Tilburg University
Configural Processes in Face Recognition
Rouw, R.
Publication date: 2001
Document Version
Publisher's PDF, also known as Version of record Link to publication in Tilburg University Research Portal
Citation for published version (APA):
Rouw, R. (2001). Configural Processes in Face Recognition: No "Special" Processes Underlie the "Special Case" of Face Recognition. Thela Thesis.
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal
Take down policy
e
.,
b
*4--93>' 4.3
-- ., 4'FIA: ...Jter-*f . 4 , 4
:r
ked:-1.13*...t ..t ./
, t. .7 k, 2.. :4./*
v/ Academic Publishing ServicesThela Thesis
ISBN 90-9014773-X
PROEFSCHRIFT
ter verkrijging vandegraad van doctor
aandeKatholiekeUniversiteitBrabant, op gezag van derector magnificus,
prof.dr. F.A. van der DuynSchouten,
inhet openbaarteverdedigenten overstaan
van een doorhet collegevoor promotiesaangewezen commissie in de aula vandeUniversiteit opvrijdag 18 mei 2001 om 16.15
door
Romke Rouw
geboren28september 1970teHilversum
-.
Promotor:
The Controversy/NeurologicalBasis/Whichstudies areincluded and why
Chapter One - 35
Structural encoding precludes recognition offaceparts inprosopag:nosia
Chapter Two - 61
Paradoxicalinversion effectforfacesand objectsin prosopagnosia
ChapterThree-81
Configuralfaceprocessesinacquiredanddevelopmental prosopagnosia:
evidence fortwo separatefacesystems?
Chapter Four - 99
Impairedfacerecognition doesnotpreclude intact wholeface perception
Chapter Fiue - 133
Howfacerecognitionbecomes special:
configural recognitionoffacesandobjectsin childrenandadults
Chapter Six - 185
Categorizationvs.recognitionofown-raceandother-racefaces
Conclusions - 219
I. The Controversy
The questionhowfaces,asopposed toothervisualstimuli,arerecognized has received a lotofattentioninrecent years.Facesareperhaps the mostimportantvisual
stimuli tous,presenting us with informationforestablishing andmaintainingsocial contacts. There is anintuitivelogic tothenotion that thisimportantvisual
informationisdealt withby specializedprocesses, andthat perhaps even an area of the brainisreservedforprocessing facestimuli.The degreetowhichfaces are
"special" has beenatopic indifferentresearchfields(e.g.electrophysiology,
neuropsychology, developmentalstudies, andbehavioral studies).Onecould say that the researchfield on face recognition is in factdividedinseveral separate segments,
askingdifferentquestionsand usingavarietyof paradigms.Thereason of this disunity is thatthequestion whether thereissomething"special" aboutfaces as
stimuli intriguedmanyresearchers,from differenttheoretical and experimental
backgrounds. Asaconsequence ofthis multiple input in thefacerecognitiondebate,
it is not easy to integrate all results toaunified total. As we will show below,there
are howeversomereturningissues.
Oneparticularly striking finding infacerecognitionliteraturecomes from neuropsychological reports on"prosopagnosia"(Bodamer, 1947).Prosopagnosic
patientscannotrecognizeanindividualfacewhilerecognitionofothervisualstimuli isstillintact. Thepatientcanstillrecognizethepersonthroughnon-facial cues such as clothing or posture.Sometimes,the patientcanstill identifytheperson through the face, if the face hasaparticularstrikingfeature such as a mole,aremarkable haircut, oreyeglasses.Howevertheseprocedures are not only odd, butalsounreliable and
Introduction
normalrange.AsDavidoffand Landis (1990) haveargued,aminimumrequirement
for using thetermprosopagnosia is thatatleastthe firstvisualinput intothestream
offacerecognitionprocessesisintact. Ontheother hand, prosopagnosianeeds not be
confined toimpairedfacememory butisoften accompaniedbyimpaired visual processes as well. InfactSchweich and Bruyer (1993) havereported on the
heterogeneity ofthedeficitandshowedthat patientsdiffer intheirabilitytoperform facedecision (recognition of theface classrather thanindividualface), visualanalysis ofunfamiliarfaces,andfamiliarfacerecognition.
Differentexplanations have beenputforwardtoaccount forthispuzzling
selectivedeficit. Ofcourse,acompelling interpretation is thatthere are separate and
independentprocessesspecifically reserved forthestimulusclassoffaces. Damage to these processesresults inaselectiveimpairment tofacestimuli. In itsmostextreme
form, this explanation is one ofanimpaired'facemodule'.Onealternative explanation is thatprosopagnosia is alesssevere form ofagnosia, andasfaces are
more difficulttorecognize this isthestimulusgroupsuffering most fromsuchdeficit. However, this hypothesis hasbeenconvincinglycountered bythefindingof double dissociations.Whileprosopagnosicpatientsareselectively worse with facestimuli,
other patientsareimpairedwithobject but notfacerecognition(Farah, 1991;
McCarthy&Warrington,1986;Moscovitch,Winocur,&Behrmann, 199D.Damasio and colleaguesput forwardthe"individuationhypothesis" (seeMoscovitch et al., 1997;Damasio, Damasio, &Hoesen, 1982). Facerecognition requires within-class
discrimination (one face fromthe other),whileobjectrecognitiontestsare between-category (discriminate a car fromaflower).Theseauthors showed problems with
subtlewithin-category discrimination forprosopagnosic patients, however other prosopagnosic patientshaveshownintactwithin-classdiscrimination(DeRenzi,
1986). McNeil and Warrington(1993)reported afarmer suffering from prosopagnosia
thatfacerecognitionisspecial intheparticularcombinationofdemandsitposes on therecognitionsystem.
The issuewhetherthereissomething'special'aboutfacestimulicannot be
understoodwhile it isnotclearwhich arethe processesinvolved infacerecognition. Behavioral datafromnormal participantshaveindicated thatfacestimuli might indeedbeprocessed inadifferentmanner thanobject(non-facial) stimuli. The most
wellknown behavioral finding isthe "face inversioneffect";recognition offaces is
more impaired by inverted (upsidedown)presentation thenisrecognitionofother
stimuli.Theexplanationprovided forthis finding (see Yin, 1969) is thatbesides a
general factor offamiliaritymaking mono-orientedobjects (suchasfacesandhouses)
more difficultto recognizeupsidedown,there is anadditionalfactorofincreased
difficulty only forfaces. Theadditional difficulty forfacestimuliwouldberelated to
theparticular importance ofconfigural information infacerecognition. Presenting a stimulus upsidedownhindersextraction ofconfiguralinformation,whileanalysis of
isolatedfeatures isrelativelyspared.Other stimulusclassescanstillberecognized
fairly wellusing this featuralinformation, butfacerecognitionisseverely disturbed.
Similarly,otherbehavioralfindings of differentpatternsofperformance with face thanwithobjectstimuliaregenerally explained assomehow related to face recognition depending moreoninformation ofthewholeface configuration,while
recognitionofotherstimuli canusefeatures orparts-basedinformation.One example
of abehavioraleffectsupporting this notion is thefacecontexteffect(Homa, Haver, &
Schwartz, 1976; see alsoSuzuki&Cavanagh, 1995)whereanormal upright face
contextaidsrecognition of a facepart.TanakaandFarah(1993)found thatthiseffect of contextwasstronger with face thanwithother(house)stimuli. Afterlearning
wholefacesorhouses,recognition of only a part was moredifficultthan recognition
ofthewhole stimulus, but thiseffectwasfound stronger with face thanwithhouse
Introdtiction
Farah (1990) hasput forwardaclearmodel, withadichotomybetween face
recognition depending moreon 'holistic'(whole-based)informationandobject recognition depending moreon featural(parts-based)information. In
neuropsychology, suchsimilar assumption explains prosopagnosiaasimpaired
configuralprocesses(Levine &Calvanio, 1989). The opposite case ofonlyspared
configuralprocessesthen predictsagnosiawithoutprosopagnosia,whichhasindeed
beenreported (Moscovitch et al., 1997). In thisthesis,this assumption ofastrict
divisionbetweenconfiguralfacerecognitionprocesses on the one handand parts-based objectrecognitionprocesses on theother handwas testedin different
paradigms.
If thereisnothing "special" aboutthestimulusclassoffaces, thefacespecific
findingsmustreflectafailuretopresentthecorrectcontrolstimuli.Inthesemodels
notthestimulusclassoffacesitself,butratheracombinationofstimulus properties underlies the way in whichfacesarerecognized.There aretwo characteristics that
clearly emerge intheliteratureasbeing mostimportant.Thefirst istheexpertise we
all holdforfaces, and the second is the factthat recognition or matchingofindividual facesrequires within-classdiscrimination.Inthesemodels, it hasbeen suggested
(Diamond&Carey, 1986; Rhodes&McLean, 1990)that extractionof"norm-based"
("second-order relational")information isanimportantcomponent ofexpertise with
within-classdiscrimination ofahomogeneous stimulusclass.This norm-based or second-order relationalinformation is not onlytheconfiguration of thefaceitself, but moreover how this configuration deviates from the mean (orprototypical)
configuration ofthestimulusclass. We areall experts in recognizingfaces, and the
stimuliare alike (i.e. themembersshareaconfiguration),causingparticular
importance of configuralinformation infacerecognition.Thus,opposite to the "face
the position that thefacestimulusclassitselfisspecial, is thequestion whether it is
possible tofindobjectstimuli with which thesamebehavioral, neurological and
neuropsychological findings canbeobtained.
Findingthe exactright controlobjects tocompare withfacestimulihasproven
problematical. Itisalwayspossibletoexplainthedifferencebetween face and object stimuli as asideeffectofchoiceofcontrol object (it does not containtheright
combinationofstimulus properties). Similarly, itisdifficulttodiscusswhether
differentor separate processesareinvolved inthe "special case"offaces as long as it
is not yetclearwhichexactly are the processesinvolved infacerecognition. The main focus inthis thesistherefore istoprovidea better picture onthevisualprocesses
involved infacerecognition. Subsequentlythequestion canbe asked whetherthese
processesareinvolved inobjectrecognition as well, orarereservedexclusively for
thestimulusclassoffaces. Ifthe processes are reservedforfaces, aswe explained above,thiseitherreflectsaneffectofstimulusclass per se (a facemodule), or is a
consequence of thespecificstimulus properties combined in aface image.Particularly
thefactors"within-class processing" and "expertise"areassumed essential
underlyingthese processes. Thus,instudyingthevisualprocessesinvolved in face recognition, alsothe influence or importance ofthesefactors is considered.
Aswillbecome clearwhen readingthefollowingchapters, the role of
"configural" processes inface recognition andfacematching is a main issue in this thesis. Infacerecognition literature, "configuration- is assumedofcritical importance
infaceprocessing, asopposed to"parts-based" or"featural"information.
Unfortunately, usingtheseterms necessitatesdisentanglingthedifferent terminology
and definitions used in previous research.Toavoid confusion, please note that
"configural" canrefer tobothinformationandprocesses.First asalsoexplained in the introduction of Chapter 5 and Chapter 6, the term "configuration" is usedtorefer
Introduction
prosopagnosicpatient LH withnormaluprightfaces thanwithinvertedfaces is a
disruptinginfluence offaceconfiguration on matching performance.Second, there is
a question ofwhatvisualprocesses areinvolved inextractinginformation from the facestimulus.Differentnotions onwhatthese processesare,isreflected in the use of differentterms suchas"whole-based" or "holistic"and "configural". Intheliterature isnotalways made clear what arethedefinitions oftheseterms and how they are separate from eachother. Furthermore, itisdifficulttodetermine inacertain task whether one ortheother kindof information is used.Forexample,disturbing
relational information by moving afacefeaturealsodisturbsthe"whole face" information.Vice versa,contrasting presentation of a face partwith wholefaces not
only manipulates the presencevs. absenceof"holistic"information, but also that of relationalinformation. In ourexperiments we often donotdifferentiate between relational (configural) informationandholistic (whole-based) information.Toavoid confusion, simplythecontrast is made between on the one hand processing of the whole stimulus,therelations betweentheparts, andtherelations between parts and
whole, and ontheother handtheisolated parts or stimulusfeatures.
II. NeurologicalBasis
Each chapter inthisthesisstarts withanintroduction ofthespecificresearchquestion
at hand.Thoughliteratureofimmediaterelevanceisdiscussed, the space is too
limitedtoprovide a moreextensive discussion.In particular, intheintroductions of
thechapters thereislittle discussion oftheextensive researchfield on the
neurologicalbasis of facerecognition. This subject thoughofclearimportance in the broaderfieldofunderstandingfacerecognition,receiveslittleattention in the
and moreovertoindicatetheparallels between questionsraisedand theories
proposed in thisfield with theonesdiscussed inthefollowingchapters of thisthesis,
a short discussion of thisresearchfieldisprovidedbelow.
Area Responding Specifically to Faces?
Recentimagingstudiesreported "faceselective" activation inthefocalregion of the
fusiform gyrus, andtermed this areathefusiform face area or "FFA" (Ishai,
Ungerleider,Martin,Maisog,& Haxby, 1997;Kanwisher, McDermott, & Chun, 1997;
McCarthy, Puce, Gore,&Allison, 1997).Thedegree offacespecificitywasstudied by
comparing level of activationtofaceswithlevelofactivationtootherstimuli. FFA shows stronger activationforfaces thanforletter strings(Puce, Allison,Asgari, Gore, & McCarthy,1996),flowers orassorted objects(McCarthy et al., 1997),houses
(Kanwisher et al., 1997), andscrambledfaces(Tong,Nakayama, Moscovitch, Weinrib, & Kanwisher,2000).Sergent, Ohta,andMacDonald (1992)showed inanearlier study using PET thattheactivation found in afacerecognition task but notanobject
recognition taskwaslocated inthe
right
lingualandfusiformgyrus,therightparahippocampal gyrus, andtheanterior temporalcortex.Thisindicates that face specific activationmightbefound more strongly intheright hemisphere.Theissue of
right vs. left hemisphere involvementisseparate(though related) issue and will not
befurtherdiscussed here. For amore detailed discussion of lateralizationeffects, and
a viewpoint on how differentfacerecognition functions mightbeserved in the two
hemispheres, see DeGelder and Rouw (inpress),Laeng and Rouw (inpress), and de Schonen& Mathivet (1989).
Introduction
Nakayama(1998)found activation inthefusiform face areatouprightor inverted
grayscale facestimuliand uprighttwo-tone"Mooney"faces.However, whenstimuli
werenolongerrecognizable as a face (thesametwo-tonefacespresentedupside
down) no FFA activationisfound. Similarly,Puce,Alison, Gore,andMcCarthy (1995)
comparedfacestimuli with stimulicontaining thesamecomponents butwiththeir
featuresrearranged.GreatestfMRIactivationtofaceswasfound inthemidfusiform
gyrus, and more posteriorandlateral (occipitotemporalsulcusandinferiortemporal gyrus). There wasnosignificant difference between right and left hemisphere activation. Presentationofscrambled facesactivateddifferentareas,namely the
laterallingualgyrus andthecollateralsulcus. Theauthors of this study conclude that
the findings do confirm theface sensitivity ofareas but"thedegree towhich it is face
selectiveremains tobedetermined."
McCarthy et al.(1997)demonstrated intheirfMRIstudy that bothfaces and
flowers activatelarge and partiallyoverlapping regions ofinferiorextrastriatecortex. The facesor flowers were viewed amongnon-objectsor amongobjects.While flowers
amongnon-objects evoked bilateralfusiformactivation, flowers among objects did not result in any flower-specific activation. Presentingfacesamong non-objects resultedinface-specificactivation in bilateral regions oftheposteriorfusiformgyrus.
Evenfacesamongobjectsshowedface-specificactivation (though smaller than in the
non-object presentation condition), namely in thefocalright fusiformregion.Taken
togetherthe findingsshowed activation specificallytofaces,primarily intheright
lateralfusiformgyrus.Theauthors concludethatfacesareperceivedatleast in part
by a separate processing streamwithintheventralobjectrecognitionsystem.Still, the
findingsalsoindicatethat activation to one butnotanotherclassof stimuli can also be found withotherstimuli butfacesandother areas than the "face area".
Thepossibleneurologicalbasis of face processes and the existence of face
Electrophysiologicalstudiescomparingfaceswithotherstimulihavereported an
earlynegativepotential(N200)responding preferentiallytofaces but nottoother
stimuli suchasscrambledfaces,cars, andbutterflies (Allison, Ginter, McCarty,
Nobre, Puce, Luby, &Spencer,1994a).Thiscomponentwasrecordedatlocations in the leftand right fusiformandinferiortemporalgyri.Furthermore, electrical stimulation of thesameregion sometimes produceda temporary inability to name
familiarfaces(Allison etal.,1994a). Puce et al.(1995)observed that thefacesensitive areas (inthefusiformandinferiortemporal gyri) reportedintheirfMRIstudy were approximately coextensive with thefaceareasidentified intheseelectrophysiological studies (Allison etal., 1994a;Allison etal., 1994b).Similarly, Puce et al.'sfindings
with scrambledfacestimuliwere correspondingwithevokedpotentials inthestriate
andperistriate cortex inresponse tohighspatial frequency stimuli such as
checkerboards and scrambledfaces(Allison,McCarty,Nobre, Puce,&Belger, 1994b). Thesecongruentfindingsfromdifferentmethodsto measurebrain activityprovide a
particularconvincing argumentinfavor ofsome degree of "facespecificity" in the
brain.Inclinicalstudies,Puce, Allison,Spencer, Spencer,&McCarthy, G. (1997) also studiedface-specific ERPs as well asfMRIactivationto faces intwoepileptic patients
with intracranialelectrodes. Theearly surface-negative potential (N200) and the fMRI
activation inresponse tofacestimuliwere foundatcorresponding locations. In both patients, activationwasfoundin
right
fusiform gyri. Thetwopatients showed additional activation inthelateral cortex, one on the left side andtheother at the right side.Thesefindings provideclearevidenceforareas of thebrain responding more
stronglytofaces thantootherclassesofstimuli. Thisdoeshowevernotestablish
whethertheseareas arefaceselective, andrespondexclusively tofaces.Asecond issue
is that even
if
there isaselectivebrainarearesponding exclusivelytofaces, this doesIntl'oditction
functions.Perhaps thefunctional specialization ofthebrainallowsspecificactivation tostimuli other thanfaces.
Tile finding that the FFA respondstoinvertedfacesled Kanwisher et al. (1998) to
propose that thisareamightbeinvolvedin detection of theface classrather than
identification oftheindividualface.Behavioral studieshaveshown that inverted
faces aremuchharder torecognize (andmightberecognized inadifferentmanner)
thanuprightfaces, butthis difference between uprightandinvertedfaces was not
reflectedin activation ofFFA. Bentin,Allison,Perez,Puce, & McCarthy (1996) reportedacorresponding finding with thefacespecific negative component (N170) responding preferentially to humanfacesandisolatedhumaneyes, but not to a
variety ofcontrol stimuli. This component was not muchaffectedbyalteration of the location oftheinner components or byfaceinversion(Bentin et al., 1996). Thus, both
theearly negative componentasface-specificactivationfound in the FFAappears
onceastimulusisrecognized as a face(Kanwisher et al., 1998) but isnotdependent
on further identificationprocesses. Even if theearly component isnotrelated to
identification,later componentsmighthoweverbe.Bentin andDelouell (2000) reportedbesidesthis early component (N170)alater negativefacecomponent (rangingbetween 250 and 500 ms). In linewith theirproposal thatN170 reflects an
earlystage offaceperceptionwhilethelater componentreflectslaterprocesses of
identification, they found thattheearly component (unlikethelater N400) was not
modulated bythefamiliarity of the face, nor influenced by task (Bentin & Delouell,
2000).However, other findings point to a role of the FFA inidentificationrather than categorization or detection.Recentstudiesexaminedbrainactivity patterns in prosopagnosicpatients, and did notfind activation in the FFA inresponse to face
stimuli.(DeGelder& Kanwisher,1999).However,theactivation found in what has been termedtheparahippocampalplace area (PPA), an arearespondingtoscenes
In conclusion, in literatureclearsupport forface specificareasisprovided.
However, this does notimply afaceselective areasurrounded by generalobject
recognitionareas. A morejust conclusion holds thatthe areas areonlyrelatively face
specific, and are moststrongly activated byfacesascomparedwithotherstimuli.
First,studiesfound thatfacesactivate specificareas,butotherareasareactivated by
flowersandbutterflies (McCarthy et al., 1997).Second,though theareacalled
"fusiform face area"isreported to showselectivefaceactivation, boththelocation
and intensity oftheactivationisinfluenced by task andevenstimulussettings. The
FFA thusresponds notselectively, but morestronglytofaces than tootherkinds of
stimuli. Althoughno clear consensus exists, facespecific processes seemlocated in
the areaoffusiformandinferiortemporal gyrus. Tentatively, we suggest that the earlyfacenegativecomponentreported in electrophysiological studies, and the
face-specific activation in the FFA reportedin brainimagingstudiesreflectasimilar
system. Thissystemisinvolvedwithearly encoding of thefaceshape.
"Face Specific" us. "Stimulus Properties"
Theprevious overviewshowedsupport forthenotion ofa"face area" inthebrain.
The degreeofspecificitywasindicated bytherelative activation tofacesascompared with activationintheseareastootherstimuli.Asdiscussedpreviously with
behavioral findings,thefindings of"facespecificity" canbeexplainedeither as
reflectingthe special status of thestimulusclassoffaces (facemodule), or as a function ofaparticularcombinationofstimulus properties.
Gauthierand colleaguesargued that"facespecific" activation in an area in fact reflectsthesensitivity oftheseareas tothefactorsofexpertiseandsubordinate level (within-category) matching. Thislatter factorisrelatedtolevelofprocessing
(recognizing "a face"issuperordinatetorecognizing "Jan's face"), but also to the
Introduction
compared withthesimilarity ofanarbitrarygroup ofobjects.Facespecificfindings are explainedasfailuretocontrolforlevelofcategorizationatwhichstimuli are typically recognized. For example, Gauthier, Tarr, Anderson, Skudlarski, and Gore
(1999) pose that both thestudy of McCarthy et al. (199DandKanwisher et al. (1997)
didnotrequire subordinate level processing ofthecontrolobjects, because the task was oneofpassiveviewing orbecausepresentation times weretooshort. According to these authors, passiveviewing ofastreamoffacesandnon-face objectsresults in subordinate levelprocessingoffaces(theindividualperson)whileobjects are processed at amorebasiclevel.Gauthierand colleaguesfound, usingobjectstimuli,
additional activationtoverifyasubordinatelevel ofapicture over its subordinate level inthefusiformandinferiortemporal gyri as wellastemporalpoles. Gauthier, Anderson,Tarr,Skudlarski, & Gore(1997)concluded"theadditionalvisual processing requiredtoaccessthesubordinatelevel ofanon-faceobject over and
aboveitsbasic level, engages theregion ofthebrain previouslydefined as the face-selective area."
Some questions arise on the processesunderlyingtheactivation level of the fusiform inferiortemporal region. Possibly themeasurementshaveincluded more than thevisualprocessesof particularimportance infaceprocessing,however, as the volumeofactivation in this area wasnotlarger inthevisual than inasemantic task.
Anotherquestion involves howtheregions of interest (ROI) aredetermined. Gauthier,Skudlarski, Gore, & Anderson(2000)studied subordinatelevelprocessing
using bothaveraged andindividually definedROIs.These areasresponded more stronglytofacesthanobjects inapassiveviewing task. Inaword-to-object matching
task theseROIsshowedadditionalactivationtosubordinate levelobjectmatching
("sparrow")overbasiclevelobjectactivation("bird").Furthermore,theadditional
activation offacesover objects didnotdiffer significantly fromtheadditional
that the FFA does notrespond tofacestimuliexclusively, it also does not show that it
is a"subordinate level matching area". There were manyareasresponding more
stronglytosubordinate thanto basiclevel matching:the"visualsubordinate minus basic"condition elicited activation ofthemiddleand posterior fusiform gyri, but also
theentire occipital gyri as wellasposteriormiddletemporal gyri. Finally, note that activation of the"face areas" bystimuliother thanfaces (seealsoMartin, Wiggs, Ungerleider, &Haxby,1996;Schacter et al.,1995) refutesastrictly modular
explanation, but does notputasideamodelofrelative specialization.
Asdiscussedpreviously inthisIntroduction, the secondproperty of particular importance inthe face-objectstimulicomparison lies intheexpertise we all hold for face recognition. Does thefacespecificactivation found in the FFAincrease (or is
even dependenton)expertise with the stimulusclass?Gauthier et al.(2000)found
indeed that bird andcar expertsshowed FFAactivity fortheir subjectofexpertise.
Furthermore, FFA activation was even foundwithnonsense objects called
"Greebles", but only
if
expertisewiththesestimuliwasacquired throughtraining(Gauthier et al., 1999).
This brings us tothefinal point in this model, namelytheinteraction between the factors homogeneity, categorization level andexpertise.Thisinteractionisassumed necessary for "facespecific"findings toemergewithotherstimuli.Gauthier and Tarr
(1997) studied theimportance of thecombination of stimulussettingswith"Greeble"
stimuli.Two groups, either "novices" or "experts", were presented witharecognition
procedurewith uprightandinvertedGreebles.Only "experts" were participants trainedwithGreeble gender,familyandindividualGreebles.Both noviceandexpert
participants learnedthenames of6parts and6individualGreebles,followed by a
recognition task oftheparts eitherinisolation,instudied configuration (the
individualGreeble) orintransformed configuration (upperfeaturesrotated). The
Introduction
and experts were faster andlessaccurate in the'isolated parts' than inthe'studied
configuration' condition,regardlessof stimulus orientation. Accuracy showed only
forexpertspresentedwith uprightGreeblesasignificant better performance with
studied configuration than with isolatedparts.Ignoringthelatencyfindings(experts
weresignificantly fasterwith uprightisolated parts) in this condition,theauthors conclude thatthis replicates the objectsuperiorityeffectasmeasuredwithfaces. Further,onlyexpertsshowedasignificantadvantage (faster responses)ofstudied over transformed configuration, and only intheuprightcondition. Thisfinding is not
unexpected, asonlyexperts were trainedextensively with this configuration. The conclusion ofthis study was that notthestimulusclassoffaces,butexpertise with
visually similarobjectsproducesconfigural sensitivity.
Theoppositestandpointasserts thatfaceperception involvescognitive and neuralmechanismsspecializedforprocessing ofthestimulusclassoffaces, and argues against anexplanation based on a combination ofstimulusproperties. Such counter argumentwasprovided inarecentfMRI study by Tong et al. (2000). They locatedperindividual theFFA(Kanwisher et al., 1997), the arearesponding strongest
tohuman faces but notto objects.Takingthe human face as one extreme(strong FFA
response) and objects as theother extreme (noFFAactivation), they comparedthese conditionswithstimuli suchasanimalfacesandschematicfaces.Thesimilarity
betweenfindings inthis study and theonesarguing against modularity is that the FFA didnotrespond inamannerexpectedfollowinga strictly modularaccount,
namely activation onlytohuman faces but nottootherstimuli.Thestudy of Tong et
al. contrastedwiththese studies,however, infindingFFAactivationforanimalfaces. The participants had no expertisewiththese faces, and theactivationwasfound
equally in a taskofpassivewatching as inadelayed matchingtask.Therefore, the authors argued, activation in the FFA canbefound inabsenceof within-category
essential for FFAactivationinGauthier's model. Further, Tong etal. arguedthat the
FFA responds to both featuralasconfiguralaspects ofafacial pattern: the FFA respondstoschematicfaces(only configuralinformationwithoutface features) and
to onlyeyes(featurewithoutthecontext oftheface), but is muchlessactivated by thesestimuli thantowholefaces,catfacesorfaceswithouteyes.Further arguments
againstthe"stimulusproperties" model wereprovided byKanwisher and colleagues.
Kanwisher et al.(1997)argued that FFA activation isnotrelatedtofine-grain
computationsnecessaryfor within-categorydiscrimination, as FFA activation found
with faces wasnotfoundwith discrimination ofhands.Furthermore, FFA activation isnotdependent on task, as faceFFAactivationwasequally presentin discrimination
as in passiveviewingtasks(asimilar findingandconclusionwasfoundwithhouse
stimuli by Tong et al.,2000).Finally, Kanwisher argues that FFA activation is not
dependent onexpertise: whereasthereisconsiderable expertisewithalphanumeric
characters FFA responseisextremely low.
An important notion ofGauthier et al. (2000) isthat within-category recognition
in interactionwith expertiseisimportanttostronglyengage the FFA.Indeed, they showedincreased FFAactivation in car andbirdexperts.Kanwisher(2000)however argues "whenthefusiform face area wasdefined usingthestandardcriteria adopted
in my lab,faceareascouldbefound intherighthemisphere in only 5 of the 19
participants in this study;ofthese,the magnitudes oftheexpertise and
categorization-leveleffects arequite small, andthe response tofacesremainatleast
twiceasstrong as that totheexpert category."Furthermore,shechallenges the use of
Greeblesascontrol stimuli,asthesestimuli look likefaces.Gauthier et al. (2000)
replied to this argumentbypointing outthatwhileGreebleexperts showed
Introduction
.
In conclusion, thedebate on the exactfunctionand specificity of the "face area has notbeensettled yet. Afinalremark onthe"stimulusproperties" model is that not
asingle factorissufficienttoexplainthedifferentlevelofactivationto faces as to
objects. To find thesamelevelofactivation inthe"face area"to objects astofaces, an
interactionoffactors, orasimultaneousoptimalsettingoffactorsisneeded. This raisestheobvious questionwhichobjects
fulfill
theserequirements. So far, it has provendifficult to find controlnon-facestimuli thatshow equally strong activation ofthis area as facestimuli have.Most convincingfindingswere reported with the "Greeble" stimuli,howeverGreebleswere specificallydesigned as "facecontrol".
Greeblesresemblefaces not onlyintermsofsubordinate level matching orexpertise,
but also in many other stimulusfactors (nameableparts,individual as wellasfamilies
of Greebles) and even in general shape (3D surfacewith protrudingfeatures). If no "real
world"
controlobjects canbefound that givesimilarFFAactivationasfaces, theface specificity of thisareamight intheorystillbeexplained bythe specific
combination of propertiespresentwithfaces.However,thegreat numberoffaces and
littlenumberofobjectsactivating this area would then makeitlegitimate to call it a "face area" .
As in behavioralresearch, areturningissue istherelativedependence on
"configural information"(though noconsensusseemstoexist on how this term is defined).Perhapsexpertise doesnotalways lead toconfigural processing. This explanationissupported bythestudy ofTanaka andGauthier(1997)showing that
expertisewithcells or dogs does not leadtobehavioraleffects expected with "holistic
processing".TheseeffectswerehoweverfoundwithGreeble experts,indicating
thateitheradditionalstimulus propertiesare necessaryand "expertise" and
airplanes,afeatural-basedapproachleads tobetter performance thataholistic
approach.Thus,depending on what is moreefficientexpertise withsomestimulus classeswillleadto increasedwhole-basedorconfiguralprocessing,whileexpertise withother stimulusclassesleadstostrongdependenceon extractionofcertain
informativedetails.
BrainModels
Underlyingthedebate of theneurologicalbasis offacerecognitionprocesses, of course amore generalquestion onbrainorganizationisaddressed. Asdiscussed
previously inthis introduction,onetheorydescribes separate andindependent"brain
modules"(Fodor, 1983).Facerecognitionprocesses have beenexplicitlymentioned as an example ofacomponent or "module" inthebrain.Amoduleislocalized
(neurologically separate), and its functionsareinnate,separate,independent and
mandatory. Thus, not only arefacerecognitionprocessesneurologicallyseparate
(located ataparticular brainarea),buttherearedifferentseparateandindependent functions involved in facevs. objectrecognition.
A verydifferentnotion of howfacerecognitionisbased inabrainorganization
underliesthe"object formtopologyhypothesis"(Ishai,Ungerleider,Martin,
Schouten,&Haxby, 1999; Chao,Haxby, & Martin, 1999). In this model, theventral
temporalcortex hasatopological arrangementwith continuous representation of informationonobjects. Therepresentationsof differentcategories aredistributed and overlapping, astheinformation is onobjectformrather thanobject class. Thus,
characteristics ofan objectcluster together.
A thirdmodelwastermed by Tarrand Gauthier (2000) the"flexible
process-map". Rather thandividingthebrainaccordingtostimulusclassorstimulus form, in
this modelareas arebestsuited foracertaincomputation orprocess.Anassortment
Introduction
experience an objectcategory recruits (viarecognitionstrategies)certain components
ofthe process map. For example,we learnto recognize faces at theindividuallevel
(for faces itisnecessary tofindsubtle differences betweensimilarobjects)while most
objects areassociated withahigher "basic-level" recognition.
Obviously it is not yetclearwhichmodeldescribes mostclosely the way thebrain works.Theoverviewprovided previouslycastssomedoubt onthestrongest version ofthe"modular"theory, asitmakesstrongpredictions ontheseparation between
areasinvolvedrespondingtofacesandotherareasrespondingto non-faces. For one,
although the "facearea"probably respondsmoststrongly tofacestimuli, it does show
activationto severalotherclassesof stimuli as well. Similarly,presenting a face does not leadexclusively to "facearea" activation,butseveralareasthroughoutthebrain show activation tofacestimuli.
Further evaluation ofthethreemodelsdescribed above ismadedifficult by the
fact thatnospecificationisprovided onthenatureof functions andinformation
processesinvolved inprocessing a face image.What functions comprise the face
module? What isthe natureof informationstoredinobjectformtopology? Which
processesconstitute the separatecomponents ofthe process map?Furthermore, it is
possible thatadifferentdegreeof"specificity" or dedication tothestimulusclass of facesmightexistforseparate levelsofvisualprocessesinvolved infacerecognition. However, withthe exactnatureof information or functions involved yet unclear it is not possibletosolvethisissue. Inthisthesisthe natureof visualprocessesinvolved in
face recognitionisstudied. Additionallythequestionisaskedwhatfactorsinfluence theseprocesses (e.g. are the processesbound tothestimulusclassofface, dependent
III.Which studiesareincluded and why
The chapters inthethesis areallsubmittedasarticles andtherefore can also be read
independently ofthewhole.The chapters areinchronological order, that is, earliestwritingfirst and thelatest last.Thisstructurewas chosen asitallows the readertofollowthelogicalcourseofdevelopmentofideasandexperiments over the years. Inthisparagraph we
will
introducethechapters of thisthesis.Wedescribe thequestionsaskedand generalparadigmused,without giving awayallresultsobtained (these are discussedin detail inthe separate chapters).
As wesaid before, studieswere performed to studythevisualprocessesinvolved
infacerecognition. Thus,theexperiments presented toprosopagnosic patients were
not set upasclinicalstudiesmapping outwhatunderlies this
deficit
butrather were designedto singleoutvisualprocessesinvolved infacerecognition.The'clinical'approachwould imply thatthelevelof performanceofprosopagnosic patients is
directlycompared with that ofnormal(control) participants.Instead,we studied the
pattern ofpe«ormanceoftheprosopagnosicpatients andcompared that with the patternofperformance of normal controls.Thenormalparticipants weretested to see
whether withourchoiceofstimuli andtest settings, the expected"normal"pattern wouldbeobtained.
We studiedthepuzzling finding of "inversion superiority",where prosopagnosic patientsshowworse performancewith stimuli in upright than with stimuli in
inverted orientation(apattern opposite to thatof normalparticipants). This
phenomenon is ofparticularinterestbecause of theapparent contradictionitposes.
The patients showimpaired performance infacematching tasks, butthedifferential
performance betweenuprightandinvertedstimuliindicatesatleast someintact
Introduction
common assumption that impairedfacerecognitionabilityreflectsimpaired
"configural" processes.
The effectof worse performance with upright than invertedfaceswasfirst reported byFarah,Wilson,Drain,and Tanaka (1995)withprosopagnosic patient LH. In Chapter 1, wefurtherstudy this phenomenonwithpatient LH.Weexamine the generality oftheeffectby presenting to LH morphed stimuli,createdfrom fusing animalandhumanfaces, inboth uprightandinverted orientation. Furthermore, we presented a taskwithhumanfacestimuli, invertedfaces, and also scrambled faces (withthe eyes at theposition ofthemouth andvice versa). Theselatterstimuli are
particularly fittoexaminethephenomenon as onlytheintactvs.impaired face configuration differentiatesbetween thenormalandscrambled facestimuli. The relation betweenintactandimpairedprocesses wasfurtherstudied by presenting a
task specificallyrequiringmatching of a face part. Does thistask,encouraging a parts-based analysis,together withLH'sintactabilitytofocus on a part, help him to overcomehisdifficulty withnormal,uprightfaces?
Intheinitialreport on inversion superiority,Farah et al. (1995)account for the
effect as animpaired"facemodule". Recognition offacestimuliis severelyimpaired
but processing ofobjectsornon-objects (suchasinvertedfaces) can rely onalternative routes.Ourstudies donotsupportthisexplanation ofFarahandcolleagues. In Chapter 1 weprovided logicalreasons why we did not find the "facemodule" explanation likely, andthestudies reported in Chapter 2 were designedtoprovide a
convincing empirical counter argument.We againstudiedthepattern of performance of prosopagnosicpatient LHascompared with thatofnormal controls, but this time we presented notonlyfaces but also objectsinuprightand inverted orientation. The questions onthe"specificity" ofconfigural andfaceprocessesindeed logicallyraise theinverse questionofwhetherthe processesinvolved infacerecognitionare shared
and objectsinuprightandinvertedorientation showedaninversioneffectwithobject
stimuli as well (as wasalsofound inthe famous study by Yin,1969)although the inversioneffect wassmallerwithobjectsthanfaces. Doesthis indicate the use of configuralprocessesinobjectrecognition as well? Conversely,theexplanation of
impaired "facemodule"isdifficulttomaintain
if
inversionsuperiorityeffect is alsofoundwithobjectsinChapter 2.
In Chapter 3, we add a new dimension tothediscussionbycontrasting a case of acquiredprosopagnosia with a caseofdevelopmental prosopagnosia, again with a particularfocusonconfiguralfaceprocessing.Thepurpose of this studyistwofold.
Firsttheexperimentsprovidedaninteresting contrast in matching performance between thetwo patients.This againindicates thattherelationship between
configuralprocesses andfacerecognitionability is notanall-or-nothingsituation.
Whiletheacquiredpatient showed inversionsuperiority, thedevelopmental case did
not. Thisfindingindicates thattherelationshipbetweenconfiguralprocesses and face
stimuli may infact differ, depending on whetheratleast somenormaldevelopment
hastaken place. (This idea of how the use of configuralprocesses relates to
developmental patternwasfurtherstudied inChapter 5.) The second issue in
Chapter3involved extendingthe contrastbetween thetwopatientstorecognition of
the faceclass(superordinate levelrecognition of "a face"), rather thandiscrimination ofindividualfaces(subordinate level recognition of "John's face"). From literature on prosopagnosia, itisknownthatimpairedrecognition ofindividualfaces can be
accompaniedby intact recognition of thefaceclass, and even some spared face processingwithimpairedfacedetection(Schweich& Bruyer,1993).However, the previousstudies also assumed thatimpairedfacerecognitionisaccompanied by
parts-based processing,thereforethe recognition of a face as "aface"couldproceed
inslowparts-basedanalysis. Thesestudies were set up inamanner(e.g.providing
Introduction
strategy. In ourfacedetectionstudiesweincluded tasks with very short presentation time that donotallowafeat:ural analysisbutforcesarapid decisionbased on the
wholestimulusinstead.Though severelyimpaired inthematchingtasks, both
patients showed goodabilitytoperformthese tasks, and evenstillabovechance with 50 ms presentation time. Furthermore, thepatientsshowed averydifferent
performance pattern in thefacematching task, but thiscontrastbetween the two patients wasnotobserved in thefacedetection task.
Inthepreviousstudies, theinversionsuperiorityeffectshown by prosopagnosic patientsisinterpreted assome sparedprocessing of thefaceconfiguration. This
entailsaconclusion that is incontradiction withacommon assumption in face recognition literature, on therelationship betweenconfiguralprocesses and face
recognitionability. Iffacerecognitionabilityisimpaired whilepatientsarestill able
to process thefaceconfiguration, thisgoesagainstthenotion thatfacerecognition
depends on whole-based orconfiguralinformation, whileobjectrecognition depends
on parts-basedinformation. Furthermore, this is in contradiction with the most
common explanation ofprosopagnosia,whichisimpaired configuralprocesses
forcingthepatienttocompletely depend on parts-based analysis instead.Chapter 4 is
aimed specifically at thisissue: Does facerecognition depend on configuralprocesses
whileobjectrecognition dependon parts-based processes?Similarly, doesimpaired
processing of thefacestimulus reflectaninabilityto process thefaceconfiguration?
Theexperimentspresented toprosopagnosic patient RPinChapter 4 were designed specifically totearaparttheinfluence of aface configuration(thenormal face
stimulus) on the one hand, andamannerofprocessing(configuralvs. featural) on the other hand.
Asmentionedbefore,Chapter5presentsdevelopmentalstudies.Ourprevious
divisionbetweenfacerecognitionabilityand configuralprocessingability is by examining how thetwodevelop inrelation toeachother. If thetwo processes were the same, onewouldexpect thatthepatternofrelativedependenceonconfigural information infaceprocessing doesnotchange. Rather,configuralprocessingability
keeps pace withthedevelopment offacerecognitionability.This relationship between configuralprocessing andprocessing members ofthestimulusclassoffaces was studiedinchildandadult performance on visual matchingandrecognition tasks.Furthermore, we studied theassumption that childrendiffer fromadults in their recognition or perception offaces duetoinsufficient availability of configural
processes.Chapter5directlytested whetherchildren dependonfeaturalor
parts-based processesrather thanconfiguralprocesses.In contrast with previousstudies,
thattestedchildrenwithalready quitesomedevelopmentandvisual recognition experience (from 5- or6-years-old), we tested also withveryyoung children (3- or 4-years-old).
Anotherreason toincludethedevelopmental studies in thisthesis isthatchildren
andadultsnaturally differ intheiramountofexperience with processingafacial
image. Expertise has beenproposed to bean essentialfactor, with the use of configuralinformationdepending onexpertise withastimulusclass(Gauthier & Tarr, 1997;Diamond&Carey, 1986). Ifthisindeed were the case, onewouldexpect that the useofconfiguralinformationincreases withage. Adults, with more
experience andthus moreexpertise than children, then show stronger dependence of configural information than children. This samerelationshipbetweenconfigural
processes,facerecognitionabilityand expertisewasexamined again, but with a very differentparadigm, inChapter 6. Here, we presented DutchandTaiwanese
Introduction
Dixon, 1995;Levin, inpress). The useof configuralprocesses infacerecognition is reflected intheworse performancewithinverted thanuprightfaces('inversion effect', Yin, 1969). Thehypothesis ofacloselinkbetweenfacerecognitionability, configuralprocesses,andexpertisepredictsdependencebetweentheown-race
advantage andtheinversioneffect withtheinversioneffectlarger forownrather thanforother race of face.
Furthermore, in thisfinalchapter we returned tothe issue raisedin Chapter 3. In the study of prosopagnosic patients,tworoutes fortherecognition of thefaceclass
and therecognition oftheindividual facewere proposed. In particular, intact or impaired useofconfiguration inoneroute doesnot influencetheother route. The
cross-raceeffectpresents us withanopportunityto againexaminethisissue. Is the
better performance (higher recognition ability) with own race thanother race of face
inthematchingtaskmirrored inanown-race advantage when detecting or categorizing own-vs.other-racefaces?
References
Allison,T.,Ginter,H.,McCarty,G.,Nobre, A., Puce, A., Luby, M., &Spencer, D. D.(1994a)
Face recognitionin human extrastriatecortex.JournalOfNeurophysiology,71(2), 821-825.
Allison,T.,McCarthy,G.,Nobre, A., Puce, A.,&Belger,A.(1994b).Human extrastriate visual
cortex and the perception of faces, words, numbers, andcolors. Cerebral Cortex,
5,544-554.
Bentin,S.,Allison,T.,Perez, E., Puce, A.,& McCarthy, G.(1996).Electrophysiologicalstudies offaceperception inhumans. Journal of CognitiveNeuroscience,8,551-565.
Bentin, S.,&Deouell, L. Y. (2000).Structural encodingandidentification infaceprocessing:
ERP evidence for separate mechanisms. Cognitive Neuropsychology,17,35-54.
Bodamer, J. (194D.Die prosop-Agnosie.ArchivfurPsychiatrie und Nervenkrankheiten, 179, 6-53. Chao, L.L.,Haxby, J. V.,&Martin, A.(1999).Attribute-based neuralsubstratesin temporal
cortex for perceiving and knowing aboutobjects.Nature Neuroscience, 2, 913-919.
Davidoff, J., & Landis, T. (1990). Recognition of unfamiliarfaces. Neuropsychologia,28, 1141-1161.
De Gelder, B.,& Kanwisher, N.(1999).Absence ofafusiform face area inaprosopagnosic
patient.NeuroImage, 9(6), S604.
De Gelder, B., & Rouw, R. (in press).Beyond localisation: A dual routeaccount of face recognition.ActaPsychologica.
DeRenzi, E.(1986).Currentissuesin prosopagnosia. In H. Ellis, M.A.Jeeves,F.Newcombe, & A. W. Young(Eds.),Aspectsoffaceprocessing.Dordrecht:Martinus Nijhoff.
De Schonen, S.,&Mathiver, E.(1989).Firstcomefirstserved.Ascenarioabout development of hemispheric specialization infacerecognitionduringinfancy.European Bulletin of Cognitive Psychology, 9,3 - 44.
Diamond, R.,&Carey,S.(1986).Whyfaces are and arenotspecial:Aneffectofexpertise. Journal of ExperimentalPsychology, 115, 107-117.
Epstein, R.&Kanwisher, N. (1998).A cortical representation ofthe localvisual environment. Nature,392,598-601.
Farah, M. J.(1990) Visualagnosia:Disorders ofobject recognition and what they tellusabout
normal vision.Cambridge, MITPress.
Farah, M. J.(1991).Patternsofco-occurrence among the associative agnosias:Implications for
visualobject representation. CognitiveNeuropsychology, 8, 1-19.
Farah, M.J.,Wilson,K.,Drain, H.,&Tanaka, J.(1995).The invertedfaceinversion effect in prosopagnosia: Evidencefor mandatory,face-specificperceptualmechanisms. Vision
Research, 35(14), 2089 - 2093.
Fodor, J.(1983).ModularityOfMind.MITPress,Cambridge,Massachusetts.
Gauthier,I.,Anderson, A. W., Tarr, M.J.,Skudlarski, P, & Gore, J. C.(1997).Levels of
categorization in visual recognition studied with functional MRI.Current Biology, 7,645-651.
Gauthier,I.,Skudlarski, Gore, J. C.,& Anderson, A. W.(2000).Expertise for cars andbirds
recruitsbrainareasinvolved infacerecognition.NatureNeuroscience, 3, 191-197.
Gauthier, I. & Tarr, M. J.(1997).Becominga"greeble"expert:exploringmechanisms for face recognition.VisionResearch, 37(12), 1673 - 1682.
Gauthier, I., Tarr, M.J.,Anderson, A.
W.
Skudlarski, P., & Gore, J. C.(1999).Activation of themiddle fusiform 'facearea' increaseswithexpertise inrecognizing novelobjects. Nature
Neuroscience,2,568-573.
Homa, D.B.,Haver, B.,&Schwartz, T.(1976).Perceptibility ofschematicfacestimuli: Evidence foraperceptualgestalt.MemoryandCognition,4,176-285.
Ishai, A.,Ungerleider,L.,Martin, A.,Maisog, J. M.,&Haxby, J. V. (1997). fMRI reveals differential activation intheventralobjectvision pathway duringtheperception offaces,
houses, and chairs.Neuroimage,5, S149.
Ishai,A.,Ungerleider,L.,Martin, A.,Schouten, J. L.,&Haxby, J. V.(1999).Distributed
representation of objects in the human ventral visual pathway.Proceedings of the National Academy of Science, USA, 96,9379 - 9384.
Kanwisher, N. (2000). Domain specificity in face perception.NatureNeuroscience,3(8),
Introduction
Kanwisher,N.,McDermott J., & Chun, M. M. (1997).Thefusiformfacearea:amodule in human extrastriate cortex specialized for faceperception. Journal ofNettroscience, 17(11), 4302-4311.
Kanwisher, N., Toni F., & Nakayama, K. (1998). Theeffect offaceinversion onthehuman fusiform facearea,Cognition,68,Bl-Bll.
Laeng, B. & Rouw, R.(inpress). Canonicalviews offaces andthe cerebral hemispheres. Lateralization.
Levin, D. T. (inpress). Race asavisualfeature:Using visualsearchandperceptual
discriminationtasks tounderstandfacecategories and the cross racerecognition deficit.
Journal of Experimental Psychology: General.
Levine, D. N.,&Calvanio, R.(1989).Prosopagnosia:Adefectinvisual configuralprocessing. Brainand Cognition,10,149-170.
Martin,A., Wiggs, C.L.,Ungerleider, L. G.,&Haxby, J. V.(1996).Neuralcorrelates of
category-specifcknowledge.Nature, 379,639 - 652.
McCarthy, G., Puce, A., Gore, J. C.,&Allison, T.(1997).Face-specificprocessing inthehuman fusiformgyrus.Journal of Cognitive Neuroscience, 9(5), 604-609.
McCarthy, R., &Warrington, E. K.(1986).Visualassociative agnosia:A clinico-anatomical study of a singlecase.Journal of Neurology, Nettrosurgery and Psychiatry,49, 1233-1240.
McNeil, J.,&Warrington, E.(1993).Prosopagnosia:Aface-specificdisorder.Quarterly Journal
of Experimental Psychology: Human Experimental Psychology, 46A(1), 1-10.
Moscovitch,M.,Winocur, G.,&Behrmann, M. (199D· Whatisspecialaboutfacerecognition: Nineteen experiments onapersonwith visualobject agnosia anddyslexia but normal facerecognition.Journal of Cognitive Neuroscience, 9,555 - 604.
Puce,A.,Allison,T.,Asgari, M., Gore, J. C.,&McCarthy, G.(1996).Differentialsensitivity of
human visual cortex to faces, letterstrings, and textures: A functional MRI study. journal
of Neuroscience, 16,5205-5215.
Puce,A., Allison, T., Gore, J. C.,& McCarthy, G.(1995).Face-sensitiveregions in human extrastriatecortexstudiedbyfunctionalMRI. Journal of Neurophysiology, 74(3),1192-1199. Puce,A.,Allison,T., Spencer, S.S.,Spencer, D. D.,& McCarthy, G.(1997).Comparison of
cortical activation evokedbyfacesmeasured byintracranialfieldpotentials and
functional MRI: twocasestudies. Human Brain Map;,ing, 5,298 - 305.
Rhodes, G.,&McLean, I. G.(1990).Distinctivenessand expertise effectswithhomogeneous stimuli: Towards a model of configural coding.Perception, 19,773-794.
Schacter, D.L.,Reiman,E.,Uecker,A., Polster, M. R., Yun, L. S.,& Cooper, L. A.(1995).Brain
regions associated with retrieval of structurally coherent visual information.Nature, 376, 587 - 590.
Schweich, M.,&Bruyer, R.(1993).Heterogeneity inthecognitive manifestations of
prosopagnosia: thestudy ofagroup ofsinglecases.Cognitive Neuropsychology, 10(6), 529 - 547.
Sergent, J., Otha, S.,&McDonald, B.(1992).Functional neuroanatomy of faceand object
processing:A positronemissiontomography study.Brain,115, 15-36.
Suzuki, S.,&Cavanagh, P.(1995).Facialorganisation blocksaccessestolow-levelfeatures: An object inferiorityeffect. journal of Experimental Psychology: Human Perception and
Tanaka, J. W.,&Farah, M. J.(1993).Parts andwholes infacerecognition. Quarterly Journal Of Experimental Psychology,46A, 225-245.
Tanaka, J. W.& Gauthier, I.(1997).Expertise inobject andfacerecognition. ThePsychology Of
LearningandMotivation, 36, 83- 125.
Tarr, M. J.&Gauthier, I. (2000). FFA:aflexible fusiform areaforsubordinate-level visual processingautomatized byexpertise. NatureNeuroscience,3(8),764-769.
Tong,F.,Nakayama, K.,Moscovitch,M., Weinrib, O., & Kanwisher, N.(2000).Response properties ofthehumanfusiform facearea.CognitiveNeuropsychology,17(1/2/3/), 257
-279.
Valentine,T.,Chiroro, P.,&Dixon, R. (1995).Anaccount oftheown-race bias and the contact hypothesisbased on a "face space"model offacerecognition. InT.Valentine (Ed.), CognitiveandComputationalAspects OfFaceRecognition:Explorations inFaceSpace (pp. 69-94).London:Routledge.
Chapter 1
Structural Encoding
Precludes
Recognition
of
Face
Parts
in
Prosopagnosial
The extent andtheimpactofsparedprocessingoffacialstimuli intheprosopagnosicpatient LH is
examinedusingthe inversion e#ect and thefacecontext e#ect. Our studyasked howthedeficit in
individualface recognition isrelated to two perceptual abilities thatarespared in thispatient but
between whichthereisinterference when bothareapplied to thefacestimulus,i.e.structuralencoding
Of the.filceand parts-basedmatchingprocedures. Threeexperimentsstudiedthis relationship with task
demands and stimulus properties designed to trigger the parts-based processes. In the first experiment,
human and animalfacesare presenteduprightorinverted with good performance onlyfortheinverted condition. In Experiment 2 normal participants show a clear face context effect (matching of upright
faces easier than scrambledor inverted ones) in thefullface matching task whereas in the parts matching task the face superiority effect disappears. In contrast, LH shows a face inferiority effect when
matchingfullfaces but alsowhen matchingan isolated.face part to a.face part inafullfacecontext. The results show thatstructuralencodingOftheface overrules parts-based proceduresthatcould otherwise be helpful totell individualfacesapart.
1
Introduction
Prosopagnosia isadeficit infacerecognition(Bodamer, 194D,whereby the face no
longerelicits anysenseof familiarityalthoughthepatient continuestorecognize
familiarvoices or gait.Howspecifictofacesthis disorder is, isstillcontroversial,
partlybecause veryfewcasesofprosopagnosia havebeenstudied in such a way that
the possibility ofat least somemild deficit inother areas like word orobject
recognition canbeentirely excluded(Bruce&Humphreys,1994;Farah, 1990;
Gauthier,Behrmann, & Tarr, 1999).Thedebate isnow broadenedbycontributions fromelectrophysiologicalstudies (seeJeffreys, 1996) andfrombrain imaging methods (Gauthier, Tarr, Anderson, Skudlarski, & Gore, 1999;Kanwisher, McDermoth, & Chun, 1997).Recentreportshaveprovidedevidence that lossof normal face
recognitioncanmanifest itself not just as a loss ofthenormal patternofperformance
- forexample,better performancewith upright than withinverted faces - but as its opposite,superior performancewithinverted in contrasttouprightfaces(De Gelder, 1999;De Gelder,Bachoud-Levi, & Degos,1998;Farah,Wilson,Drain, &Tanaka, 1995).In other words, thesepatientspresent us withareversal ofthenormal pattern.
This datasuggests that loss offaceprocessingability isnotsimplyamatteroflosing the abilityto processacertaincategory ofstimuli(faces) norof losingacertain
processingstyle(onewhichtargetsthestimulus configuration), but thatthere is an
interaction betweendamagedandintactskills.Inorder tofocus onthis interaction we refer totheintactaspects offaceprocessingas"structuralencoding" of the face. It is
important to note that in this paper theterm"structuralencoding" doesnotrefer specifically to one or another theory offacerecognition.Thepresent paper
Structuralencoding precludes recognitionofface partsinprosopagnosia
toapplypartsandfeaturebasedmatchingstrategies tofaces. The robustness and
generality of thiseffect isshownin three experiments.
A perspectivecommon tomany studiesofprosopagnosia is thatthedeficit is situated atthewithin-category level and thatfacecategorization itself ortheability to
make afacedecisionisintact.Interms ofthepopularmodelofBruceandYoung (1986), loss offace recognitionabilitycorrespondsto damage to the "facerecognition
units",leaving intacttheearlierstages offaceprocessing. The factthat recognition at
the individualor exemplar leveliscritical forfacerecognition led to the
"individuation" theoryofprosopagnosia (Moscovitch, Winocur,&Behrmann, 1997)
or the view that prosopagnosia isadeficitofwithin-category discriminations, defended by Damasio(Damasio,Damasio, & VanHoesen, 1982;Damasio,Tranel, &
Damasio, 1990). Like themodelofBruceand Young, thisviewassumesthatprior to individual identityrecognition, visualface processingisintactinprosopagnosics. The
conceptofstructural encoding will beusedthroughoutthis papertorefer to this
initialfacecategorizationstage because it ismore general than some ofthespecific
notions advanced to explainfaceprocessing(seefollowing)
A family of more orlessrelated theoriesofnormalfacerecognitionhasfocused
onwithin-categoryprocesses of facerecognition. A common theme is thatthewhole face is more thanthe parts,butthere isnoconsensus as to what isexactlymeant by "whole". One view is that the face
initially
consistsofclearlyseparateparts orprimaryfeatures,whichwhen integrated give rise tothesecond-order features or to
recognition of the face asaconfiguration(thespatial relations betweentheindividual
features), asargued byRhodesandcolleagues (Rhodes,1988;Rhodes,Brake, &
Atkinson, 1993).Astronger claim made byFarahand collaborators, is that face recognition does not start fromtheencodingofseparatefaceparts orinitialparsing but that the faceisrepresentedholistically such thatits parts are not represented
developmentalistshaveargued that, attheentrylevel, facesareencoded thesame
way asanyotherobjectby attending totherelations betweentheparts or to the
overallconfiguration. Fromtheredevelopstheability tousesecond-order facial information, whichunderliesindividualfacediscrimination(Carey&Diamond, 1994; Diamond &Carey, 1986).As Moscovitch et al.(1997) remark,thesedifferenttheories
each addressslightlydifferentquestions.Nevertheless, fine detailsaside, each of them suggests justwhatmight be lost in prosopagnosia: loss of configuration-based
processes inthesenseofRhodes et al., loss of the facemodule or ofholistic processing
inthesenseofFarah et al., or loss of faceexpertise relatedtosecond-order representation ofindividualdifferences.
A critical questionforgrasping the differences betweenthesethreeviews concerns what isthen sparedinprosopagnosia.Ifsecond-orderor
configuration-basedprocessing is lost iswhatremainstherecognition of isolatedfaceparts? Or,
once thefacemodule is lost,arefaces processedlikeobjects, inaparts-based way? Or again, isfaceexpertise - ortheability tousesecond-order relational
information-lost butare facesstillprocessedasbundlesoffirst-orderrelationalinformation just
like objects?Thislatter view isin line withthe consensus in theliterature that
prosopagnosic patients have lostthe abilitytodiscriminate betweenfaces but
continueto categorizefacesnormally. Thus,thenotionthatfirst-order information is
sparedandsecond-orderinformation is lost (in Carey'sterms) reflectsacertain
consensusconcerningthepatternofspared and lost skillsinprosopagnosia.
Recentstudiesof prosopagnosic patients have looked at thisissue inmore detail, complementingtraditional clinicaltestsofintactfacedecisionwithbehavioraltasks
thathaveshown strongface-specificperceptualprocessingeffects, liketheinversion
effect innormal participants. But studieshave revealedaparadox, because some
Structuralencoding precludes recognitionoffaceparts inprosopagnosia
TheParadoxOfInversion Superiority
Theinversioneffect,firstreported by Yin (1969)fornormal participants, hasoften beenreferred to as abenchmarkfor establishingnormalfaceprocessing (Yin, 1970).
Normal participantsarebetteratrecognizing, matching,and remembering a pair of faceswhenthesearepresentedupright thanwhen upside-down (Yin, 1969).Recently Farah et al. (1995) havestudiedtheinversioneffectin prosopagnosic patient LH.
Followingthe resultsobtainedwithright-hemisphere deficitstheprediction was that
LHwouldperform the same waywith uprightandinvertedfaces.But instead, LH showedfaceinversion superiority. Wehaveconfirmed this finding and at the same
time extended itto objectswithagnosicpatient AD(DeGelder et al., 1998).
Subsequently thesameeffectswereobserved with LH(DeGelder, 1999). This reversedfaceinversion effectcannot bereducedtoabsence ofthenormal pattern. Thisfindingchallenges thenotion thattheability tousesecond-orderrelational information is lost andissubsequently compensated forbyusing intact feature-based
routinestodiscriminatefaces.
Instead, theparadoxofinversion superiority is thatindividualfacerecognition is lost but thatsome aspects offaceprocessingarestillactiveandinterferewithreliance upongeneralvisualroutinesinordertodiscriminateindividualfaces. Thus, when the
normal pattern of better performancewith uprightfacesisreversed,moreseems to be at stake thanjustspared facecategorization inthe presence oflost second-order or within-category discrimination. Somehowthesepatientsarehandicappedbytheir
sparedfacecategorization and preventedfromusing intactparts-based processes
with faces.Thelatterare successfully usedwithinvertedfaces but areclearly of no
use to deal withanuprightface.Presumablyinverting afacemakesit object-like and no longer triggersface-specific processes,thereforegivingachancetopart-based routines.Thepresent studyreports experiments designed to testthe robustness and