% of the objects as real and only

Etcoff; et al. 31

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.1991.3.1.25 by guest on 28 September 2021 11% of the nonobjects as real. Equally of interest, he were presented one at a time on a computer screen and performed this task quickly, and saw the humor and he identified them by pressing one of three labeled keys. surreal quality of the nonobjects. For example, shown a To further ensure that recognition could not be based combination gun-trumpet, he laughed and said “talk on some local feature, the other four shapes from the about firing out a song!” set were randomly mixed in with the series, and LH was required to press a foot pedal when he saw one. Each of the three studied shapes was presented 24 times; each Spatial Layouts of the four distractors was presented six times. LH We gave LH only one test of purely spatial perception, learned the shapes readily, and during the computerized and like previous investigators, found this realm to be presentations his accuracy rate was 96%. untouched by his injuries. LH performed flawlessly on a standard test of judging the orientation of lines (Benton, Dot Patterns Hamsher, Varney, & Spreen, 1983). In this task, the sub- ject must match the orientation of a line to its counterpart To establish the generality of LH’s relatively preserved Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/3/1/25/1755723/jocn.1991.3.1.25.pdf by guest on 18 May 2021 in a display of 11 numbered radii, spaced at 18 degree ability to distinguish similar nonbiological shapes we intervals from the point of origin. Levine et al. (1985) used a second set of shapes that were similar in parts reported that LH can draw maps and describe the spatial and overall arrangement, differing only in relative posi- layouts of his room and neighborhood from memory. tions: three configurations of nine dots. Here there were Farah et al. (1988) found a clear dissociation between no topological attachment relations to encode; discrim- LH’s performance on tasks of visual versus spatial imag- inations had to be based on continuous dimensions of ery, with a deficit only on the former. relative position alone. In addition, each shape was not unique but consisted of a family of minor variants, cre- ated by taking a “prototype” configuration and then gen- Shapes Varying along Multiple Continuous erating nine distortions of each (see Posner & Keele, Dimensions 1968, 1970). LH was shown four distortions of one of the Most objects can be distinguished by the presence or prototypes, and was told that they were “A’s”; this was absence of a particular part or by the arrangement of repeated with the other two configurations, labeled “B” parts. Not so for faces that have the same parts in the and “C.” He was given a minute to study the 12 shapes. same arrangement; they comprise a category of shapes Then the 12 cards were shuffled and LH was asked to that varies continuously along multiple quantitative di- sort them from memory into piles corresponding to each mensions. Prosopagnosics, however,. usually report ana- of the three letters. He did this with 100% accuracy twice. lyzing faces as one would a common object. They search He also performed at 100% accuracy when asked to sort for parts (glasses, moles, beards, distinctive hairstyles), a new random pile consisting of previously seen shapes, and look for telling details that might enhance recogni- new shapes distorted from the prototype by the same tion. Perhaps prosopagnosia is a deficit in distinguishing amount as the training pictures, and the actual prototype, among any set of objects whose members are qualita- seen for the first time; in addition, he correctly sorted tively alike and that vary continuously along multiple 83% of new shapes that were distorted from the proto- dimensions, not just faces. type by different amounts than the training pictures. Clearly LH can not only distinguish shapes consisting of the same parts in different relative positions, but he can Rectilinear Objects ignore minor irrelevant variation in exact locations (anal- Tarr and Pinker (1989) designed sets of objects that are ogous to ignoring differences in facial expression when built around a single plan and composed of the same recognizing individuals) and focus on those that differ- kind of parts (line segments and sets of local subconfig- entiate one configuration from another (see also Etcoff urations of segments such as intersections and right angle et al., 1991). bends; see Fig. 5). Thus they can be discriminated only by encoding the configuration of parts within each object: Degraded or Distorted Depictions their attachment relations and relative lengths. As such they can be used to test whether LH’s deficit is one of Noncanonical Views general encoding of configural relations. Three of the Eighteen objects were photographed both from a can- seven shapes were shown to LH, each in conjunction onical or typical vantage point (e.g., a bucket photo- with a nonsense name. He was required to learn the graphed so that a side was visible) and from an unusual shape-name pairs by tracing each shape and repeating vantage point (e.g., a bucket photographed from above). its name five times, then drawing them from memory The test was modeled after that of Warrington and Taylor and recalling their names. If an error occurred it was (1973, 1978). The noncanonical views were presented pointed out and LH had to redraw them, until he drew first, followed by the 18 photographs of objects from the all three correctly twice in succession. Then the shapes canonical perspective. LH peformed normally, achieving

32 JournuE of Cognitive Neuroscience Volume 3, Number 1

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.1991.3.1.25 by guest on 28 September 2021 Figure 5. Examples of recti- linear objects taught to LH.

1 2 3 4 Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/3/1/25/1755723/jocn.1991.3.1.25.pdf by guest on 18 May 2021

I 5 6 7

92% correct overallkone object was not recognized in parts of a whole. To see if LH’s inabilit71 to recognize either condition and one additional object was misiden- fragmented objects is due to an absence of this contour tified in the noncanonical view. Warrington and Taylor completion process, we administered a task that taps it (1973, 1978) found a recognition deficit for unconven- directly.’ Kanizsa’s (1976, 1979) “illusory contour” figures tional views of objects among patients with right poste- are visual patterns in which the phenomenon of contours rior damage. Recall that they suggested that such patients without physical intensity discontinuities is very compel- suffer a deficit in what they call perceptual categorization, ling to the normal perceiver. For example, Figure 6 could and suggest that such a deficit may underlie the “aper- be described as consisting of three angles and three black ceptive” form of visual agnosias. Note that by such a circular figures. Rarely, normal subjects with a strongly definition, LH would again escape the definition of an analytical set have initially described it this way (Kanizsa, apperceptive agnosic with a domain-general impairment. 1979). The vast majority of perceivers see the white area in the center as an opaque triangle partly superimposed over the circles. Kanizsa noted several interesting prop- Fragmented Depictions In the second test, we presented line drawings of objects segmented into fragments (Hooper, 1958). LH was asked to assemble these fragments in his mind and name the whole object. LH performed poorly at this task, recog- nizing only 18 of 30. Normal adults usually recognize no fewer than 25. Levine et al. (1989) also found that LH was severely impaired on a similar test where he had to identify visual items (objects and words) that were pre- sented either embedded in visual noise, or as silhouettes from which parts were deleted; the parts remaining were never identifiable individually. These tests of visual clo- sure were virtually impossible for LH who could not recognize the objects, even after being given their names. Remember also that LH was impaired on the Benton- Van Allen (1968) face-matching task when the faces ap- peared fragmented due to poor lighting.

Illusoy Contours One of the most important subprocesses involved in recognizing objects from fragmented pictures is inter- polating contours that join up the separate fragments as Figure 6. Example of Kanizsa illusory contour figure shown to LH

Etcofi et al. 33

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.1991.3.1.25 by guest on 28 September 2021 erties of these figures: they are perceived as brighter regular blobs LH tended to give them names, such as than the contiguous regions; they seem displaced in the “banana” or “potato.”LH did not see the illusory contours third dimension “in front of” or “over” the rest of the on figures where normals do not (when the edges are figure; they possess a clear border that separates them closed). Like the normal control, he saw these figures as from neighboring regions, even though there is no phys- flat and remarked that although a “triangle could be ical intensity change that might suggest the presence of there, it doesn’t seem so.” a border. Finally, they have functional as well as percep- tual effects: they can cause visual illusions such as the Summary and Discussion Ponzo illusion (where two lines of equal length sur- rounded by converging lines-in this case, illusory In sum, LH is severely prosopagnosic 22 years after his ones-appear unequal). Note that this task is a paradigm injury. Faces cannot be identified and do not elicit even case of visual completion. a hint of familiarity. He retains information about people LH was presented 20 patterns taken from Kanizsa (Kan- and can access it via other routes such as sound or smell. izsa 1976; Kanizsa & Gerbino, 1982). Of these patterns, LH can also derive information from his visual inspection Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/3/1/25/1755723/jocn.1991.3.1.25.pdf by guest on 18 May 2021 10 contained illusory contours, including triangles, rec- of a face: he can distinguish faces from nonfaces even tangles, and irregular blobs. The rest were either groups when distinctions are subtle, and can judge the person’s of shapes, shapes that had illusions of depth but no sex, relative age, and likeability (at least, in a similar illusory contours, or crude outlines of figures (2 human, manner to that of a normal control). He has some diffi- 1 animal). LH was asked to describe the shapes he saw, culty recognizing specific facial expressions of emotion to outline them with his finger, and to say whether the though he can distinguish them and group together ex- image was flat or in depth. If the figures appeared in pressions across different faces. Although LH cannot rec- depth, he was asked to say which were on top of which. ognize a face, he can discriminate one face from another, Finally, if he mentioned a figure formed by illusory con- matching identical views of faces flawlessly, and tolerat- tours, he was asked to rate it on a -5 to +5 scale on its ing changes of expression and angle reasonably well. He brightness relative to the rest of the figure and on how does not have a general visual agnosia, and his recog- compelling it was (- 5 being the least bright compared nition of objects, even within categories, is quite good, to the surround, or the least compelling). Finally, one as is his ability to learn and recognize abstract objects figure for which normals usually show the Ponzo illusion that are formed out of the same sets of simple parts and was included. One normal male control was tested using that differ only in the relative lengths, positions, and the same procedures. distances among the parts. LH does, however, have dif- LH clearly saw the illusory figures and showed the ficulty recognizing animals, although nonvisual infor- Kanizsa effect. For the first pattern he was shown, Figure mation about animals is well-preserved. Purely spatial 6, he initially seemed to perform like the rare “analytical” perception is intact. normal. He saw the triangle that was depicted with actual In sum, whatever perceptual abilities are needed for interrupted lines (he “completed” it) but he did not at reading, recognizing spatial layouts for navigation, rec- first mention the figure defined exclusively by illusory ognizing patterns of dots or connected lines, recognizing contours: “I see a triangle in the middle, 3 circles with nonanimate objects, matching faces (even across differ-

parts missing . . . (5 sec pause) , . , and a white triangle ent views and expressions), matching emotions (even in the middle. That now seems like the main object across different people), and recognizing a face’s sex, though I saw it last , , , it all of a sudden popped out at age, and likeability, are functioning adequately in LH. It me . . . I think I was distracted by the solid stuff.” LH seems difficult, then, to classify LH as any kind of “ap- likened his perception of this display to that of an am- perceptive” agnosic. Rather, his deficit appears to be biguous figure because he could “flip back and forth” specific to associating visual input with specific people, between the image of the single triangle and three circles and, to a lesser extent, facial expressions, and animals. and the image of the two triangles and circles, but this Furthermore, in both our own studies of face and voice happened only for a short time. LH spontaneously men- perception and in the studies of Farah et al. (1988) of tioned that the white triangle seemed to “stand above visual and spatial imagery, LH’s imagery abilities parallel the other” and seemed brighter. For the other nine dis- his recognition abilities, and as Farah et al. (1988, p. 455) plays, he reported seeing the illusory figures immedi- note, the most parsimonious explanation for such a pat- ately. Correlations of his ratings with that of the normal tern is that the long-term visual memories required for control showed that he also perceived the figures in a both higher perceptual functioning and mental imagery similar way-their ratings of the relative brightness of have been damaged. the illusory contour figures and of how compelling the effect was for each figure were significantly correlated Alternative Accounts Positing a Perceptual Deficit (p < ,001 for the former, andp < .05 for the latter). He also was susceptible to the Ponzo illusion, induced by One could raise two objections to this claim. First, though an illusory triangle. When the illusory shapes were ir- LH had high success rates on most of the perception

34 Journal of Cognitive Neuroscience Volume 3, Number 1

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.1991.3.1.25 by guest on 28 September 2021 tasks, in many cases he did the task slowly, and reaction whatever object representations they are most consistent time data would surely distinguish him from normal with. This cycle continues until a mutually consistent controls. Note that we are not claiming that LH has intact object hypothesis and geometric analysis of the input perception. However, the perceptual abilities he retains emerge. Crucially, stored knowledge of objects suffi- are sufficient to accomplish the multifarious difficult tasks ciently detailed so that some are analyzed as more con- listed above, while he performs poorly on the face, emo- sistent with the degraded input than others is required. tion, and animal recognition tasks, euen with great effort As such, any deficit in perceiving degraded depictions and unlimited time. As such there is a qualitative differ- of objects cannot be attributed to a purely perceptual ence between the two kinds of tasks. deficit. The second possible objection is that there is a family Second, tests with unusual views involve a confound. of tasks that LH did poorly on that have been called The three-dimensional geometry of the depicted object “perceptual” in previous studies: matching faces and rec- is typically difficult to recover from bottom-up processing ognizing objects that are fragmented or appear frag- because of foreshortening and occlusion, and the object mented owing to extreme high contrast lighting (the is seldom encountered in that orientation. Tarr and Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/3/1/25/1755723/jocn.1991.3.1.25.pdf by guest on 18 May 2021 altered illumination condition of the Benton-Van Allen, Pinker (1989) and Tarr (1989) have shown that sheer 1968, task, the Hooper Visual Organization Test, 1958, familiarity with a specific object at a specific orientation and tests of Levine et al., 1989, of visual closure). War- enhances recognition of it at that orientation, suggesting rington and Taylor (1978) and Benton and Van Allen that people store view-specific representations of shapes. (1968) suggest that both fragmentation and unusual Hence a patient with various perceptual or associative views invoke a single perceptual process, involving the deficits could recognize an unusual view if a represen- construction of a structural representation of the percep- tation specific to that object in that view had been spared. tual input prior to matching against memory, and Levine This is probably impossible for high-contrast fragmen- et al. conclude that LH has a more general apperceptive tation: people are unlikely to have stored representations deficit underlying his prosopagnosia. LH’s performance specific to some unusual kind of photographic depiction. on our tasks renders such suggestions paradoxical: he But the most cpmpelling evidence that LH’s poor rec- recognized objects in unusual views, and performed well ognition of fragmented figures does not indicate a pri- on numerous tasks that would seem to require the com- marily perceptual deficit is that he successfully perceived putation of a structural representation of the perceptual illusory contours, presumably the most important purely input; only fragmented-appearing faces, words, and ob- perceptual component in the recognition of fragmented jects were difficult. Further complicating the interpre- figures, in Kanizsa displays. These figures have been tation of LH’s performance on these tasks is the fact extensively studied, and they are believed to provide a that the tests of Levine et al. were difficult even for relatively pure test of gestalt perception with no concep- normal subjects: Of the three tasks, the normal con- tual or memory component (Kanizsa, 1979, Kanizsa & trols recognized only 76% of the figures (15.2 of 20) Gerbino, 1982). Furthermore, they are known to be gov- in one (“Gestalt completion” of fragmented shapes), erned by prestriate area 18 (V2) of the visual cortex (Von 47% (23.6 of 50) on another (“Concealed”-actually der Heydt, Peterhans, & Baumgartner, 1984). fragmented-words), and 24% (5.7 of 24) on a third In sum, there is no evidence that LH has a general (“Snowy pictures”). apperceptive deficit of which his prosopagnosia is a sim- We suggest that contrary to earlier authors, it is un- ple consequence. It would be unwarranted to conclude likely both that the two kinds of distortions (noncanon- that LH has an apperceptive deficit from his poor rec- ical views and contrast-induced fragmentation) tap ognition of fragmented pictures, given that such recog- mechanisms that are exclusively perceptual, and that the nition clearly requires associative knowledge in some mechanisms they tap are necessarily the same. First, complex manner that is poorly understood even for nor- when bottom-up processing of object geometry cannot mals, and that the component of these tasks that is most take place successfully owing to degraded input (fore- purely perceptual, interpolating virtual contours from shortening, occlusion, unusual patterns of high-contrast fragments, was found to be present. Furthermore, in lighting), recognition is thought to succeed with the help direct tests of his abilities to perceive information rele- of top-down processes tapping stored knowledge of ob- vant to faces (face-matching, ignoring irrelevant emo- jects’ shapes (see Marr & Nishihara, 1978; Pinker, 1984; tional information and angle of view, judging sex, age, Finke, Pinker, & Farah, 1989, for discussion). Specifically, and likeability) and his ability to perceive and recognize a large set of object representations that crudely match other objects (even in unconventional views), no quali- the input may be activated, and each may strengthen tative deficit is present. Note that we are not claiming tentative analyses of features of the input that are consis- that LH’s perceptual abilities are identical to those of tent with them (e.g., local contours, bits of figure-ground normals, only that there is no way to explain his proso- assignment, analyses of surface curvature). Those acti- pagnosia as a direct consequence of them, given his vated features that are geometrically consistent with each successful performance on other tasks with equivalent other will enhance each other, which in turn activate perceptual demands.

Etcofi et al. 35

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.1991.3.1.25 by guest on 28 September 2021 TESTS OF COVERT RECOGNITION thetic skin responses to deep inspiration were 1062 (SD 413) pV for LH, and 850 (SD 430) pV for a representative In three experiments, we took subtle measures of LH’s control subject. Electrical stimulation elicited response behavior in response to familiar and unfamiliar faces to amplitudes of 1416 (SD 155) pV for LH, and 1063 (SD see whether he would give evidence of covert recogni- 411) pV for a representative control subject. These are tion of faces when he is consciously unaware of their within the normal range as reported by Shahani, Hal- identity. perin, Boulu, & Cohen (1984), with LH somewhat ex- ceeding our controls and those of Shahani et al. in his Sympathetic Skin Response response to electrical stimulation. For each normal control subject, we compared the Method mean sympathetic skin response for Unfamiliar faces, LH was shown 44 black and white photographic slides Familiar faces, and Famous faces. For each subject an of faces. The faces were of roughly uniform size and Analysis of Variance was conducted to establish the sta- were all in frontal view. No identifying clothing or back- tistical reliability of the difference, using the responses Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/3/1/25/1755723/jocn.1991.3.1.25.pdf by guest on 18 May 2021 ground was visible. Thirty-three of the faces were of to the different face slides within each condition as the people LH had never met (the unfamiliar set) and 11 random factor. The difference among the three condi- were of individuals well-known to LH (the familiar set). tions was significant at the .05 level for each of the control The 11 familiar individuals consisted of family members subjects. This effect seems to reflect the difference be- and close friends. Four were people he had known prior tween the Familiar faces (mean 695 pV) and those in the to his injury, and seven were people he had met since. other two conditions; Famous (mean 395 pV) and Un- The unfamiliar set was chosen to match the familiar set familiar (mean 388 pV) faces were not distinguished in in age and sex. The slides were shown in random order the responses. In addition, we averaged each subject’s under the constraint that two familiar faces could not be responses to each of the faces within a condition and shown in succession, to prevent habituation of any effect compared the conditions in an Analysis of Variance using of familiarity. Subjects as the random factor; the effect of Condition The sympathetic skin response was recorded with a was statistically significant [F(2,6) = 12.34,p < ,011. Thus pair of standard surface disk EMG electrodes placed on the difference between Familiar and other faces is reli- the palm and the dorsal surface of the hand. The band- able over samples of stimuli and samples of normal pass filter was set at 0.5-2000 Hz. The sweep velocity subjects. was 500 msec per division. All responses were recorded For LH, in contrast (see Fig. 7) the mean responses to on a Nicolet Viking LE EMG machine. Prior to showing Familiar and Unfamiliar faces were virtually identical the faces, we elicited sympathetic skin responses to deep (233 and 240 FV, respectively), and did not differ in an inspiration and mild electric shock. Analysis of Variance using face stimuli as the random Slides were presented for 15 sec at intervals of 20-30 factor [F(1,42) < 11. sec. No response was required. The first two slides (both As can be seen in Figure 7, LH’s responses to visual of unfamiliar faces) were treated as practice and were stimuli in general were lower in amplitude than those not entered into the data analysis. Slides were presented of controls. This raises the question df whether LH’s in three blocks: 20 women (including the two initial responses simply reflect floor effects. Interestingly, one practice trials), 20 men, and 6 children. Brief rest periods of our control subjects had response amplitudes similar were given between blocks. Following presentation of to LH’s for Unfamiliar faces (mean of 221 pV). Nonethe- all the faces, the slides were shown a second time and less, this subject did show a significant response to Fa- LH was asked if he could identify any or if any seemed miliar faces (mean of 490 pV), [F(2,29) = 7.93,p < ,0031. familiar. Four neurologically intact adults, ages 30-40, served as control subjects. They were run through a procedure identical to that used with LH, except for the slides shown. Control subjects were shown pictures of unfa- Pupillometry miliar individuals, individuals well known to them (fa- In a second experiment, LH’s pupil size was monitored miliar set), and of prominent political and entertainment while he viewed slides of familiar and unfamiliar faces, figures (famous set).2 and slides of neutral and of emotionally provocative or interesting visual scenes. Like the sympathetic skin re- sponse, pupil size is governed by the autonomic nervous Results system. It has been used in previous experiments to Peak-to-peak amplitude was measured manually by an index the interest value and emotional impact of visual assistant blind to which trials the traces came from. Re- stimuli (see Hess, 1972). To our knowledge, this is the sults indicate that both LH and control subjects show the first time it has been used to index recognition or fa- expected responses to nonvisual stimulation. Sympa- miliarity.

36 Journal of Cognitive Neuroscience volume 3, Number 1

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.1991.3.1.25 by guest on 28 September 2021 Figure 7. Sympathetic skin response amplitudes to famil- iar and unfamiliar faces for LH and control subjects. Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/3/1/25/1755723/jocn.1991.3.1.25.pdf by guest on 18 May 2021

3- LH

b

Unfamiliar Faces Familiar Faces

Method tween the eyebrows. Faces were shown for a maximum of 10 sec. Trials were aborted if there was a blink, head LH was shown 46 black and white photographic slides movement, or saccade away from the fixation. A face of faces, all in frontal view and without visible identifying slide would never appear until a 5-sec fixation on the clothing or background. Thirty-three of these faces were cross-hairs and gray field was successfully completed. of individuals unfamiliar to LH; 13 of these faces were of An ISCAN model FX-416 pupil tracker was used to individuals well-known to him, As in the previous ex- measure pupil diameter. The eye tracker received an periment, the familiar ser included both faces of people image of the subject’s right eye from an RCA TC2000 known prior to his injury (5) and those met since (8). video camera with a close-focus lens and an infrared All photographs were new; however, 3 of the familiar filter. A table light fitted with an infrared filter illuminated individuals (1 preinjury, 2 postinjury) had been shown the subject’s eye, and a chin rest and forehead restraint to LH in the previous study. Since pupil response is held the subject’s head in place. The pupil tracker re- highly sensitive to the overall brightness and contrast of ceived a new video iniage of the subject’s eye from the pictures, photographs of faces were first digitized with a camera 60 times each second. In each video image, the monochrome CCD camera and a Datacube digitizing pupil tracker used an algorithm implemented in hard- frame store. Both the mean and variance of each image ware to locate the pupil by identifying a large dark re- were then normalized to standard values, and the images gion. It then calculated the diameter of this region, and were displayed on a gamma-corrected monitor. Slides transmitted the value to the IBM XT microcomputer that were then created from these normalized images. As in was controlling the experiment. Experiment 1, LH was shown the faces in three blocks One neurologically intact adult male, age 39, was given (women, men, children) with rest periods between the same procedure as a control. Four of the “unfamiliar” blocks. faces (for LH) were of individuals well-known to this In a separate procedure, LH was shown 12 black and subject. The control subject saw the initial 29 pictures white photographic slides of visual scenes. Nine were of shown to LH (the rest were omitted because they were “neutral” scenes such as landscapes or home interiors. all unfamiliar). He had no interest in hockey, but was Three were emotionally provocative scenes-a nude highly familiar with one of the neutral slides for LH, the woman, a woman in a low-cut swimsuit, and a hockey interior of the home of a relative, which served as a third game (hockey is a passionate interest for LH). Like the interesting stimulus for him. faces, all slides were normalized in contrast and bright- ness digitally. On each trial, LH looked at cross-hairs affixed to a rear ResuIts projection screen. For 5 sec pupil size was monitored while LH looked at a uniform gray field that was of the Percent change from prestimulus size to stimulus size same overall brightness as the projected slides. After this was calculated for each slide. We analyzed these data for 5-sec period a slide appeared, replacing the gray field. the period 1 through 5 sec poststimulus onset so as not Each face was positioned so that the cross-hairs fell be- to exclude too many trials that were aborted after this

EtcoB et al. 37

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.1991.3.1.25 by guest on 28 September 2021 period, and to exclude the initial constriction-dilation al., 1983; De Haan et al., 1987) have tested prosopagno- response to the sound of the slide projector advancing. sics who learn true pairings faster than untrue pairings. For the normal control subject, the mean change in pupil diameter for familiar faces (1.03) was larger than Method for unfamiliar faces (.99), a difference that was statistically significant in an Analysis of Variance using faces as the LH was shown six photographs of familiar faces, all taken random factor [F(1,27) = 5.074,~< ,041. For LH, there from the yearbook of the small private high school he was no such difference [mean .985 versus .995 F(1,44) < attended. He was given a name to associate with each 11. The results are presented in Figure 8. To compare face and had a minute to learn these pairings. Three of the performance of LH and the control subject directly, the faces were presented with their correct names and we entered just the data from those faces shown both to three with incorrect names. All names were of individuals LH and the control (four of these were familiar to the from his high school class, and all were determined control, nine to LH) into an Analysis of Variance using previously to be equally and highly familiar to LH. On faces as the random factor. There was no main effect for each trial, the faces were presented in front of LH. The Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/3/1/25/1755723/jocn.1991.3.1.25.pdf by guest on 18 May 2021 person (LH versus control) nor for face familiarity, but a six names were read one at a time and he had to point significant Person by Familiarity interaction [F( 1,27) = to the corresponding face. There were 10 trials, with the 15.05,p < .OOl], with only the control showing greater order of names and the positions of faces randomized pupil dilation to familiar faces. before each one. This failure to discriminate familiar from unfamiliar Results are in Table 1. LH did not show facilitation for faces cannot be attributed to an absence of emotion- true pairings: In all but the second trial, the number relevant pupil responses. Both the control subject and correct was the same for true and untrue pairings. He LH had significantly larger pupillary changes to emotion- was 100% accurate on two face-name pairs, one correct, ally evocative slides than to neutral slides: for the control one incorrect, and confused the rest on virtually all the subject, 1.054 versus. .999, F(1,lO) = 10.074,p = .01; for trials. LH, 1.029 versus ,979, F(1,lO) = 8.82,p < .02 (pictures as random factor). These data are shown in Figure 9. Discussion LH does not show any evidence of covert recognition of familiar faces. His autonomic responses to these stimuli Paired Associate Learnfng are no greater than his responses to completely unfa- Finally, covert recognition was investigated through a miliar faces, and he does not show facilitation in the behavioral measure. LH was given a paired-associate task learning of true versus untrue name-face pairs. He thus in which he was required to learn six face-name pairs. joins a small group of prosopagnosics who do not rec- Half of the faces were paired with their correct names ognize faces even when recognition is tested implicitly. and half with incorrect names. Two studies (Bruyer et The fact that LH’s autonomic nervous system does re-

Figure 8. Percent change in pupil dilation to familiar and PUPIL DILATION: 1.06 - unfamiliar faces for LH and the FACES control subject. 1.05 -

1.04 - J W CONTROL N - gZ 1.03- Em . $2 1.02- 2z2s . 8 i, 1.01 - K n.

1 .oo -

0.99 -

098 ! UnfamiliarFaces Familiar Faces I

38 Journal of Cognitiue Neuroscience Volume 3, Number I

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.1991.3.1.25 by guest on 28 September 2021 Figure 9. Percent change in pupil dilation to neutral and PUPIL DILATION: emotionally provocative pic- 1.06- tures for LH and the control subject.

1.04 -

YN sz 1.02 - E2 W-l a3 za qF 0: 1.00- apg Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/3/1/25/1755723/jocn.1991.3.1.25.pdf by guest on 18 May 2021

0.98-

Neutral Emotional

Table 1. Number of Correct Responses: True versus Untrue Pairings

~ Trial #

I 2 3 4 5 6 7 8 9 I0 True pair 1 2 1 2 1 2 1 1 1 2 Untrue pair 1 1 1 2 1 2 1 1 1 2

spond normally to peripheral stimuli (electrical stimu- essing), to an apperceptive deficit (i.e.,preserved infor- lation) and to interesting visual but nonfacial stimuli mation about faces which cannot be accessed because of suggests that the results are not due to floor effects in imperfectly analyzed perceptual input), or with an in- autonomic responding. ability to recognize complex or subtly varying shapes (i.e., loss or degradation of shape memory in generalJ3 Since his deficit includes both the loss of memories of GENERAL DISCUSSION previously learned faces and the ability to learn new We have presented a patient who cannot recognize fa- ones, the deficit is not mere erasure of old memory miliar faces or report that they are familiar, nor answer information, but the loss of the mechanism in which the questions about the faces from memory, though he can memories are stored. (1) recognize common objects and novel subtly varying Our findings have at least two implications for further shapes, (2) match faces while ignoring irrelevant infor- research. First, since LH appears to lack stored memory mation such as angle of view and emotional expression, information for certain kinds of objects, not just access (3) recognize sex, age, and likeability from faces, and (4) to such information, it suggests that there is a modicum recognize people by a number of nonfacial channels. His of localization of such information. Unfortunately, LH's only other marked deficits are in recognizing animals numerous cortical and white matter lesions make him a and emotional expressions, though even here they are poor candidate for attempts at localization of function, not completely comparable to his catastrophic loss of although we can say that he does not have the classic face recognition. Using three different measures, we pattern of damage bilaterally to inferotemporal-occipital gathered converging evidence that LH's lack of face rec- cortices. Despite his patchwork of injuries, comparisons ognition is not restricted to a lack of awareness; he shows with extensively tested prosopagnosics who do show no signs whatsoever of possessing visual information covert recognition might prove instructive when areas of about the identity of familiar faces. Thus prosopagnosia nonoverlap are examined. Studies of possible homolo- cannot always be attributed solely to an absence of gous areas in nonhuman primates would also be fruitful. awareness (i.e., preserved information about faces whose Second, our evidence suggests a certain degree of output is disconnected from conscious cognitive proc- content-specificity for visual shape representations, as

Etcofi et al. 39

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.1991.3.1.25 by guest on 28 September 2021 LH’s deficit does not extend to most classes of objects Albert, M. S., Butters, N., & Levin, J. (1979). Temporal gra- (words, tools, furniture, fruits and vegetables, dot pat- dients in the retrograde amnesia of patients with alcoholic Korsakoff’s disease. Brain, 36, 211-216. terns, rectilinear shapes, etc.). However the reasons that Bauer, R. M. (1984). Autonomic recognition of names and three different families of shapes were affected-faces, faces in prosopagnosics: A neuropsychological application emotional expressions, and animals-are unclear. One of the guilty knowledge test. Neuropsychologia, 22, 457- possibility is that there are separate localizable stores for 469. each of these classes (they are all evolutionarily signifi- Bauer, R. M. (1986). The cognitive psychophysiology of proso- pagnosia. In H. D. Ellis, M. A. Jeeves, F. Newcombe, & cant to humans), and that LH’s lesions happened to dam- A. Young (Eds.), Aspects of faceprocessing. Dordrecht: Mar- age all three. If so, one should be able to find a patient tinus Nijhoff, pp. 253-267. with a recognition deficit akin to LH’s but that is specific Benton, A. L., Hamsher, K. deS., Varney, N. R., & Spreen, 0. to one or two of the three categories that gave him (1983). Contributions to neuropsychological mement. trouble. Alternatively, there may be a neural substrate New York: Oxford University Press. Benton, A. L., & Van Allen, M. W. (1968). Impairment in facial that underlies all three of these classes to various degrees

recognition in patients with cerebral disease. Cortex, 4, Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/3/1/25/1755723/jocn.1991.3.1.25.pdf by guest on 18 May 2021 by virtue of some commonality in the geometric infor- 344-358. mation needed to distinguish among members of the Bodamer, J. (1947). Die Prosopagnosie. Archiv fur Psychiutrie classes. Such a “zoomorphic” superclass might consist of und Nenenkrankheiten, 179, 6-53. shapes defined by smooth elastic two-dimensional sur- Bornstein, B. (1963). Prosopagnosia. In L. Halpern (Ed.), Problems in dynamic neurology. Jerusalem: Hadassah Med- faces drawn over a three-dimensional skeleton and hr- ical School. ther warped and filled out by underlying soft connectors Bruyer, R., Laterre C., Seron, X., Feyereisen, P., Strypstein, E., and pads (muscle and fat). Further research in anatomy, Pierrard E., & Rectem, D. (1983). A case of prosopagnosia geometry, and the computational study of vision could with some preserved covert remembrance of familiar faces. reveal whether there is indeed a common geometric Brain and Cognition, 2, 257-284. Carey, S. (1982). : Anomalies of development. characterization of members of these classes for which In S. Strauss (Ed.), U-Shaped behavioral growth. New York: the might have specific circuits. Academic Press. Damasio, A. R., Damasio, H., & Van Hoesen, G. W. (1982). Prosopagnosia: Anatomic basis and behavioral mechanisms. Acknowledgments Neurology, 32,331-341. Damasio, A. R., Damasio, H., & Tranel, D. (1986). Prosopagno- This research was supported by NIH Grant DC00565 to N.L.E. sia: Anatomic and physiological aspects. In H. D. Ellis, M. A. and by NIH Grant RO1 HD22166 and NSF Grant 85-18774. We Jeeves, F. Newcombe, & A. Young (Eds.), Aspects of face thank Steven Pinker, David Mumford, and Lynn Hillger for processing. Dordrecht: Martinus Nijhoff, pp. 279-290. comments and suggestions on this manuscript, Michael Alex- De Haan, E. H. F., Young, A,, & Newcombe, F. (1987). Face ander and Marcel Mesulam for comments on an earlier version, recognition without awareness. Cognitive Neuropsychology, and Susan Carey for helpful discussions on face perception. We are grateful to Suzanne Corkin for introducing us to LH; 4, 385-415. DeRenzi, E. (1986). Current issues in prosopagnosia. In H. D. Edward Adelson, John Magee, Paul Finn, Mike Sokolov, and Ellis, M. A. Jeeves, F. Newcombe, & A. Young (Eds.), Aspects Catherine Thomas for technical assistance; Dan Bub, Susan of face procesing. Dordrecht: Martinus Nijhoff, pp. 243- Carey, Suzanne Corkin, Rhea Gendzier, Michael Tarr, and Diana 252. Van Lancker for the use of their stimulus materials; and LH for Ekrnan, P., & Friesen, W. V. (1975). Pictures of facial affect. his cooperation and interest. Parts of this research were pre- Palo Alto, CA: Consulting Psychologists Press. sented at the 42nd Annual Meeting of the Academy of Neurol- Etcoff, N. L. (1984). Selective attention to facial identity and ogy, Miami, February, 1990. facial emotion. Neuropsychologia, 22, 281-295. Etcoff, N. L., Tarr, M. J., & Carey, S. (1991). Prosopagnosia Reprint requests should be sent to N. L. Etcoff, Department of Is a specijic deficit in recognizingfaces or general deficit in Brain and Cognitive Sciences, E10-237A, MIT, Cambridge, a distinguz3hing objects with similar parts? Paper to be pre- MA 02139 USA. sented to the International Neuropsychological Society Meetings, San Antonio, Texas, February. Farah, M. J., Hammond, K. M., Levine, D. N., & Calvanio, R. Notes (1988). Visual and spatial mental imagery: Dissociable sys- 1. We thank David Mumford for suggesting this task. tems of representation. Cognitive Psychology, 20, 439-462. 2. The famous faces were not shown to LH because it was Ferro, J. M., & Santos, M. E. (1984). Associative visual agnosia: found that the control subjects did not show a differential A case study. Cortex, 20, 121-134. response to them. Finke, R. A., Pinker, S., & Farah, M. J. (1989). Reinterpreting 3. Of course, it is logically possible that LH’s face storage region visual patterns in mental imagery. Cognitive Science, 13, is intact but inaccessible by any pathway, either overt or covert. 51-78. It is not clear how this possibility could be tested; there is no Garner, W. R. (1974). The processing of ir2formation and evidence for it. structure. Hillsdale, NJ: Erlbaum. Hecaen, H., Goldblum, M. C., Masure, M. C., & Ramier, A. M. (1974). Une nouvelle observation d’agnosie d’objet. Deficit REFERENCES de I’association ou de la categorisation specifique de la modalite visuelle? Neuropsychologia, 12, 447-464. Albert, M. L., Reches, A,, & Silverberg, R. (1975). Associative Hess, E. H. (1972). Pupillometrics: A method of studying, visual agnosia without alexia. Neurology, 25, 322-326. mental, emotional, and sensory processes. In N. S. Green-

40 Journal of Cognitive Neuroscience Volume 3, Number 1

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.1991.3.1.25 by guest on 28 September 2021 field & R. A. Sternbach (Eds.), Handbook ofpsychophysiol- Posner, M. I., & Keele, S. W. (1970). Retention of abstract ogy. New York: Holt, Rinehart & Winston, pp. 491-534. ideas. Journal of Experimental Psychologp, 83, 304-308. Hooper, H. E. (1958). The Hooper Visual Organization Test. Renault, B., Signoret, J-L., Debruille, B., Breton, F., & Bolgert, Los Angeles: Western Psychological Services. F. (1989). Brain potentials reveal covert facial recognition Kanizsa, G. (1976). Subjective contours. Scient@cAmerican, in prosopagnosia. Neuropsychologia, 27, 705-712. 234,48-52. Schacter, D. L., McAndrews, M. P., & Moscovitch, M. (1988). Kanizsa, G. (1779). Organization in vision: Esays in Gestalt Access to consciousness: dissociations between implicit and perception. New York: Praeger. explicit knowledge in neuropsychological . In L. Kanizsa, G., & Gerbino, W. (1782). &nodal completion: Weiskrantz (Ed.), Thought without language. Oxford: Ox- Seeing or thinking? In J. Beck (ed.), Organization and rep- ford University Press. resentation in perception. Hillsdale, NJ: Erlbaum, pp. 167- Sergent, J., & Villemure, J.-G. (1989). Prosopagnosia in a right 190. hemispherectomized patient. Brain, 112, 975-775. Landis, T., Cummings, J. L., Christen, L., Bogen, J. E., & Imhof, Shahani, B. T., Halperin, J. J., Boulu, P., & Cohen, J. (1984). H.-G. (1986). Are unilateral right posterior cerebral lesions Sympathetic skin response-a method of assessing unmye- sufficient to cause prosopagnosia? Clinical and radiological linated axon dysfunction in peripheral neuropathies.Jour-

findings in six additional patients. Cortex, 22, 243-252. nal of Neurology, Neurosurgery, and Psychiatry, 47, 536- Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/3/1/25/1755723/jocn.1991.3.1.25.pdf by guest on 18 May 2021 Levine, D. N., & Calvanio, R. (1989). Prosopagnosia: A defect 542. in visual configural processing. Brain and Cognition, 3 0, Snodgrass, J. G., & Vanderwart, M. (1780). A standardized set 149-170. of 260 pictures: Norms for name agreement, image agree- Levine, D. N., Calvanio, R., & Wolf, E. (1980). Disorders of ment, familiarity, and visual complexity. Journal of Eqeri- visual behavior following bilateral posterior cerebral le- mental Psycbology: Human Learning and Memoq, 6, 174- sions. Psychological Research, 41, 217-234. 215. Levine, D. N., Warach, J., & Farah, M. (1985). Two visual sys- Tarr, M. J. (1989). Orientation dependence in three-dimen- tems in mental imagery: Dissociations of “what” and sional object recognition. Doctoral Dissertation, MIT. “where” in imagery disorders due to bilateral posterior Tarr, M. J., & Pinker, S. (1987). Mental rotation and orienta- cerebral lesions. Neuroloa, 35, 1010-1017. tion dependence in shape recognition. Cognitive Psychol- Lissauer, H. (1890). Ein fall von seelenblindheit nebst einem OQ, 21, 233-282. beitrage zur theorie derselben. Archiv fur Psychiatric und Teuber, H.-L. (1968). Alteration of perception and memory in Neruenkrankheiten, 21, 222-270. man. In L. Weiskrantz (Ed.), Analysis of behavioral change. Marr, D. (1982). Vision. San Francisco: Freeman. New York: Harper & Row. Marr, D., & Nishihara, H. K. (1978). Representation and recog- Thompson, P. (1980). Margaret Thatcher: A new illusion. Per- nition of the spatial organization of three-dimensional ception, 9, 483-484. shapes. Proceedings of the Royal Society of London, 200, Tranel, D., & Damasio, A. R. (1985). Knowledge without 269-294. awareness: An autonomic index of facial recognition by Marslen-Wilson, N. D., & Teuber, H.-L. (1975). Memory for prosopagnosics. Science, 228, 1453-1454. remote events in anterograde amnesia: Recognition of pub- Tranel, D., & Damasio, A. R. (1988). Nonconscious face recog- lic figures from news photographs. Neuropsychologia, 13, nition in patients with face agnosia. Behavioural Brain Re- 347-352. search, 30, 235-249. McCarthy, R. A,, & Warrington, E. K. (1986). Visual associative Van Lancker, D., & Kreiman, J. (1987). Voice discrimination agnosia: a clinico-anatomical study of a single case. Journal and recognition are separate abilities. Neuropsychologia, of Neurology, Neurosurgery and Psychiaty, 49, 1233- 25,829-834. 1240. Von der Heydt, R., Peterhans, E., & Baumgartner, G. (1784). Michel, E, Perenin, M. T., & Sieroff, E. (1986). Prosopagnosie Illusory contours and cortical neuron responses. Science, sans hemianopsie apres lesion unilaterale occipito-tempor- 224, 1260-1262. ale droite. Revue NeurologtQue, 142, 545-549. Warrington, E. K., & Taylor, A. M. (1973). The contribution of Newcombe, E, Young, A. W., & De Haan, E. H. F. (1989). Pro- the right to object recognition. Cortex, 9, 152- sopagnosia and object agnosia without covert recognition. 164. Neuropsychologia, 27, 179-191. Warrington, E. K., & Taylor, A. M. (1978). Two categorical Pinker, S. (1984). Visual cognition: An introduction. Cogni- stages of object recognition. Perception, 7, 695-705. tion, 18, 1-63. Young, A, W. (1988). Functional organization of visual recog- Perrett, D. I., Smith, P. A. J., Potter, D. D., Mistlin, A. J., Head, nition. In L. Weiskrantz (Ed.), Thought without Language. A S., Milner, A. D., & Jeeves, M. A. (1984). Neurones re- Oxford: Clarendon Press, pp. 78-107. sponsive to faces in the temporal cortex: Studies of func- Young, A. W., Hellawell, D., & De Haan, E. H. F. (1988). Cross- tional organization, sensitivity to identity and relation to domain semantic priming in normal subjects and a proso- perception. Human Neurobiology, 3, 197-208. pagnosic patient. The Quarterly Journal of Experimental Posner, M. I., & Keele, S. W. (1968). On the genesis of ab- psycho lo^, 40A, 561-580. stract ideas. Journal of Experimental Psychology, 77, 353- Young, A. W., & Ellis, H. D. (1789). Childhood prosopagnosia. 363. Brain and Cognition, 9, 16-47.

Etcofi et al. 41

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn.1991.3.1.25 by guest on 28 September 2021