Adaptive face space coding in congenital prosopagnosia: Typical figural aftereffects but abnormal identity aftereffects

This article appeared in a journal published by Elsevier. The attachedcopy is furnished to the author for internal non-commercial researchand education use, including for instruction at the authors institution

and sharing with colleagues.

Other uses, including reproduction and distribution, or selling orlicensing copies, or posting to personal, institutional or third party

websites are prohibited.

In most cases authors are permitted to post their version of thearticle (e.g. in Word or Tex form) to their personal website orinstitutional repository. Authors requiring further information

regarding Elsevier’s archiving and manuscript policies areencouraged to visit:

http://www.elsevier.com/copyright

Author's personal copy

Neuropsychologia 49 (2011) 3801– 3812

Contents lists available at SciVerse ScienceDirect

Neuropsychologia

jo u rn al hom epa ge : www.elsev ier .com/ locate /neuropsychologia

Adaptive face space coding in congenital prosopagnosia: Typical figuralaftereffects but abnormal identity aftereffects

Romina Palermoa,b,c,∗, Davide Rivoltac,1, C. Ellie Wilsonc,2, Linda Jefferyb,d

a Department of Psychology, The Australian National University, Canberra, ACT, Australiab Australian Research Council Centre of Excellence in Cognition and its Disorders (CCD), Australiac Macquarie Centre for Cognitive Science (MACCS), Macquarie University, Sydney, NSW, Australiad School of Psychology, The University of Western Australia, Crawley, WA, Australia

a r t i c l e i n f o

Article history:Received 12 July 2011Received in revised form21 September 2011Accepted 26 September 2011Available online 1 October 2011

Keywords:Face recognitionProsopagnosiaIdentity aftereffectFigural aftereffectDevelopmentalAdaptive face codingNorm-based coding

a b s t r a c t

People with congenital prosopagnosia (CP) report difficulty recognising faces in everyday life and performpoorly on face recognition tests. Here, we investigate whether impaired adaptive face space coding mightcontribute to poor face recognition in CP. To pinpoint how adaptation may affect face processing, a groupof CPs and matched controls completed two complementary face adaptation tasks: the figural aftereffect,which reflects adaptation to general distortions of shape, and the identity aftereffect, which directly tapsthe mechanisms involved in the discrimination of different face identities. CPs displayed a typical figuralaftereffect, consistent with evidence that they are able to process some shape-based information fromfaces, e.g., cues to discriminate sex. CPs also demonstrated a significant identity aftereffect. However,unlike controls, CPs impression of the identity of the neutral average face was not significantly shiftedby adaptation, suggesting that adaptive coding of identity is abnormal in CP. In sum, CPs show reducedaftereffects but only when the task directly taps the use of face norms used to code individual identity.This finding of a reduced face identity aftereffect in individuals with severe face recognition problems isconsistent with suggestions that adaptive coding may have a functional role in face recognition.

© 2011 Elsevier Ltd. All rights reserved.

1. Introduction

People with prosopagnosia have severe difficulty recognisingthe identity of familiar faces, despite intact low-level vision andunimpaired general cognitive abilities. Individuals with acquiredprosopagnosia lose their ability to recognise faces after suffering ahead injury, such as a stroke (e.g., Barton, 2008; Ramon, Busigny, &Rossion, 2010). In contrast, individuals with congenital prosopag-nosia (CP; also referred to as developmental prosopagnosia) donot have a known brain injury but rather appear to have failedto develop adequate face recognition skills (Behrmann & Avidan,2005; Duchaine & Nakayama, 2006a). Though individuals with CPdo not report brain trauma, face recognition difficulties in CP areassociated with reduced brain volumes (Behrmann, Avidan, Gao, &Black, 2007; Furl, Garrido, Dolan, Driver, & Duchaine, 2011; Garrido

∗ Corresponding author at: Department of Psychology, The Australian NationalUniversity, Canberra, ACT 0200, Australia. Tel.: +61 2 6125 5545;fax: +61 2 6125 0499.

E-mail address: Romina.Palermo@anu.edu.au (R. Palermo).1 Now at: Department of Neurophysiology, Max Planck Institute for Brain

Research, Frankfurt, Germany.2 Now at: Department of Forensic and Neurodevelopmental Brain Sciences,

Institute of Psychiatry, King’s College London, UK.

et al., 2009) and compromised white matter tracts (Thomas et al.,2009) in occipital and temporal areas involved in face processing.CP may affect as many as 2.5% of the educated population (Bowleset al., 2009; Kennerknecht et al., 2006) and often runs in fami-lies (Duchaine, Germine, & Nakayama, 2007; Grüter et al., 2007;Lee, Duchaine, Wilson, & Nakayama, 2010; Schmalzl, Palermo, &Coltheart, 2008). The inability to reliably recognise the faces of closefriends, work colleagues and family members can have significantpsychosocial consequences for individuals with CP, such as anxietyand avoidance of social situations (Yardley, McDermott, Pisarski,Duchaine, & Nakayama, 2008).

The source of CPs difficulty with face recognition is not yet clear.One possibility is that the key perceptual mechanisms of face recog-nition are disrupted in CP. There is evidence that CPs, as a group,show deficits in one of these mechanisms, namely holistic cod-ing (Avidan, Tanzer, & Behrmann, 2011; Palermo et al., 2011) inwhich visual information across the whole of the face is integratedinto a unified representation (Maurer, Le Grand, & Mondloch, 2002;McKone & Yovel, 2009; Rossion, Dricot, Goebel, & Busigny, 2011).However not all individual cases of CP appear to have a deficit withthis perceptual mechanism (Le Grand et al., 2006; Schmalzl et al.,2008; Susilo et al., in press).

A second mechanism integral to the processing of faces, onewhich we examine in this study, is adaptive face coding (for reviews

0028-3932/$ – see front matter © 2011 Elsevier Ltd. All rights reserved.doi:10.1016/j.neuropsychologia.2011.09.039

Author's personal copy

3802 R. Palermo et al. / Neuropsychologia 49 (2011) 3801– 3812

see Rhodes & Leopold, 2011; Webster & MacLeod, 2011). Adap-tive face coding mechanisms are believed to generate face norms,which represent the average characteristics of faces that have beenexperienced (Leopold, O‘Toole, Vetter, & Blanz, 2001; Rhodes &Leopold, 2011). Individual faces are coded relative to these norms,which are continually updated by experience. Adaptive coding isreflected by face aftereffects, in which prolonged exposure to a facebiases subsequent perception in the opposite direction. Afteref-fects are not simply the adaptation of low-level properties, but alsoinvolve some high-level, likely face-specific, properties (e.g., Afraz& Cavanagh, 2008; Rhodes, Evangelista, & Jeffery, 2009; Susilo,McKone, & Edwards, 2010). Adaptation may calibrate the face cod-ing system to the current “diet” of faces so that discrimination isbest among the kinds of faces an individual is currently experi-encing (Chen, Yang, Wang, & Fang, 2010; Oruc & Barton, 2011;Rhodes & Leopold, 2011; Rhodes, Watson, Jeffery, & Clifford, 2010;Wilson, Loffler, & Wilkinson, 2002; Yang, Shen, Chen, & Fang, 2011).In this study, we examine whether abnormal adaptive face codingis associated with the face identity recognition impairments seenin CP.

Face aftereffects occur when perception of a face is influencedby exposure to a preceding face or faces. For example, adaptationto distorted faces (e.g., highly centre-contracted faces) causes aface figural aftereffect in which undistorted faces that are viewedsubsequently appear slightly distorted in the opposite way (e.g.,slightly centre-expanded) (see Fig. 1). Likewise, in the face identityaftereffect, adaptation to an individual face results in perceptionbeing biased toward an identity with opposite characteristics (seeFigs. 2 and 3). Adaptation has been found for every facial attributetested, including attributes that vary naturally among faces, suchas sex, race, expression (Webster, Kaping, Mizokami, & Duhamel,2004), identity (Leopold et al., 2001) and age (Schweinberger et al.,2010), and more arbitrary distortions of face shape (e.g., Carbon &Leder, 2006; Rhodes, Jeffery, Watson, Clifford, & Nakayama, 2003;Webster & MacLin, 1999). A variety of evidence suggests that faceaftereffects reflect adaptation of high-level coding mechanismsincluding face-specific coding mechanisms (Leopold et al., 2001;Rhodes et al., 2009, 2003; Watson & Clifford, 2003, 2006; Zhao &Chubb, 2001), though they also reflect contributions from low-levelretinotopic adaptation (Afraz & Cavanagh, 2008, 2009; Dickinson,Almeida, Bell, & Badcock, 2010) and mid-level shape adaptation(Susilo et al., 2010), all of which contribute to face coding. Faceaftereffects are therefore thought to reveal the underlying natureof the mechanisms of face perception.

We asked whether CPs experience two types of face aftereffect,a figural aftereffect and an identity aftereffect. Figural afteref-fects reflect adaptation of general shape information in faces andmeasure changes in perception without requiring identificationor discrimination of individual faces, and as such do not explic-itly tap face recognition processes. Figural aftereffects affect bothperceptual judgements and early brain responses (190–260 msafter stimulus onset as measured with event-related potentials,Burkhardt et al., 2010). These aftereffects have not previously beenstudied in CP and so it is currently unknown whether impairmentswill be evident. In fact, there are several reasons to think that CPsmay display typical figural aftereffects. First, most CPs are able toprocess some face shape information, such as those face attributesused as cues to discriminate sex (Le Grand et al., 2006; Nunn,Postma, & Pearson, 2001). Second, although face-specific codingdoes contribute to figural face aftereffects, mid-level shape cod-ing mechanisms can contribute substantially (Susilo et al., 2010).Therefore CPs may show typical figural aftereffects because thisaftereffect taps aspects of shape and face processing that are rel-atively unimpaired in CP. Either way, investigating face figuralaftereffects will provide important insights into the nature of theface coding deficits in CP.

In contrast, identity aftereffect tasks require discriminationbetween individual faces and thus more directly tap the mecha-nisms involved in the recognition of facial identity (Rhodes et al.,2009). In these tasks, the perception of the identity of the face,rather than its general appearance, is changed by adaptation. Thiseffect is most dramatically demonstrated when adaptation to a facecauses a previously “neutral” average face to resemble the oppo-site identity. Therefore, CPs might be expected to show atypicalidentity aftereffects, given that the primary deficit in CP is recog-nition of individual faces. To date, one study has investigated faceidentity aftereffects in a group of CPs. In contrast to expectations,Nishimura, Doyle, Humphreys and Behrmann (2010) reported thatface identity aftereffects for a group of CPs (n = 6) did not differ fromthose of typical participants. However, given the apparent hetero-geneity of CP (Le Grand et al., 2006; Schmalzl et al., 2008), it maybe premature to conclude that adaptive face coding is unimpairedon the basis of results from one study with a small sample of CPs.

We examined whether a larger sample of CPs show typical figu-ral aftereffects (n = 9; Experiment 1) and typical identity aftereffects(n = 14; Experiment 2). We predicted that CPs would show typicalfigural aftereffects because this aftereffect taps aspects of shapeand face processing that appear to be relatively unimpaired in CP.However, we predicted that CPs would show atypical identity after-effects because this aftereffect taps the specific processes used toindividuate faces. We used adaptation tasks in which the adaptand test faces were different sizes to minimise the contributionof low-level, retinotopic adaptation (Zhao & Chubb, 2001). It ispossible that the apparently typical aftereffects Nishimura et al.(2010) found for CPs were driven by adaptation of low-level visualattributes because adapt and test images were the same size.

2. Methods and results

2.1. Participants

2.1.1. Congenital prosopagnosicsThe CP group comprised 14 people (4 males, aged between 20 and 60 years,

M = 37.93, SD = 13.74) who reported everyday face recognition difficulties andshowed impaired performance on tests of facial identity recognition. Most con-tacted us via our online prosopagnosia register: http://www.maccs.mq.edu.au/research/projects/prosopagnosia/. Most of this group completed the screening testsused to confirm prosopagnosia (detailed below) at MACCS, although a few weretested at their home and one was assessed at the Australian National University(ANU). Data has been reported for some of these individuals previously (Bowleset al., 2009; Palermo et al., 2011; Rivolta, Palermo, Schmalzl, & Coltheart, in press;Susilo et al., in press).

All 14 CPs completed the identity aftereffect task, and ten of these also com-pleted the figural aftereffect task (the latter task was not administered to F30, F33,F23-1, F50). However, the data from M53 was excluded from analysis of the figuralaftereffect as no maxima could be obtained for one curve (see Experiment 1, Section3.3). Ten CPs completed both the figural and identity aftereffect tasks at MACCS andwere reimbursed at the rate of $30, one CP completed the identity aftereffect task atANU and was reimbursed $15 and three CPs completed the identity aftereffect taskat their home under the supervision of one of the researchers.

The CPs reported normal or corrected-to-normal vision, and demonstrated nor-mal range contrast sensitivity (Functional Acuity Contrast Test [FACT], Vision SciencesResearch Corporation, 2002), colour perception (Ishihara Test for Colour Blindness,Ishihara, 1925) and ability to judge the length, size and orientation of lines (Birm-ingham Object Recognition Battery [BORB], Riddoch & Humphreys, 1993).3 None hadany difficulty with basic level object recognition, as assessed with the BORB. IQ,as measured with the Raven Coloured Progressive Matrices (Raven, Raven, & Court,1998), was within the normal range. No CP reported any psychiatric or neurologicalproblems. Given that face processing problems are common in Autism SpectrumDisorders (ASD), the Autism Spectrum Quotient ([AQ], Baron-Cohen, Wheelwright,Skinner, Martin, & Clubley, 2001), a self-report questionnaire assessing the numberof autistic traits was also adminstered. No CP scored 32 or above, which would beindicative of an ASD.

As in previous studies (e.g., Palermo et al., 2011; Rivolta et al., in press), prosopag-nosia was determined via performance on two tests of face identity memory (MACCS

3 The FACT was not administered to F47 and M57; The Ishihara Test for ColourBlindness was not administered to F23-1.

Author's personal copy

R. Palermo et al. / Neuropsychologia 49 (2011) 3801– 3812 3803

Fig. 1. Figural aftereffect (Experiment 1). (A) Sample adapting face with extreme contraction (−50). (B) Sample test faces for one individual showing distortion level changesin 10% steps (from −50, most contracted on the left to 50, most expanded, on the right. The undistorted face is in the middle).

Famous Face Test 2008 [MFFT]; Cambridge Face Memory Test [CFMT] Duchaine &Nakayama, 2006b); and one of face perception (Cambridge Face Perception Test[CFPT], Duchaine et al., 2007). As can be seen in Fig. 4, the individuals in the CPgroup performed at least two standard deviations (SDs) below control norms onone or both tests of face memory, and some also showed significant impairment onthe test of face perception.

MACCS Famous Face Test 2008 [MFFT]. The MFFT assesses memory for facescommonly seen in print and digital media in Australia. The task is described indetail in the Supplementary Materials. Briefly, 40 faces are presented (20 famous,

20 unfamiliar) and participants initially judge whether it is familiar or not. Anindex of familiarity, d′ , was calculated, with higher d′ scores associated with betterperformance (i.e., more hits and fewer false alarms) (Macmillan & Creelman, 1991).The d′ scores of the CPs were significantly lower than those of the control groupsused as a comparison in these experiments, indicating a reduced sense of familiarityfor famous faces (Table 1).

For each of the 20 famous faces, participants are also asked to identify the faceby providing its name or other specific autobiographical information, such as thename of a film or TV show they were in, the sport that they played. Following this,

Fig. 2. A simplified face-space showing an average face, two target faces, Dan and Jim, and their opposite “antifaces”. “Weaker” versions of each target are made by morphingeach target with the average face by different amounts e.g., 60% Dan (0.60) and 30% Dan (0.30) as shown here. Adapting to antiDan facilitates recognition of Dan, so that“weaker” versions of Dan are more accurately identified, but does not facilitate recognition of Jim.

Reprinted with kind permission from Jeffery et al. (in press).

Author's personal copy

3804 R. Palermo et al. / Neuropsychologia 49 (2011) 3801– 3812

Fig. 3. Identity aftereffect (Experiment 2). Sample adapt and test stimuli for one target pair, Dan and Jim. (A) Adapting faces, antiDan and antiJim. (B) Test stimuli rangingfrom 90% Jim (−0.9) through to 90% Dan (0.9).

Reprinted with kind permission from Jeffery et al. (2011).

Table 1Mean scores (SDs in parentheses) for age, Autism Quotient (AQ) (Baron-Cohen et al., 2001), MACCS Famous Face Test 2008 (MFFT), Cambridge Face Memory Test (CFMT)(Duchaine & Nakayama, 2006b), Cambridge Face Perception Test (CFPT) (Duchaine et al., 2007), for the prosopagnosic and control groups completing the (a) identity, and (b)figural aftereffect tasks. Independent samples t-tests confirmed that age and AQ did not differ between the prosopagnosic and control groups, whereas performance on thetests of facial identity recognition did. See text for more details on the tests and calculation of z and d′ scores.

Prosopagnosics (n = 9) Controls (n = 19) t-Test

(a) Figural aftereffectAge 38.00 (14.87) 37.95 (13.54) t(26) < 1AQ 17.89 (8.45) 13.74 (5.34) t(26) = 1.59, p > .1MFFT familiarity d′ 1.52 (1.19) 2.58 (.77) t(26) = 2.84, p < .01MFFT identification Z-score −2.58 (1.14) −.35 (.85) t(26) = 5.82, p < .001CFMT Z-score −2.14 (.35) .20 (.83) t(26) = 8.10, p < .001CFPT Z-score −1.47 (1.08) .11 (.74) t(26) = 4.57, p < .001

Prosopagnosics (n = 14) Controls (n = 23) t-Test

(b) Identity aftereffectAge 37.93 (13.74) 40.22 (13.43) t(35) < 1AQ 17.29 (7.35) 15.13 (6.12) t(35) < 1MFFT familiarity d′ 1.69 (.99) 2.49 (.80) t(35) = 2.69, p < .02MFFT identification Z-score −2.61 (.93) −.30 (.83) t(35) = 7.84, p < .001CFMT Z-score −2.08 (.43) .18 (.85) t(34) = 10.65, p < .001a

CFPT Z-score −1.68 (1.48) .18 (.73) t(17) = 16.88, p < .001a

MFFT = MACCS Famous Face Test.CFMT = Cambridge Face Memory Test (Duchaine & Nakayama, 2006b).CFPT = Cambridge Face Perception Test (Duchaine et al., 2007).

a t adjusted for unequal variance.

-6.00

-5.00

-4.00

-3.00

-2.00

-1.00

0.00

1.00

M60M57M53M20F50F47F40F37aF37F33F30F23aF23F21

Z-score

Prosopagnosics

MFFT CFMT CFPT

Fig. 4. Age and sex-adjusted Z-scores for the 14 prosopagnosics (labelled by sexand age), on three tests of facial identity recognition – the MACCS Famous Face Test(MFFT), the Cambridge Face Memory Test (CFMT, Duchaine & Nakayama, 2006b)and the Cambridge Face Perception Test (CFPT, Duchaine et al., 2007).

the famous person’s name and relevant autobiographical information (e.g., occupa-tion) are presented, and participants report whether the famous person was actuallyknown to them. People that are unknown are excluded from further analyses, sothat accuracy is calculated as the percentage of correctly recognised faces of knownfamous people. Age-appropriate Z-scores were calculated for each of the 14 CPs (seeSupplementary Materials for details), and vary from −0.95 to −4.39 (see Fig. 4). TheCPs were substantially and significantly poorer at identifying the faces of famouspeople than the age-matched controls in this study (Table 1).

Cambridge Face Memory Test [CFMT] (Duchaine & Nakayama, 2006b). The CFMTassesses face learning and memory. Participants learn the faces of six people, andthen select the learnt faces from two similar distractors on trials in which the testfaces are either identical to the learnt images, varied in lighting and viewpoint, ordegraded with visual noise. The CFMT (and the CFPT) were administered, followingstandard instructions, on a 15-in. MacBook or 17-in. e-mac, with participants seateda comfortable distance from the screen. Total scores on the upright CFMT weretransformed to age-adjusted Z-scores (using age-based norms reported in Bowleset al., 2009), with the CPs scoring between −1.39 and −2.83 below the Australiansample (see Fig. 4),4 significantly poorer than the controls in this study (Table 1).

4 The CFMT z-score (−1.39) for F30-1 likely overestimates her face recogni-tion skills as it was the second time she completed the test (the initial score six

Author's personal copy

R. Palermo et al. / Neuropsychologia 49 (2011) 3801– 3812 3805

Cambridge Face Perception Test [CFPT] (Duchaine et al., 2007). The CFPT assessesthe ability to perceptually discriminate between similar faces. On each trial, par-ticipants are required to order a series of six morphed front-view faces in terms oftheir similarity to a target face photographed from a three-quarter view. For the CPs,Z-scores for upright faces calculated using the age- and sex-based norms in Bowleset al. (2009), ranged from 0.53 to −5.17 (see Fig. 4), significantly poorer than thoseof controls in this study (Table 1).

2.1.2. Control participantsThe control group comprised 235 Caucasian individuals (10 males), aged

between 19 and 62 years (M = 40.22, SD = 13.43), recruited from Macquarie Uni-versity and the general community. They reported normal, or corrected-to-normalvision, and did not report any brain injury or other neurological or psychologicalcondition likely to affect face recognition. Data from all 23 controls were assessedfor the identity aftereffect, and a subset of 19 for the figural aftereffect (data fromfour controls were excluded; one because no maxima could be obtained for onecurve, and three due to poor fit: R2 < 0.5). In addition to completing the figural andidentity aftereffect tasks, the controls also completed tests of identity recognition(MFFT, CFMT & CFPT), and the AQ (all controls scored below the cut-off of 32 thatindicates a possible autism spectrum disorder). Independent samples t-tests con-firmed that mean age and AQ scores of the CP and control groups did not differ,whereas performance on the tests of facial identity recognition did (Table 1). Theparticipants were assessed at MACCS and reimbursed $30 for the 2-h session.

3. Experiment 1: figural aftereffect

The figural aftereffect task we used, similar to that devised byRhodes et al. (2003), involves rating faces for attractiveness, bothbefore and after adaptation to a set of contracted faces (see Fig. 1).In typical adults, consistent exposure to contracted faces shifts themost attractive image toward the adapting distortion, so that con-tracted faces are rated as more attractive after adaptation thanbefore, suggesting that perceptual adaptation is rapidly recalibrat-ing what looks average and what looks attractive (Rhodes et al.,2003). This task is suited for use with CPs, as their ratings of attrac-tiveness are generally consistent with those of typical participants(Carbon, Grüter, Grüter, Weber, & Lueschow, 2010, but see Le Grandet al., 2006 for evidence that some individual CPs may be impairedat judging attractiveness when the features of faces are displaced).

3.1. Stimuli

Twenty grayscale photographs of young adult females wereplaced within black oval masks that covered the hair but not theinner hairline or face outline. Ten of these were used in the adapt-ing phase, and were distorted by −50% using the spherize functionin Adobe Photoshop, so that the centre of the face was contracted(Fig. 1A). The faces approximated 5 × 7 degrees of visual angle whenviewed at an approximate distance of 50 cm, on a 17-in. e-mac. Theother ten faces were used in the test phase, with each face usedto form a morphed continua containing 11 images that varied sys-tematically in distortion level, by setting the spherize function to−50, −40, −30, −20, −10, 0, +10, +20, +30, +40, and +50 (Fig. 1B).The test faces were reduced in size by 25%, to control for retino-topic adaptation, and were approximately 3.5 × 5 degrees of visualangle. Stimulus presentation was controlled with SuperLab Pro 1.75(Cedrus Corp.).

3.2. Procedure

In the pre-adaptation rating phase, each of the 110 test faceswas presented for 1500 ms, in a random order, in a box with theword “rate” at the top and bottom. Participants were instructed

weeks prior was not recorded due to computer malfunction) and controls, on aver-age, improve 6.3 percentage points when re-tested approximately 6-months later(Wilmer et al., 2010).

5 Twenty-seven control participants were initially tested; data of four wereexcluded due to poor performance on one or more of the tests of face recognition:<−1.7 SD on the CFMT; <−2.7 SD on the CFPT, <−4.4 SD on the MFFT-08.

to keep looking at the face on the screen for the duration ofthe presentation. A 9-point rating scale was presented after eachface, which participants used to rate the attractiveness of the face(1 = unattractive, 9 = attractive). Participants were asked to try anduse the full range of the scale. It was also noted that the faces weredistorted to varying degrees and hence may all look moderatelyunattractive, relative to everyday faces, but that they should try anddiscriminate among this set. In the subsequent adaptation phase,participants adapted to the contracted (−50%) distorted faces byviewing the 10 adapting faces, in a random order, for a total of5 min. Each face was shown for 4 s, with a 200 ms interstimulusinterval. In the final post-adaptation rating phase, the 110 test faceswere once again rated for attractiveness on the 9-point scale. Theywere randomly presented, in a box with the word “rate” at the topand bottom. Adaptation to contracted images was maintained byalternating the test faces (shown for 1500 ms) to be rated with ran-domly chosen adapting faces (shown for 8 s). A 500 ms blank screenseparated the adapting and test images.

3.3. Results and discussion

We calculated the mean attractiveness rating for each distortionlevel both before and after adaptation for each participant. Giventhe small and unequal sample sizes we used a non-parametric test,the Mann–Whitney U, when comparing the two groups.

Fig. 5(a) and (b) shows the mean distortion ratings for CPs andControls. Both groups show the predicted shift to rate contractedfaces as more attractive after adaptation. To estimate the distortionlevel of the most attractive faces both before and after adaptationthird-order polynomials were fitted to each participant’s data (seeFig. 5(c) and (d)) and the curve maxima taken as the “most nor-mal” distortion. Fits (R2) were good for both groups (CPs: M = .82,Range = 0.51–0.99, Controls: M = .84, Range = 0.52–0.97).

The size of the aftereffect for each participant was calculated asthe difference between the distortion level rated as most attractivebefore and after adaptation (after–before) so that a positive scoreindicates an aftereffect in the predicted direction (SupplementaryFig. 1 shows the size of the aftereffect for each control (a) andCP (b) participant). Both groups showed significant aftereffectsi.e., differences greater than 0 (CPs: M = 12.56, SE = 3.91, t(8) = 3.21,p = .012, d = 1.07, Controls: M = 13.56, SE = 2.26, t(18) = 6.00, p < .001,d = 1.38). The size of the CPs’ aftereffects did not differ significantlyfrom those of controls (U = 83.00, z = .12, p = .923, r = .02). CPs andcontrols also rated the attractiveness of the stimuli similarly (seeFig. 5) and perceived similar distortions to be the most attractivebefore and after adaptation (Before: CPs M = 5.28, SE = 2.22, Con-trols M = 6.59, SE = 2.08, U = 75.00, z = 0.517, p = .629, r = .10, After:CPs M = −7.28, SE = 3.50, Controls M = −6.97, SE = 1.60, U = 80.00,z = 0.271, p = .809, r = .05).

These results show, for the first time, that CPs experience figuralface aftereffects and that these aftereffects do not differ in strengthfrom those of typical controls. Further, CPs rated the attractive-ness of faces similarly to controls in the absence of adaptation,with similar levels of distortion rated as most attractive and similarsensitivity to variations in the amount of distortion. This is consis-tent with other work showing that CPs rating of attractiveness aresimilar to those of controls (Carbon et al., 2010).

4. Experiment 2: identity aftereffect

Here, we use an identity aftereffect task, that more directly tapsinto the mechanisms used to code facial identity. In this task, adapt-ing to an individual face (e.g., “antiDan” in Fig. 2) temporarily shiftsthe norm closer to that face, causing the average (identity-neutral)face to look more like its computationally opposite identity (i.e.,Dan) (c.f., Leopold et al., 2001).

Author's personal copy

3806 R. Palermo et al. / Neuropsychologia 49 (2011) 3801– 3812

Fig. 5. Figural aftereffect (Experiment 1). Mean attractiveness as a function of distortion level, before (open circles) and after (closed circles) adaptation to contracted faces(a) pooled over all CPs (n = 9), (b) pooled over all controls (n = 19), (c) for one example CP, and (d) for one example control. Fitted third-order polynomials are shown.

Participants were randomly assigned a pair of target faces tolearn, either Dan and Jim or Ted and Rob. We use Dan and Jim asexamples. Participants first learned to identify the pair of targetfaces (e.g., Dan and Jim) and then weaker versions of these targets(e.g., Dan and Jim’s “brothers”). This task was designed to examineidentity aftereffects in children and takes the format of a game inwhich participants are instructed that they will see “robbers” (e.g.,adapt faces – antiDan or antiJim; Fig. 3) and then see the face ofthe “catcher” (e.g., test faces – Dan or Jim or one of their broth-ers) (Jeffery et al., in press; Nishimura, Maurer, Jeffery, Pellicano, &Rhodes, 2008; Pellicano, Jeffery, Burr, & Rhodes, 2007; Pimperton,Pellicano, Jeffery, & Rhodes, 2009). The task of the participant wasto identify the briefly presented “catcher” so that a reward could beallocated to the correct team. As adapting to antiDan should biasparticipants to see “Dan” whereas adapting to antiJim should biasparticipants to see “Jim”, the proportion of “Dan” responses shouldbe higher after adapting to antiDan than after adapting to antiJim.

The task (using general procedure of Jeffery et al., in press) wasas in as Nishimura et al. (2010) except for four variations. First,and most importantly, a size change between adapt faces and testfaces was included to remove low-level retinotopic contributions.Second, there were small differences in test-face identity strengthvalues (−0.9, −0.6, −0.3, 0.0, 0.3, 0.6 and 0.9 in the current study;−0.8, −0.6, −0.4, −0.2, 0.0, 0.2, 0.4, 0.6, 0.8 in Nishimura et al.). Third,we included Ted–Rob, as well as Dan–Jim pairs. Finally, there wasno baseline condition in our design.

4.1. Stimuli

The stimuli (or a subset of the stimuli) have been used inprevious studies (Rhodes & Jeffery, 2006; Pellicano et al., 2007;Nishimura et al., 2008, 2010; Jeffery et al., in press). The target

faces were grayscale photos of four male faces, labelled “Dan”, “Jim”,“Rob” and “Ted”. We use Dan and Jim as examples. Reduced identitystrengths of each target face were made by morphing each targetface with an average male face (composed of 20 male faces averagedby placing 187 landmark points on each face), so that each targetface contributed either 90%, 60%, 30% or 0%, using Gryphon Morph2.5 (Maxwell, 1994). These faces were used as test faces (additional40% morphs were made in the same way for use in training trials).Thus, test stimuli were presented at 7 different identity strengths(−0.9, −0.6, −0.3, 0.0, 0.3, 0.6 and 0.9), which ranged from stronglyJim (90% Jim or “−0.9”) through to the average (0) and through tostrongly Dan (90% Dan or “0.9”) (see Fig. 2).

For each target face (e.g., Dan and Jim) an opposite or “antiface”(e.g., antiDan and antiJim) was created to serve as an adaptor (seeFig. 2). Antifaces were created by morphing away from the aver-age face along the same identity trajectory but in the directionopposite to each target face by 80%, using Gryphon Morph (afterLeopold et al., 2001). For example, if Dan has a large chin (relativeto average), antiDan will display a proportionately smaller-than-average chin. Grey oval masks obscured the hair, upper hairlineand the outer edge of the ears of each image (see Fig. 3). The adapt-ing anti-faces faces were 25% larger than the test faces; the testfaces were approximately 5 × 6, and the adapt faces 7 × 8 degreesof visual angle, when viewed from an average distance of 50 cm.Stimulus presentation, on a 17-in. e-mac or 15-in. mac powerbook,was controlled with SuperLab 4 (Cedrus Corp.).

4.2. Procedure

4.2.1. Training and practiceAn initial phase ensured that all participants were able to iden-

tify the two identities (e.g., Dan and Jim) reliably and that the

Author's personal copy

R. Palermo et al. / Neuropsychologia 49 (2011) 3801– 3812 3807

participants demonstrated sensitivity to identity strength. As inNishimura et al. (2010), the CPs were all able to do this relativelysimple discrimination. However, it is worth noting that we do notknow whether the CPs discriminated the identities using the sameprocesses as controls, or rather used atypical compensatory strate-gies that are sufficient for this type of task.

Participants were initially shown a printout of the “team cap-tains”, e.g., Dan and Jim, side by side, labelled with their names. Thepair of target faces was then shown on the computer screen, againlabelled with their names. Participants were asked to press a bluekeyboard key labelled “Dan” whenever they saw Dan and a red keylabelled “Jim” whenever they saw Jim. Participants were then givenpractice at identifying the targets, which were initially shown untila response was made, and then for only 400 ms. Auditory feedbackwas used to provide feedback on accuracy. Participants were thenfamiliarised with lower strength versions of the target faces (60%and 40% versions of each target face), which were introduced asDan’s and Jim’s brothers, and shown as printouts and then imageson the computer screen. A practice block followed in which par-ticipants were asked to press the key corresponding to the teamleader (Dan or Jim) whenever they saw a member of his team (i.e.,the target or one of his brothers). As before, the targets were ini-tially shown until a response was made, and then only shown for400 ms, and feedback was provided.

4.2.2. Adaptation taskParticipants were told that they would see the face of a “robber”

(adapting stimulus) followed by the face of the person who caughtthe robber. Their task was to identify which team (Dan or Jim) theperson who caught the robber belonged to. The faces of the tworobbers were shown to participants and two practice trials werecompleted.

Adapting antifaces (antiDan or antiJim) were shown for5000 ms, followed by a 150 ms interstimulus interval, followed bya 400 ms test stimulus (“catcher”) and then a blank grey screen thatremained until a response was made. Participants were instructedthat it was very important to look at the adapting antiface for theentire time it was shown as otherwise they may miss the face ofthe “catcher”, and the experimenter monitored compliance withthis instruction at the beginning of the experiment. Pressing thespacebar initiated the next trial, which commenced after 300 ms.There were 84 trials, with 6 trials presented at each of the 7 iden-tity strengths (−0.9, −0.6, −0.3, 0.0, 0.3, 0.6 and 0.9), for each ofthe two adapting conditions (antiDan or antiJim). To avoid buildingup adaptation to one face, the trials were presented in a pseudo-random order in which the same adapting face did not appear onmore than two consecutive trials.

4.3. Results

For each participant, we calculated the proportion of “Dan”responses at each identity strength for both adaptation condi-tions (adapt antiDan, adapt antiJim) (as per Nishimura et al., 2010;Pellicano et al., 2007). Fig. 6 shows the group means (a and b)and data from two representative individual participants (c and d)(Supplementary Fig. 1 displays the data for each CP (c) and control(d) participant). Adaptation to antiDan should result in a greaterproportion of “Dan” responses than adaptation to antiJim, hencethe curve for adapt antiDan should be left of that for adapt antiJim.Both groups show this pattern.

We measured the size of the aftereffect by taking the differencebetween the Point of Subjective Equality (PSE) for the two adaptingconditions (e.g., PSE adapt antiJim-PSE adapt antiDan) for eachparticipant, so that a positive score indicated an aftereffect in thepredicted direction. The PSE represents the identity strength thatwas classified as “Dan” and “Jim” equally and therefore represents

the identity strength that is perceived as “neutral” for identity.To determine the PSE we fit cumulative Gaussians to each partic-ipant’s data for each adaptation condition and took the mean ofeach function as the PSE. Fits (R2) were good for both groups (CPs:M = 0.97, SE = 0.01, Range = 0.74–1.00, Controls: M = 0.98, SE = 0.04,Range = 0.73–1.00). Both groups showed significant aftereffects(CPs: M = 0.13, SE = 0.05, t(13) = 2.89, p = .013, d = 0.77, Controls:M = 0.17, SE = 0.04, t(22) = 4.38, p < .001, d = 0.91). CPs’ aftereffectswere numerically smaller than those of controls but the groupsdid not differ significantly, U = 140.5, z = 0.643, p = .526, r = .11.

We also assessed the effects of adaptation by examining howperception of the identity-neutral average face (0 identity strength)was affected by adaptation condition. The average face is uniqueamong the test stimuli because it does not contain a greater pro-portion of one identity over the other. Following Pellicano et al.(2007) we compared the proportion of times the average face wasidentified as the identity opposite the adapting face, in the twogroups. CPs showed a significantly reduced adaptation effect rel-ative to controls (CPs: M = .55, SE = .03, Controls: M = .64, SE = .03,U = 93.50, z = 2.14, p = .033, r = .35). Further, CPs’ identifications ofthe average did not differ from chance (0.5), t(13) = 1.66, p = .122,d = 0.44, whereas controls identified the average as the identityopposite the adapting face significantly more often than chance,t(22) = 4.05, p = .001, d = 0.84. These data indicate that controls werestrongly biased to perceive the average face as the identity oppositethe adapting face whereas CPs were not.6

Inspection of the group curves in Fig. 6 suggests that both CPsand controls showed similar patterns of identification for faces ofdiffering identity strengths, with very good performance for highidentity strength faces and progressively less accurate identifica-tions as identity strength approached zero. These data suggest thatboth groups had learned to discriminate the target identities. How-ever, the groups appear to differ in discrimination precision, withcontrols showing steeper slopes indicating more precise discrim-ination. We quantified slope using a statistic derived from thecurve-fitting analysis: the standard deviation of the cumulativeGaussian. A smaller standard deviation indicates a steeper slopeand more sensitive discrimination of the targets (e.g., Dan andJim). We compared the groups on their mean standard deviation,pooled across adaptation conditions, and found that controls weremarginally more precise than CPs (CPs: M = 0.24, SE = 0.03, Con-trols: M = 0.19, SE = 0.03, U = 99.5, z = 1.93, p = .053, r = .32). This mayindicate that CPs had learnt to discriminate the faces in a differentmanner to controls (e.g., using less holistic processing; Avidan et al.,2011; Palermo et al., 2011), with greater reliance on compensatorystrategies, such as focusing on the shape of local features (e.g., eye-brows). Some CPs have noted using local features, such as eyebrows,in laboratory tasks of face recognition (McKone et al., in press).Attention to a single local feature may led to successful discrimina-tion in laboratory tasks such as this, where only two test faces needto be discriminated and considerable training is provided. However,in everyday life, such strategies may be less successful because theshapes of local features often vary with view, lighting, makeup and

6 The aim of this study was to examine aftereffects in prosopagnosia, which istypically defined as the inability to recognise faces from their identity, regardless ofwhether the origin of this deficit is a face perception and/or a face memory problem.Nonetheless, it is possible that impaired identity aftereffects may be associated withdeficits of perception, rather than memory (although note that some face afteref-fects can be enduring, suggesting that they can affect memory representations, e.g.,Carbon & Ditye, 2010), and some of the prosopagnosics were not severely impairedon our test of face perception (CFPT, see Fig. 4). However, for the group of prosopag-nosics, there was no correlation between the perception of the average face and CFPTz-scores, whereas the correlation between the perception of the average face and oneof the tests of face memory (MFFT identification z-scores) approached significance(see Supplementary Table 1).

Author's personal copy

3808 R. Palermo et al. / Neuropsychologia 49 (2011) 3801– 3812

Fig. 6. Identity aftereffect (Experiment 2). Mean proportion of “Dan” responses at each level of identity strength as a function of adapting condition, fit with a cumulativeGaussian, (a) pooled over all CPs (n = 14), (b) pooled over all controls (n = 23), (c) for one example CP, and (d) for one example control. An aftereffect is indicated by the curvefor adapt antiDan (filled circles and black line) being left of the curve for adapt antiJim (open circles and dashed line). Error bars show one standard error either side of themean.

expression. Moreover, one must discriminate any given face fromamong the hundreds or thousands of known people (e.g., a workcolleague in the mall could be a work colleague, a friend, a fam-ily member, etc.) rather than just one or other identity, as in thepresent task.

The results of Experiments 1 and 2 indicate that CPs show adeficit in the identity aftereffect but not the figural aftereffect.However, sample sizes were larger in the identity aftereffect task(CPs = 14, Controls = 23) than the figural aftereffect task (CPs = 9,Controls = 19) raising the possibility that we failed to detect a deficitin the figural aftereffect due to reduced power or due to other dif-ferences in the samples for the two experiments. We first note thatin the figural aftereffect task the patterns of results and means forCPs and controls were very similar and the numerical differencebetween the groups is smaller than in the identity aftereffect study.It is therefore unlikely that boosting the power would substantiallychange the findings. Second, it is possible that the five additionalCPs in the identity aftereffect study may have different profiles ofabilities, given the considerable heterogeneity in CP (Le Grand et al.,2006; Schmalzl et al., 2008), and this could have led to different

results for the two experiments. However, Inspection of Table 1suggests that this is not the case, with the CPs in both studies show-ing similar profiles. Finally, to further rule out this possibility weexamined whether the identity aftereffect deficit was still appar-ent when only data from the participants who also completed thefigural aftereffect task were examined. The pattern of results wasvery similar to that obtained with the larger samples. CPs’ per-ception of the identity neutral average faces did not differ fromchance (M = 0.53, SE = 0.04, t(8) = 0.63, p = .55, d = 0.21) whereas con-trol participants were biased to see the average as the face oppositethe adapt identity (M = 0.62. SE = 0.04, t(18) = 3.05, p = .007, d = 0.70).The difference between the groups was marginally significant(U = 49.0, z = 1.82, p = .069, r = 0.34). For the PSE measure, aftereffectswere numerically larger for controls than CPs though the groups didnot differ significantly (CPs: M = 0.05, SE = 0.04, Controls: M = 0.15,SE = 0.04, U = 54.5, p = .127, z = 1.53, r = 0.29). However, only the con-trols group’s aftereffect differed significantly from zero [Controls:t(18) = 3.68, p = .002, d = 0.84, CPs: t(8) = 1.22, p = .256, d = 0.41]. Wealso examined whether the size of the identity and figural after-effects were correlated and found no significant relationships for

Author's personal copy

R. Palermo et al. / Neuropsychologia 49 (2011) 3801– 3812 3809

Table 2Pearson correlations (r) and Kendall’s Tau (�) between the figural aftereffect and both identity aftereffect measures, for both participant groups.

Figural aftereffect

Identity aftereffect measures Prosopagnosics (n = 9) Controls (n = 19)

PSE shift r = −.52, p = .12, � = −.42, p = .12 r = −.02, p = .94, � = −.02, p = .89Perception of average face r = .20, p = .60, � = .17, p = 53 r = .03, p = .92, � = 16, p = .35

either groups, as might be expected if face aftereffects reflectedonly adaptation of the same low-level mechanisms (see Table 2).

5. Discussion

The group of CPs experienced a figural face aftereffect that wasindistinguishable from that of controls. The CPs also experienceda face identity aftereffect. However, the CPs did not show typi-cal adaptation to facial identity because their impression of theidentity of the neutral average face was not significantly alteredby adaptation. This is the first study to show an atypical faceidentity aftereffect in CP. This study is also the first to rule outlow-level retinotopic adaptation as the source of CPs’ aftereffectsbecause adapt and test stimuli differed in size. Therefore CPs’ faceaftereffects must reflect adaptation of higher order shape codingmechanisms. Overall, our results indicate that CPs do not show ageneral perceptual impairment in adaptive coding of face shape,because they show typical figural aftereffects, but rather they showa specific deficit in adaptive coding of face identity.

The ability of CPs to adapt to shape information from faces isconsistent with other evidence that CPs are able to accurately pro-cess some shape-based cues from faces, such as those used to judgesex (Le Grand et al., 2006) and expression (Palermo et al., 2011).Although impaired at face recognition, CPs are also able to discrim-inate between identity to some extent (e.g., they could discriminatebetween Dan and Jim in the current study). However, there isincreasing evidence that CPs are not making these judgements withthe same mechanisms as controls. For instance, CPs rely less onholistic processing for the recognition of expression and identity(Avidan et al., 2011; Palermo et al., 2011) and instead may relyupon mid-level shape and/or high-level object coding mechanismsthat are not specific to faces. Thus, it is possible that CPs’ aftereffectslargely reflect adaptation of mechanisms that are not face-specific,namely mid-level shape and/or high-level object coding mech-anisms. Adaptation of such non face-specific mechanisms couldproduce robust aftereffects, particularly in figural aftereffect tasks,which typically reflect less contribution from face-specific adap-tation than identity aftereffect tasks (Rhodes et al., 2009; Susiloet al., 2010). The reduced identity aftereffect for CPs could there-fore indicate that face-specific identification mechanisms were notadapted or were adapted less so than for controls. One way to exam-ine this in future studies would be to use aftereffect tasks that havebeen designed to minimise the contribution of generic adaptationto shape (e.g., Dennett, McKone, Edwards & Susilo, 2011; Susiloet al., 2010).

Adaptation may have important functional benefits, calibratingthe face coding system so that discrimination is best amongthe kinds of faces an individual is currently experiencing (Chenet al., 2010; Rhodes & Leopold, 2011; Rhodes et al., 2010; Wilsonet al., 2002; Yang et al., 2011). Non-optimal adaptation of facecoding mechanisms could therefore result in less sensitive facediscrimination. This is consistent with recent evidence that CPshave difficulty estimating the average face when shown ensemblesof upright (but not inverted) faces (Leib, Puri, Bentin, Whitney, &Robertson, 2011). Our finding that adaptation to facial identity isreduced in CP supports this view and is consistent with evidenceof reduced adaptive coding of face identity in ASD (Pellicano et al.,2007), another developmental disorder in which face processing is

impaired (e.g., Wilson, Palermo, Brock, & Burton, 2010; see reviewsby Jemel, Mottron, & Dawson, 2006; Sasson, 2006). Importantly,CP is not simply a weak form of ASD, and CPs do not exhibit theadditional deficits in social cognition that are characteristic ofASD (Duchaine, Murray, Turner, White, & Garrido, 2009; Wilson,Palermo, Schmalzl, & Brock, 2010). No CP in the current studyscored above a cut-off indicative of ASD on a questionnaire assess-ing autistic traits (AQ, Baron-Cohen et al., 2001), and the AQ scoresof the CPs and controls did not differ (Table 1). It is thereforeunlikely that autistic traits underlie the reduced face identityaftereffect in CP. However, we note that there was a significantnegative association between AQ scores and the amount of adap-tation to the identity-neutral average face for the CPs (r = −.65,p = .013; � = −.43, p = .043) but not controls (r = .19, p = .39; � = −.01,p = .96) [note there was no association between AQ scores and facerecognition skills on the CFMT, CFPT or MFFT]. Thus, it is possiblethat, for people with developmental face recognition difficulties,social interaction difficulties are associated with abnormal identityaftereffects (i.e., Pellicano et al., 2007 found that weaker identityaftereffects in ASD were associated with greater social and com-munication difficulties). This could be because CPs with higher AQscores may have (or had) less social interest, leading to reducedexposure to faces, but this interesting hypothesis remains to betested.

There is now evidence that two developmental disorders involv-ing impaired face recognition – ASD (Pellicano et al., 2007) and CP(the current study) – are associated with impaired identity afteref-fects. However, it is important to note that poorer face recognitionis not always associated with reduced face aftereffects. Young chil-dren, who are worse than adults at recognising faces, do not showsmaller identity aftereffects than adults (Nishimura et al., 2008;Pimperton et al., 2009), and on a nearly identical task to that usedin the current study children aged 5–7 years show larger identityaftereffects than adults (Jeffery et al., in press). Larger aftereffectsin children than adults have also been seen in other tasks that havenot assessed face identity (Hills, Holland, & Lewis, 2010; Vida &Mondloch, 2009). However, face recognition performance in child-hood may not reflect immaturity in face coding mechanisms andmay rather reflect immature general perceptual and cognitive abili-ties (see Crookes & McKone, 2009; McKone, Crookes, & Kanwisher,2009; Mondloch, Maurer, & Ahola, 2006, for discussion). The factthat children do not generally show reduced face aftereffects sug-gests that CPs are not developmentally delayed.

Face identity aftereffects are thought to reflect adaptation ofmechanisms that code identity in a multi-dimensional face-space(see Rhodes & Leopold, 2011, for a review) so reduced identityaftereffects in CP may indicate that CPs face-space has atypi-cal properties. The fact that CPs in both the current study andNishimura et al. (2010) did show an identity aftereffect is con-sistent with Nishimura et al.’s proposal that CPs have at leasta coarse face-space, that allows the overt recognition of somefaces (e.g., the CPs in this study were able to recognise, on aver-age, 35% [SD = 16] of famous faces) and the covert (or implicit)recognition of famous faces they are unable to recognise (Avidan& Behrmann, 2008; Rivolta et al., in press; Rivolta, Schmalzl,Coltheart, & Palermo, 2010). However, face space may be organ-ised differently in CP, perhaps with different or fewer dimensions,for example.

Author's personal copy

3810 R. Palermo et al. / Neuropsychologia 49 (2011) 3801– 3812

The only previous study to investigate identity aftereffects in CPconcluded that the aftereffects of CPs do not differ from those ofcontrols (Nishimura et al., 2010). As noted in the Introduction, dif-ferences between CPs and controls were more likely to be evidentin our study as we tested a larger group of CPs (n = 14 vs. 6) and usedan adaptation task that minimised the influence of low-level visualattributes. We found similar patterns of results in the identity after-effect task with both larger (n = 14) and smaller (n = 9) samples ofCPs, indicating that the size of the sample is not a crucial difference.Rather, our conclusion that identity aftereffects are abnormal in CPis based upon analysis of adaptation to the identity-neutral aver-age face, an analysis not conducted by Nishimura et al. (although wenote that reduced adaptation to the identity-neutral average faceby CPs is also evident in Nishimura et al.’s results in a visual com-parison of Fig. 2a and b). The identity-neutral face is unique amongthe test stimuli because it does not contain a greater proportionof one identity over the other (i.e., there is no correct response atthis one level) and it is consequently the stimulus for which adap-tation to both antifaces is likely to produce the largest effects onidentification performance. Indeed, a number of previous studieshave only examined adaptation to the identity-neutral average testface (e.g., Leopold, Rhodes, Müller, & Jeffery, 2005; Rhodes, Jeffery,Clifford & Leopold, 2007). We suggest that, in the present paradigm,the average face is the most sensitive point on the continuum andthis may explain why we observed a significant difference betweenthe groups in the aftereffect for this stimulus. It is possible thatdetecting group differences measured with the shift in the PSE mayrequire greater power. However, it is also possible that CPs are lesssensitive to small differences in identity strength, as suggested bymarginally lower precision in discriminating the identities. Suchreduced sensitivity could make it difficult to detect changes in CPsperception of the identity of the average face because the aftereffectlikely results in perception of only a very low strength version ofDan, for example. Nevertheless, we note that the end result wouldstill be that CPs judgements of the identity of a face are less readilymodified by adaptation than those of controls.

On a related point one might ask whether a group differencewould be found for the figural aftereffect if we also measured thisaftereffect using only the undistorted (0) test stimuli. Use of sucha measure is problematic in the figural aftereffect paradigm weemployed because it is difficult to attribute pre–post increases inattractiveness ratings for any one distortion level purely to adap-tation effects. This is because attractiveness ratings commonlyincrease with repeated exposure to the same faces, and is why weused a shift in the peak of the curve as the measure of the aftereffect(i.e., this measure is not affected by general increases in attractive-ness ratings). Further, inspection of Fig. 5 indicates that adaptationeffects are largest for highly contracted faces and diminish as thefaces approach undistorted. Therefore, in this paradigm, the undis-torted faces are not the most sensitive point to measure adaptation.However, we also note that when pre–post shifts in attractivenesswere calculated for each level of distortion, the groups did not differsignificantly for any distortion level.

One important brain region for identity aftereffects may be theanterior inferotemporal cortex, with neuron firing patterns in thisbrain region of macaque monkeys suggestive of norm-based cod-ing of facial identity (Leopold, Bondar, & Giese, 2006). Althoughhomology between species is difficult to establish, we note thatface recognition deficits in CP are associated with smaller ante-rior fusiform volumes (Behrmann et al., 2007), reduced grey mattervolume in the right anterior inferior temporal lobe (Garrido et al.,2009) and less face-selective activation in the right anterior tem-poral lobe (Rivolta, Schmalzl, Palermo, & Williams, 2011).

Finally, it is important to note that we do not wish to argue thatadaptive face coding is the only, or even the most important, per-ceptual mechanism disrupted in CP. As noted earlier, two recent

studies have demonstrated that groups of CPs display deficits inholistic face processing (Avidan et al., 2011; Palermo et al., 2011).Deficits in the ability to recognise facial features and the spac-ing between features have also been observed (Yovel & Duchaine,2006). In addition, some individuals with CP do not show abnormalidentity aftereffects (specifically, F23 in the current study showedtypical performance with the identity aftereffect task reportedhere, and also another four aftereffect tasks; Susilo et al., in press),indicating that deficits in adaptive face coding may not be sufficientfor CP.

To summarise, we compared the performance of a group of CPsand matched controls on two tests of adaptive face coding. CPs dis-played both figural and identity aftereffects, but their aftereffectswere significantly reduced in size when they judged the identityof the neutral average face after adaptation. Typical figural, butabnormal identity, aftereffects indicate that CPs do not have a gen-eral perceptual impairment in adaptive face coding but only showreduced aftereffects when the task directly taps the use of facenorms used to code individual identity. In addition, the observa-tion of abnormal identity aftereffects in a group of people with facerecognition impairments is consistent with other evidence fromtypical individuals which suggests that adaptive face coding mayplay a functional role in face recognition (e.g., Oruc & Barton, 2011;Rhodes et al., 2010).

Funding sources

This research was supported by funding from Macquarie Univer-sity and the Australian National University, and by the AustralianResearch Council’s Discovery Projects funding schemes (projectnumber: DP110100850 to RP, DP0770923 to LJ and ARC Centre ofExcellence Grant CE110001021).

These funding sources played no role in study design; in thecollection, analysis, and interpretation of data; in the writing of thereport; or in the decision to submit the paper for publication.

Acknowledgements

The face identity aftereffect task was developed with DaphneMaurer and Mayu Nishimura. Thank you to all the prosopagnosicsfor their time, and to Tirta Susilo for collecting the data for F23 onthe identity aftereffect task.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, inthe online version, at doi:10.1016/j.neuropsychologia.2011.09.039.

References

Afraz, A. & Cavanagh, P. (2009). The gender-specific face aftereffect is based in retino-topic not spatiotopic coordinates across several natural image transformations.Journal of Vision, 9(10), 1–17.

Afraz, S.-R. & Cavanagh, P. (2008). Retinotopy of the face aftereffect. Vision Research,48, 42–54.

Avidan, G. & Behrmann, M. (2008). Implicit familiarity processing in congenitalprosopagnosia. Journal of Neuropsychology, 2, 141–164.

Avidan, G., Tanzer, M. & Behrmann, M. (2011). Impaired holistic processing in con-genital prosopagnosia. Neuropsychologia, 49, 2541–2552.

Baron-Cohen, S., Wheelwright, S., Skinner, R., Martin, J. & Clubley, E. (2001). TheAutism Spectrum Quotient (AQ): evidence from asperger syndrome/high func-tioning autism, males and females, scientists and mathematicians. Journal ofAutism and Developmental Disorders, 31, 5–17.

Barton, J. J. S. (2008). Structure and function in acquired prosopagnosia: Lessonsfrom a series of 10 patients with brain damage. Journal of Neuropsychology, 2,197–225.

Behrmann, M. & Avidan, G. (2005). Congenital prosopagnosia: face-blind from birth.Trends in Cognitive Sciences, 9, 180–187.

Author's personal copy

R. Palermo et al. / Neuropsychologia 49 (2011) 3801– 3812 3811

Behrmann, M., Avidan, G., Gao, F. & Black, S. (2007). Structural imaging revealsanatomical alterations in inferotemporal cortex in congenital prosopagnosia.Cerebral Cortex, 17, 2354–2363.

Bowles, D. C., McKone, E., Dawel, A., Duchaine, B., Palermo, R., Schmalzl, L., et al.(2009). Diagnosing prosopagnosia: Effects of ageing, sex, and participant-stimulus ethnic match on the Cambridge Face Memory Test and Cambridge FacePerception Test. Cognitive Neuropsychology, 26, 423–455.

Burkhardt, A., Blaha, L. M., Jurs, B. S., Rhodes, G., Jeffery, L., Wyatte, D., et al. (2010).Adaptation modulates the electrophysiological substrates of perceived facialdistortion: Support for opponent coding. Neuropsychologia, 48, 3743–3756.

Carbon, C.-C. & Ditye, T. (2010). Sustained effects of adaptation on the perception offamiliar faces. Journal of Experimental Psychology: Human Perception and Perfor-mance, doi:10.1037/a0019949. Advance online publication

Carbon, C.-C. & Leder, H. (2006). The Mona Lisa effect: Is ‘our’ Lisa fame or fake?Perception, 35(3), 411–414.

Carbon, C.-C., Grüter, T., Grüter, M., Weber, J. E. & Lueschow, A. (2010). Dissociation offacial attractiveness and distinctiveness processing in congenital prosopagnosia.Visual Cognition, 18, 641–654.

Chen, J., Yang, H., Wang, A. & Fang, F. (2010). Perceptual consequences of face view-point adaptation: Face viewpoint aftereffect, changes of differential sensitivityto face view, and their relationship. Journal of Vision, 10, 1–11.

Crookes, K. & McKone, E. (2009). Early maturity of face recognition: No childhooddevelopment of holistic processing, novel face encoding, or face-space. Cogni-tion, 111(2), 219–247.

Dennett, H. W., McKone, E., Edwards, M. & Susilo, T. (2011 May). Do face adaptationaftereffects predict face recognition? Evidence from individual differences. In11th Annual vision sciences society meeting Naples, FL, USA,

Dickinson, J. E., Almeida, R. A., Bell, J. & Badcock, D. R. (2010). Global shape aftereffectshave a local substrate: A tilt aftereffect field. Journal of Vision.

Duchaine, B. & Nakayama, K. (2006a). Developmental prosopagnosia: A win-dow to content-specific face processing. Current Opinion in Neurobiology, 16,166–173.

Duchaine, B. & Nakayama, K. (2006b). The Cambridge Face Memory Test: Resultsfor neurologically intact individuals and an investigation of its validity usinginverted face stimuli and prosopagnosic participants. Neuropsychologia, 44,576–585.

Duchaine, B., Germine, L. & Nakayama, K. (2007). Family resemblance: Ten familymembers with prosopagnosia and within-class object agnosia. Cognitive Neu-ropsychology, 24, 419–430.

Duchaine, B., Murray, H., Turner, M., White, S. & Garrido, L. (2009). Normal social cog-nition in developmental prosopagnosia. Cognitive Neuropsychology, 26, 620–634.

Furl, N., Garrido, L., Dolan, R. J., Driver, J. & Duchaine, B. C. (2011). Fusiform gyrus faceselectivity relates to individual differences in facial recognition ability. Journalof Cognitive Neuroscience, 23, 1723–1740.

Garrido, L., Furl, N., Draganski, B., Weiskopf, N., Stevens, J., Tan, G. C.-Y., et al. (2009).Voxel-based morphometry reveals reduced grey matter volume in the temporalcortex of developmental prosopagnosics. Brain, 132, 3443–3455.

Grüter, M., Grüter, T., Bell, V., Horst, J., Laskowski, W., Sperling, K., et al. (2007).Hereditary prosopagnosia: The first case series. Cortex, 43, 734–749.

Hills, P. J., Holland, A. M. & Lewis, M. B. (2010). Aftereffects for face attributes withdifferent natural variability: Children are more adaptable than adolescents. Cog-nitive Development, 25, 278–289.

Ishihara, S. (1925). Tests for color-blindness (5th ed.). Tokyo: Kanehara.Jeffery, L., Rhodes, G., McKone, E., Pellicano, E., Crookes, K., & Taylor, E. Distinguishing

norm-based from exemplar-based coding of identity in children: Evidence fromface identity aftereffects. Journal of Experimental Psychology: Human Perceptionand Performance, in press.

Jemel, B., Mottron, L. & Dawson, M. (2006). Impaired face processing in autism: Factor artifact? Journal of Autism and Developmental Disorders, 36(1), 91–106.

Kennerknecht, I., Grüter, T., Welling, B., Wentzek, S., Horst, J., Edwards, S., et al.(2006). First report of prevalence of non-syndromic hereditary prosopagnosia(HPA). American Journal of Medical Genetics Part A, 140A, 1617–1622.

Le Grand, R., Cooper, P. A., Mondloch, C. J., Lewis, T. L., Sagiv, N., de Gelder, B.,et al. (2006). What aspects of face processing are impaired in developmentalprosopagnosia? Brain & Cognition, 61, 139–158.

Lee, Y., Duchaine, B., Wilson, H. R. & Nakayama, K. (2010). Three cases of devel-opmental prosopagnosia from one family: Detailed neuropsychological andpsychophysical investigation of face processing. Cortex, 46, 949–964.

Leib, A. Y., Puri, A., Bentin, S., Whitney, D. & Robertson, L. (2011 May). Ensemblecoding in congenital prosopagnosia. In 11th Annual vision sciences society meetingNaples, FL, USA.

Leopold, D. A., Bondar, I. V. & Giese, M. A. (2006). Norm-based face encoding by singleneurons in the monkey inferotemporal cortex. Nature, 442, 572–575.

Leopold, D. A., O‘Toole, A. J., Vetter, T. & Blanz, V. (2001). Proto-type-referenced shapeencoding revealed by high-level aftereffects. Nature Neuroscience, 4, 89–94.

Leopold, D. A., Rhodes, G., Müller, K.-M. & Jeffery, L. (2005). The dynamics of visualadaptation to faces. Proceedings of the Royal Society B: Biological Sciences, 272,897–904.

Macmillan, N. A. & Creelman, C. D. (1991). Detection theory: A user’s guide. New York:Cambridge University Press.

Maurer, D., Le Grand, R. & Mondloch, C. J. (2002). The many faces of configuralprocessing. Trends in Cognitive Sciences, 6, 255–260.

Maxwell, D. (1994). Gryphon Morph 2.5 [Computer software]. Gryphon SoftwareCorporation.

McKone, E. & Yovel, G. (2009). Why does picture-plane inversion sometimesdissociate perception of features and spacing in faces, and sometimes not?

Toward a new theory of holistic processing. Psychonomic Bulletin & Review, 16,778–797.

McKone, E., Crookes, K. & Kanwisher, N. (2009). The cognitive and neural devel-opment of face recognition in humans. In M. S. Gazzaniga (Ed.), The cognitiveneurosciences (4th ed., pp. 467–482). Cambridge, MA, USA: Massachusetts Insti-tute of Technology.

McKone, E., Hall, A., Pidcock, M., Palermo, R., Wilkinson, R., Rivolta, D., et al. Faceethnicity and measurement reliability affect face recognition performance indevelopmental prosopagnosia: Evidence from the Cambridge Face Memory Test– Australian. Cognitive Neuropsychology, in press.

Mondloch, C. J., Maurer, D. & Ahola, S. (2006). Becoming a face expert. PsychologicalScience, 17(11), 930–934.

Nishimura, M., Doyle, J., Humphreys, K. & Behrmann, M. (2010). Probingthe face-space of individuals with prosopagnosia. Neuropsychologia, 48,1828–1841.

Nishimura, M., Maurer, D., Jeffery, L., Pellicano, E. & Rhodes, G. (2008). Fitting thechild’s mind to the world: adaptive norm-based coding of facial identity in 8-year-olds. Developmental Science, 11, 620–627.

Nunn, J. A., Postma, P. & Pearson, R. (2001). Developmental prosopagnosia: Shouldit be taken at face value? Neurocase, 7, 15–27.

Oruc , I. & Barton, J. J. S. (2011). Adaptation improves discrimination of face identity.Proceedings of the Royal Society B: Biological Science, 278, 2591–2597.

Palermo, R., Willis, M. L., Rivolta, D., McKone, E., Wilson, C. E. & Calder, A. J. (2011).Impaired holistic coding of facial expression and facial identity in congenitalprosopagnosia. Neuropsychologia, 49, 1226–1235.

Pellicano, E., Jeffery, L., Burr, D. & Rhodes, G. (2007). Abnormal adaptive face-codingmechanisms in children with autism spectrum disorder. Current Biology, 17,1508–1512.

Pimperton, H., Pellicano, E., Jeffery, L. & Rhodes, G. (2009). The role of higher leveladaptive coding mechanisms in the development of face recognition. Journal ofExperimental Child Psychology, 104, 229–238.

Ramon, M., Busigny, T. & Rossion, B. (2010). Impaired holistic processing of unfamil-iar individual faces in acquired prosopagnosia. Neuropsychologia, 48, 933–944.

Raven, J., Raven, J. C. & Court, J. H. (1998). Manual for Raven’s progressive matrices andvocabulary scales. San Antonio, Texas, USA: Harcourt Assessment.

Rhodes, G. & Leopold, D. A. (2011). Adaptive norm-based coding of face identity.In A. J. Calder, G. Rhodes, M. H. Johnston, & J. V. Haxby (Eds.), Handbook of faceperception. Oxford: Oxford University Press.

Rhodes, G., Evangelista, E. & Jeffery, L. (2009). Orientation-sensitivity of face identityaftereffects. Vision Research, 49, 2379–2385.

Rhodes, G. & Jeffery, L. (2006). Adaptive norm-based coding of facial identity. VisionResearch, 46, 2977–2987.

Rhodes, G., Jeffery, L., Clifford, C. W. G. & Leopold, D. A. (2007). The timecourse ofhigher-level face aftereffects. Vision Research, 47, 2291–2296.

Rhodes, G., Jeffery, L., Watson, T. L., Clifford, C. W. G. & Nakayama, K. (2003). Fit-ting the mind to the world: Face adaptation and attractiveness aftereffects.Psychological Science, 14, 558–566.

Rhodes, G., Watson, T. L., Jeffery, L. & Clifford, C. W. G. (2010). Perceptual adaptationhelps us identify faces. Vision Research, 50, 963–968.

Riddoch, M. J. & Humphreys, G. W. (1993). BORB: The Birmingham object recognitionbattery. Hove, UK: Lawrence Erlbaum Associates.

Rivolta, D., Palermo, R., Schmalzl, L., & Coltheart, M. Covert face recognition in con-genital prosopagnosia: A group study. Cortex, in press.

Rivolta, D., Schmalzl, L., Coltheart, M. & Palermo, R. (2010). Semantic information canfacilitate covert face recognition in congenital prosopagnosia. Journal of Clinicaland Experimental Neuropsychology, 32, 1002–1016.

Rivolta, D., Schmalzl, L., Palermo, R. & Williams, M. A. (2011 August). Multi voxelpattern analysis of fMRI data reveals abnormal anterior temporal lobe activityin congenital prosopagnosia. In 34th European conference on visual perceptionToulouse, France,

Rossion, B., Dricot, L., Goebel, R. & Busigny, T. (2011). Holistic face categorizationin higher order visual areas of the normal and prosopagnosic brain: Toward anon-hierarchical view of face perception. Frontiers in Human Neuroscience, 4,1–30.

Sasson, N. J. (2006). The development of face processing in autism. Journal of Autismand Developmental Disorders, 36, 381–394.

Schmalzl, L., Palermo, R. & Coltheart, M. (2008). Cognitive heterogeneity ingenetically-based prosopagnosia: A family study. Journal of Neuropsychology, 2,99–117.

Schweinberger, S. R., Zäske, R., Walther, C., Golle, J., Kovács, G. & Wiese,H. (2010). Young without plastic surgery: Perceptual adaptation tothe age of female and male faces. Vision Research, 50(23), 2570–2576.doi:10.1016/j.visres.2010.08.017

Susilo, T., McKone, E. & Edwards, M. (2010). Solving the upside-down puzzle: Whydo upright and inverted face aftereffects look alike? Journal of Vision, 10, 1–16.

Susilo, T., McKone, E., Dennett, H., Darke, H., Palermo, R., Hall, A., et al. Normal holisticprocessing and face space coding in a case of developmental prosopagnosia, inpress.

Thomas, C., Avidan, G., Humphreys, K., Jung, K.-J., Gao, F. & Behrmann, M. (2009).Reduced structural connectivity in ventral visual cortex in congenital prosopag-nosia. Nature Neuroscience, 12, 29–31.

Vida, M. D. & Mondloch, C. J. (2009). Children’s representations of facial expressionand identity: Identity-contingent expression aftereffects. Journal of ExperimentalChild Psychology, 104, 326–345.

Watson, T. L. & Clifford, C. W. G. (2003). Pulling faces: An investigation of the face-distortion aftereffect. Perception, 32(9), 1109–1116.

Author's personal copy

3812 R. Palermo et al. / Neuropsychologia 49 (2011) 3801– 3812

Watson, T. L. & Clifford, C. W. G. (2006). Orientation dependence of the orientation-contingent face aftereffect. Vision Research, 46(20), 3422–3429.

Webster, M. A. & MacLeod, D. I. A. (2011). Visual adaptation and face perception.Philosophical Transactions of the Royal Society B, 366, 1702–1725.

Webster, M. A. & MacLin, O. H. (1999). Figural aftereffects in the perception of faces.Psychonomic Bulletin & Review, 6, 647–653.

Webster, M. A., Kaping, D., Mizokami, Y. & Duhamel, P. (2004). Adaptation to naturalface categories. Nature, 428, 557–561.

Wilmer, J. B., Germine, L., Chabris, C. F., Chatterjee, G., Williams, M., Loken, E., et al.(2010). Human face recognition ability is specific and highly heritable. Pro-ceedings of the National Academy of Science of the United States of America, 107,5238–5241.

Wilson, C. E., Palermo, R., Schmalzl, L. & Brock, J. (2010). Specificity of impaired facialidentity recognition in children with suspected developmental prosopagnosia.Cognitive Neuropsychology, 27, 30–45.

Wilson, C. E., Palermo, R., Brock, J. & Burton, A. M. (2010). Learning new faces intypically developing children and children on the autistic spectrum. Perception,39, 1645–1658.

Wilson, H. R., Loffler, G. & Wilkinson, F. (2002). Synthetic faces, face cubes, and thegeometry of face space. Vision Research, 42(27), 2909–2923.

Yang, H., Shen, J., Chen, J. & Fang, F. (2011). Face adaptation improves gender dis-crimination. Vision Research, 51, 105–110.

Yardley, L., McDermott, L., Pisarski, S., Duchaine, B. & Nakayama, K. (2008). Psychoso-cial consequences of developmental prosopagnosia: A problem of recognition.Journal of Psychosomatic Research, 65, 445–451.

Yovel, G. & Duchaine, B. (2006). Specialized face perception mechanisms extract bothpart and spacing information: Evidence from developmental prosopagnosia.Journal of Cognitive Neuroscience, 18, 580–593.

Zhao, L. & Chubb, C. (2001). The size-tuning of the face-distortion after-effect. VisionResearch, 41(23), 2979–2994.