subliminal convergence of kanji and kana words: further ... · mean accuracy (± sd ) for the...

Subliminal Convergence of Kanji and Kana WordsFurther Evidence for Functional Parcellation of the

Posterior Temporal Cortex in Visual Word Perception

Kimihiro Nakamura12 Stanislas Dehaene3 Antoinette Jobert3Denis Le Bihan3 and Sid Kouider34

Abstract

amp Recent evidence has suggested that the human occipito-temporal region comprises several subregions each sensitiveto a distinct processing level of visual words To furtherexplore the functional architecture of visual word recognitionwe employed a subliminal priming method with functionalmagnetic resonance imaging (fMRI) during semantic judg-ments of words presented in two different Japanese scriptsKanji and Kana Each target word was preceded by asubliminal presentation of either the same or a differentword and in the same or a different script Behaviorally wordrepetition produced significant priming regardless of whetherthe words were presented in the same or different script At

the neural level this cross-script priming was associated withrepetition suppression in the left inferior temporal cortexanterior and dorsal to the visual word form area hypothesizedfor alphabetical writing systems suggesting that cross-scriptconvergence occurred at a semantic level fMRI also evidenceda shared visual occipito-temporal activation for words in thetwo scripts with slightly more mesial and right-predominantactivation for Kanji and with greater occipital activation forKana These results thus allow us to separate script-specificand script-independent regions in the posterior temporallobe while demonstrating that both can be activatedsubliminally amp

INTRODUCTION

Several lines of evidence have consistently suggestedthat the left inferior temporal cortex constitutes a cere-bral substrate for the perception of visual words (Tar-kiainen Helenius Hansen Cornelissen amp Salmelin1999 Puce Allison Asgari Gore amp McCarthy 1996Nobre Allison amp McCarthy 1994) Although this andadjacent cortical regions were initially thought to repre-sent a multimodal brain area active in a variety oflanguage operations such as phonological retrieval(Price 1998) imagery and writing (Nakamura et al2001) more recent work has suggested that the neuralprocesses for invariant visual word recognition are asso-ciated with the middle part of the left occipito-temporalsulcus a region which has been termed the visual wordform area (VWFA) This particular subregion may beattuned specifically to the visual representations ofwords or more precisely to abstract (shape-indepen-dent) representations of visual letter strings (CohenLehericy et al 2002 Dehaene Le Clec et al 2002)

Behavioral evidence for the existence of such abstractrepresentations has been provided by the demonstra-

tion of cross-case repetition priming (Forster amp Davis1984 Evett amp Humphreys 1981) Visual recognition of atarget word is known to be facilitated when it is preced-ed by the same word irrespective of whether thisprimendashtarget pair is written in the same or different case(eg radiondashradio vs radiondashRADIO) and regardless ofwhether the prime and the target are visually similar ordissimilar (eg kissndashKISS vs readndashREAD) This is thecase even when the prime word is masked such thatparticipants are unaware of its presence (subliminalpriming)

Using functional magnetic resonance imaging (fMRI)Dehaene Naccache et al (2001) have demonstratedthat at the neural level this subliminal repetition prim-ing effect is visualized as an attenuated response of theleft occipito-temporal sulcus area The VWFA showsreduced activation upon repetition of words irrespectiveof whether the prime and the target are presented insame- or different-case This and a subsequent study fur-ther suggested that the occipito-temporal cortex is orga-nized from posterior to anterior for storing increasinglymore abstract representations of visual words rangingfrom physical features in the extrastriate cortex to loca-tion- and case-invariant letters and words more anteriorlyin the fusiform region (Dehaene Jobert et al 2004)

To further explore the functional architecture of thisneural system for reading the present fMRI study

1EHESSCNRSENS Paris France 2University of Tokyo 3ServiceHospitalier Frederic Joliot CEADSV Orsay France 4INSERMU421 Faculte de Medecine Paris XII-IM3 Creteil France

D 2005 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 176 pp 954ndash968

employed the subliminal priming method for a languagewhich uses two radically different writing systems to writethe same word (Figure 1) That is we focused on therepetition priming effect that may occur between twowriting systems of Japanese Kanji (logographic) and Kana(syllabic) when participants are engaged in a semanticcategorization task Importantly when a given word iswritten in the two scripts there is no one-to-one corre-spondence at the sublexical level For instance the wordlsquolsquocigarettersquorsquo is printed as lsquolsquo rsquorsquo in Kanji and lsquolsquo rsquorsquo inKana respectively while both are pronounced as ta-ba-ko Hence the cross-script repetition priming if anyshould represent a perceptual phenomenon distinctfrom the case-independent priming observed in alpha-betical systems attributable only to phonological lexicalor semantic activations beyond the orthographic repre-sentations of words As such this particular form ofpriming might be mediated by a set of areas distinct fromthe VWFA

The advantages of studying within- and cross-scriptpriming in Japanese readers with fMRI are threefoldFirst the combined use of subliminal masked primingand fMRI enables us to examine the initial stages ofreading and their neural correlates while avoiding po-tential strategic and attentional changes elicited byawareness of the primendashtarget relation (Naccache ampDehaene 2001a) That is this method allows us to focuson bottom-up activations of the neural code associatedwith the word stimuli while presumably preventing thetop-down re-entry of information (Dehaene Jobertet al 2004 Lamme 2003 Lamme Zipser amp Spekreijse2002) It may therefore provide precise informationabout which type of local word code is used in a givenbrain area without contamination by the distant globalbroadcasting that may occur when words cross thethreshold of consciousness

Second this design allows us to further evaluate thedepth of processing involved during unconscious word

perception Although it is widely accepted that cross-case repetition priming is extremely robust undermasking conditions and thus provides clear evidencefor the existence of subliminal orthographic processingthere is still a controversy as to whether one can observea similar phenomenon for higher processing stages(phonological lexical and semantic) correspondingto whole-word processing (Kouider amp Dupoux 2004Forster Mohan amp Hector 2003) That is although someresearchers have suggested that subliminal word prim-ing is restricted to an analysis of subword fragments orletters instead of reflecting lexical processing (Abramsamp Greenwald 2000) other studies have found repeti-tion priming for words translated into two differentlanguages even when written with very different scripts(Jiang 1999 Gollan Forster amp Frost 1997) Howeverbecause prime awareness was not measured in theselatter studies it is still not entirely clear whether prim-ing goes beyond the orthographic level under genuinelysubliminal conditions Evidence of masked cross-scriptpriming between Kanji and Kana together with thedemonstration that prime stimuli remained below thethreshold of awareness would provide further unequiv-ocal evidence that unconscious perception can reflectprocessing at an advanced level Also the anatomicallocalization of brain areas associated with cross-scriptpriming may in turn help interpret the cognitive pro-cess underlying this behavioral phenomenon

Third the design of the present experiment withorthogonal manipulations of prime script and targetscript incidentally allows us to examine the overlapand the differences between the neural substrates ofvisual word recognition in Kanji and in Kana Thisissue has long been a major concern for neurolinguis-tic studies of Japanese Focal brain damage oftenaffects the reading of Kanji and Kana to a differentdegree which has lead to the idea that visual words inthe two formats must have partially distinct neuralrepresentations Specifically the inferior temporal cor-tex and a more dorsal part of the occipito-temporalregion in the left hemisphere have been thoughtimportant for reading of Kanji and Kana respectively(Iwata 1986) Also several lines of studies have sug-gested the contribution of the right hemisphere in thevisual recognition of logograms (Yamaguchi ToyodaXu Kobayashi amp Henik 2002 Kamada et al 1998Nakagawa 1994 Sugishita Yoshioka amp Kawamura1986 Hatta 1977 Sasanuma Itoh Mori amp Kobayashi1977) Nonetheless the issue seems not fully resolvedas for instance a more recent multiple case studyfound no consistent correlation between specific pat-terns of impairment and lesion sites (Sugishita OtomoKabe amp Yunoki 1992) In the brain imaging literaturemoreover although a few previous studies have exam-ined brain activity during reading of Japanese (SakuraiMomose Iwata Sudo et al 2000 Sakurai MomoseIwata Ishikawa et al 1996 Sakurai Momose Iwata

Figure 1 Sequence of events used for the behavioral tasks Each trialconsisted of a visible target word ( ) preceded by a masked prime

( ) The primendashtarget relation (same or different word) prime

script and target script were manipulated independently Participantswere requested to determine whether the latter represented a natural

or artificial object as quickly as possible

Nakamura et al 955

Watanabe et al 1993) none of them provided adirect statistical comparison of the effect of scripttype

EXPERIMENT 1

Results

Behavioral Results

Mean accuracy (plusmnSD) for the semantic categorizationtask was 9483 (plusmn375) The forced-choice test forprime visibility revealed that participants were unableto see the prime words [accuracy = 5229 correctt(15) = 060 p = 56] and that the accuracy level didnot differ significantly between the two script types[mean = 5286 and 5105 for Kanji and Kana respec-tively t(15) = 049 p = 63]

The behavioral priming effect for each script type issummarized in Table 1 Overall the repetition of samewords produced a highly significant effect of primingacross the four types of script alternations [F(115) =4483 p lt 001] By contrast neither the script type ofprimes nor that of targets affected the participantsrsquoperformance significantly [F(115) = 337 p = 09F(115) = 145 p = 25 respectively] There was anonsignificant trend of interaction between prime scriptand target script [F(115) = 408 p = 06] None of theother interactions were significant [F(115) = 306p = 10 for repetition and prime script F(115) =064 p = 44 for repetition and target script respec-tively] The triple interaction was not significant either[F(115) = 296 p = 11]

In further analysis the priming effect was examinedwith respect to the script type of primes and targets Thepairwise comparisons indeed revealed that the repeti-tion of words accelerated responses systematicallyacross four types of script-changes between primesand targets [F(115) = 2454 p lt 001 for prime Kanjiand target Kanji F(115) = 3733 p lt 001 for primeKanji and target Kana F(115) = 778 p lt 01 for prime

Kana and target Kana and F(115) = 1454 p lt 002 forprime Kana and target Kana respectively]

Imaging Results

Script-specific effects Regardless of their script typevisual words produced left-predominant activation ofthe peri-sylvian areas fronto-parietal junction and occi-pito-temporal area relative to the word-absent baselineThese activation sites were distributed quite similarly forboth types of script (Figure 2)

Activations that differ according to script type areillustrated in Figure 3 and summarized in Tables 2and 3 Within the set of brain regions activated by Kanjitargets only a small bilateral region in the fusiform gyrusresponded more greatly when target words were pre-sented in Kanji than in Kana particularly in the righthemisphere Conversely target words in Kana relative tothose in Kanji exhibited activation of the bilateral occip-ital pole and left inferior parietal area including thesupramarginal gyrus and inferior parietal lobule Incontrast no region in this neural circuit respondedmore greatly to subliminal primes in Kanji relative tothose in Kana while conversely the left inferior parietallobule exhibited greater response to the latter relative tothe former

In the set of brain areas activated by Kana targetsvisible targets and subliminal primes in Kanji eachproduced no significant activation relative to their equiv-alents in Kana Targets in Kana relative to those in Kanjiactivated the bilateral occipital pole the left lateraloccipital cortex and the left inferior parietal areawhereas similarly subliminal primes in Kana activatedthe left inferior parietal lobule and thalamus relative tothose in Kanji

Repetition priming effects Brain regions showingsignificant repetition suppression are summarized inTable 4 Repetition of words in Kanji produced twodistinct clusters in the left posterior temporal areaone in the anterior superiormiddle temporal gyri andthe other in the medial fusiform gryus (Figure 4A)Note that this latter cluster overlaps the region show-ing greater activation to Kanji relative to Kana Bycontrast only the anterior portion of the middletemporal gyrus ( y = 39) exhibited a significantrepetition suppression when words were repeated inthe prime Kanjindashtarget Kana condition (Figure 4B andTable 4) In fact prime words in Kanji produced anattenuated response at this location regardless ofwhether targets were written in Kana or in Kanji(x = 44 y = 35 z = 2 Z = 336 seeFigure 4C) This site showed a greater effect of prim-ing when primes were presented in Kanji than in Kana(Z = 383) In contrast to those positive effects ofKanji primes the priming effect induced by primes in

Table 1 Behavioral Priming Effects in Experiment 1

Reaction Time (msec)PrimeScript

TargetScript Repeated Unrepeated

Effect Size(msec)

Kanji Kanji 596 624 28

Kanji Kana 606 630 23

Kana Kanji 603 614 12

Kana Kana 598 620 22

p lt 05

p lt 01

p lt 001

956 Journal of Cognitive Neuroscience Volume 17 Number 6

Kana did not survive the current statistical criteriairrespective of the script type of targets However anonsignificant trend of repetition suppression wasdetected in a close vicinity of the above left temporalregion for the prime Kanandashtarget Kanji condition (x =44 y = 51 z = 4 Z = 242)

To clarify the role of this region in cross-script prim-ing we further examined the priming effect by comput-ing a linear combination of prime Kanjindashtarget Kana andprime Kanandash target Kanji conditions (inclusively maskedby both contrasts at p lt 05) This additional analysisrevealed that this anterior part of the left middle tem-poral gyrus exhibited a significant effect of repetitionsuppression when words were repeated in differentscript irrespective of the direction of script alternations(x = 48 y = 43 z = 2 Z =365 see Figure 4D)

On the other hand when the priming effect wascompared between within- and cross-script conditions(ie prime Kanjindashtarget Kanji and prime KanandashtargetKana vs prime Kanjindashtarget Kana and prime Kanandashtarget Kanji) the same part of the left middle temporalgyrus showed a greater trend of priming when wordswere presented in different script than in same script(Z = 282) However no brain region exhibited asignificant priming effect when words were repeated insame scripts

Additionally we examined the priming effect by com-puting the across-participant mean of percent signalchange within a 5-mm-radius spherical volume of inter-est (VOI) centered at the VWFA as identified withalphabetic letter-strings (Cohen Dehaene et al 2000)The VOI analysis revealed that this region was activeduring perception of both Kanji and Kana relative to the

baseline while showing greater activation for words inKana than those in Kanji [t(15) = 257 p = 02] Thepriming effect however was not significant at thislocation for any of the four types of script alternationsTo summarize the imaging results the anatomical loca-tions and priming effects for three posterior temporalregions of interest are illustrated in Figure 5

EXPERIMENT 2

A second behavioral experiment was designed toestablish whether the subliminal cross-script effectobserved in Experiment 1 reflects a repetition primingeffect or whether it results from response primingrelated to primendashtarget congruity (Damian 2001) Theissue arises because all the unrepeated trials in Exper-iment 1 comprised only primes and targets that be-longed to opposite categories (eg prime naturaltarget artifact) This feature of the design introduceda confound between primendashtarget repetition (repeatedor different words) and response congruity (congruentor incongruent responses to the prime and target)Thus the observed repetition priming effect might infact be imputable to a motor conflict Processing ofthe prime would lead to covert preparation of anappropriate motor response and such response biaswould need to be overcome by the overt motorresponse on targets Note that this response competi-tion might occur for two reasons either the motorresponse induced by the primes resulted from subli-minal semantic processing (Abrams Klinger amp Green-wald 2002 Naccache amp Dehaene 2001b) or more

Primes in Kanji

R L

Primes in Kana

R L

Targets in Kanji

R L

Targets in Kana

R L

Figure 2 Brain regions activated by primes and targets relative to the word-absent baseline Irrespective of their script type primes and targets

similarly produced left-predominant activation of the peri-sylvian areas fronto-parietal junction and occipito-temporal area

Nakamura et al 957

simply it resulted from stimulusndashresponse associationswhich could develop because the subliminal primeswere also practiced as targets during the experiment(Damian 2001) To control for this confound and

examine the behavioral effect of word repetition inde-pendently of response factors we performed a secondbehavioral experiment in which primes and targets al-ways belonged to the same category in both repeated

Figure 3 Statistical

parametric maps showing

differential activation to thescript type Within the set of

brain regions active during

reading of Kanji (A) visibletargets produced activation of

the bilateral fusiform region

more greatly for Kanji whereas

those written in Kana activatedthe bilateral occipital pole and

the left inferior parietal region

more greatly Subliminal

primes in Kanji produced nosignificant activation relative to

those in Kana whereas the

latter activated the left inferiorparietal lobule relative to the

former All these effects of

script type except the one in

the fusiform region for Kanjitargets were similarly

observed within the neural

network active during reading

of Kana (B)


and nonrepeated trials thereby eliminating any differ-ence in response congruity

Results

The level of accuracy was almost the same as in theprevious experiment (mean plusmn SD = 9467 plusmn 300)Response time for each condition is summarized inTable 5 The main effect of repetition although dimin-ished in size across the four conditions of prime and target script remained significant even when the con-

tribution of the response priming was discounted[F(121) = 3957 p lt 001] Participants respondedsignificantly faster when the targets were written in Kanji[F(121) = 5437 p lt 001] whereas the script type ofprimes did not affect the response time [F(121) = 210p = 16] The effect of repetition did not interact withthe script type of primes [F(121) = 125 p = 28] norwith that of targets [F(121) = 050 p = 49] None ofthe other interactions reached statistical significance

Furthermore pairwise comparisons revealed that thepriming effect was significant in the four conditionsof prime and target scripts [F(121) = 2392 p lt 001for prime Kanji and target Kanji F(121) = 633p = 02 for prime Kanji and target Kana F(121) = 511p = 04 for prime Kana and target Kanji and F(121) =551 p = 03 for prime Kanji and target Kana]

Joint Behavioral Analysis of Experiments 1 and 2

The effects of repetition and response congruity werecontrasted by pooling the two sets of behavioral dataobtained from Experiments 1 and 2 The combinedanalysis therefore consisted of a 2 2 2 2 factorialdesign where the effects of repetition (repeated and

Table 2 Effects of Script Type in Brain Regions Active duringReading of Kanji

Coordinate

Brain RegionNo ofVoxels

ZValue x y z

Prime Kanji gt Prime Kana

(No suprathreshold clusters)

Prime Kanji lt Prime Kana

Left inferior parietal lobule 9 416 51 44 48

Target Kanji gt Target Kana

Right fusiform gyrus 16 354 32 48 20

Left fusiform gyrus 7 322 29 41 13

Target Kanji lt Target Kana

Left occipital pole 18 394 16 89 2

Right occipital pole 19 390 16 89 1

Left supramarginal gyrus 8 303 48 33 35

Table 3 Effects of Script Type in Brain Regions Active duringReading of Kana

Coordinate


ZValue x y z





Left thalamus 6 278 28 30 13






Left lateral occipital area 9 329 40 78 4


Table 4 Brain Regions Showing Reduced Response to theRepetition of Words

Coordinate


ZScore x y z

Prime KanjindashTarget Kanji

Left superiormiddletemporal gyri

26 400 55 38 20

Left medial occipital lobe 15 382 20 58 10

Right medial frontal area 17 368 16 9 52

Right middle temporal gyrus 20 368 48 20 6


Right superior temporal gyrus 24 360 63 34 16

Left medial frontal area 8 338 16 1 59

Left anterior cingulate gyrus 36 314 24 14 40

Right lateral occipital area 23 306 12 73 7

Left middle temporal gyrus 8 291 51 61 14

Prime KanjindashTarget Kana


Prime KanandashTarget Kana


Prime KanandashTarget Kanji


Nakamura et al 959

Figure 4 Brain regions showing the repetition priming effect

The left lateral occipito-temporal cortex exhibited reduced activationwhen both primes and targets were written in Kanji (A) whereas a

mid-lateral part of the left middle temporal gyrus showed reduced

response when primes and targets were written in Kanji and Kana

respectively (B) This same region exhibited reduced activation forprimes in Kanji regardless of whether targets were written in Kana

or in Kanji (C) while no brain area showed the activation reduction

when primes were presented in Kana script Only the mid-lateralportion of the left temporal gyrus exhibited a significant repetition

priming effect when words were repeated in different scripts

irrespective of the direction of script change (D)

Figure 5 Magnitude of activation reduction (plusmnSEM ) in posterior

temporal regions of interest Top The left middle temporal gyrus

exhibited a cross-script priming effect irrespective of the direction ofscript change (yellow) This region also exhibited a significant priming

effect when primes were presented in Kanji Middle The left medial

fusiform gyrus showed greater activation to Kanji than to Kana (blue)

The priming effect was significant at this location only when wordswere repeated in the former script Bottom The left middle fusiform

gyrus or VWFA identified for European alphabetic scripts was active

during reading of Kanji and Kana relative to the baseline (green) This

region did not produce significant activation reduction for either typeof script although there is a nonsignificant trend of priming for the

prime Kanjindashtarget Kanji condition


unrepeated) prime script (Kanji and Kana) and targetscript (Kanji and Kana) were within-participant factorswhereas that of experiment (1 or 2) was a between-group factor The analysis of variance revealed that themain effects of repetition and target script were both

significant across the two experiments [F(136) = 7990p lt 001 and F(136) = 2392 p lt 001 respectively]whereas that of prime script was not [F(136) = 017p = 68] The main effect of experiment was notsignificant [F(136) = 088 p = 35] By contrast therewas a significant interaction between repetition andexperiment [F(136) = 1814 p lt 001] suggesting agreater priming effect in Experiment 1 that is a sub-stantial contribution from response priming The ef-fects of prime script and target script interacted witheach other [F(136) = 688 p = 01] There was asignificant interaction between repetition and primescript [F(136) = 434 p = 04] suggesting that theeffect of priming was larger when prime words werepresented in Kanji A significant interaction was alsofound between prime script and experiment [F(136) =593 p = 02] and between target script and exper-iment [F(136) = 435 p = 04] suggesting a greatereffect of experiment (ie response congruity forwords presented in Kanji script) The triple interaction




Effect Size(msec)




Kana Kana 586 593 7

p lt 05

p lt 001

Figure 6 A tentative model of

the pathways traversed byconscious and subliminal Kanji

and Kana words in our task

Thick continuous dashed and

dotted lines representconnections of strong weak

or very weak strength Top

Conscious reading of Kana

words engages principally aphonologically mediated

route and conscious reading

of Kanji words a lexical route

In both cases the response isemitted mostly from the

lexico-semantic level because

the task requires semanticclassification However

because the same words are

repeatedly used direct motor

associations can also developfrom the other levels of

representations Bottom The

lexico-semantic route and the

response associations remainavailable during subliminal

processing (Dehaene

Naccache Le Clec et al 1998)but the phonological route is

largely unavailable (Kouider

et al submitted) The overlap

between the representationsactivated by primes and targets

can then explain the size and

location of the observed

behavioral and fMRI primingeffects as a function of prime

and target script (see text

for details)

Nakamura et al 961

was significant only between repetition prime scripttarget script [F(136) = 496 p = 03] All the otherinteractions were nonsignificant

GENERAL DISCUSSION

Commonalities and Differences in Reading ofKanji and Kana

The present imaging data indicate that words printed inKanji and in Kana activate largely overlapping brainregions including the fronto-temporal junction inferiorparietal and occipito-temporal cortices and that theseactivations are more extensive in the left hemisphere forboth scripts We also found that the left middle fusiformgyrus a hypothesized neural substrate of orthographicrepresentation (Cohen Dehaene et al 2000) is activeduring reading for both types of script

Amid this common neural circuit for reading wefound small differences between scripts the most nota-ble of which was a greater activation of the bilateralmedial fusiform area (BA 37) for words in Kanji relativeto those in Kana To date this script-specific activationfor Kanji has never been demonstrated using a directstatistical comparison although a similar observationwas reported by a previous PET study (Sakurai MomoseIwata Sudo et al 2000) The activation focus in the leftfusiform gyrus (peak at y = 41) although smaller inspatial extent than in the right homologous area occu-pied the medial surface of the ventral temporal cortex Itis unlikely that the script-specific effect at this locationcan be attributed simply to differences in physicalfeatures between words in Kanji and those in Kanabecause the anterior division of the occipito-temporalcortex (50 lt y lt 40) has been shown to beinsensitive to the retinal size and position of the visualstimuli presented (Vuilleumier Henson Driver amp Do-lan 2002 Grill-Spector et al 1999) Rather we envisagetwo possible interpretations of this finding

First the partial specialization for Kanji reading maybe seen in the larger context of specialization fordifferent categories of visual stimuli such as faces hous-es words or objects within the inferotemporal visualstream Hasson et al (2002) have shown that thisspecialization relates in part to a retinotopic bias Objectspecialization would reflect the requirements placed onfoveal processing by each category of visual stimuli withobjects requiring high precision (words faces) fallingwithin lateral regions biased towards foveal processingand stimuli requiring more global retinal processing(places) falling within mesial regions biased towardsthe lateral field In this context it might be suggestedthat the Kanji script requires slightly more global pro-cessing than alphabetic or Kana script hence that itsidentification tends to be supported by neurons biasedtowards the more lateral sectors of the retina located ina more mesial portion of the fusiform gyrus

A second possible interpretation is that reading ofwords in Kanji induces greater activation of regionsassociated with higher-order abstract knowledge ofwords especially semantics Indeed a meta-analysis ofprevious neuroimaging data by Cohen Dehaene et al(2002) suggested that the anterior portion of the leftfusiform region ( y = 43 on average) is sensitive toincreased task demand for semantic processing of visualor auditory words The close linkage of Kanji andsemantics in the left basal temporal cortex is supportedby neuropsychological studies of alexia in Japanesereaders Those studies confirm that damage to the leftinferior temporal area can affect the reading of Kanjiwhile leaving that of Kana almost intact and furtherdemonstrate this particular form of alexia often appearsin conjunction with anomia (Sakurai Sakai Sakuta ampIwata 1994 Yokota Ishiai Furukawa amp Tsukagoshi1990 Soma Sugishita Kitamura Maruyama amp Imanaga1989 Kawahata Nagata amp Shishido 1988) This mayreflect the fact that naming of objects and reading ofKanji both place a particular emphasis on semanticaccess prior to phonological retrieval

Regardless of whether the visual or semantic interpre-tation is correct the more extensive activation of theright fusiform area by Kanji than by Kana may representa neuroanatomical substrate of the long-standing hy-pothesis of a right hemisphere advantage in processingof Kanji Their logographic nature was long thought toallow readers to extract meaning directly from theirwritten forms without the mediation of phonologicalrecoding The rightward asymmetry for the recognitionof Kanji has been suggested by several lines of studiesincluding behavioral (Nakagawa 1994 Hatta 1977 Sa-sanuma et al 1977) neuropsychological (Sugishitaamp Yoshioka 1987 Sugishita Yoshioka et al 1986)electrophysiological (Yamaguchi et al 2002 HayashiKayamoto Tanaka amp Yamada 1998) and magneto-encephalographic (Kamada et al 1998) data Althoughthese studies have located the right-predominant activityin different brain regions such as fronto-central (Yama-guchi et al 2002) parietal (Hayashi et al 1998) andoccipito-temporal cortices (Kamada et al 1998) thepresent fMRI results suggest that the hemispheric spe-cialization differs between the two scripts early in theventral visual pathway

In the converse direction there were a few regionswhere Kana yielded greater activation than Kanji Firstreading of Kana relative to Kanji produced activation inbilateral retinotopic areas around the occipital pole ( y ~90 mm) and in the left anterior lateral occipital area( y = 74 mm) Both differences are likely to arise fromthe slight asymmetry in retinal size between the twoscripts Second visual words in Kana not only targetsbut also primes activated the left inferior parietal lobemore than those in Kanji A body of neuropsychologicaland neuroimaging data have associated this regionespecially the left supramarginal gyrus with the transla-


tion from orthography to phonology (eg Price 1998Price Moore Humphreys amp Wise 1997) The left pa-rietal activation may also relate to the deployment ofvisuospatial attentional processes that are needed toserially assemble phonology from the spatially extendedseries of characters that together represent a wordUnder both of these interpretations the greater re-sponse to Kana at this location would reflect the greateruse of phonological decoding that has been postulatedfor Kana over Kanji (Nomura 1981) If this interpreta-tion is correct interestingly our findings suggest thatthe neural processes for converting orthography to pho-nology may proceed to a certain extent even in the ab-sence of awareness (see below for further discussion)

Cross-Script Priming in the Left LateralTemporal Cortex

Behavioral results from Experiment 1 suggested that therepetition of subliminal words facilitates participantsrsquoresponse not only when they are written in same scriptbut also when written in different script This primingeffect although reduced remained significant when thecontribution from response priming was discounted inExperiment 2 Many previous experiments have demon-strated subliminal orthographic priming across lower-case and uppercase presentations of the same words inalphabetic scripts (Bowers et al 1998 Humphreys et al1990 Forster amp Davis 1984 Evett amp Humphreys 1981)The present results however suggest that subliminalprocessing goes beyond the orthographic level as nographic code is shared between words written in Kanjiand in Kana The cross-script priming effect implies theactivation of abstract script-independent representa-tions over and above the orthographic knowledge ofwords namely at the phonological lexical or semanticlevels These results replicate those previously obtainedwith cross-language repetition priming for translationequivalents ( Jiang 1999 Gollan et al 1997) and showthat they can be obtained under conditions of totalnonconsciousness of the masked primes FurthermorefMRI reveals the cerebral correlates of this abstractpriming effect We found that cross-script repetitionpriming correlates with an attenuated response in amid-lateral part of the left middle temporal cortex(x = 48 y = 43 z = 2)

Previous neuropsychological and brain imaging stud-ies indicate that the VWFA the hypothetical site of aninvariant orthographic code is located at x = 43y = 54 z = 12 in the left fusiform gyrus with astandard deviation of ~5 mm in each coordinate axis(Cohen Dehaene et al 2000) The present cross-scriptpriming site however is more anterior and superiorapproximately 18 mm further along in the posteriortemporal cortex Taken together these results suggestan anterior-to-posterior progression in word processing(Dehaene Jobert et al 2004) This may correspond to a

progressive abstraction process as also proposed forobject recognition whereby raw visual features of stim-uli are transformed progressively from perceptualto conceptual (Lerner Hendler Ben-Bashat Harel ampMalach 2001 van Turennout Ellmore amp Martin 2000Henson Shallice amp Dolan 2000) In fact this part of theleft middle temporal gyrus has been associated with thesemantic network by both functional brain imaging(Chao Haxby amp Martin 1999 Buchel Price amp Friston1998 Vandenberghe Price Wise Josephs amp Fracko-wiak 1996) and neuropsychological studies (ChertkowBub Deaudon amp Whitehead 1997) More recent workfurther suggested that the same region is invariablyactive during semantic judgment of objects irrespectiveof their categories (Tyler et al 2003) Using fMRIThompson-Schill DrsquoEsposito and Kan (1999) also ob-served activation reduction of the left inferolateral tem-poral cortex associated with the repeated retrieval ofsemantic knowledge Moreover a recent fMRI study byDevlin Jamison Matthews and Gonnerman (2004) hasreported that this region shows reduced activation tosemantically related word-pairs relative to unrelatedpairs Collectively these data suggest that the left pos-terior temporal cortex is fractionated into several dis-tinct subcomponents and that an abstract plausiblysupramodal semantic representation in the left temporallobe can be contacted by subliminal primes

In addition some caution may be needed in interpret-ing the present imaging results as the experimentaldesign included the effect of motor congruity as apotential confound which can be inflated by lsquolsquoresponseassociationrsquorsquo learned through repeated exposure to thesame items In fact a recent study by Dobbins SchnyerVerfaellie and Schacter (2004) have suggested that suchresponse learning lead to the activation reduction ofthe posterior temporal cortex It is our position how-ever that the observed repetition suppression in the leftlateral temporal cortex should have only a negligiblecontribution from the response congruity per se be-cause (1) response priming and competition have beenassociated with fronto-parietal regions (Hazeltine Pol-drack amp Gabrieli 2000 Botvinick Nystrom FissellCarter amp Cohen 1999 Carter et al 1998 DehaeneNaccache Le Clec et al 1998) (2) the lateralizedpriming effect was detected in the present study irre-spective of the actual side of manual response and (3)two recent fMRI studies have shown that repetitionpriming for visual words in the occipito-temporal cortexunder the condition where no response congruity waspresent between repeated and nonrepeated trials (De-haene Jobert et al 2004 Devlin et al 2004)

Asymmetries in Priming

Both behavioral and fMRI analyses suggested that scripttype affected priming differently across conditions no-tably (1) subliminal primes produced greater behavioral

Nakamura et al 963

priming when written in Kanji than in Kana (2) themagnitude of behavioral priming decreased in cross-script conditions especially when the prime was in Kanaand the target in Kanji (3) participants responded morequickly to target words in Kanji irrespective of the scripttype of primes (4) the effect of response congruity waslarger when words were presented in Kanji both forprimes and targets and (5) fMRI priming in the lefttemporal cortex was mostly due to Kanji primes

The faster response to targets in Kanji may be relatedto a potential difference in visual familiarity between thetwo formats That is despite the initial assumptionderived from the previous behavioral data (Amano ampKondo 2000) it is possible that the current stimuli weresubstantially lower or at least more varied in familiaritywhen presented in Kana as suggested by their slightlylower plausibility ratings Note that although this poten-tial familiarity bias can explain the faster processing ofKanji targets it can hardly account for the larger priminginduced by Kanji primes Masked repetition priming isusually insensitive to word frequency (Forster amp Davis1984 Evett amp Humphreys 1981) Even if it was sensitiveto frequencyfamiliarity as it is the case for unmaskedrepetition priming the reverse asymmetry should bepredicted because repetition priming is larger for low-than for high-frequency words (Bowers 2000 Forster ampDavis 1984)

More plausibly the asymmetries between Kana andKanji can be interpreted as reflecting differences indepth of phonological and semantic activation inducedby the two scripts Kanji words may predominantlyactivate a visualndashlexical route whereas lexical processingof Kana words may require the mediation of phonolog-ical codes to a greater degree especially when they arelower in visual familiarity (Besner amp Hildebrandt 1987Hirose 1984 Feldman amp Turvey 1980) The smallerpriming induced by Kana words might then be imput-able to the lesser automaticity of the phonological routeThat is although activation of phonological representa-tion is thought to be mandatory during visual wordrecognition (Brysbaert 2001 Lukatela Carello SavicUrosevic amp Turvey 1998 Van Orden Johnston amp Hale1988) recent behavioral studies using a primed lexicaldecision task and controlling for prime awareness sug-gest that visual words produce a phonological effectonly when participants consciously perceive the primestimuli (Kouider Peereman amp Dupoux submitted)Those findings suggest that phonological decodingmay not occur automatically for subliminal words atleast when the task emphasizes lexico-semantic process-ing rather than graphemendashphoneme conversion pro-cesses Indeed cross-task comparisons have shownthat masked phonological priming is absent in wordrecognition tasks such as lexical decision but that itoccurs when participants are engaged in explicit pho-nological production such as in the naming task (Shen ampForster 1999)

Assuming that lexical access requires phonologicalmediation to a greater extent for Kana than Kanji wordsand that subliminal presentation considerably reducesthe availability of phonological codes the observedasymmetries in priming may be explained as follows(Figure 6) First when primes and targets are writtenin Kanji they both activate lexico-semantic codesthrough an orthography-to-lexicon route This resultsin a strong priming effect reflecting shared primendashtargetactivation at orthographic lexical and semantic levelsSecond when the prime is in Kanji and the target inKana both stimuli activate the lexico-semantic systemSubliminal primes in Kanji activate it through the or-thography-to-lexicon route as in the previous conditionBy contrast visible targets in Kana activate it mainly byway of the phonological route Because shared lexico-semantic codes are ultimately activated a large cross-script effect is observed Third when the prime is inKana and the target in Kanji there is also some sharedlexico-semantic activation resulting in a cross-script ef-fect However this effect is weaker because of theweakness of the lexical route for Kana and because thisroute cannot be compensated by the alternative phono-logical route when observers are unaware of the Kanastimuli Fourth by the same token the presentation ofprimes and targets in Kana results in orthographicactivation as well as a small amount of lexico-semanticactivation thus resulting in a shallower but still signifi-cant form of repetition priming

This framework allows us to interpret the imagingresult that the repetition suppression in the left tempo-ral region putatively associated with lexico-semanticprocessing was significant only when primes were pre-sented in Kanji Furthermore assuming that all levels ofrepresentations transmit some activation to the re-sponse decision stage (Jaskowski Skalska amp Verleger2003 Jaskowski van der Lubbe Schlotterbeck amp Ver-leger 2002 Schmidt 2002 Eimer amp Schlaghecken1998) the model can explain why strong responsepriming is observed in all conditions of primes eventhough Kana primes are assumed to achieve a shallowerlevel of word processing than Kanji primes

Conclusions

Subliminal repetition priming when combined withfMRI provides a powerful method to dissect the levelsof representation involved in word processing and theircerebral substrates (Dehaene Jobert 2004 Naccache ampDehaene 2001a) Here we replicated with Japanesesubjects the repetition priming effects previously ob-served in readers of an alphabetic script (French) Thisallowed us to demonstrate an additional abstract levelof subliminal priming occurring at the convergence ofprocessing streams for Kanji and Kana scripts and as-sociated with the left temporal lobe The present imag-ing data suggest that distinct cortical subsystems within


the ventral stream contribute to visual word processingaccording to the decoding procedures required by theorthographic systems used This serial posterior-to-anterior axis of the ventral visual system appears to bestructured similarly across readers of different orthog-raphies and is only partially modulated by the specificrequirements of each script The most notable effect isthat different scripts put a differential emphasis onphonological versus lexico-semantic routes as previous-ly suggested by comparisons of brain activations inreaders of Italian or English (Paulesu et al 2000) Takentogether these results begin to dissect the universalorganization of posterior reading pathways and theiradaptation to specific cultural requirements

METHODS

Experiment 1

Participants

Sixteen right-handed native Japanese speakers (agerange 25ndash38 years) volunteered to participate in thepresent fMRI experiment None had a previous historyof neurological or psychiatric disease All of them gavewritten informed consent prior to the imaging experi-ment The protocol of this study was approved by theregional ethical committee

Materials and Procedures

The visual stimuli consisted of 40 Japanese nouns ofhigher familiarity (mean plusmn SD = 604 plusmn 032 accordingto the NTT database Amano amp Kondo 2000) None ofthe words had homophones Half of them representednatural objects (eg cat peach) and the other halfartifacts (eg telephone cigarette) These words werecomposed of one to two characters when written inKanji (mean orthographic plausibility plusmn SD = 483 plusmn017) and two to three characters when written in Kana(mean orthographic plausibility plusmn SD = 376 plusmn 031)[Here the orthographic plausibility is a 5-point scaleprovided by the same database indicating the likelihoodthat a specific script is used for writing a given word Forexample words normally written in Kanji are usuallyhigh in plausibility for Kanji but much lower in that forKana and vice versa Importantly behavioral data byAmano and Kondo (2000) have shown that the responsetime required for lexical decision does not differ be-tween words written in Kanji (mean plausibility plusmn SD =484 plusmn 017) and their transcriptions in Kana (meanplausibility plusmn SD = 328 plusmn 028) suggesting that normaladults are equally familiarized with the two forms ofwords when their plausibility level is sufficiently high foreach script]

Each trial consisted of a precisely timed sequence of amasked prime and a visible target (Figure 1) The maskswere created by semirandom arrangement of circle and

square shapes with the same line thickness as characterfonts covering up the central area of the screen Partic-ipants were not informed about the presence of primewords They were requested to decide as quickly and asaccurately as possible whether target words denotednatural objects or artifacts In lsquolsquounrepeatedrsquorsquo trials (iewhen the prime and target word differed) they alwaysbelonged to different categories (one natural and theother artificial) and shared no character in common atthe same location The word length was always the samebetween the prime and the target in lsquolsquosame scriptrsquorsquoconditions (on average 328 for Kana and 28 for Kanjirespectively) whereas that of Kana forms was longer byone character (~088) in lsquolsquodifferent scriptrsquorsquo conditionsThe experiment was therefore arranged in a 2 2 2factorial design where the main effects of interest wereprimendashtarget repetition (same or different word) primescript (Kanji or Kana) and target script (Kanji or Kana)In addition a ninth type of trials comprising the samesequence of masks without prime and target words wasused as a baseline to measure the event-related activa-tion with fMRI Participants performed four experimen-tal sessions each comprising five initial training trialsfollowed by 200 trials (20 trials for each event typeexcept for the lsquolsquoword-absentrsquorsquo baseline with 40 trialsall in random order)

Immediately after the imaging sessions participantsperformed a forced-choice test inside the scanner (64trials) designed to evaluate the visibility of primes Theywere told about the presence of hidden primes at thisstage Each trial comprised the same sequence of masksand words as in the activation task followed by a pair ofchoices (a same word as the prime and a distractor)presented left and right of the fixation Participants wereasked to determine which of the two items corre-sponded to the prime word within the preceding eventsequence

fMRI Procedures

The experiment was conducted using a 3-T whole-bodysystem (Bruker Germany) using a standard head coiloptimized for a gradient-echondashecho-planar imagingsequence (26 contiguous axial slices thickness 45 mmwith 05 mm gap TR = 2400 msec TE = 40 msec flipangle = 908 field-of-view =192 256 mm2 64 64pixels) High-resolution anatomical images were ob-tained prior to the main experiment Four scanningsessions each lasting 8 min 12 sec and giving 205volumes were performed

Data Analysis

After image reconstruction the functional images wereprocessed using the SPM99 software (Wellcome Depart-ment of Cognitive Neurology London UK) Five initial

Nakamura et al 965

images were discarded to eliminate nonequilibriumeffects of magnetization Images were corrected for headmotion resampled every 4 mm using sinc interpolationnormalized to the standard brain space (Friston Ash-burner et al 1995) and spatially smoothed with anisotropic Gaussian filter (7 mm full width at half maxi-mum) These images were then high-pass filtered at120 sec and smoothed with a 4-sec Gaussian kernelGroup-based statistical inference was made using a ran-dom effect model (Friston Holmes amp Worsley 1999)For each participant a weighted-mean image for eachcontrast was computed by fitting each voxel time-serieswith the known time-series of the nine event typesconvolved with a canonical hemodynamic responsefunction with time and dispersion derivatives The ef-fect of repetition priming or repetition suppression wascalculated as activation reduction in repeated trialsrelative to nonrepeated trials (Henson amp Rugg 2003)Brain activation specific to the script type was examinedseparately for primes and targets To discount voxelsinactive during reading we first determined brain re-gions activated by target words relative to the lsquolsquoword-absentrsquorsquo baseline for Kanji and Kana respectively(thresholded at voxelwise p lt 005) The resultingstatistical parametric maps served as a masking imagefor comparing lsquolsquotarget Kanjirsquorsquo with lsquolsquotarget Kanarsquorsquo andlsquolsquoprime Kanjirsquorsquo with lsquolsquoprime Kanarsquorsquo respectively The twoscript-specific masks were used separately because it isunknown whether and to what extent Kanji and Kanaactivate overlapping brain regions Unless stated other-wise all the effects of interest were tested voxelwiseat p lt 005 Only regions with contiguous clusters ofmore than five voxels corresponding to an activatedvolume of 320 mm3 were interpreted

Experiment 2

Participants

Twenty-two right-handed Japanese speakers (age range20ndash34 years) were recruited None of them participatedin Experiment 1

Materials and Procedure

Using the same set of words as Experiment 1 40 primendashtarget pairs were created for nonrepeated trials so thatboth of them belonged to the same category eithernatural or artificial Hence primes were semanticallycongruent with targets irrespective of whether wordswere repeated or not The same stimulus materials andevent sequence as Experiment 1 were used The ex-perimental design was a 2 2 2 factorial arrange-ment in which repetition prime script and targetscript were within-participant factors Participants wereseated approximately 50 cm from a computer monitor ina dimmed room and performed a total set of 640 trials

without the word-absent baseline trials giving 80 trialsfor each condition

Acknowledgments

We thank Laurent Cohen and two anonymous reviewers fortheir helpful comments This work was supported by INSERMCEA a centennial fellowship of the McDonnell foundation toSD a postdoctoral fellowship from Fondation de la RechercheMedicale to KN and a Grant-in-Aid for Young Scientists (B)15700252 by the Japan Ministry of Education Culture SportsScience and Technology to KN

Reprint requests should be sent to Kimihiro Nakamura MDDepartment of Speech and Cognitive Neuroscience GraduateSchool of Medicine University of Tokyo 3-7-1 Hongo Tokyo113-0033 Japan or via e-mail kimihiromu-tokyoacjp

The data reported in this experiment have been depositedin the fMRI Data Center (wwwfmridcorg) The accessionnumber is 2-2004-1178P

REFERENCES

Abrams R L amp Greenwald A G (2000) Parts outweigh thewhole (word) in unconscious analysis of meaningPsychological Science 11 118ndash124

Abrams R L Klinger M R amp Greenwald A G (2002)Subliminal words activate semantic categories (notautomated motor responses) Psychonomic Bulletin andReview 9 100ndash106

Amano S amp Kondo T (2000) Nihongo-no goitokusei [lexicalproperties of Japanese] (Vol 2) Tokyo Sanseido

Besner D amp Hildebrandt N (1987) Orthographic andphonological codes in the oral reading of Japanese KanaJournal of Experimental Psychology Learning Memoryand Cognition 13 335ndash343

Botvinick M Nystrom L E Fissell K Carter C S amp CohenJ D (1999) Conflict monitoring versus selection-for-actionin anterior cingulate cortex Nature 402 179ndash181

Bowers J S (2000) The modality specific and non-specificcomponents of long-term priming are frequency sensitiveMemory and Cognition 28 406ndash414

Bowers J S Vigliocco G amp Haan R (1998) Orthographicphonological and articulatory contributions to maskedletter and word priming Journal of ExperimentalPsychology Human Perception and Performance 241705ndash1719

Brysbaert M (2001) Prelexical phonological coding of visualwords in Dutch Automatic after all Memory and Cognition29 765ndash773

Buchel C Price C amp Friston K (1998) A multimodallanguage region in the ventral visual pathway Nature 394274ndash277

Carter C S Braver T S Barch D M Botvinick M MNoll D amp Cohen J D (1998) Anterior cingulate cortexerror detection and the online monitoring of performanceScience 280 747ndash749

Chao L L Haxby J V amp Martin A (1999)Attribute-based neural substrates in temporal cortexfor perceiving and knowing about objects NatureNeuroscience 2 913ndash919

Chertkow H Bub D Deaudon C amp Whitehead V (1997)On the status of object concepts in aphasia Brain andLanguage 58 203ndash232

Cohen L Dehaene S Naccache L Lehericy SDehaene-Lambertz G Henaff M A amp Michel F (2000)


The visual word form area Spatial and temporalcharacterization of an initial stage of reading in normalsubjects and posterior split-brain patients Brain 123291ndash307

Cohen L Lehericy S Chochon F Lemer C Rivaud S ampDehaene S (2002) Language-specific tuning of visualcortex Functional properties of the Visual Word Form AreaBrain 125 1054ndash1069

Damian M F (2001) Congruity effects evoked by subliminallypresented primes Automaticity rather than semanticprocessing Journal of Experimental Psychology HumanPerception and Performance 27 154ndash165

Dehaene S Jobert A Naccache L Ciuciu P Poline J BLe Bihan D amp Cohen L (2004) Letter binding andinvariant recognition of masked words Behavioral andneuroimaging evidence Psychological Science 15307ndash313

Dehaene S Le Clec H G Poline J B Le Bihan D ampCohen L (2002) The visual word form area A prelexicalrepresentation of visual words in the fusiform gyrusNeuroReport 13 321ndash325

Dehaene S Naccache L Cohen L Bihan D L Mangin J FPoline J B amp Riviere D (2001) Cerebral mechanisms ofword masking and unconscious repetition priming NatureNeuroscience 4 752ndash758

Dehaene S Naccache L Le Clec H G Koechlin EMueller M Dehaene-Lambertz G van de Moortele P Famp Le Bihan D (1998) Imaging unconscious semanticpriming Nature 395 597ndash600

Devlin J T Jamison H L Matthews P M ampGonnerman L M (2004) Morphology and the internalstructure of words Proceedings of the National Academy ofSciences USA 101 14984ndash14988

Dobbins I G Schnyer D M Verfaellie M amp Schacter D L(2004) Cortical activity reductions during repetition primingcan result from rapid response learning Nature 428316ndash319

Eimer M amp Schlaghecken F (1998) Effects of masked stimulion motor activation Behavioral and electrophysiologicalevidence Journal of Experimental Psychology HumanPerception and Performance 24 1737ndash1747

Evett L J amp Humphreys G W (1981) The use of abstractgraphemic information in lexical access Quarterly Journalof Experimental Psychology 33A 325ndash350

Feldman L B amp Turvey M T (1980) Words written in Kanaare named faster than the same words written in KanjiLanguage and Speech 23 141ndash147

Forster K I amp Davis C (1984) Repetition priming andfrequency attenuation in lexical access Journal ofExperimental Psychology Learning Memory andCognition 10 680ndash698

Forster K I Mohan K amp Hector J (2003) The mechanics ofmasked priming In K Kinoshita amp S J Lupker (Eds)Masked priming State of the art (pp 3ndash37) Hove UKPsychology Press

Friston K J Ashburner J Poline J B Frith C DHeather J D amp Frackowiak R S J (1995) Spatialregistration and normalization of images Human BrainMapping 2 165ndash189

Friston K J Holmes A P amp Worsley K J (1999) How manysubjects constitute a study Neuroimage 10 1ndash5

Gollan T H Forster K I amp Frost R (1997) Translationpriming with different scripts Masked priming with cognatesand noncognates in HebrewndashEnglish bilinguals Journal ofExperimental Psychology Learning Memory andCognition 23 1122ndash1139

Grill-Spector K Kushnir T Edelman S Avidan GItzchak Y amp Malach R (1999) Differential processing

of objects under various viewing conditions in the humanlateral occipital complex Neuron 24 187ndash203

Hasson U Levy I Behrmann M Hendler T amp Malach R(2002) Eccentricity bias as an organizing principle forhuman high-order object areas Neuron 34 479ndash490

Hatta T (1977) Recognition of Japanese kanji in the left andright visual fields Neuropsychologia 15 685ndash688

Hayashi M Kayamoto Y Tanaka H amp Yamada J (1998)Semantic activation by Japanese kanji Evidence fromevent-related potentials Perceptual and Motor Skills 86375ndash382

Hazeltine E Poldrack R amp Gabrieli J D (2000) Neuralactivation during response competition Journal ofCognitive Neuroscience 12 118ndash129

Henson R N amp Rugg M D (2003) Neural responsesuppression haemodynamic repetition effects andbehavioural priming Neuropsychologia 41 263ndash270

Henson R N Shallice T amp Dolan R (2000) Neuroimagingevidence for dissociable forms of repetition primingScience 287 1269ndash1272

Hirose T (1984) The effect of script frequency on semanticprocessing of Kanji and Kana words Shinrigaku Kenkyu 55173ndash176

Humphreys G W Evett L J amp Quinlan P T (1990)Orthographic processing in visual word identificationCognitive Psychology 22 517ndash560

Iwata M (1986) Neural mechanism of reading and writingin the Japanese language Functional Neurology 143ndash52

Jaskowski P Skalska B amp Verleger R (2003) How the selfcontrols its lsquolsquoautomatic pilotrsquorsquo when processing subliminalinformation Journal of Cognitive Neuroscience 15911ndash920

Jaskowski P van der Lubbe R H Schlotterbeck E ampVerleger R (2002) Traces left on visual selective attentionby stimuli that are not consciously identified PsychologicalScience 13 48ndash54

Jiang N (1999) Testing explanations for asymmetry incross-language priming Bilingualism Language andCognition 2 59ndash75

Kamada K Kober H Saguer M Moller MKaltenhauser M amp Vieth J (1998) Responses to silentKanji reading of the native Japanese and German in tasksubtraction magnetoencephalography Brain ResearchCognitive Brain Research 7 89ndash98

Kawahata N Nagata K amp Shishido F (1988) Alexiawith agraphia due to the left posterior inferior temporallobe lesionmdashneuropsychological analysis and itspathogenetic mechanisms Brain and Language 33296ndash310

Kouider S amp Dupoux E (2004) Partial awareness creates thelsquolsquoillusionrsquorsquo of subliminal semantic priming PsychologicalScience 15 75ndash81

Kouider S Peereman R amp Dupoux E (submitted)Phonological activation during reading Time-course orconscious awareness

Lamme V A (2003) Why visual attention and awareness aredifferent Trends in Cognitive Sciences 7 12ndash18

Lamme V A Zipser K amp Spekreijse H (2002) Maskinginterrupts figurendashground signals in V1 Journal of CognitiveNeuroscience 14 1044ndash1053

Lerner Y Hendler T Ben-Bashat D Harel M ampMalach R (2001) A hierarchical axis of object processingstages in the human visual cortex Cerebral Cortex 11287ndash297

Lukatela G Carello C Savic M Urosevic Z amp Turvey M T(1998) When nonwords activate semantics better thanwords Cognition 68 B31ndashB40

Nakamura et al 967

Naccache L amp Dehaene S (2001a) The priming methodImaging unconscious repetition priming reveals an abstractrepresentation of number in the parietal lobes CerebralCortex 11 966ndash974

Naccache L amp Dehaene S (2001b) Unconscious semanticpriming extends to novel unseen stimuli Cognition 80215ndash229

Nakagawa A (1994) Visual and semantic processing in readingKanji Journal of Experimental Psychology HumanPerception and Performance 20 864ndash875

Nakamura K Honda M Hirano S Oga T Sawamoto NHanakawa T Inoue H Ito J Matsuda T Fukuyama Hamp Shibasaki H (2001) Modulation of the visual wordretrieval system in writing A functional MRI study on theJapanese orthographies Journal of Cognitive Neuroscience14 104ndash115

Nobre A C Allison T amp McCarthy G (1994) Wordrecognition in the human inferior temporal lobe Nature372 260ndash263

Nomura Y (1981) The information processing of Kanji Kanascript The effects of data-driven and conceptually-drivenprocessing on reading Japanese Journal of Psychology 51327ndash334

Paulesu E McCrory E Fazio F Menoncello LBrunswick N Cappa S F Cotelli M Cossu G Corte FLorusso M Pesenti S Gallagher A Perani D Price CFrith C D amp Frith U (2000) A cultural effect on brainfunction Nature Neuroscience 3 91ndash96

Price C J (1998) The functional anatomy of wordcomprehension and production Trends in CognitiveSciences 2 281ndash288

Price C J Moore C J Humphreys G W amp Wise R J S(1997) Segregating semantic from phonological processesduring reading Journal of Cognitive Neuroscience 9727ndash733

Puce A Allison T Asgari M Gore J C amp McCarthy G(1996) Differential sensitivity of human visual cortex tofaces letterstrings and textures A functional magneticresonance imaging study Journal of Neuroscience 165205ndash5215

Sakurai Y Momose T Iwata M Ishikawa T Sato T ampKanazawa I (1996) Regional cerebral blood flow in thecovert reading of kana words A comparison with thestudy of reading aloud tasks European Neurology 36237ndash239

Sakurai Y Momose T Iwata M Sudo Y Ohtomo K ampKanazawa I (2000) Different cortical activity in readingof Kanji words Kana words and Kana nonwords BrainResearch Cognitive Brain Research 9 111ndash115

Sakurai Y Momose T Iwata M Watanabe T Ishikawa Tamp Kanazawa I (1993) Semantic process in kana wordreading Activation studies with positron emissiontomography NeuroReport 4 327ndash330

Sakurai Y Sakai K Sakuta M amp Iwata M (1994)Naming difficulties in alexia with agraphia for kanjiafter a left posterior inferior temporal lesion Journalof Neurology Neurosurgery and Psychiatry 57609ndash613

Sasanuma S Itoh M Mori K amp Kobayashi Y (1977)Tachistoscopic recognition of kana and kanji wordsNeuropsychologia 15 547ndash553

Schmidt T (2002) The finger in flight real-time motor controlby visually masked color stimuli Psychological Science 13112ndash118

Shen D amp Forster K I (1999) Masked phonological primingin reading Chinese words depends on the task ProcessingEast Asian languages [Special issue] Language andCognitive Processes 14 429ndash459

Soma Y Sugishita M Kitamura K Maruyama S amp ImanagaH (1989) Lexical agraphia in the Japanese language Pureagraphia for Kanji due to left posteroinferior temporallesions Brain 112 1549ndash1561

Sugishita M Otomo K Kabe S amp Yunoki K (1992) Acritical appraisal of neuropsychological correlates ofJapanese ideogram (kanji) and phonogram (kana) readingBrain 115 1563ndash1585

Sugishita M amp Yoshioka M (1987) Visual processes in ahemialexic patient with posterior callosal sectionNeuropsychologia 25 329ndash339

Sugishita M Yoshioka M amp Kawamura M (1986) Recoveryfrom hemialexia Brain and Language 29 106ndash118

Tarkiainen A Helenius P Hansen P C Cornelissen P L ampSalmelin R (1999) Dynamics of letter string perception inthe human occipitotemporal cortex Brain 122 2119ndash2132

Thompson-Schill S L DrsquoEsposito M amp Kan I P (1999)Effects of repetition and competition on activity in leftprefrontal cortex during word generation Neuron 23513ndash522

Tyler L K Bright P Dick E Tavares P Pilgrim LFletcher P Greer M amp Moss H (2003) Do semanticcategories activate distinct cortical regions Evidence for adistributed neural semantic system CognitiveNeuropsychology 20 541ndash559

Vandenberghe R Price C Wise R Josephs O ampFrackowiak R S (1996) Functional anatomy of a commonsemantic system for words and pictures Nature 383254ndash256

Van Orden G C Johnston J C amp Hale B L (1988) Wordidentification in reading proceeds from spelling to sound tomeaning Journal of Experimental Psychology LearningMemory and Cognition 14 371ndash386

van Turennout M Ellmore T amp Martin A (2000)Long-lasting cortical plasticity in the object naming systemNature Neuroscience 3 1329ndash1334

Vuilleumier P Henson R N Driver J amp Dolan R J (2002)Multiple levels of visual object constancy revealed byevent-related fMRI of repetition priming NatureNeuroscience 5 491ndash499

Yamaguchi S Toyoda G Xu J Kobayashi S amp Henik A(2002) Electroencephalographic activity in a flankerinterference task using Japanese orthography Journal ofCognitive Neuroscience 14 971ndash979

Yokota T Ishiai S Furukawa T amp Tsukagoshi H (1990)Pure agraphia of kanji due to thrombosis of the Labbe veinJournal of Neurology Neurosurgery and Psychiatry 53335ndash338


employed the subliminal priming method for a languagewhich uses two radically different writing systems to writethe same word (Figure 1) That is we focused on therepetition priming effect that may occur between twowriting systems of Japanese Kanji (logographic) and Kana(syllabic) when participants are engaged in a semanticcategorization task Importantly when a given word iswritten in the two scripts there is no one-to-one corre-spondence at the sublexical level For instance the wordlsquolsquocigarettersquorsquo is printed as lsquolsquo rsquorsquo in Kanji and lsquolsquo rsquorsquo inKana respectively while both are pronounced as ta-ba-ko Hence the cross-script repetition priming if anyshould represent a perceptual phenomenon distinctfrom the case-independent priming observed in alpha-betical systems attributable only to phonological lexicalor semantic activations beyond the orthographic repre-sentations of words As such this particular form ofpriming might be mediated by a set of areas distinct fromthe VWFA

The advantages of studying within- and cross-scriptpriming in Japanese readers with fMRI are threefoldFirst the combined use of subliminal masked primingand fMRI enables us to examine the initial stages ofreading and their neural correlates while avoiding po-tential strategic and attentional changes elicited byawareness of the primendashtarget relation (Naccache ampDehaene 2001a) That is this method allows us to focuson bottom-up activations of the neural code associatedwith the word stimuli while presumably preventing thetop-down re-entry of information (Dehaene Jobertet al 2004 Lamme 2003 Lamme Zipser amp Spekreijse2002) It may therefore provide precise informationabout which type of local word code is used in a givenbrain area without contamination by the distant globalbroadcasting that may occur when words cross thethreshold of consciousness

Second this design allows us to further evaluate thedepth of processing involved during unconscious word

perception Although it is widely accepted that cross-case repetition priming is extremely robust undermasking conditions and thus provides clear evidencefor the existence of subliminal orthographic processingthere is still a controversy as to whether one can observea similar phenomenon for higher processing stages(phonological lexical and semantic) correspondingto whole-word processing (Kouider amp Dupoux 2004Forster Mohan amp Hector 2003) That is although someresearchers have suggested that subliminal word prim-ing is restricted to an analysis of subword fragments orletters instead of reflecting lexical processing (Abramsamp Greenwald 2000) other studies have found repeti-tion priming for words translated into two differentlanguages even when written with very different scripts(Jiang 1999 Gollan Forster amp Frost 1997) Howeverbecause prime awareness was not measured in theselatter studies it is still not entirely clear whether prim-ing goes beyond the orthographic level under genuinelysubliminal conditions Evidence of masked cross-scriptpriming between Kanji and Kana together with thedemonstration that prime stimuli remained below thethreshold of awareness would provide further unequiv-ocal evidence that unconscious perception can reflectprocessing at an advanced level Also the anatomicallocalization of brain areas associated with cross-scriptpriming may in turn help interpret the cognitive pro-cess underlying this behavioral phenomenon

Third the design of the present experiment withorthogonal manipulations of prime script and targetscript incidentally allows us to examine the overlapand the differences between the neural substrates ofvisual word recognition in Kanji and in Kana Thisissue has long been a major concern for neurolinguis-tic studies of Japanese Focal brain damage oftenaffects the reading of Kanji and Kana to a differentdegree which has lead to the idea that visual words inthe two formats must have partially distinct neuralrepresentations Specifically the inferior temporal cor-tex and a more dorsal part of the occipito-temporalregion in the left hemisphere have been thoughtimportant for reading of Kanji and Kana respectively(Iwata 1986) Also several lines of studies have sug-gested the contribution of the right hemisphere in thevisual recognition of logograms (Yamaguchi ToyodaXu Kobayashi amp Henik 2002 Kamada et al 1998Nakagawa 1994 Sugishita Yoshioka amp Kawamura1986 Hatta 1977 Sasanuma Itoh Mori amp Kobayashi1977) Nonetheless the issue seems not fully resolvedas for instance a more recent multiple case studyfound no consistent correlation between specific pat-terns of impairment and lesion sites (Sugishita OtomoKabe amp Yunoki 1992) In the brain imaging literaturemoreover although a few previous studies have exam-ined brain activity during reading of Japanese (SakuraiMomose Iwata Sudo et al 2000 Sakurai MomoseIwata Ishikawa et al 1996 Sakurai Momose Iwata

Figure 1 Sequence of events used for the behavioral tasks Each trialconsisted of a visible target word ( ) preceded by a masked prime

( ) The primendashtarget relation (same or different word) prime

script and target script were manipulated independently Participantswere requested to determine whether the latter represented a natural

or artificial object as quickly as possible

Nakamura et al 955


EXPERIMENT 1

Results

Behavioral Results





Imaging Results








Effect Size(msec)





p lt 05

p lt 01

p lt 001







EXPERIMENT 2


Primes in Kanji

R L

Primes in Kana

R L

Targets in Kanji

R L

Targets in Kana

R L



Nakamura et al 957





















of Kana (B)



Results







Coordinate


ZValue x y z













Coordinate


ZValue x y z














Coordinate


ZScore x y z



26 400 55 38 20
















Nakamura et al 959


























Effect Size(msec)




Kana Kana 586 593 7

p lt 05

p lt 001




















processing (Dehaene









for details)

Nakamura et al 961


GENERAL DISCUSSION

















Nakamura et al 963






Conclusions




METHODS

Experiment 1

Participants







fMRI Procedures


Data Analysis


Nakamura et al 965


Experiment 2

Participants





Acknowledgments




REFERENCES






















































Nakamura et al 967

































EXPERIMENT 1

Results

Behavioral Results





Imaging Results








Effect Size(msec)





p lt 05

p lt 01

p lt 001







EXPERIMENT 2


Primes in Kanji

R L

Primes in Kana

R L

Targets in Kanji

R L

Targets in Kana

R L



Nakamura et al 957





















of Kana (B)



Results







Coordinate


ZValue x y z













Coordinate


ZValue x y z














Coordinate


ZScore x y z



26 400 55 38 20
















Nakamura et al 959


























Effect Size(msec)




Kana Kana 586 593 7

p lt 05

p lt 001




















processing (Dehaene









for details)

Nakamura et al 961


GENERAL DISCUSSION

















Nakamura et al 963






Conclusions




METHODS

Experiment 1

Participants







fMRI Procedures


Data Analysis


Nakamura et al 965


Experiment 2

Participants





Acknowledgments




REFERENCES






















































Nakamura et al 967





































EXPERIMENT 2


Primes in Kanji

R L

Primes in Kana

R L

Targets in Kanji

R L

Targets in Kana

R L



Nakamura et al 957





















of Kana (B)



Results







Coordinate


ZValue x y z













Coordinate


ZValue x y z














Coordinate


ZScore x y z



26 400 55 38 20
















Nakamura et al 959


























Effect Size(msec)




Kana Kana 586 593 7

p lt 05

p lt 001




















processing (Dehaene









for details)

Nakamura et al 961


GENERAL DISCUSSION

















Nakamura et al 963






Conclusions




METHODS

Experiment 1

Participants







fMRI Procedures


Data Analysis


Nakamura et al 965


Experiment 2

Participants





Acknowledgments




REFERENCES






















































Nakamura et al 967




















































of Kana (B)



Results







Coordinate


ZValue x y z













Coordinate


ZValue x y z














Coordinate


ZScore x y z



26 400 55 38 20
















Nakamura et al 959


























Effect Size(msec)




Kana Kana 586 593 7

p lt 05

p lt 001




















processing (Dehaene









for details)

Nakamura et al 961


GENERAL DISCUSSION

















Nakamura et al 963






Conclusions




METHODS

Experiment 1

Participants







fMRI Procedures


Data Analysis


Nakamura et al 965


Experiment 2

Participants





Acknowledgments




REFERENCES






















































Nakamura et al 967

































Results







Coordinate


ZValue x y z













Coordinate


ZValue x y z














Coordinate


ZScore x y z



26 400 55 38 20
















Nakamura et al 959


























Effect Size(msec)




Kana Kana 586 593 7

p lt 05

p lt 001




















processing (Dehaene









for details)

Nakamura et al 961


GENERAL DISCUSSION

















Nakamura et al 963






Conclusions




METHODS

Experiment 1

Participants







fMRI Procedures


Data Analysis


Nakamura et al 965


Experiment 2

Participants





Acknowledgments




REFERENCES






















































Nakamura et al 967

























































Effect Size(msec)




Kana Kana 586 593 7

p lt 05

p lt 001




















processing (Dehaene









for details)

Nakamura et al 961


GENERAL DISCUSSION

















Nakamura et al 963






Conclusions




METHODS

Experiment 1

Participants







fMRI Procedures


Data Analysis


Nakamura et al 965


Experiment 2

Participants





Acknowledgments




REFERENCES






















































Nakamura et al 967

































GENERAL DISCUSSION

















Nakamura et al 963






Conclusions




METHODS

Experiment 1

Participants







fMRI Procedures


Data Analysis


Nakamura et al 965


Experiment 2

Participants





Acknowledgments




REFERENCES






















































Nakamura et al 967








































Nakamura et al 963






Conclusions




METHODS

Experiment 1

Participants







fMRI Procedures


Data Analysis


Nakamura et al 965


Experiment 2

Participants





Acknowledgments




REFERENCES






















































Nakamura et al 967





































Conclusions




METHODS

Experiment 1

Participants







fMRI Procedures


Data Analysis


Nakamura et al 965


Experiment 2

Participants





Acknowledgments




REFERENCES






















































Nakamura et al 967

































METHODS

Experiment 1

Participants







fMRI Procedures


Data Analysis


Nakamura et al 965


Experiment 2

Participants





Acknowledgments




REFERENCES






















































Nakamura et al 967

































Experiment 2

Participants





Acknowledgments




REFERENCES






















































Nakamura et al 967







































































Nakamura et al 967
































subliminal convergence of kanji and kana words: further ... · mean accuracy (± sd ) for the...

Documents