Investigating the Relation between Imagery and Perception: Evidence ...

10 downloads 23 Views 230KB Size Report
Department of Psychology, University of Glasgow, Scotland. Shigeru ..... Half of the target celebrities had dark hair (e.g. Hugh Grant, Joan Collins) and half had ...
TH E QUARTE R LY JOU RNAL O F E XPE R IM E NTAL PS YCH OL OG Y, 1997 , 50A (2), 274 ± 289

Investig ating the Relatio n between Im agery an d Perception: E vidence from Face P rim ing Roberto Cabeza Rotma n R esea rch I nstitute of B a ycrest Centre, U niversity of Toronto, Ca na da

A. M ike Burton and S tephen W. K elly Depa rtment of Psychology, U niversity of Gla sgow, Scotla nd

S higeru Akam atsu A TR H uma n I nforma tion P rocessing R esea rch La bora tories, K yoto, J a pa n

T h e relation b etw een im agery and percep tio n w as inve stig ated in face p rim ing. Two exp erim en ts are rep orted in w h ich su bjects eith er saw or im agined th e faces of celebrities. T h ey were later given a s peed ed p ercep tual test (fam iliarity judg em en t to p ictu res of celebrities) or a sp eed ed im agery test (in w h ich th ey w ere told th e n am es of celebrities an d asked to m ake a decision abou t th eir app ear an ce). S eeing faces pr im ed th e p ercep tu al test, an d im aging faces prim ed th e im age ry test; h ow ever, th ere w as n o prim ing betw een seeing an d im agin g faces. T h ese resu lts sh ow th at p erce ption an d im agery can be d issociated in no r m al su bje cts. In tw o fu rth er exp erim en ts, w e exam ined th e effects of im aging faces on a su bsequ ent face-n am ing task an d on a task requ irin g fam iliarity jud gem en ts to p artial faces. Both th ese tasks w ere facilitated by p rior im agin g of faces. T h ese resu lts are d iscus sed in relation to th os e of M cD er m ott & Roed iger (1994), w ho fou nd th at im agery pr om oted object p rim ing in a percep tu al test involving n am ing p ar tial lin e d raw ings. T he im p lication s for m od els of face re cogn itio n are also d iscus sed.

W hen o ne im agines a w ell-know n face, such as the face of a friend or a fam ous actor, the ex perien ce ap pears to b e sim ilar to that of actually seeing the face. E ven though the im age is no rm ally less clear, o ne can ``inspect’ ’ the im age of the face just as o ne can inspect the actual face. T his close resem blance betw een imagery an d p ercep tio n is the origin of the notion that im agery is a w eak form of perception. H ow ever, the difference b etw een imagery an d perception could b e qualitative, rather than m erely quan titative. T he objective of the present research w as to investig ate this issue by com parin g the effects of imagery an d perception o n face prim ing.

Requests for reprints sh ould be se nt to M ike Burto n, D epartme nt of Psyc hology, Un ive rsity of Glasgow, G lasgo w G 12 8Q Q , U.K . E m ail: m ike@ psy.gla.ac. uk We w ish to than k Vick i Bru ce, Sh lom o Bentin, and M orris M oscovitch for their valuable comm ents about this pape r. We are also indebted to Yoh’ ichi Tohkura for the provisio n of AT R ± HIP researc h facilitie s. This researc h was supported in part by an ES RC grant to Vick i Bruc e and M ike Burto n (ES RC G R 0002 3 4573) .

Ó

199 7 The E xp erim ental Psyc holo g y S ociety

IM A G E R Y A N D P E R C E P T IO N IN FA C E P R IM IN G

275

T he notion th at im agery an d perception are closely related m ental processes is supported by cognitive an d neuropsychological evidence (for a review, see F inke, 1985). Fo r ex am ple, it has been dem onstrated that im agery can produce perceptual ad ap tation (e.g. Finke, 1979) an d that im agery can activate visu al areas of the cortex (e.g. Farah , Weisberg, M on heit, & Peronnet, 1988; K osslyn et al., 1993). A t the sam e tim e, there is evidence sug gesting that im agery an d perception involve different neural m echan ism s. O n the o ne h an d, th ere are brain-dam aged patients w ho are impaired in imagery b ut not in perception (e.g. Farah & L evine, 1988; Riddoch , 1990). Fo r ex am ple, Farah an d L evine (1988) reported the case of a patient w ith an infarct w ho had an isolate d de® cit in visual imag ery but displayed ``perfect’ ’ (p. 16 1) perfor m an ce in picture an d ob ject identi® catio n. O n the other han d, there are also b rain -dam aged patients w ho are im paired in perceptio n, but not in im agery. Behr m an n, W inocur, an d M oscovitch (1992), for in stance, have recently presen ted th e case of a patient w h o had su ffered a closed head injury w ho disp layed excellent m ental im agery but w as dram atically im paired in object recognition. T h us, even though im agery an d percep tio n are closely related processes, they can be dissociated in brain-dam aged patients. C an im agery an d perceptio n also be dissociated in nor m al subjects? To an swer this questio n, w e investig ated face p rim ing, a pheno m enon ® rst studied by Bruce an d Valen tine (1985), w ho found that seeing a face of a celebrity speeds a later decision ab out w hether or not the face is fam iliar (the face fam iliarity decision task). M ost face prim ing stu dies, like that of Bruce an d Valentine (198 5), have investig ated p rim ing of fam iliar faces in th e face fam iliarity d ecision task, p artly becau se face priming is dif® cult to obtain o n other tests or w ith unfam iliar faces. Fo r exam ple, face prim in g does not read ily occur on sex or exp ression decisio n tests (E llis, Young, & Flude, 1990), an d prim ing of unfam iliar faces is a pictu re-dependent an d short-lived phenom enon (e.g. Bentin & M oscovitch, 1988; Roberts & B ruce, 1989). T he m ain properties of face priming in the fam iliarity d ecisio n task can be sum m ariz ed as follow s (for a review, see Bruce, B ur ton, C arson, H an na, & M aso n, 1994): (1) it occurs between different photog rap hs of th e sam e person (Bruce & Valentine, 1985; E llis, Young, Flude & H ay, 1987); (2) it tran sfers betw een one part of an im age of a face (e.g. inter nal or ex ter nal features) an d the w hole image of the face (Brun as, Young, & E llis, 1990); (3) it does n ot cross dom ains: neith er nam es (B ru ce & Valentine, 1985; E llis et al., 1987) nor pictures of b odies (E llis et al., 1987) produce signi® can t prim ing of faces; (4) it is not affected by how the ® r st face is processed: fam iliarity, sex, or expressio n judgem ents prod uce sim ilar am oun ts of prim in g on a subseq uen t face fam iliarity task (E llis et al., 1990 ); an d (5) it does no t occur un less the face is identi® ed spontaneously at study (B runas-Wagstaff, Youn g, & E llis, 1992). Face prim ing has usually been interpreted in relation to activatio n m odels of face processing, such as those of H ay an d Youn g (1982) an d Bruce and Young (1986). A ccording to the latter m odel, recognizing a fam iliar face proceeds independently of expressio n an d facial speech processing an d involves three sequential step s: activatio n of its face recognitio n u nit (F RU ), activation of its person iden tity node (P IN ), an d n am e retrieval. T here is o ne F RU for each fam iliar face, it is view -independent, an d it responds only to faces. P IN s, in contrast, respo nd also to voices, nam es, an d so on, integ rating in this w ay different form s of perso nal infor m ation .

276

CA BE ZA ET A L.

Burton , Bruce, an d John ston (1990) im plem ented an d extended parts of the B ruce an d Young (1986) m odel u sing an interactive activation an d com petitio n architecture (IAC ). W ithin the IAC m od el, repetition priming is described as follow s. First, the presen tatio n of a fam iliar face activates its F RU, an d this passes activation to the associated PIN. If the PIN ’ s activatio n reaches a given threshold, the face is judged fam iliar. S eco nd, the sim ultaneous activatio n of an FRU an d a PIN strengthens the link between them , so that w hen the sam e face is presented ag ain, the activatio n ¯ ow s faster to the PIN an d the fam iliarity decision is sp eeded. T his is consistent w ith the fact that prim ing does not cross stim ulus dom ains. To recognize a perso n by (for exam ple) a nam e would involve a different route into th e P IN. H en ce, there w ill be no subsequent adv an tage for th e FRU to PIN route used in recognizing this p erso n from a face. T he IAC m odel, derived from the Bru ce an d Young (1986) fram ew ork, can account for m an y face priming results. F irst, it can explain w hy face priming is dif® cult to obtain o n sex an d expression judgem ents an d w ith unfam iliar faces: sex an d expression are processed in dependently of the identi® catio n route in w hich priming occurs, an d unfam iliar faces d o not have F RU s an d PIN s. Second, it can account for m ost of the p roperties of fam iliar face prim ing in the fam iliarity decisio n task: (1) it occurs between d ifferent photog rap h s o f the sam e perso n becau se they activate the sam e F RU an d PIN ; (2) it tran sfers between par t an d w hole face because the F RU ± P IN link is strengthened w henever there is sim ultaneous FRU an d P IN activatio n; (3) it does not cro ss dom ains, because nam es, pictures of bo dies, an d so o n activate th e PIN but not the FRU ; (4) it is not affected by the w ay in w hich the prim e face is p rocessed becau se the ¯ ow of activatio n from the FRU to the PIN is m an datory (an d so identi® cation of a presented fam iliar face is m an datory); an d (5) it does not occur w hen the ® rst face is not recognized spontaneo usly, as the effect depends on a simultaneou sly active FRU an d P IN , an d the u nreco gn iz ed face has not activated a PIN above threshold (by de® nition). W hat predictions are possible concer ning th e effects of im agery on face priming? A lthough there is a large literature on face recognition, there is rather little know n ab out face im agery. C raigie an d H an ley (1993) provide one of the few pap ers o n this topic. Working w ithin the B ruce an d Young fram ew ork, C raigie an d H an ley propose that there is a dedicated im agery m odule for recalling facial infor m ation. In th e rem aind er of this pap er, w e explore the p redictio ns of an alter native prop osalÐ that face imagery m ight rely on top-dow n activation of the sam e represen tatio ns that are used to recogniz e faces. Bruce an d Young (1986) an ticipated this proposal by sug gesting that FRU s can b e activated topdow n , for exam ple by exp ectation (p. 312). If this is true, it seem s reason able to assum e that a voluntary effort to form a m ental image of the perso n’ s face m ight also activate an FRU. W ithin the IAC m odel, all FRU ± PIN links are bi-directio nal, an d so it is possible that FRU s can be activated top-dow n from their correspond ing PIN s by processes su ch as imagery. If this proposal is consistent, face prim ing should occur not o nly as a result of perceiving a face, bu t also as a result of imagining a face. A n alternative IAC -like account is also possible. If FRU ± PIN links are sensitive not only to the simultaneous activation of an FRU an d its cor respond ing P IN but also to the direction in w hich activatio n ¯ ow s alo ng the link, then the effects of imagery an d perception sh ould be d issociable: perceptio n (bottom -up) would produce m ore prim ing than imagery (top-dow n) in a test em ph asizing perceptio n (b ottom -up), w hereas the op posite

IM A G E R Y A N D P E R C E P T IO N IN FA C E P R IM IN G

277

would occur in a test em p hasizin g im agery (top -dow n). T he sam e prediction can be derived from the tran sfer-ap propriate p rocessing p rinciple (M or ris, B ran sford, & Fran ks, 1977 ), as is discussed at the end of the p ap er. T he ob jective of E xperiments 1a an d 1b w as to test this prediction. Subjects saw or imagined faces of celebrities an d then received either a test involvin g perception (face fam iliarity decisio n te st) or a test involving imagery (face im agery decision test). In the face fam iliarity decision test, faces are presented an d subjects have to decide w hether or not each face is fam iliar, w hereas in the face imagery decision test nam es are presented an d su bjects have to im agine the cor resp onding faces in order to decid e w hether or not each face has a particular attribute. If imagery is a weak for m of perception (a qu an titative difference), the ``see’ ’ condition should produce m ore p rim ing than the ``im agine’ ’ cond itio n in b oth tests. In contrast, if imagery and p ercep tion are qualitatively different pro cesses, it is possible that th e ``im agine’ ’ co nditio n w ould surpass the ``see’ ’ cond ition w hen the test involves imagery. Such a resu lt w ould be consistent w ith a con® guration of the IAC m odel in w hich FRU s an d PIN s are con nected by separate bottom -up an d top -dow n links.

EX P ER IM E N T 1A M eth o d Subjects T h irty-six Jap an ese u n d erg r ad u ates p articipated in th e exp erim en t as volu n teers or in retu r n for a sm all p aym en t.

M aterials T h e target m aterials were p ho tog rap h s of 48 Jap an ese celebrities an d th eir n am es. For each of these celebrities, th ere w ere tw o ph otogr ap h s d iffering in view an gle an d /or h airstyle, on e to b e u sed at stud y an d on e at test. T h e 48 celebrities s elected h ad th e h igh est ratin gs in a p ilot stu dy in w h ich 15 su bjects w ere p resented w ith a list of 102 celebrities’ n am es an d h ad to r ate ho w w ell th ey cou ld im agine th eir faces u sin g a 5-point scale (0 = can n ot im agin e th e face or d o n ot kn ow th e p erso n; 1 = p oor im age; 2 = fair im age; 3 = good im age; 4 = excellen t im age). T h e set of 48 celebrities w as d ivided into 3 sets of 16 (8 m ale; 8 fem ale) m atch ed in im agery ratings (3.59, 3.60, an d 3.59, respe ctively). A dd ition ally, p h otogr ap h s of th e faces of a fur ther 28 celebrities an d of 68 u nfam iliar p eop le w ere also em p loye d. A ll th e ph otogr ap h s were cu t from m ag azines, an d the view an gle varied betw een fron t an d 3/4. T h e p ho tog rap h s w ere scan n ed at 72 d p i, an d th e im ages were en larged or red uced so th at th e im age d isp lay h ad a h eigh t of 6 cm on th e com p u ter screen u sed for th is experim en t. W idth varied slightly betw ee n im ages, but w as ap p roxim ately 4 cm . In all th e p h otogr ap h s, everyth ing excep t th e face and th e hair w as er ased u sin g g rap h ics so ftw are. D u ring th e experim en t, all im ages w ere d isp layed as grey-level. V iew ing d istan ce w as 60 cm .

Design and Procedure T h e exp erim ent w as c on du cted on a p ers on al com p u ter. T h e gener al instr uction s o n the screen told su bje cts th at th e object of the exp erim en t w as to investig ate th eir kno w led ge about celebrities, an d th at it co n sisted of th ree tas ks. T h ese th ree tasks cor respo n ded to th e ``im agine’ ’ an d ``see’ ’ stu d y

278

CA BE ZA ET A L.

tasks (a w ith in-su bjects m anip u lation ) an d th e test (fam iliarity d ecisio n or im agery d ecision , a betw een -su bjects m an ipu latio n). D u rin g th e s tu d y p h ase, n o m em ory test w as m ention ed , an d h en ce lear n ing w as inciden tal. D u rin g th e test p ha se, n o referen ce w as m ad e to infor m atio n p reviou sly p rese nted , an d con sequen tly th e tests w ere im p licit. E ach stu dy task con sisted of 1 critical set of 16 celebrities p lus 8 ® ller celeb rities (3 at the beginnin g of th e list and 5 at the en d of the list). O n e ach tr ial of b oth stu dy tasks, th e com p u ter p rese nted a stim u lus for 8 sec, b eep ed , and th en d isp layed a scale u n til a resp on se w as m ad e. In th e ``im agine’ ’ stu dy task, th e stim u lus w as a ce lebrity’ s n am e, an d th e su bjects’ task w as to close th eir eyes an d im agine th e celeb rity’ s face d urin g th e 8 sec. W hen th e com p u ter beep ed , they op ene d th eir eyes an d rated th e m en tal im age on an im age-qu ality scale (th e sam e u sed in th e p ilot stu d y). In th e ``see’ ’ stu dy task, the stim u lus w as a celebrity’ s face, an d th e su bjects’ task w as to insp ect the celebrities’ faces d u ring th e 8 sec. W h en th e com p uter beeped , th ey se lected th e occu p atio n of the celebrity am on g 5 alter n atives (u nk n ow n , actor, singer, com ed ian, oth er). G iven the fact th at all the celebrities u sed w ere very w ell kn ow n in the sub je ct p op u latio n, selecting a profession w as a trivial task th at w as u n likely to d istrac t th eir atten tion from insp ectin g th e p h otog rap h . T h is task w as includ ed on ly to h ave su bjects m ake a resp on se afte r the ``see’ ’ task. In the face fam iliarity d ecision test, 48 target p h otog r aph s of cele brities (16 im agined , 16 seen, an d 16 n on -stu died ), 16 ® ller p h otog rap h s of ce lebrities, an d 64 ph otogr ap h s of u n fam iliar faces w ere p rese nted m ix ed, in a differen t r an d om orde r for each su bject. In th e case of seen faces, stu dy an d test p ho tog rap h s w ere d ifferen t. For each face, su bjects h ad to ind icate as qu ickly as p ossible w h eth er th e face w as of a kn ow n per so n (press th e ``/’ ’ key w ith th e righ t h and ) or of an u n kn ow n p erso n (p ress the ``z’ ’ key w ith th e left h an d ). L aten cies of m ore th an 2 sec w ere n ot ana lysed . Su bjects w h o w ere n ot fam iliar en ou gh w ith th e celebrities (u su ally becau se th ey did n ot w atch televisio n) an d h en ce m ad e too m any er ro r s (m ore th an 12.5% ) w ere exclud ed an d rep laced. In th e face im agery d ecision test, su bjects w ere p resen ted w ith th e w ritten na m e of a celebrity an d asked to d ecide as qu ic kly as p ossible w h eth er th is p erso n h as a m ole on his/h er face. A s no celebrity in th e stud y list w as fam ou s for h avin g a m ole, sub jects could n ot m ake this d ecision o n th e basis of their gener al kn ow ledg e abou t th e celebrities; in order to m ake th e d ecision , they h ad to con stru ct a m en tal im age of the celebrity’ s face an d scan it for th e p resence of a m ole. It is very d if® cu lt to d eter m ine w h eth er or no t each cele brity in th e list actu ally ha s a m ole on th e face, an d h en ce w e d id n ot try to equ ate th e n um ber of cor rect ``yes’ ’ an d ``n o’ ’ resp on ses or to assess th e accu racy of su bjects’ resp on ses. Reaction tim e in this test re¯ ects th e tim e su bjects requ ired to for m an im age an d scan it u n til th ey felt con ® d ent en ou gh to m ake th e resp on se. A ccordin gly, reactio n tim es re¯ ect su bjects’ im agery p erfor m an ce accord ing to th eir ow n criteria. T h e im agery d ecision test d oes n ot involve u n fam iliar faces, an d h en ce it h ad on ly h alf as m an y trials as th e face fam iliarity d ecisio n test. In both test con d ition s, th e ord er of the tw o stu dy tasks and th e assign m en t of th e th ree critical sets of celebr ities to th e ``see’ ’ , ``im agin e’ ’ , an d n on -stu died con d itio ns wer e com p le tely cou n terbalan ced across su bjects.

R esu lts an d D iscu ssion T he top pan el of Table 1 show s the m ean reaction tim es in the fam iliarity an d imagery decision tests of E xperim ent 1a. A s in previous research (e.g. Br uce & Valentine, 1985; E llis et al., 1 987), er ror rates in the fam iliarity decision test w ere sm all (see: 3.8% ; imagine: 6.6% ; no n-studied: 5 .9% ) an d were not an alysed further. First, as the reactio n tim es were co nsiderably longer in th e im agery decision test than in the fam iliarity decision test, a separate o ne-way A N OVA (see vs. imagine vs. non-stu died)

IM A G E R Y A N D P E R C E P T IO N IN FA C E P R IM IN G

279

TABLE 1 a E x p e rim en ts 1a a n d 1b : M e an R e sp o n s e L at en c ies in th e Fa c e Fa m ilia rity D ec is io n an d F ac e Im a g e ry D e cis io n T e s ts

Encoding Conditions Test Exp eriment 1a Exp eriment 1b

a

face fam iliarity decision face im agery dec ision face fam iliarity decision error s (% ) face im agery dec ision error s (% )

I ma ge

S ee

Non-studied

803 154 1 817 24 100 3 21

773 164 3 748 16 115 6 17

82 0 170 4 86 2 23 120 3 24

I n milliseconds.

w as perfor m ed on each test. In the fam iliarity decisio n test, the m ain effect of the see± imagine± no n-studied m an ipu lation w as signi® can t, F(2, 34) = 4.58, p < .0 5. Prim ing (studied ver sus no n-studied) w as sig ni® can t in the see con ditio n, F(1, 34) = 8.92, p < .01 , but not in the im agine cond ition , F(1, 34) = 1.1 5, p > 0.2. T he difference betw een these two condition s w as m argin ally signi® cant, F(1, 34) = 3.66, p < .065. In the imagery decision test, th e m ain effect of the see± imagine± no n-studied m an ipulatio n w as signi® can t, F(2, 34) = 9.07; p < .01. Prim in g w as signi® can t in the im agine co ndition. F(1, 34) = 17.75; p < .01, but not in the see condition, F(1, 34) = 2.46; p > .1. T he difference between the see an d imagine co nditions w as signi® can t, F(1, 34) = 6.99; p < .0 5. In sum , priming in th e face fam iliarity decision test w as robust fo r seen faces but non-signi® can t for imagined faces, w hereas prim ing in the im agery decisio n test show ed an o pp osite p atter n: it w as robust for imagined faces but no n-signi® can t for seen faces. Seco nd, a 2 (study task: see vs. im agine) 3 2 (test: fam iliarity decisio n, imagery decision) A N OVA w as co nducted o n priming scores (no n-studied RTsÐ stud ied RTs). T he m ain result of this an alysis w as a signi® can t interactio n between study task an d test factors, F(1, 34) = 8.8 9; p < .01, re¯ ecting the fact that perceptio n produced m ore prim ing than im agery in the face fam iliarity d ecision test, w hereas im agery produced m ore priming than perceptio n in th e im agery decision test. T hese results ap pear to support the id ea that seeing an d im aginin g faces involve separate processes. T he results show that seeing a face w ill prim e sub sequently seeing the face, an d imaging a face w ill prim e subsequent imaging of a face. Prim in g crossing between seeing an d imaging, in either directio n, w as not signi® cant in E xp erim ent 1a. T hese results seem to b e inconsistent w ith th e assum ption in the IAC m odel th at FRU ± PIN links are b i-directional an d eq uivalent. Instead , the data are consistent w ith separate, an d separately primeable, top-dow n an d bottom -up links b etw een PIN s an d FRU s. U nfortunately, the interpretation of the results of E xperim ent 1a is com plicated by the fact that the type of im agery decision test (the ``m ole’ ’ task) used in this ex periment does not allow an objective m easure of subjects’ accu rac y in im agining faces. We assum e that faster RTs in this test re¯ ect a facilitation to im agine

280

CA BE ZA ET A L.

previou sly encountered faces (priming), but w e cannot dem o nstr ate that sub jects were actually imagining target faces. A num ber of p ossible tests of imagery are available w hich m ight allow o ne to test the accur acy of the sub jects’ imagery. Fo r ex am ple, sub jects m ight be asked w hether celebrities have glasses or wear beards (Craigie & H an ley, 1993) or w hether their hair is light or d ark. N one o f these m an ipulations is plausible w ith Jap an ese celebrities an d subjects. T herefore, in E xperim ent 1b w e replicated E xp eriment 1a using C au casian sub jects an d celeb rities, an d a light hair/dark hair imagery task at test.

E XP E R IM E N T 1 B M eth o d Subjects Twen ty fou r su bjects w ere recru ited from the U n iver sity of G lasgow. S u bjects received a sm all p aym en t for th eir p ar ticipation .

M a terials T h e face s of 64 C au casian celebrities (48 critical an d 16 ® ller) an d 64 u n fam iliar C au casian face s w ere u sed in th is exp erim en t. T h e celebrities w ere selected from a p ilot stu d y of 92 celebrities. T h e m ost im ageable cele brities w ere chosen from th e 92 w ith the ad d ition al criterio n that p ilot su bjects be in a m inim um of 70% agreem en t as to w h ethe r a sp eci® c celebr ity’ s h air w as ``ligh t’ ’ or ``dark ’ ’ . H alf of th e targ et celebrities h ad d ark h air (e.g. H u gh G ran t, Jo an C ollin s) an d h alf h ad ligh t h air (e.g. D ar ry l H an n ah , Steve M artin), an d th ese lists w ere split into 50% m ale:50% fem ale. T hese 48 critical celebrities w ere d ivided into 3 lists of equal im ageability (rated 3.41, 3.42, an d 3.43 on th e sam e 0± 4 scale u sed in E xp erim en t 1a). P ho togr ap h s w ere scan n ed, and ver tical im age size w as ad justed to 6 cm . H orizon tal im age size varied sligh tly betw een im ages bu t w as ap p roxim ately 4 cm . E very th ing except the face an d hair w as er ased u sin g gr ap h ics so ftw are. D urin g th e experim en t, all im ages w ere d isp layed as g rey-leve l an d view ing distan ce w as 50 cm .

Design and Procedure T h e d esign and p ro ced u re w ere iden tical to E xp erim en t 1a, w ith the exc eption th at th e im agery task at test requ ired s ub jects to state w h ether th e n am ed celeb rity had ligh t or d ar k h air. A s in E xp erim ent 1a, laten cies above 2 sec w ere d iscarded , alth ou gh su bjects w ith er ror rates above 12.5% w ere n ot excluded .

R esu lts an d D iscu ssion T he RT results of E xperiment 1a are clear an d echo th ose of E xperim ent 1b (see Table 1). H owever, before w e co nsider the RT data, we should ® rst exam ine the er ror rates in this ex periment. E rror rate s were h igh, though relatively consistent in this study. Fo r the fam iliarity decision test, erro r r ates in each of the conditio ns w ere: see, 16% ; im agine, 24% ; not

IM A G E R Y A N D P E R C E P T IO N IN FA C E P R IM IN G

281

stu died, 23% . Fo r the im agery decision test, er ror rates w ere: see 17% ; im agine, 21% , not stu died, 24% . Separate A N OVAs w ere conducted on er rors fro m the two tests, each revealing no signi® can t differences (on the fam iliarity decision, F(2, 22) = 3.26, p > 0.0 5; on the imagery decisio n, F(2, 22) = 2.15, p > 0.05). Both the nature of the experim ent an d the restrictio n on stimulus choice lead to these high error rates. Er ror s are calculated fro m test phase only. It w ould be possible to co nditionalize er rors at S tage 2 by kn ow n-only item s at Stage 1, an d this reduces the er ror rates to those m ore norm ally reported in face recognitio n experim ents. H owever, standard izing in th is w ay lead s to disproportionate errors in the ``not studied’ ’ g roup, as all unknow n celebrities ap pear as er rors at Stage 2. Fo r this reason w e have chosen to present er rors at test, ignoring perfor m an ce on the item s at study. N ote also th at the restrictions on attributed hair colour also lead to er rors. It is relatively dif® cult to ® n d fam ous faces w hose hair is consistently held to be light or dark, an d hence the fam iliarity of the faces w as low er than w ould ideally b e the case. H aving said this, there is n o evidence that errors w ere unevenly distribu ted between the cond ition s, an d so w e now tur n to an alysis of reactio n tim es. RT data for cor rect decisions o nly are show n in th e bottom pan el of Table 1. T hese results show the sam e patte rn as those of Ex perim ent 1a. F irst, a separate A N OVA w as co nducted on each test. In the fam iliarity decisio n test, the m ain effect of the see± imagine± no n-studied m an ipulation w as signi® can t, F(2, 22) = 8.2, p < .01. Prim in g (studied vs. no n-studied) w as signi® can t in the se e co ndition, F(1, 22) = 16.2, p < .01, but not in the im agine cond ition , F(1, 22) = 2.62, p > 0.05. T he d ifference between these two co nditions w as reliable, F(1, 22) = 5.74, p < .05. In the im agery decisio n test, the m ain effect of the see± imagine± n on-studied m an ipulation w as signi® can t, F(2, 22) = 5.02; p < .05. Prim ing w as reliable in the im agine condition, F(1, 22) = 9.12; p < .01, bu t not in the see co nditio n, F < 1. T he difference between the see an d im agine conditions w as signi® can t, F(1, 22) = 5.4; p < .05. Seco nd, a 2 (study task: see vs. im agine) 3 2 (test: fam iliarity decisio n, imagery decision) A N OVA w as conducted on priming scores (RTs for no n-studied item s m inus RTs for studied item s). In this A N OVA, the m ain effect o f study task, F(1 , 22) < 1, an d test, F(1, 22) < 1, were non-signi® can t. T hese results indicate th at, over all, the am ou nt of prim ing w as sim ilar in the two study conditions an d in the two tests. H owever, the effect of the study m an ipulatio n on th e two tests w as very differen t, as indicated by a highly signi® can t interaction between study task an d test, F(1, 22) = 6.56; p < .01. T h is signi® cant interaction re¯ ects a cross-over dissociation betw een face fam iliarity an d face imagery decision tests as an effect of the see-im age study m an ipulation. In a test involving perceptio n of faces (face fam iliarity decision), priming w as larger for faces that were seen at study than for faces th at w ere im agined at study. In a test req uiring su bjects to m ake a decision requiring a face to be imagined, the opposite patter n of results w as found, w ith prim ing b eing larger for im agined faces th an for faces th at w ere actually seen at study. T he results of E xperim ent 1b are co nsistent w ith the results of E xperim en t 1a an d su pp ort the idea that there is a qualitative difference between im agery an d perceptio n. If imagery w ere m erely a w eak for m of perception, the see co nditio n should have p roduced m ore prim ing in both tests, but it did not. In terpreted in ter m s of the IAC m odel of face processing (Burto n et al., 1990), the results of E xperim ent 1a an d 1b sup port the n otio n

282

CA BE ZA ET A L.

that between an F RU an d a PIN there are two separate links: a bottom -up FRU ® P IN link, w hich is th e one primarily involved in identi® cation, an d a top-dow n PIN ® F RU link, w hich is the o ne prim arily tapped by imagery; these two links seem to be independen tly primeable.

E X PE R IM EN TS 2A A N D 2B T he ® nd ing of E xperim ents 1a an d 1 b that imagery produced non-signi® can t priming o n the fam iliarity decisio n test is ap parently inconsistent w ith the results of M cD er m ott an d Roediger (1994), w h o found that an imagery study condition produced rob ust prim ing in a percep tual test. In the study phase of their experim ents, sub jects read w ords (e.g. read ``b utter¯ y’ ’ ), saw line draw ings (e.g. see a picture of a butter¯ y ), or read w ords an d imagined their referents (e.g. read ``butter¯ y’ ’ an d im agine a lin e dr aw in g of a butter¯ y). T hey also rated the p leasan tness of the w ords, drawings, an d im ages. A t test, subjects had to nam e brie¯ y presented fragm ented pictu res of the referent ob jects. P rim ing w as signi® cant in th e see condition but not in the read condition. M ore important, prim ing in the imagery cond ition w as ap proximately m idw ay between the see an d read co nditions an d w as sign i® can t (E xperiments 1, 2a, 2b, an d 4). T he image study cond ition in M cD er m ott an d Roediger’ s (1994) exp eriments w as very sim ilar to the one in E xperim ents 1a an d 1b. In b oth cases, subjects read a w ord (th e nam e of an object or the nam e of a celebrity), constructed a m en tal im age (an image of a line draw ing of the object or an image of a celebrity’ s face), an d rated the im age (a pleasan tness or vividn ess rating of the object image, or a q uality rating of th e face im age). It is therefore su rprising that an essentially iden tical imagery condition produced a co nsiderable am ount of priming in M cD erm ott an d Roediger’ s experiments bu t not in our E xperiments 1a an d 1b. O n e sim ple explanation is that line drawings of o bjects are easier to im agine than are faces o f celebrities, an d h ence subjects generated better images at encoding in M cD er m ott an d Roediger’ s exp eriments than in E xperim ents 1a an d 1b. Yet, the m ean rating in the imagery study task in E xperim ent 1a w as 3.59 an d in Exp erim ent 1b w as 3.42 (3 = good image, 4 = excellent image), w hich ind icates that subjects evaluated their ow n im ages as very good. M oreover, the sam e im agery condition yielded a rob ust am ount of prim ing in the im agery decision test. If the n atu re of the test can produce such a rem arkable chan ge on the effects of imagery, it is m ore reaso nable to attribute the inconsistency between M cD er m ott an d Roediger’ s results an d ours to differences between th e picture fragm ent identi® cation test an d the face fam iliarity decision test. T h ere are two differences th at m igh t be relevan t. First, the pictu re fragm ent identi® cation test involves nam ing, w hereas the face fam iliarity decision test d oes not. Becau se the imagine condition involves seeing a nam e, it is possible that part o f the priming generated by this cond ition in the pictu re fragm ent identi® catio n test re¯ ects verb al tran sfer. In this case, the difference between the effects of the im agine co ndition in the fam iliarity decision test an d in picture fragm ent identi® catio n test w ould not be in imaginal priming but in verb al priming. Seco nd, the test stim uli in the picture fragm ent id enti® cation test are deg rad ed, w hereas in the face fam iliarity decision test th ey are not. T he reason w hy this difference could be relevan t is that identifying fragm ented stimuli m ay req uire a m ental reconstruction of the

IM A G E R Y A N D P E R C E P T IO N IN FA C E P R IM IN G

283

fragm ents, a process that m ay be similar to imagery. T h us, it is possible that the im age co ndition produced robust prim ing in th e picture fragm ent identi® cation test b ut not in the face fam iliarity decisio n test, becau se th e ® rst test involves an important im aginal co m ponent w hereas the seco nd is m ore pu rely perceptual. T he face fam iliarity decisio n test w ould primarily involve a bottom -up activation ¯ ow, w hereas the picture fragm ent co m pletio n would involve bo th bo ttom -u p an d top-dow n ¯ ow s of infor m ation. In sum , it is possible that im agery can generate su bstantial prim ing on a perceptual test if the test involves nam ing or stim ulus deg rad ation. A ccord ingly, E xperim ent 2a exam ined imaginal prim ing in a percep tual test involving nam in g an d E xperiment 2b in a perceptual test involvin g identi® catio n of deg rad ed stim uli. T he test investig ated in E xperiment 2a w as a face identi® cation test an d similar to th e w ord id enti® cation test that is w idely used in the verb al priming dom ain: subjects have to nam e brie¯ y p resented faces of celebrities, an d the dependent variable is nam ing accuracy. T his test differs from the face fam iliarity decision test not only in nam ing, but also in that stim ulus exposure is very brief. In order to deter m ine w heth er this second difference is relevan t, exposu re tim e w as also m an ipulated (20 m sec vs. 5 0 m sec). T he test inv estig ated in Ex periment 2b w as a ver sion of the face fam iliarity decision test in w hich the test p ictures included o nly the inter nal features of th e face (eyebrow s, eyes, nose, an d m outh). T he focus of E xperim ents 2a an d 2b w as w hether im agery can produce signi® can t prim ing o n perceptual tests involving nam ing or stim ulus deg rad ation. It w as therefore necessary to introduce a co ntrol con ditio n in w h ich the nam es of celebrities are read b ut no im ages are for m ed (labelled the ``read ’ ’ co ndition).

M eth o d E igh teen Jap an ese un d erg rad u ates w ere p aid to p articip ate in E xp erim en t 2a, an d a fu rth er 18 in E xp erim ent 2b. T h e m aterials w ere th e sa m e as in E xp erim en t 1a. In bo th experim en ts, th e stu dy tasks w ere: (1) th e ``im age’ ’ tas k, u sed in E xperim en ts 1a an d 1b; an d (2) a ``re ad’ ’ task, in w h ich su bjects are asked to m ake an occu p ation d ecision to a cele br ity’ s n am e. Task (2) rep laces th e ``see’ ’ task from E xp erim ents 1a an d 1b. To keep th ese tasks as sim ilar as p ossible, celeb rity n am es w ere presen ted for 8 sec, follow ing w h ich an occu p ation d ecisio n w as m ad e. O n ce again , read /im agine p rim ing con d ition s w as a w ith in-su bjects m an ipu latio n. T h e face iden ti® catio n test (E xp erim en t 2a) includ ed th e 48 c ritical ph otog r ap hs of celebrities (16 im agine con dition , 16 read con d ition , an d 16 n o n-stu d ied ) p lus 16 ® ller p h otog rap h s of celebrities m ixed in a ran d om ord er. S u bjects w ere told th at faces of celebrities w ou ld ap p ear brie¯ y on th e screen and that th ey h ad to sa y th eir n am es, gu essing w h en n ot su re. E ach trial con sisted of a w ar n ing sign al (an oval of h eigh t 7.5 cm ) for 500 m sec, the p ictu re of a face for 20 or 50 m sec, a m ask (a 7.5cm oval ® lled w ith a h atch ing p atter n available w ith in th e g rap h ics so ftw are u sed ) for 500 m sec, an d the sen ten ce ``A n swer p lease’ ’ u n til th e resp on se w as m ad e. A ll the se even ts ap p eared in th e sam e central p osition of th e screen . S u bjects said th e n am es, and the exp erim en ter en ter ed the respo ns e in the com p u ter as eith er cor rect or incor rect. T h e inter val betw een trials w as 1.5 sec. B efore the test there w ere 4 p ractice tr ials w ith ® ller celebrities, an d th e ® rst 4 trials of th e test w ere also ® llers. T h e face fam iliarity d ecision test (E xp e rim en t 2b) w as as in E xp erim ent 1a, w ith tw o excep tion s. F irst, th e p ictu res of celebrities an d u n fam iliar faces in th e test includ e d o nly the inter nal featu res of the face. U sing a gr ap hics prog ram , an oval inclu d ing on ly the eyebro w s, eyes, n ose, an d m ou th w as cu t from each face. T h e p recise d im en sio ns of th is oval varied from face to face, bu t it w as c on -

284

CA BE ZA ET A L.

str u cted to be of th e sm allest siz e po ssible th at w ou ld includ e eyes, no se, an d m ou th. S econ d , in th e test each face w as p reced ed by a be ep so u n d an d rem ain ed on the screen u n til su bjects pressed th e / key (fam iliar) or th e Z key (u n fam iliar). A 1500-m sec ISI follow ed th e respo n se.

R esu lts an d D iscu ssion T he results of the face nam ing test in E xperiment 2a are show n in Table 2. T he m ain effect of exposure w as reliable, F(1, 31) = 3 9.74, p < .01, re¯ ecting overall higher accuracy in the 50-m sec cond ition th an in the 20-m sec co nditio n. T he m ain effect of the im agine± read ± no n-studied factor w as signi® can t, F(2, 62) = 3.72, p < .05, but the E xposure 3 Item Type interaction w as n ot, F < 1. Prim ing w as signi® can t in the im age co ndition, F(1, 70) = 5.69, p < .05, but not in the read condition, F(1 , 70) = 2.37, p > .1. T he difference between the image an d read conditions w as non-signi® can t, F < 1. T he accur acy an d RT data of the inter nal-feature face fam iliarity d ecision test in E xperiment 2b are show n in Table 3. In the accuracy data, there w as a signi® can t effect of the image± read ± non-studied factor, F(2, 34) = 3.2, p < .05, an d pairw ise com parisons indicated th at prim ing in th e image co ndition w as reliable, F(1, 34 ) = 3.9, p < .05, w hereas priming in the read condition w as not, F < 1. T he difference between the im age an d read conditions w as signi® cant, F(1, 34) = 5.7, p < .05. A nalysis of th e RT data prod uced a signi® cant m ain effect of the im age± read ± n on-studied factor, F(2, 34) = 5.8, p < .01. Priming w as reliable in the image co nditio n, F(1, 34) = 11.47, p < .01, but not in the read condition, F(1, 34) = 2.2, p > .1, w ith no reliable difference between them , F(1, 34) = 3.63, p > .05 . In E xperim ent 2a, the sam e im age co ndition that produced alm ost no priming in the face fam iliarity decisio n test in E xperiments 1a an d 1b yielded a signi® can t am oun t of prim ing in the face identi® cation test. P rim ing by reading a nam e w as not signi® can t, though it suggests a trend in th at direction. T hese results sugg est that nam ing a t test is affected by prior im agery, th ough this result m ay depend in pa rt o n the fact that su bjects read a nam e as part of the imagery prim in g task. In E xp eriment 2b, the image co nditio n generated a robust am oun t of prim ing in the inter nal-feature fam iliarity decision test, w hereas the sam e condition produced alm ost no prim ing in the full-face fam iliarity decisio n te st in E xperim ents 1a an d 1b. T h is difference is especially striking becau se th e two tests w ere essentially identical, w ith the o nly relevan t difference b eing that the test stim uli w ere deg rad ed in o ne case but not in the other. T his result is consistent w ith th e hypoth esis that stim ulus d eg radatio n can enhan ce im aginal TA B LE 2 M e a n A c c u ra c y o f N am in g Fa c es as a Fu n c tio n o f E n c o d in g T a sk a n d E xp o s u re o f T e st Ite m

Encoding Condition Test 20-m sec exposure 50-m sec exposure

I ma ge

R ea d

N on-studied

.32 .55

.2 9 .5 3

.24 .49

IM A G E R Y A N D P E R C E P T IO N IN FA C E P R IM IN G

285

T A B LE 3 M e a n A c c u r ac y an d R e s p o n s e La ten c ie s fo r a F am ilia rity D ec is io n T ak e n to the In te rn al F e atu res o f Fa c es

Encoding Conditions Test

I ma ge

decision accurac y reaction times

.76 141 0

Rea d

N on-studied

.66 1538

.6 8 1638

prim ing o n perceptual tests, b ecau se processing deg rad ed stimuli involves a process of m ental reconstructio n that is close to imagery operatio ns. In sum , the results of E xperim ents 2a an d 2b suggest that both nam ing an d stim ulus deg rad ation at test can enhan ce the am ount of priming obtained by prior imagery generated from a nam e. We postulate that the reasons for th is effect m ight b e different for nam ing (2a) an d stim ulus d eg rad atio n (2b). N am ing m ay receive enhan ced prim ing from imagery in part due to som e verbal prim ing, th ough this rem ains speculative. Identifying deg rad ed stimuli, o n the oth er han d , m ay receive priming from imagery b ecau se imagery is itself involved in this particular percep tual test. T he results of E xp eriments 2a an d 2b can account for the ap paren t inco nsistenc y between the effect of the im agery condition o n the face fam iliarity decisio n test in E xperim ents 1a an d 1b an d o n the pictu re fragm ent identi® cation test in M cD er m ott an d Roediger (1994), because the latter test involves both nam ing an d stim ulu s deg rad ation.

G EN E R A L D IS C U S S IO N To sum m arize the p resent research, the results of E xperim ents 1a an d 1b show ed a crossover dissociatio n between imagery an d perception o n face priming tests: perceptio n prod uced prim ing in a test involving perception, w hereas im agery generated prim ing in a test involving im agery. T he resu lts of E xperim en ts 2a an d 2b indicate that th e am ount of p rim ing generated by image g eneration from a nam e to a perceptual test increases w hen the test involves n am ing or image deg rad ation. T he presen t results have direct im plications for theoretical accounts of face priming. T he IAC m odel of Burto n et al. (1990) postulates that priming re¯ ects strengthening of the con nections between FRU s an d PIN s. T he cross-over dissociatio n foun d in E xperim ents 1a an d 1b is co nsisten t w ith the hypothesis that each F RU ± PIN pair is connected by two separate links: a bottom -up F RU ® PIN link an d a top-dow n PIN ® F RU link. Perceiving a face primarily strengthens th e bo ttom -u p FRU ® PIN link an d prod uces a co nsiderable am ount of p rim ing on tests involving the sam e link, such as the fam iliarity decision test, but little prim ing o n tests tapping th e top-dow n PIN ® FRU link. In contrast, im agining a face stren gth ens in particular the top-dow n PIN ® FRU link an d yields robu st prim in g on tasks tapping this co nnection , like the imag ery decisio n test, but little prim ing o n tasks b ased on the bottom -up FRU ® PIN link. If this interpretation is correct, there are im portant im plications for the structu re of the IAC m odel. A t present links are strengthened th rough a H eb bian up date m echan ism :

286

CA BE ZA ET A L.

w henever two co nnected units are simultaneously active, the strength of their connecting link is increased (Burton & Bruce, 1993; Burto n, 1994). H owever, if F RU ± P IN links are not sym m etrical, then a new m echan ism m ust be introdu ced: it is not suf® cient to know that two units are sim ultan eously active, o ne needs also to kn ow w hich b ecam e active ® rst. A lthough com putationally quite tractable, th is proposal w ill req uire extra resources not available to the m odel as it cur rently exists. T he results of E xperim ents 1a and 1b are also consistent w ith processing accounts of m em ory perfor m an ce (G raf & Ryan , 1990; M oscovitch, Vrie z en, & G oshe-G ottstein, 1993 ; Roediger, Weldon , & C hallis, 1989). T hese accounts are based on the tran sferap propriate processing principle (M or ris et al., 1977), w hich states that m em ory perfor m an ce is a functio n of the overlap between study an d test operatio ns. O n the b asis of this principle, the dissociation between the fam iliarity decisio n test an d the im agery decisio n task as an effect of the see-im age m an ipulatio n can be explained as follow s: the m ental oper ation s tap ped by the fam iliarity d ecisio n test have a larger overlap w ith the operations involved in the ``see’ ’ study task than w ith the ones eng aged by the imag ery study task, w hereas th e converse is true fo r the imagery decisio n te st. O ne problem for this explanatio n, how ever, is that M cD er m ott an d Roediger (1 994) em ployed th e sam e p rinciple to account for signi® cant prim in g effects between the im age stu dy task an d the picture fragm ent identi® catio n test. A ccording to these au thor s, these signi® can t priming effects re¯ ect the im portant overlap between imagery an d perceptio n. In fact, w e have show n that these results can be explain ed by overlap in th e particular tasks used to exam ine these ph enom en a. First, th e results of E xperim ent 2a an d 2b su ggest that a signi® can t am ount of the priming prod uced by imagery o n picture fragm ent identi® cation tests is p artly a co nsequence of nam ing at test an d partly a co nsequence of using a perceptual test designed to incor porate som e imagery. Seco nd, the idea that imagery an d p erception share com m on operatio ns an d the evidence that they can som etim es prim e each oth er does not imply that they can not be dissociated w hen they are pitted ag ainst each other. In fact, differen t types of perceptual processing (C ab eza, 1995; Weldon & Roediger, 1987) an d different types of conceptual processing (Cab eza, 1 994) can be dissociated w hen the precise discrepan cies between th em are directly m an ipulated in th e experim ent. A lthough b oth processing an d structural accounts can be brought to bear on the data presented here, we have focused on a structural interpretation, an d in particular on the IAC account of prim ing. It seem s to us that converging evidence favours a structural interp retatio n of these effects. T he resu lts of E llis et al. (1990) clearly dem onstrated that fam iliarity d ecision s cou ld be p rim ed equally w ell by a prior fam iliarity, sex, or expressio n judgem ent. H ow ever, sex or expression jud gem ents do not prime subsequent sex or ex pression judgem ents. Further, E llis, B urton, Young, an d F lude (in press) dem onstrated that recognition of a part-face primes subsequent recognitio n of a no n-overlapping partface of the sam e personÐ a result that seem s m ore naturally to favour a structural account of p erson recognition prim ing. O f course, we do not claim to have elim inate d the possibility of a processing accoun t of these p henom ena, an d data from an y individ ual experim ent are explicable in these ter m s. H ow ever, w ithin the theoretical fram ew ork of face recognitio n the co nverg ing eviden ce ap p ears to favou r stru ctural theories, an d w e have

IM A G E R Y A N D P E R C E P T IO N IN FA C E P R IM IN G

287

therefore co ncentrated on the im plications of these results for the structural th eory developed around the IAC architectu re. Finally, the presen t results are also relevan t for neuropsychological research co ncer ning impair m ents o n face p rocessing. Burto n, Young, Bruce, Johnston, an d E llis (1 991) used the IAC m odel to prop ose an explanatio n for covert recognitio n in prosopagnosia. Brie¯ y, this is a disorder in w hich a patient is unable to recogn iz e an y faces explicitly b ut can be show n to recogniz e them w hen tested imp licitly. E xam ples of im plicit tests include sem an tic prim ing an d interference effects (e.g. Young, H ellawell, & D e H aan, 1988). Burto n et al. (1991) proposed th at this set of phenom ena could be explained if the links between F RU s an d PIN s w ere attenuated. So, although the F RU becom es active, it can only pass som e of its activation on to the PIN. In particular, activation at the PIN m ay be insuf® cient to trig ger a recognition response b ut m ay nevertheless be suf® ciently large to in¯ uence other sim ultaneous processes. If the IAC account of th e data in the p resent pap er is accurate, th en the links in two directio ns (FRU ® PIN an d PIN ® FRU ) are dissociable an d could selectively be d am aged. T herefore, it should b e p ossible to ® nd patients w ho, desp ite b eing unable to recogn iz e faces, are no netheless able to im age them an d, co nversely, patients w ho are impaired in imagery but not in perception of faces. C onsistent w ith this prediction , Young, H um phreys, Riddoch, H ellaw ell, an d de H aan (1994 ) describe a p rosop agnosic patient w ho is severely impaired in face recognition but show s relatively p reserved imagery of faces. A lthough w e are unaw are of any patient w ho is impaired in face imagery but not in face recognition, if su ch a patient w ere fou nd, this w ould lend ad ditio nal support to the structur al m odel of repetitio n prim ing w e have describ ed here. In sum m ary, w e h ave argued that recog nitio n an d im agery of faces do share som e of the sam e representatio ns. In ter m s of the IAC m odel, w e have im plicated the sam e (F RU an d PIN ) represen tational units in both processes. H ow ever, w e have argued (here an d elsew here) that prim ing re¯ ects ch an ges to the connections b etw een representational units. T he experim ents described here suggest that these connectio ns m ay be separate an d dissociable in th e case of face recognition an d im agery.

R E FE R E N C E S Behrm an n, M ., W inocur, G., & M oscovitch, M . (1992) . D issociation between m ental imagery an d object recognition in a br ain-dam aged patient. N a ture, 359 , 636± 63 7. Bentin, S., & M o scovitch, M . (198 8). Th e tim e cou rse of repetitio n effects for words an d unfamiliar faces. J ourna l of Experimenta l P sychology : Genera l, 117 , 14 8± 160. Bruce, V., B urton, A.M ., Car so n, D., H an na, E., & M ason, O. (1994) . Repetitio n prim ing of face recognition. In C. U milta & M . M oscovi tc h (E ds.), Attention a nd performa nce X V (p p. 179 ± 210). Cambr idge, M A: M IT P ress. Bruce, V., & Valentine, T. (1985) . Identity priming in th e recognitio n of fam iliar faces. B ritish J ourna l of P sychology, 76, 373± 383 . Bruce, V., & You ng, A. (1986 ). Understanding face recog nition. B ritish J ourna l of P sychology, 77, 305± 327. Brunas, J., Young, A.W., & Ellis, A.W. (1990 ). Repetition prim ing for incom plete faces: Evidence for par t to whole com pletion. B ritish J ourna l of P sychology, 81, 43± 56 . Brunas-Wagstaff, J., Young, A.W., & Ellis, A.W. (1992) . Repetitio n prim ing follo ws spon taneous b ut not prom pted recognitio n of fam iliar faces. Qua rterly J ourna l of Experimenta l P sychology, 44A , 423± 454 .

288

CA BE ZA ET A L.

Burton, A.M . (1994) . L ear ning new faces in an interactiv e activatio n an d comp etitio n mod el. Visual Cognition, 1, 313± 348. Burton, A.M ., & B ruce, V. (19 93) . N am ing faces and nam ing nam es: Exploring an interactiv e activati on mod el of face recognition. M emory, 1, 457± 480. Burton, A.M ., Bruce, V., & Johnsto n, R.A. (199 0). Understanding face recognition with an inte ractive activation m od el. B ritish J ourna l of P sychology, 81, 86 1± 880. Burton, A.M ., Young, A.W., Bruce, V., Johnston, R.A., & Ellis, A.W. (1991) . U nderstanding covert recognition. Cognition, 39 , 129± 166. Cabeza, R. (1994) . A dissociatio n betwee n tw o implicit co nceptu al tests supports th e distinctio n between types of conceptu al processing. P sychonomic B ulletin & R eview, 1, 505± 508. Cabeza, R. (1995 ). Investig ating the mixture an d subd ivision of perceptu al and conceptual processing in Japan ese m emory tests. M emory & Cognition, 23, 153± 165. Craigie, M ., & Han ley, J.R . (1993) . Access to visual infor matio n from a nam e is co ntinge nt upo n identity-speci® c seman tic infor matio n. M emory, 1, 367± 391. Ellis, A.W., Burton, A.M ., Young, A.W., & Flude, B.M . (in press). Repetitio n prim ing betwee n par ts an d wholes: Tests of a com putatio nal mod el of fam iliar face recognition. B ritish J ourna l of P sychology. Ellis, A.W., Young, A.W., & Flude, B.M . (1990 ). Repetition prim ing an d face processing: P rim ing occurs within th e system th at respon ds to th e identity of face. Qua rterly J ourna l of Experimenta l P sychology, 42A , 495± 51 2. Ellis, A.W., Young, A.W., F lude, B.M ., & Hay, D.C. (1987) . Repetition prim ing of face recognitio n. Qua rterly J ourna l of Experimenta l P sychology, 39A , 193± 210. Farah , M .J., & L evine, D.N. (1988) . A case stu dy of mental imagery de® cit. B ra in a nd Cognition, 8, 147± 164. Farah , M .J., Weisberg, L .L ., M o nheit, M .A., & P eron net, F. (1988) . Brain activity underlying mental imagery : Eve nt-related potentials during mental ima ge generatio n. J ourna l of Cognitiv e Neuroscience, 1, 302± 316. Finke, R.A. (19 79) . Th e fu nctional equivalence of m ental im ages and errors of mov em ent. Cognitive P sychology, 11, 235± 264 . Finke, R.A. (1985) . T heories relating m ental ima gery to perception. P sychologica l B ulletin, 98 , 236± 23 9. Graf, P., & Ryan, L . (1990) . Transfer-appropriate processing for im plicit an d explicit m emory. J ourna l of Experimenta l P sychology : Lea rning, M emory, a nd Cognition, 16, 97 8± 992. Hay, D.C., & Young, A.W. (1982) . T he human face. In A.W. Ellis (E ds.), N orma lity a nd pa thology in cognitiv e functions. Lo ndon : Academ ic Press. Kosslyn, S.M ., Alpert, N.M ., T hom pson, W.L ., M aljkovic, V., Weise, S.B., Chabris, C.F., H am ilto n, S.E., Rau ch, S.L ., & Buon an no, F.S. (19 93) . Visual m ental im agery activate s topograp hically org an ized visual cortex : P ET investigato rs. J ourna l of Cognitiv e Neuroscience, 5, 263± 287 . M cD erm ott , K .B., & Roediger, H.L. (1994) . Effects of imagery o n perceptu al im plicit memory tests. J ourna l of Experimenta l P sychology: Lea rning, Memory, a nd Cognition, 20, 1379 ± 1 390 . M or ris, C.D., Bransford, J.D., & Franks, J.J. (1977) . L evels of processing versus transfer ap propriate processing. J ourna l of Verba l Lea rning a nd Verba l B eha viour, 16 , 51 9± 533. M oscovitch, M ., Vriezen, E., & Goshe-G ottstein, Y. (19 93) . Im plicit tests of mem or y in patients with foca l lesions an d degenerativ e br ain disorders. In H . S pindler & F. B oller (E ds.), H a ndbook of neurospsychology . Amsterdam : Elsevier. Riddoch, M .J. (19 90) . L oss of visual im agery: A generation de® cit. Cognitiv e N europsychology, 7, 249± 273. Roberts, T., & B ruce, V. (1989) . Repetitio n prim ing of face recognition in a serial choice reaction-time task. B ritish J ourna l of P sychology, 80, 20 1± 211. Roe diger, H .L ., Weldon , M .S., & Challis, B.H. (1989) . Exp lain ing dissociatio ns between im plicit an d explicit m easures of retention: A processing account. In H .L . Roediger & F.I.M . Craik (E ds.), Va rieties of memory a nd consciousness: Essa ys in honour of Endel Tulving (pp. 3± 41) . Hillsdale, NJ: L aw ren ce Erlbaum Associates, Inc. Weldon , M .S., & Roediger, H .L . (1987) . Altering retrieval deman ds reverses th e picture superiority effe ct. Memory & Cognition, 15, 26 9± 280.

IM A G E R Y A N D P E R C E P T IO N IN FA C E P R IM IN G

289

Young, A.W., H ellawell, D., & D e H aan, E.H.F. (1988) . Cross-dom ain sem an tic prim ing in nor m al sub jects an d a prosop agnos ic patient. Qua rterly J ourna l of E xperimenta l P sychology, 40A , 561± 580. Young, A.W., Hum phreys, G.W., Riddoch, M .J., Hellawell, D.J., & de Haan, E.H.F. (199 4). Recognition impair m ents and face imagery. N europsychologia , 32, 693 ± 702. Origina l ma nuscript received 1 Ma y 199 5 Accepted revision receiv ed 25 J uly 199 6