- Email: [email protected]
Configurational coding, familiarity and the right hemisphere advantage for face recognition in sheep
Neuropsychologia 38 (2000) 475±483
www.elsevier.com/locate/neuropsychologia
Con®gurational coding, familiarity and the right hemisphere advantage for face recognition in sheep J.W. Peirce*, A.E. Leigh, K.M. Kendrick Laboratory of Cognitive and Developmental Neuroscience, Babraham Institute, Babraham, Cambridge CB2 4AT, UK Received 4 September 1998; received in revised form 7 April 1999; accepted 9 June 1999
Abstract This study examined characteristics of visual recognition of familiar and unfamiliar faces in sheep using a 2-way discrimination task. Of particular interest were eects of lateralisation and the dierential use of internal (con®gurational) vs external features of the stimuli. Animals were trained in a Y-maze to identify target faces from pairs, both of which were familiar (same ¯ock as the subjects) or both of which were unfamiliar (dierent ¯ock). Having been trained to identify the rewarded face a series of stimuli were presented to the sheep, designed to test for the use of each visual hemi®eld in the discriminations and the use of internal and external facial cues. The ®rst experiment showed that there was a left visual hemi®eld (LVF) advantage in the identi®cation of `hemifaces', and `mirrored hemifaces' and `chimeric' faces and that this eect was strongest with familiar faces. This represents the ®rst evidence for visual ®eld bias outside the primate literature. Results from the second experiment showed that, whilst both familiar and unfamiliar faces could be identi®ed by the external features alone, only the familiar faces could be recognised by the internal features alone. Overall the results suggest separate recognition methods for socially familiar and unfamiliar faces, with the former being coded more by internal, con®gurational cues and showing a lateral bias to the left visual ®eld. # 2000 Elsevier Science Ltd. All rights reserved. Keywords: Asymmetry; Hemispheric lateralisation; Chimeric; Face processing; Expertise; Internal features
In humans, recognition of faces has been found to display a robust visual ®eld bias with faces presented to the left of ®xation being identi®ed faster [4] and more accurately [10,15] than those to the right. Similar results are found for children [4,40] and commisurotomy patients [22,33]. The left visual hemi®eld (LVF) also plays a disproportionate role in judgements of aective state with chimeric faces, where half of the face is smiling and half is frowning [20,21]. It is thought that this asymmetry in perception arises from the neural lateralisation which has been reported by various human brain imaging studies such as Positron Emission Topography (PET) [34], functional Magnetic Resonance Imaging (fMRI) [16,29] and Magneto-Ence* Corresponding author. Tel.: +44-1223-496364; fax: +44-1223496028. E-mail address: [email protected] (J.W. Peirce).
phalography (MEG) [35] in normal subjects, and by lesion localisation in prosopagnosic patients [7 for review]. All of these lines of research have implicated occipitotemporal cortex in face processing with activation predominantly (although not exclusively) in the right hemisphere. This region may not be dedicated to face perception per se but rather to some visual process on which face recognition relies heavily. Diamond and Carey [8] claim that faces are merely a category of objects which have a common con®guration and that through experience we have learnt to identify subtle variations in this common con®guration. They tested the hypothesis with the use of the inversion eect (whereby humans [e.g. 38,39] monkeys [28,37] and chimpanzees [26] are particularly bad at identifying faces presented upsidedown) and found that dog-show judges demonstrate an equivalent eect for dogs. Thus they suggest that
0028-3932/00/$ - see front matter # 2000 Elsevier Science Ltd. All rights reserved. PII: S 0 0 2 8 - 3 9 3 2 ( 9 9 ) 0 0 0 8 8 - 3
476
J.W. Peirce et al. / Neuropsychologia 38 (2000) 475±483
any visual objects with a common feature con®guration can show face-like eects if they are discriminated frequently. This theory is supported by neuropsychological data which ®nds that most prosopagnosic patients do not show agnosia purely for faces but also for other intra-category discriminations. For example there are reported prosopagnosic farmers who could no longer discriminate between their cows [1,3] or ornithologists who could no longer identify dierent birds [2]. There are a few cases, however, where the de®cit is found for human faces despite an intact ability to identify other animal faces such as cows [5] or sheep [23]. Rhodes [31] examined the hypothesis that expertise leads to lateralisation and con®gurational1 coding. She measured the ability to recognise faces of dierent ethnic backgrounds (Caucasian and Chinese) in conjunction with visual ®eld asymmetry and inversion (as a measure of con®gurational coding). Whilst verifying that con®gurational coding was correlated to hemi®eld bias, it was not dependent on the degree of familiarity with the ethnic background of the faces as predicted by the expertise hypothesis. It should be noted that the use of inversion may not be a true measure of con®gurational coding. Whilst it is frequently cited as the most likely cause for the eect there is evidence that inversion of individual features could also give rise to inversion de®cits [30]. Thus, in the present study, con®gurational coding is tested for directly with faces where the internal features have been masked or rearranged rather than inverted. In non-human primates evidence for an asymmetry in either neural activation or behavioural responses has been mixed. Dittrich [9] used line-drawings of faces and required their macaques to indicate the face which was showing a certain gesture. They then produced asymmetric chimeric faces and jumbled-feature versions of these stimuli to identify whether the internal and external, and the left and right sides of the face were being used dierentially. The results suggest that the right hemi®eld was being used most (contrary to human subjects) and also that, whilst the facial outline was the most important visual cue, the eye region and mouth were also used. Overman and Doty [25] used mirrored hemifaces (where one visual hemi®eld is mirrored onto the other) as stimuli and found an advantage for the left±left face in human subjects, as expected, but no equivalent bias in macaque subjects. Another method has used split brain macaques where presentations can be made purely to one brain hemisphere, and this did yield an LVF advantage [14] although previous attempts to do 1 Some authors use the term `con®gural' rather than con®gurational.
so by the same group had failed [12,13]. Finally, Morris and Hopkins [24] found a bias in two out of three chimps (pan troglodytes) from the Savage±Rumbaugh project, performing Levy's [21] happy/sad discrimination on chimeric faces. This appears to be the strongest evidence to date for such perceptual asymmetry in a non-human species. At an anatomical level, the functional mapping of the primate brain during face recognition tasks has not been attempted and, at a cellular level, electrophysiology has shown little evidence for a right hemisphere processing bias. One paper even reports more face sensitive cells in the left hemisphere than the right [27], although this may be an artefact of the technique which lends itself better to the descriptions of neuronal ®ring characteristics than to the quanti®cation of cell populations. Sheep have been shown to use visual cues to discriminate between the species or breed of faces presented in a Y-maze [4] as well as identifying faces of individual sheep [18]. In both studies signi®cant classic inversion eects were found for the faces of a familiar breed, and especially for socially familiar individuals [18]. In sheep, neither electrophysiological nor behavioural studies have previously attempted to identify any asymmetry, although recent data from in-situ hybridisation of c-fos has shown stronger right hemisphere activation during a face recognition task [19]. Thus, the ®rst experiment of the present paper was designed to test for a visual hemi®eld bias during face recognition. This diers slightly from the facial expression recognition used by a number of primate [e.g. 10,24] and human studies [e.g. 21] but is necessary since sheep show very few facial expressions. Whilst the recognition of facial identity may be a separable process from gesture recognition, it has also been shown to demonstrate visual ®eld bias [e.g. 31]. The second experiment tested directly for the use of con®gurational coding in sheep face recognition. Both experiments compared familiar and unfamiliar individuals to test for expertise eects. 1. Method 1.1. Subjects The animals used were 10 polled Dorset ewes. This is a hornless breed with high facial homogeneity. They had been extensively trained in the Y-maze procedure whereby they were given food reward for choosing a target stimulus and were habituated to the maze environment and the experimenters. They were kept amongst a ¯ock of 20±30 sheep. Food (grass or hay) and water was provided ad libitum throughout the study.
J.W. Peirce et al. / Neuropsychologia 38 (2000) 475±483
477
Fig. 1. Examples of stimuli used in Experiment 1: (a) a pair of training stimuli; (b) left hemifaces; (c) left-left mirrored faces; (d) chimeric faces. Note the high homogeneity between the stimuli.
478
J.W. Peirce et al. / Neuropsychologia 38 (2000) 475±483
Fig. 2. Examples of stimuli used in Experiment 2: (a) internal features only; (b) external features only; (c) scrambled features; (d) swapped-feature stimuli.
J.W. Peirce et al. / Neuropsychologia 38 (2000) 475±483
1.2. Stimuli (Experiment 1) Stimuli were created from pictures taken by digital camera (Canon DC50) of sheep within the same ¯ock as the subjects (familiar) or another ¯ock (unfamiliar). They were edited on a Pentium PC computer and printed in full colour onto photographic slides which were back-projected onto a screen next in the Y-maze such that the face is seen at roughly life-size. For training stimuli the background was edited out leaving just a head on a black screen. From these, the following experimental manipulations were made (see Fig. 1 for examples): . Hemi®eld Ð Half of the face was obscured by a black mask; . Hemi®eld mirrored Ð Half of the face was mirrored onto the other half (also referred to as left±left/ right±right faces). Similar to those used by Overman and Doty [25]; . Hemi®eld chimeric Ð The traditional chimeric face stimuli [21,24] where the left and right of two separate faces (in this case the rewarded and nonrewarded) are combined to form one.
1.3. Stimuli (Experiment 2) Stimuli were prepared in the same manner as for Experiment 1. On this occasion, however, stimulus manipulations were aimed at addressing issues of con®gurational coding, i.e. the degree to which the internal features are necessary or sucient in sheep face recognition and the degree to which their con®guration is important. Stimuli included (see Fig. 2): . Internal only Ð The external features of the face were covered using a black oval mask; . External only Ð The internal features were covered with a ¯eece texture leaving only the external face shape visible; . Scrambled Ð The internal features of the face were present but in a new spatial con®guration. All stimuli were given the same overall con®guration (mouth moved to top left of face etc . . .) and the edges of the features were then smoothed into the ¯eece texture; . Swapped features Ð The inner oval for the target face was swapped with the same area of the nonrewarded stimulus to create a chimeric-type stimulus.
1.4. Procedure The Y-maze apparatus and protocol has been
479
described elsewhere [17] and diered very little in this study. In brief, the animal entered a start box from where it could see to the ends of the two arms of a Ymaze. Here were projected two stimuli on back-projection screens located next to an exit door. As the animal entered the body of the Y-maze, it broke a light beam which triggered a timer. By approaching one of the pictures a second light beam was triggered (1.5 m from the ®rst) which stopped the timer and opened the door next to the relevant stimulus. The sheep then exited into a goal pen where an experimenter was waiting with either an empty or full bucket of food (sheep nuts). After receiving a mouthful of food the animal was ushered into a further holding pen to await the start of the next trial. Position of the target stimulus was randomised across trials. Each trial (of all 10 animals) took 3±5 min to complete and 20 trials were performed consecutively on each day. Stimuli were sorted into pairs of faces (four pairs of familiar and four pairs of unfamiliar) and the sheep were trained on one pair at a time. where one of the faces was consistently associated with food. When the animals achieved 80% accuracy or above over a day's trials the testing period began. This involved one day of 20 trials for each stimulus type (including others not described here) where the same two faces were used but in their degraded format. For instance, one day the animals might be tested on left hemi®eld faces for 20 trials and on the next day the stimuli might be internal features of the faces etc . . . The order of test stimuli was randomised between stimulus pairs. When all test stimuli had been run for one stimulus pair the next training set was started. During testing the animals were still rewarded for choosing the correct face. The only exception to the general protocol was when the chimeric stimulus set (Fig. 1(d)) and swapped feature stimuli (Fig. 2(d)) were used. In these cases both stimuli contained information from the original rewarded face. Hence either was a correct choice and the animal was always rewarded. 2. Results (Experiment 1) 2.1. Accuracy Two-way Analysis of Variance in recognition accuracy for `left only' against `right only' hemi®eld faces (Fig. 3(a)) showed a signi®cant main eect of hemi®eld (left vs right) (F = 4.24, df=1, 36, p < 0.05) with no eect of familiarity (F = 0.17, df=1, 36, p > 0.05) or any interaction (F = 0.13, df=1, 36, p > 0.05). Similarly with mirrored hemi®elds (Fig. 3(b)) there was a signi®cant main eect of the stimuli (left±left or right± right) (F = 5.22, df=1, 36, p < 0.05); but none for
480
J.W. Peirce et al. / Neuropsychologia 38 (2000) 475±483
between individuals but also from day to day within subjects, depending on conditions such as temperature. A paired-means t-test did reveal a signi®cant dierence between the response times of sheep choosing between chimeric faces (Fig. 3(c)), with responses being faster in the familiar face condition (t = 6.38, df=9, p < 0.001). 3. Results (Experiment 2) 3.1. Accuracy
Fig. 3. Results from Experiment 1: (a) accuracy of recognition and response times of 10 sheep over 80 trials (per experimental condition) discriminating between left and right hemifaces; (b) accuracy and response time during recognition of left±left and right±right mirrored faces; (c) choices made towards the `left chimeric face' i.e. that which had the left of the target face combined with the right of the distractor stimulus. All error bars show standard error of the mean across subjects. p < 0.05, p < 0.01 (see text for details of tests used).
Two-way Analysis of Variance (stimulus type X familiarity) for accuracy showed a main eect of stimulus type (F = 22.62, df=1, 36, p < 0.001) with the removal of external features (to make `internal only' faces) being most damaging to discriminations in both levels of familiarity (Fig. 4(a)). There was no main eect of familiarity although there was a signi®cant interaction between that and stimulus type (F = 7.40, df=1, 36, p < 0.01) suggesting that familiar faces were less aected by external feature removal. There also appears to be a trend for the scrambled feature stimuli to impair the identi®cation of familiar faces but not to have any eect on unfamiliars. For the chimeric-type `swapped features' stimuli, the face containing the external features of the rewarded face was chosen more often for each individual sheep on familiar and unfamiliar faces ( p < 0.05 in binomial test over 80 trials). However comparing the eect for familiar and unfamiliars we ®nd that the eect is less for familiars (t = 2.40, df=9, p < 0.05) suggesting that
familiarity (F = 2.06, df=1, 36, p > 0.05); or any interaction (F = 0.93, df=1, 36, p > 0.05). In the choice task where hemi®eld chimeric stimuli were used, an eect of familiarity was observed (Fig. 3(c)). Here, the binomial test showed a signi®cant LVF bias across the population (choice=61%, n = 800, p < 0.05) and for 5 out of 10 individual animals discriminating familiar faces. In the non-familiar condition however, no signi®cant eects were seen for individuals or the population. This dierence was con®rmed by a paired-means t-test between the familiar and unfamiliar conditions (t = 2.26, df=9, p < 0.05).
2.2. Response time Two-way Analysis of Variance yielded no signi®cant main eects or interactions for the response times to hemiface or mirrored stimuli. This is probably because the measure of the response time depends not only on the ability of the animal to recognise the face but also on factors such as motivation. These vary greatly
Fig. 4. Results from Experiment 2: (a) accuracy and response times over 800 trials discriminating between the `external only', `scrambled' and `internal only' stimuli; (b) choices made towards the stimulus containing the external features of the target face and the internal features of the distractor. All error bars show standard error of the mean. p < 0.05, p < 0.01 (see text for details of tests used).
J.W. Peirce et al. / Neuropsychologia 38 (2000) 475±483
in this instance animals are in¯uenced by the presence of internal features. 3.2. Response time Again there was a signi®cant eect of familiarity on decision time during the swapped-feature stimuli, with the familiar-face condition showing faster responses (t = 2.65, df=9, p < 0.05). As with the ®rst experiment, 2-way Analysis of Variance revealed no signi®cant eects of response time with the other stimuli. 4. Discussion The results from Experiment 1 showed that there was indeed a visual ®eld asymmetry in the recognition of the faces (and potentially other visual stimuli) by sheep, using three separate stimulus formats (hemi®eld, mirrored and chimeric). This poses some of the most robust and convincing evidence of perceptual asymmetry outside the domain of human experiments and the only such data from a non-primate species. It also supports the theory that in humans the left visual ®eld bias for face recognition results from a neural lateralisation, rather than a bias in scanning direction as suggested by Vaid and Singh [36]. They claim that this bias in scanning direction arises from the fact that most of the subjects of psychology experiments are native English speakers and always read left to right. Thus when they observe a face they look at the left ®rst. Clearly the present study is not explained by this hypothesis. The hemiface stimuli do not oer any insight as to the stage of recognition at which the asymmetry occurred (i.e. during encoding or retrieval). It may have occurred because the hemi®eld seen during testing had received more attention previously, or maybe because that hemi®eld was the only one activated during retrieval. However, the fact that the mirrored stimuli showed an equivalent eect implies that the asymmetry is occurring during the encoding phase of the task since for this stimulus the retrieval stage contained a whole face, the only dierence being the hemi®eld in which it had been experienced during learning. The reason that these mirrored and hemiface stimulus types did not show eects of familiarity may have been due to a ceiling eect. Note especially the mirrored face accuracy which reached almost 80% (i.e. training criterion) on three of the groups (familiar left, familiar right and unfamiliar left). Thus the chance of familiar animals showing a stronger LVF bias is reduced by the fact that they are at optimum performance already for the RVF. With chimeric stimuli this is avoided by the fact that the animals are forced to make a direct choice between the hemi®elds. Therefore
481
if a bias exists it should be identi®ed with a direct eect of the strength of that bias. From the fact that the chimeric stimuli did show a familiarity eect, and from the non-signi®cant trend in both response-time graphs (Fig. 3(a) and (b)) with the familiar LVF stimuli being recognised faster, we conclude that there is indeed a stronger visual ®eld bias for the familiar (most experienced) faces. We suggest that the hemi®eld and mirrored stimuli were simply not sensitive enough to identify this dierence in terms of accuracy. From Experiment 2 we can see that the sheep are predominantly using the external shape of faces in two-way discriminations, consistent with Dittrich's ®ndings in macaques [9]. However, it appears that like humans [11,41], they may rely less on such features when recognising highly familiar individuals. Removing the external features of unfamiliar animals reduced recognition performance almost to chance, whilst for familiar sheep the faces were still identi®ed to within 10% of the original training criterion. Disrupting the feature con®guration had little eect on face recognition but the in¯uence of familiarity might have been masked by a ceiling eect, as in the ®rst experiment. The swapped features also showed that whilst all face recognition was using external features predominantly, this was less the case in familiar faces where the faces with the correct internal features were also chosen on occasion. Thus we suggest that initial learning processes use local features solely as cues to identity but that the more sophisticated global and con®gurational cues may be used after a great deal of experience. The ®ndings may explain Kendrick et al.'s [18] ®nding that the inversion eect of sheep face recognition is greater with familiar faces. It also supports the suggestion of others [31,32] that inversion eects are the result of con®gurational or global coding methods used extensively in the recognition of faces. The fact that the expertise eect has appeared so strongly in the present experiment may represent the fact that for sheep a great deal more exposure to a face is required for it to become truly familiar. Humans may learn to use internal cues much more quickly when presented with a new face, which makes the study of such processes in humans more dicult. Combining the data from both sets of stimuli, we ®nd that familiar faces, recognised with increasing use of internal, con®gurational features, demonstrate a stronger LVF (and presumably RH) advantage. Therefore the study supports the theory that the RH dominance of occipital±temporal cortical activity in expert face recognition is caused by the use of con®gurational and global cues to identi®cation. Finally, the present study also sheds light on the debate between the development of motor vs perceptual asymmetries. In a recent review, Corballis [6]
482
J.W. Peirce et al. / Neuropsychologia 38 (2000) 475±483
claims that the advantage of motor asymmetry (in terms of handedness) and the left hemisphere localisation of language function has driven other forms of neural lateralisation such as the perceptual asymmetry described here. The present study describes an animal with no apparent need for motor asymmetry, and certainly no language, for which a perceptual asymmetry is manifest. For these animals the need to quickly identify cohorts appears to be the only task complex enough to require hemispheric specialisation. The model may give us a method by which we can examine more closely the relationship between the cerebral and perceptual asymmetries, and may also enable us to study the development of hemispheric lateralisation in general.
References [1] Assal G, Favre C, Anderes JP. [Nonrecognition of familiar animals by a farmer. Zooagnosia or prosopagnosia for animals]. Revue Neurologie 1984;140:580±4. [2] Bornstein B. Prosopagnosia. In: Halpern L, editor. Problems of dynamic neurology. Jerusalem: Hadassah Medical School, 1963. [3] Bornstein B, Sroka H, Munitz H. Prosopagnosia with animal face agnosia. Cortex 1969;5:164±9. [4] Broman M. Reaction-time dierences between the left and right hemispheres for face and letter discrimination in children and adults. Cortex 1978;14:578±91. [5] Bruyer R, Laterre C, Seron X, Feyereisen P, Strypstein E, Pierrard E, et al. A case of prosopagnosia with some preserved covert remembrance of familiar faces. Brain and Cognition 1983;2:257±84. [6] Corballis MC. Cerebral asymmetry: motoring on. Trends in Cognitive Sciences 1998;2:152±7. [7] De Renzi E, Perani D, Carlesimo GA, Silveri MC, Fazio F. Prosopagnosia can be associated with damage con®ned to the right-hemisphere Ð an MRI and PET study and a review of the literature. Neuropsychologia 1994;32:893±902. [8] Diamond R, Carey S. Why faces are and are not special Ð an eect of expertise. Journal of Experimental Psychology Ð General 1986;115:107±17. [9] Dittrich W. Representation of faces in longtailed macaques (Macaca fascicularis). Ethology 1990;85:265±78. [10] Ellis HD, Shepherd JW. Recognition of upright and inverted faces presented in the left and right visual ®elds. Cortex 1975;11:3±7. [11] Ellis HD, Shepherd JW, Davies GM. Identi®cation of familiar and unfamiliar faces from internal and external features: some implications for theories of face recognition. Perception 1979;8:431±9. [12] Hamilton CR. An assessment of hemispheric specialization in monkeys. Annals of the New York Academy of Science 1977;299:222±32. [13] Hamilton CR. Lateralization for orientation in split-brain monkeys. Behavioural Brain Research 1983;10:399±403. [14] Hamilton CR, Vermeire BA. Complementary hemisphericspecialization in monkeys. Science 1988;242:1691±4. [15] Hilliard RD. Hemispheric laterality eects on a facial recognition task in normal subjects. Cortex 1973;9:246±58. [16] Kanwisher N, Chun MM, McDermott J, Ledden PJ. Functional imaging of human visual recognition. Cognitive Brain Research 1996;5:55±67.
[17] Kendrick KM, Atkins K, Hinton MR, Broad KD, Fabrenys C, Keverne B. Facial and vocal discrimination in sheep. Animal Behaviour 1995;49:1665±76. [18] Kendrick KM, Atkins K, Hinton MR, Heavens P, Keverne B. Are faces special for sheep Ð evidence from facial and object discrimination-learning tests showing eects of inversion and social familiarity. Behavioural Processes 1996;38:19±35. [19] Kendrick KM, Mimmack ML, da Costa APC, Broad KD. Cfos mapping of neural substrates involved in face discrimination by sheep. Society for Neuroscience (Abstract) , 1998. [20] Levine SC, Levy J. Perceptual asymmetry for chimeric faces across the life span. Brain and Cognition 1986;5:291±306. [21] Levy J, Heller W, Banich MT, Burton LA. Asymmetry of perception in free viewing of chimeric faces. Brain and Cognition 1983;2:404±19. [22] Levy J, Trevarthen C, Sperry RW. Perception of bilateral chimeric ®gures following hemispheric disconnection. Brain 1972;95:61±78. [23] McNeil JE, Warrington EK. Prosopagnosia Ð a face-speci®c disorder. Quarterly Journal Of Experimental Psychology Section A Ð Human Experimental Psychology 1993;46:1±10. [24] Morris RD, Hopkins WD. Perception of human chimeric faces by chimpanzees Ð evidence for a right-hemisphere advantage. Brain and Cognition 1993;21:111±22. [25] Overman WH, Doty RW. Hemispheric specialization displayed by man but not macaques for analysis of faces. Neuropsychologia 1982;20:113±28. [26] Parr LA, Dove T, Hopkins WD. Why faces may be special: evidence of the inversion eect in chimpanzees. Journal Of Cognitive Neuroscience 1998;10:615±22. [27] Perrett DI, Mistlin AJ, Chitty AJ, Smith PJ, Potter DD, Broennimann R, et al. Specialized face processing and hemispheric-asymmetry in man and monkey Ð evidence from single unit and reaction-time studies. Behavioural Brain Research 1988;29:245±58. [28] Phelps MT, Roberts WA. Memory for pictures of upright and inverted primate faces in humans (Homo sapiens), Squirrel monkeys (Saimiri sciureus), and pigeons (Columba livia). Journal of Comparative Psychology 1994;108:114±25. [29] Puce A, Allison T, Gore JC, McCarthy G. Face-sensitive regions in human extrastriate cortex studied by functional MRI. Journal of Neurophysiology 1995;74:1192±9. [30] Rakover SS, Teucher B. Facial inversion eects: parts and whole relationship. Perception & Psychophysics 1997;59:752±61. [31] Rhodes G. Con®gural coding, expertise, and the right hemisphere advantage for face recognition. Brain and Cognition 1993;22:19±41. [32] Rock I. The perception of disoriented ®gures. Scienti®c American 1974;230:78±85. [33] Sergent J, Bindra D. Dierential hemispheric processing of faces Ð methodological considerations and reinterpretation. Psychological Bulletin 1981;89:541±54. [34] Sergent J, Signoret JL. Functional and anatomical decomposition of face processing Ð evidence from prosopagnosia and PET study of normal subjects. Philosophical Transactions of The Royal Society of London Series B Ð Biological Sciences 1992;335:55±62. [35] Swithenby SJ, Brautigam S, Bailey AJ, Hari R, Tesche CD. Processing of faces in the human brain: a multi-task MEG study. International Journal of Psychophysiology 1997;25:61. [36] Vaid J, Singh M. Asymmetries in the perception of facial aect Ð is there an in¯uence of reading habits. Neuropsychologia 1989;27:1277±87. [37] Wright AA, Roberts WA. Monkey and human face perception: inversion eects for human faces but not for monkey faces or scenes. Journal of Cognitive Neuroscience 1996;8:278±90.
J.W. Peirce et al. / Neuropsychologia 38 (2000) 475±483 [38] Yin RK. Looking at upside-down faces. Journal of Experimental Psychology 1969;81:141±5. [39] Yin RK. Face recognition by brain-injured patients: a dissociable ability? Neuropsychologia 1970;8:395±402. [40] Young AW, Ellis HD. An experimental investigation of developmental dierences in ability to recognise faces presented to
483
the left and right cerebral hemispheres. Neuropsychologia 1976;14:495±8. [41] Young AW, Hay DC, Mcweeny KH, Flude BM, Ellis AW. Matching familiar and unfamiliar faces on internal and external features. Perception 1985;14:737±46.
www.elsevier.com/locate/neuropsychologia
Con®gurational coding, familiarity and the right hemisphere advantage for face recognition in sheep J.W. Peirce*, A.E. Leigh, K.M. Kendrick Laboratory of Cognitive and Developmental Neuroscience, Babraham Institute, Babraham, Cambridge CB2 4AT, UK Received 4 September 1998; received in revised form 7 April 1999; accepted 9 June 1999
Abstract This study examined characteristics of visual recognition of familiar and unfamiliar faces in sheep using a 2-way discrimination task. Of particular interest were eects of lateralisation and the dierential use of internal (con®gurational) vs external features of the stimuli. Animals were trained in a Y-maze to identify target faces from pairs, both of which were familiar (same ¯ock as the subjects) or both of which were unfamiliar (dierent ¯ock). Having been trained to identify the rewarded face a series of stimuli were presented to the sheep, designed to test for the use of each visual hemi®eld in the discriminations and the use of internal and external facial cues. The ®rst experiment showed that there was a left visual hemi®eld (LVF) advantage in the identi®cation of `hemifaces', and `mirrored hemifaces' and `chimeric' faces and that this eect was strongest with familiar faces. This represents the ®rst evidence for visual ®eld bias outside the primate literature. Results from the second experiment showed that, whilst both familiar and unfamiliar faces could be identi®ed by the external features alone, only the familiar faces could be recognised by the internal features alone. Overall the results suggest separate recognition methods for socially familiar and unfamiliar faces, with the former being coded more by internal, con®gurational cues and showing a lateral bias to the left visual ®eld. # 2000 Elsevier Science Ltd. All rights reserved. Keywords: Asymmetry; Hemispheric lateralisation; Chimeric; Face processing; Expertise; Internal features
In humans, recognition of faces has been found to display a robust visual ®eld bias with faces presented to the left of ®xation being identi®ed faster [4] and more accurately [10,15] than those to the right. Similar results are found for children [4,40] and commisurotomy patients [22,33]. The left visual hemi®eld (LVF) also plays a disproportionate role in judgements of aective state with chimeric faces, where half of the face is smiling and half is frowning [20,21]. It is thought that this asymmetry in perception arises from the neural lateralisation which has been reported by various human brain imaging studies such as Positron Emission Topography (PET) [34], functional Magnetic Resonance Imaging (fMRI) [16,29] and Magneto-Ence* Corresponding author. Tel.: +44-1223-496364; fax: +44-1223496028. E-mail address: [email protected] (J.W. Peirce).
phalography (MEG) [35] in normal subjects, and by lesion localisation in prosopagnosic patients [7 for review]. All of these lines of research have implicated occipitotemporal cortex in face processing with activation predominantly (although not exclusively) in the right hemisphere. This region may not be dedicated to face perception per se but rather to some visual process on which face recognition relies heavily. Diamond and Carey [8] claim that faces are merely a category of objects which have a common con®guration and that through experience we have learnt to identify subtle variations in this common con®guration. They tested the hypothesis with the use of the inversion eect (whereby humans [e.g. 38,39] monkeys [28,37] and chimpanzees [26] are particularly bad at identifying faces presented upsidedown) and found that dog-show judges demonstrate an equivalent eect for dogs. Thus they suggest that
0028-3932/00/$ - see front matter # 2000 Elsevier Science Ltd. All rights reserved. PII: S 0 0 2 8 - 3 9 3 2 ( 9 9 ) 0 0 0 8 8 - 3
476
J.W. Peirce et al. / Neuropsychologia 38 (2000) 475±483
any visual objects with a common feature con®guration can show face-like eects if they are discriminated frequently. This theory is supported by neuropsychological data which ®nds that most prosopagnosic patients do not show agnosia purely for faces but also for other intra-category discriminations. For example there are reported prosopagnosic farmers who could no longer discriminate between their cows [1,3] or ornithologists who could no longer identify dierent birds [2]. There are a few cases, however, where the de®cit is found for human faces despite an intact ability to identify other animal faces such as cows [5] or sheep [23]. Rhodes [31] examined the hypothesis that expertise leads to lateralisation and con®gurational1 coding. She measured the ability to recognise faces of dierent ethnic backgrounds (Caucasian and Chinese) in conjunction with visual ®eld asymmetry and inversion (as a measure of con®gurational coding). Whilst verifying that con®gurational coding was correlated to hemi®eld bias, it was not dependent on the degree of familiarity with the ethnic background of the faces as predicted by the expertise hypothesis. It should be noted that the use of inversion may not be a true measure of con®gurational coding. Whilst it is frequently cited as the most likely cause for the eect there is evidence that inversion of individual features could also give rise to inversion de®cits [30]. Thus, in the present study, con®gurational coding is tested for directly with faces where the internal features have been masked or rearranged rather than inverted. In non-human primates evidence for an asymmetry in either neural activation or behavioural responses has been mixed. Dittrich [9] used line-drawings of faces and required their macaques to indicate the face which was showing a certain gesture. They then produced asymmetric chimeric faces and jumbled-feature versions of these stimuli to identify whether the internal and external, and the left and right sides of the face were being used dierentially. The results suggest that the right hemi®eld was being used most (contrary to human subjects) and also that, whilst the facial outline was the most important visual cue, the eye region and mouth were also used. Overman and Doty [25] used mirrored hemifaces (where one visual hemi®eld is mirrored onto the other) as stimuli and found an advantage for the left±left face in human subjects, as expected, but no equivalent bias in macaque subjects. Another method has used split brain macaques where presentations can be made purely to one brain hemisphere, and this did yield an LVF advantage [14] although previous attempts to do 1 Some authors use the term `con®gural' rather than con®gurational.
so by the same group had failed [12,13]. Finally, Morris and Hopkins [24] found a bias in two out of three chimps (pan troglodytes) from the Savage±Rumbaugh project, performing Levy's [21] happy/sad discrimination on chimeric faces. This appears to be the strongest evidence to date for such perceptual asymmetry in a non-human species. At an anatomical level, the functional mapping of the primate brain during face recognition tasks has not been attempted and, at a cellular level, electrophysiology has shown little evidence for a right hemisphere processing bias. One paper even reports more face sensitive cells in the left hemisphere than the right [27], although this may be an artefact of the technique which lends itself better to the descriptions of neuronal ®ring characteristics than to the quanti®cation of cell populations. Sheep have been shown to use visual cues to discriminate between the species or breed of faces presented in a Y-maze [4] as well as identifying faces of individual sheep [18]. In both studies signi®cant classic inversion eects were found for the faces of a familiar breed, and especially for socially familiar individuals [18]. In sheep, neither electrophysiological nor behavioural studies have previously attempted to identify any asymmetry, although recent data from in-situ hybridisation of c-fos has shown stronger right hemisphere activation during a face recognition task [19]. Thus, the ®rst experiment of the present paper was designed to test for a visual hemi®eld bias during face recognition. This diers slightly from the facial expression recognition used by a number of primate [e.g. 10,24] and human studies [e.g. 21] but is necessary since sheep show very few facial expressions. Whilst the recognition of facial identity may be a separable process from gesture recognition, it has also been shown to demonstrate visual ®eld bias [e.g. 31]. The second experiment tested directly for the use of con®gurational coding in sheep face recognition. Both experiments compared familiar and unfamiliar individuals to test for expertise eects. 1. Method 1.1. Subjects The animals used were 10 polled Dorset ewes. This is a hornless breed with high facial homogeneity. They had been extensively trained in the Y-maze procedure whereby they were given food reward for choosing a target stimulus and were habituated to the maze environment and the experimenters. They were kept amongst a ¯ock of 20±30 sheep. Food (grass or hay) and water was provided ad libitum throughout the study.
J.W. Peirce et al. / Neuropsychologia 38 (2000) 475±483
477
Fig. 1. Examples of stimuli used in Experiment 1: (a) a pair of training stimuli; (b) left hemifaces; (c) left-left mirrored faces; (d) chimeric faces. Note the high homogeneity between the stimuli.
478
J.W. Peirce et al. / Neuropsychologia 38 (2000) 475±483
Fig. 2. Examples of stimuli used in Experiment 2: (a) internal features only; (b) external features only; (c) scrambled features; (d) swapped-feature stimuli.
J.W. Peirce et al. / Neuropsychologia 38 (2000) 475±483
1.2. Stimuli (Experiment 1) Stimuli were created from pictures taken by digital camera (Canon DC50) of sheep within the same ¯ock as the subjects (familiar) or another ¯ock (unfamiliar). They were edited on a Pentium PC computer and printed in full colour onto photographic slides which were back-projected onto a screen next in the Y-maze such that the face is seen at roughly life-size. For training stimuli the background was edited out leaving just a head on a black screen. From these, the following experimental manipulations were made (see Fig. 1 for examples): . Hemi®eld Ð Half of the face was obscured by a black mask; . Hemi®eld mirrored Ð Half of the face was mirrored onto the other half (also referred to as left±left/ right±right faces). Similar to those used by Overman and Doty [25]; . Hemi®eld chimeric Ð The traditional chimeric face stimuli [21,24] where the left and right of two separate faces (in this case the rewarded and nonrewarded) are combined to form one.
1.3. Stimuli (Experiment 2) Stimuli were prepared in the same manner as for Experiment 1. On this occasion, however, stimulus manipulations were aimed at addressing issues of con®gurational coding, i.e. the degree to which the internal features are necessary or sucient in sheep face recognition and the degree to which their con®guration is important. Stimuli included (see Fig. 2): . Internal only Ð The external features of the face were covered using a black oval mask; . External only Ð The internal features were covered with a ¯eece texture leaving only the external face shape visible; . Scrambled Ð The internal features of the face were present but in a new spatial con®guration. All stimuli were given the same overall con®guration (mouth moved to top left of face etc . . .) and the edges of the features were then smoothed into the ¯eece texture; . Swapped features Ð The inner oval for the target face was swapped with the same area of the nonrewarded stimulus to create a chimeric-type stimulus.
1.4. Procedure The Y-maze apparatus and protocol has been
479
described elsewhere [17] and diered very little in this study. In brief, the animal entered a start box from where it could see to the ends of the two arms of a Ymaze. Here were projected two stimuli on back-projection screens located next to an exit door. As the animal entered the body of the Y-maze, it broke a light beam which triggered a timer. By approaching one of the pictures a second light beam was triggered (1.5 m from the ®rst) which stopped the timer and opened the door next to the relevant stimulus. The sheep then exited into a goal pen where an experimenter was waiting with either an empty or full bucket of food (sheep nuts). After receiving a mouthful of food the animal was ushered into a further holding pen to await the start of the next trial. Position of the target stimulus was randomised across trials. Each trial (of all 10 animals) took 3±5 min to complete and 20 trials were performed consecutively on each day. Stimuli were sorted into pairs of faces (four pairs of familiar and four pairs of unfamiliar) and the sheep were trained on one pair at a time. where one of the faces was consistently associated with food. When the animals achieved 80% accuracy or above over a day's trials the testing period began. This involved one day of 20 trials for each stimulus type (including others not described here) where the same two faces were used but in their degraded format. For instance, one day the animals might be tested on left hemi®eld faces for 20 trials and on the next day the stimuli might be internal features of the faces etc . . . The order of test stimuli was randomised between stimulus pairs. When all test stimuli had been run for one stimulus pair the next training set was started. During testing the animals were still rewarded for choosing the correct face. The only exception to the general protocol was when the chimeric stimulus set (Fig. 1(d)) and swapped feature stimuli (Fig. 2(d)) were used. In these cases both stimuli contained information from the original rewarded face. Hence either was a correct choice and the animal was always rewarded. 2. Results (Experiment 1) 2.1. Accuracy Two-way Analysis of Variance in recognition accuracy for `left only' against `right only' hemi®eld faces (Fig. 3(a)) showed a signi®cant main eect of hemi®eld (left vs right) (F = 4.24, df=1, 36, p < 0.05) with no eect of familiarity (F = 0.17, df=1, 36, p > 0.05) or any interaction (F = 0.13, df=1, 36, p > 0.05). Similarly with mirrored hemi®elds (Fig. 3(b)) there was a signi®cant main eect of the stimuli (left±left or right± right) (F = 5.22, df=1, 36, p < 0.05); but none for
480
J.W. Peirce et al. / Neuropsychologia 38 (2000) 475±483
between individuals but also from day to day within subjects, depending on conditions such as temperature. A paired-means t-test did reveal a signi®cant dierence between the response times of sheep choosing between chimeric faces (Fig. 3(c)), with responses being faster in the familiar face condition (t = 6.38, df=9, p < 0.001). 3. Results (Experiment 2) 3.1. Accuracy
Fig. 3. Results from Experiment 1: (a) accuracy of recognition and response times of 10 sheep over 80 trials (per experimental condition) discriminating between left and right hemifaces; (b) accuracy and response time during recognition of left±left and right±right mirrored faces; (c) choices made towards the `left chimeric face' i.e. that which had the left of the target face combined with the right of the distractor stimulus. All error bars show standard error of the mean across subjects. p < 0.05, p < 0.01 (see text for details of tests used).
Two-way Analysis of Variance (stimulus type X familiarity) for accuracy showed a main eect of stimulus type (F = 22.62, df=1, 36, p < 0.001) with the removal of external features (to make `internal only' faces) being most damaging to discriminations in both levels of familiarity (Fig. 4(a)). There was no main eect of familiarity although there was a signi®cant interaction between that and stimulus type (F = 7.40, df=1, 36, p < 0.01) suggesting that familiar faces were less aected by external feature removal. There also appears to be a trend for the scrambled feature stimuli to impair the identi®cation of familiar faces but not to have any eect on unfamiliars. For the chimeric-type `swapped features' stimuli, the face containing the external features of the rewarded face was chosen more often for each individual sheep on familiar and unfamiliar faces ( p < 0.05 in binomial test over 80 trials). However comparing the eect for familiar and unfamiliars we ®nd that the eect is less for familiars (t = 2.40, df=9, p < 0.05) suggesting that
familiarity (F = 2.06, df=1, 36, p > 0.05); or any interaction (F = 0.93, df=1, 36, p > 0.05). In the choice task where hemi®eld chimeric stimuli were used, an eect of familiarity was observed (Fig. 3(c)). Here, the binomial test showed a signi®cant LVF bias across the population (choice=61%, n = 800, p < 0.05) and for 5 out of 10 individual animals discriminating familiar faces. In the non-familiar condition however, no signi®cant eects were seen for individuals or the population. This dierence was con®rmed by a paired-means t-test between the familiar and unfamiliar conditions (t = 2.26, df=9, p < 0.05).
2.2. Response time Two-way Analysis of Variance yielded no signi®cant main eects or interactions for the response times to hemiface or mirrored stimuli. This is probably because the measure of the response time depends not only on the ability of the animal to recognise the face but also on factors such as motivation. These vary greatly
Fig. 4. Results from Experiment 2: (a) accuracy and response times over 800 trials discriminating between the `external only', `scrambled' and `internal only' stimuli; (b) choices made towards the stimulus containing the external features of the target face and the internal features of the distractor. All error bars show standard error of the mean. p < 0.05, p < 0.01 (see text for details of tests used).
J.W. Peirce et al. / Neuropsychologia 38 (2000) 475±483
in this instance animals are in¯uenced by the presence of internal features. 3.2. Response time Again there was a signi®cant eect of familiarity on decision time during the swapped-feature stimuli, with the familiar-face condition showing faster responses (t = 2.65, df=9, p < 0.05). As with the ®rst experiment, 2-way Analysis of Variance revealed no signi®cant eects of response time with the other stimuli. 4. Discussion The results from Experiment 1 showed that there was indeed a visual ®eld asymmetry in the recognition of the faces (and potentially other visual stimuli) by sheep, using three separate stimulus formats (hemi®eld, mirrored and chimeric). This poses some of the most robust and convincing evidence of perceptual asymmetry outside the domain of human experiments and the only such data from a non-primate species. It also supports the theory that in humans the left visual ®eld bias for face recognition results from a neural lateralisation, rather than a bias in scanning direction as suggested by Vaid and Singh [36]. They claim that this bias in scanning direction arises from the fact that most of the subjects of psychology experiments are native English speakers and always read left to right. Thus when they observe a face they look at the left ®rst. Clearly the present study is not explained by this hypothesis. The hemiface stimuli do not oer any insight as to the stage of recognition at which the asymmetry occurred (i.e. during encoding or retrieval). It may have occurred because the hemi®eld seen during testing had received more attention previously, or maybe because that hemi®eld was the only one activated during retrieval. However, the fact that the mirrored stimuli showed an equivalent eect implies that the asymmetry is occurring during the encoding phase of the task since for this stimulus the retrieval stage contained a whole face, the only dierence being the hemi®eld in which it had been experienced during learning. The reason that these mirrored and hemiface stimulus types did not show eects of familiarity may have been due to a ceiling eect. Note especially the mirrored face accuracy which reached almost 80% (i.e. training criterion) on three of the groups (familiar left, familiar right and unfamiliar left). Thus the chance of familiar animals showing a stronger LVF bias is reduced by the fact that they are at optimum performance already for the RVF. With chimeric stimuli this is avoided by the fact that the animals are forced to make a direct choice between the hemi®elds. Therefore
481
if a bias exists it should be identi®ed with a direct eect of the strength of that bias. From the fact that the chimeric stimuli did show a familiarity eect, and from the non-signi®cant trend in both response-time graphs (Fig. 3(a) and (b)) with the familiar LVF stimuli being recognised faster, we conclude that there is indeed a stronger visual ®eld bias for the familiar (most experienced) faces. We suggest that the hemi®eld and mirrored stimuli were simply not sensitive enough to identify this dierence in terms of accuracy. From Experiment 2 we can see that the sheep are predominantly using the external shape of faces in two-way discriminations, consistent with Dittrich's ®ndings in macaques [9]. However, it appears that like humans [11,41], they may rely less on such features when recognising highly familiar individuals. Removing the external features of unfamiliar animals reduced recognition performance almost to chance, whilst for familiar sheep the faces were still identi®ed to within 10% of the original training criterion. Disrupting the feature con®guration had little eect on face recognition but the in¯uence of familiarity might have been masked by a ceiling eect, as in the ®rst experiment. The swapped features also showed that whilst all face recognition was using external features predominantly, this was less the case in familiar faces where the faces with the correct internal features were also chosen on occasion. Thus we suggest that initial learning processes use local features solely as cues to identity but that the more sophisticated global and con®gurational cues may be used after a great deal of experience. The ®ndings may explain Kendrick et al.'s [18] ®nding that the inversion eect of sheep face recognition is greater with familiar faces. It also supports the suggestion of others [31,32] that inversion eects are the result of con®gurational or global coding methods used extensively in the recognition of faces. The fact that the expertise eect has appeared so strongly in the present experiment may represent the fact that for sheep a great deal more exposure to a face is required for it to become truly familiar. Humans may learn to use internal cues much more quickly when presented with a new face, which makes the study of such processes in humans more dicult. Combining the data from both sets of stimuli, we ®nd that familiar faces, recognised with increasing use of internal, con®gurational features, demonstrate a stronger LVF (and presumably RH) advantage. Therefore the study supports the theory that the RH dominance of occipital±temporal cortical activity in expert face recognition is caused by the use of con®gurational and global cues to identi®cation. Finally, the present study also sheds light on the debate between the development of motor vs perceptual asymmetries. In a recent review, Corballis [6]
482
J.W. Peirce et al. / Neuropsychologia 38 (2000) 475±483
claims that the advantage of motor asymmetry (in terms of handedness) and the left hemisphere localisation of language function has driven other forms of neural lateralisation such as the perceptual asymmetry described here. The present study describes an animal with no apparent need for motor asymmetry, and certainly no language, for which a perceptual asymmetry is manifest. For these animals the need to quickly identify cohorts appears to be the only task complex enough to require hemispheric specialisation. The model may give us a method by which we can examine more closely the relationship between the cerebral and perceptual asymmetries, and may also enable us to study the development of hemispheric lateralisation in general.
References [1] Assal G, Favre C, Anderes JP. [Nonrecognition of familiar animals by a farmer. Zooagnosia or prosopagnosia for animals]. Revue Neurologie 1984;140:580±4. [2] Bornstein B. Prosopagnosia. In: Halpern L, editor. Problems of dynamic neurology. Jerusalem: Hadassah Medical School, 1963. [3] Bornstein B, Sroka H, Munitz H. Prosopagnosia with animal face agnosia. Cortex 1969;5:164±9. [4] Broman M. Reaction-time dierences between the left and right hemispheres for face and letter discrimination in children and adults. Cortex 1978;14:578±91. [5] Bruyer R, Laterre C, Seron X, Feyereisen P, Strypstein E, Pierrard E, et al. A case of prosopagnosia with some preserved covert remembrance of familiar faces. Brain and Cognition 1983;2:257±84. [6] Corballis MC. Cerebral asymmetry: motoring on. Trends in Cognitive Sciences 1998;2:152±7. [7] De Renzi E, Perani D, Carlesimo GA, Silveri MC, Fazio F. Prosopagnosia can be associated with damage con®ned to the right-hemisphere Ð an MRI and PET study and a review of the literature. Neuropsychologia 1994;32:893±902. [8] Diamond R, Carey S. Why faces are and are not special Ð an eect of expertise. Journal of Experimental Psychology Ð General 1986;115:107±17. [9] Dittrich W. Representation of faces in longtailed macaques (Macaca fascicularis). Ethology 1990;85:265±78. [10] Ellis HD, Shepherd JW. Recognition of upright and inverted faces presented in the left and right visual ®elds. Cortex 1975;11:3±7. [11] Ellis HD, Shepherd JW, Davies GM. Identi®cation of familiar and unfamiliar faces from internal and external features: some implications for theories of face recognition. Perception 1979;8:431±9. [12] Hamilton CR. An assessment of hemispheric specialization in monkeys. Annals of the New York Academy of Science 1977;299:222±32. [13] Hamilton CR. Lateralization for orientation in split-brain monkeys. Behavioural Brain Research 1983;10:399±403. [14] Hamilton CR, Vermeire BA. Complementary hemisphericspecialization in monkeys. Science 1988;242:1691±4. [15] Hilliard RD. Hemispheric laterality eects on a facial recognition task in normal subjects. Cortex 1973;9:246±58. [16] Kanwisher N, Chun MM, McDermott J, Ledden PJ. Functional imaging of human visual recognition. Cognitive Brain Research 1996;5:55±67.
[17] Kendrick KM, Atkins K, Hinton MR, Broad KD, Fabrenys C, Keverne B. Facial and vocal discrimination in sheep. Animal Behaviour 1995;49:1665±76. [18] Kendrick KM, Atkins K, Hinton MR, Heavens P, Keverne B. Are faces special for sheep Ð evidence from facial and object discrimination-learning tests showing eects of inversion and social familiarity. Behavioural Processes 1996;38:19±35. [19] Kendrick KM, Mimmack ML, da Costa APC, Broad KD. Cfos mapping of neural substrates involved in face discrimination by sheep. Society for Neuroscience (Abstract) , 1998. [20] Levine SC, Levy J. Perceptual asymmetry for chimeric faces across the life span. Brain and Cognition 1986;5:291±306. [21] Levy J, Heller W, Banich MT, Burton LA. Asymmetry of perception in free viewing of chimeric faces. Brain and Cognition 1983;2:404±19. [22] Levy J, Trevarthen C, Sperry RW. Perception of bilateral chimeric ®gures following hemispheric disconnection. Brain 1972;95:61±78. [23] McNeil JE, Warrington EK. Prosopagnosia Ð a face-speci®c disorder. Quarterly Journal Of Experimental Psychology Section A Ð Human Experimental Psychology 1993;46:1±10. [24] Morris RD, Hopkins WD. Perception of human chimeric faces by chimpanzees Ð evidence for a right-hemisphere advantage. Brain and Cognition 1993;21:111±22. [25] Overman WH, Doty RW. Hemispheric specialization displayed by man but not macaques for analysis of faces. Neuropsychologia 1982;20:113±28. [26] Parr LA, Dove T, Hopkins WD. Why faces may be special: evidence of the inversion eect in chimpanzees. Journal Of Cognitive Neuroscience 1998;10:615±22. [27] Perrett DI, Mistlin AJ, Chitty AJ, Smith PJ, Potter DD, Broennimann R, et al. Specialized face processing and hemispheric-asymmetry in man and monkey Ð evidence from single unit and reaction-time studies. Behavioural Brain Research 1988;29:245±58. [28] Phelps MT, Roberts WA. Memory for pictures of upright and inverted primate faces in humans (Homo sapiens), Squirrel monkeys (Saimiri sciureus), and pigeons (Columba livia). Journal of Comparative Psychology 1994;108:114±25. [29] Puce A, Allison T, Gore JC, McCarthy G. Face-sensitive regions in human extrastriate cortex studied by functional MRI. Journal of Neurophysiology 1995;74:1192±9. [30] Rakover SS, Teucher B. Facial inversion eects: parts and whole relationship. Perception & Psychophysics 1997;59:752±61. [31] Rhodes G. Con®gural coding, expertise, and the right hemisphere advantage for face recognition. Brain and Cognition 1993;22:19±41. [32] Rock I. The perception of disoriented ®gures. Scienti®c American 1974;230:78±85. [33] Sergent J, Bindra D. Dierential hemispheric processing of faces Ð methodological considerations and reinterpretation. Psychological Bulletin 1981;89:541±54. [34] Sergent J, Signoret JL. Functional and anatomical decomposition of face processing Ð evidence from prosopagnosia and PET study of normal subjects. Philosophical Transactions of The Royal Society of London Series B Ð Biological Sciences 1992;335:55±62. [35] Swithenby SJ, Brautigam S, Bailey AJ, Hari R, Tesche CD. Processing of faces in the human brain: a multi-task MEG study. International Journal of Psychophysiology 1997;25:61. [36] Vaid J, Singh M. Asymmetries in the perception of facial aect Ð is there an in¯uence of reading habits. Neuropsychologia 1989;27:1277±87. [37] Wright AA, Roberts WA. Monkey and human face perception: inversion eects for human faces but not for monkey faces or scenes. Journal of Cognitive Neuroscience 1996;8:278±90.
J.W. Peirce et al. / Neuropsychologia 38 (2000) 475±483 [38] Yin RK. Looking at upside-down faces. Journal of Experimental Psychology 1969;81:141±5. [39] Yin RK. Face recognition by brain-injured patients: a dissociable ability? Neuropsychologia 1970;8:395±402. [40] Young AW, Ellis HD. An experimental investigation of developmental dierences in ability to recognise faces presented to
483
the left and right cerebral hemispheres. Neuropsychologia 1976;14:495±8. [41] Young AW, Hay DC, Mcweeny KH, Flude BM, Ellis AW. Matching familiar and unfamiliar faces on internal and external features. Perception 1985;14:737±46.