Brain (2002), 125, 1054±1069 Language-speci®c tuning of visual cortex? Functional properties of the Visual Word Form Area Laurent Cohen,1,2 SteÂphane LeheÂricy,3 Florence Chochon,1 Cathy Lemer,1,2 Sophie Rivaud1 and Stanislas Dehaene2 1Institut de Neurologie, Ho à pital de la SalpeÃtrieÁre, AP-HP, Paris, 2INSERM U334, Service Hospitalier FreÂdeÂric Joliot, CEA/DSV, Orsay and 3Service de Neuroradiologie Fischgold, HoÃpital de la SalpeÃtrieÁre, AP-HP, Paris, France Summary The ®rst steps in the process of reading a printed word belong to the domain of visual object perception. They culminate in a representation of letter strings as an ordered set of abstract letter identities, a representation known as the Visual Word Form (VWF). Brain lesions in patients with pure alexia and functional imaging data suggest that the VWF is subtended by a restricted patch of left-hemispheric fusiform cortex, which is reproducibly activated during reading. In order to determine whether the operation of this Visual Word Form Area (VWFA) depends exclusively on the visual features of stimuli, or is in¯uenced by language-dependent parameters, brain activations induced by words, consonant strings and chequerboards were compared in normal subjects using functional MRI (fMRI). Stimuli were presented in the left or right visual hemi®eld. The VWFA was identi®ed in both a blocked-design experiment and an event-related experiment as a left-hemispheric inferotemporal area showing a stronger Correspondence to: L. Cohen, Service de Neurologie 1, Clinique Paul Castaigne, HoÃpital de la SalpeÃtrieÁre, 47/83 Bd de l'HoÃpital, 75651 Paris Cedex 13, France E-mail: laurent.cohen@psl.ap-hop-paris.fr activation to alphabetic strings than to chequerboards, and invariant for the spatial location of stimuli. In both experiments, stronger activations of the VWFA to words than to strings of consonants were observed. Considering that the VWFA is equally activated by real words and by readable pseudowords, this result demonstrates that the VWFA is initially plastic and becomes attuned to the orthographic regularities that constrain letter combination during the acquisition of literacy. Additionally, the use of split-®eld stimulation shed some light on the cerebral bases of the classical right visual ®eld (RVF) advantage in reading. A left occipital extrastriate area was found to be activated by RVF letter strings more than by chequerboards, while no symmetrical region was observed in the right hemisphere. Moreover, activations in the precuneus and the left thalamus were observed when subjects were reading RVF versus left visual ®eld (LVF) words, and are likely to re¯ect the attentional component of the RVF advantage. Keywords: alexia; asymmetry; brain mapping; language; magnetic resonance imaging; orthography; reading; visual ®eld Abbreviations: FFA = fusiform face area; fMRI = functional MRI; LVF = left visual ®eld; RVF = right visual ®eld; TC = Talairach coordinates; TE = echo time; TR = relaxation time; VWF = visual word form; VWFA = visual word form area Introduction The ®rst steps in the process of reading a printed word belong to the domain of visual object perception. Just as in the case of common objects or faces, the ultimate goal of these initial stages is to build up a representation of the input that is invariant for various irrelevant perceptual dimensions (Riesenhuber and Poggio, 1999). Letter strings can be identi®ed irrespective of their location in the visual ®eld, of the colour of the ink, of the case, size, type of font, etc. According to Warrington and Shallice, the visual identi®cation of letter strings is achieved by an abstract representation ã Guarantors of Brain 2002 that has been termed the Visual Word Form (VWF) (Warrington and Shallice, 1980). The left fusiform gyrus and the VWF system While reading involves a large network of connected cortical regions (see reviews in Price, 1997; Fiez and Petersen, 1998), we formulated the hypothesis that an area located in the midportion of the left fusiform gyrus, which activates whenever literate subjects read printed words, contributes crucially to Language-speci®c tuning of visual cortex the cerebral basis of the VWF, and accordingly proposed to label this left fusiform region the Visual Word Form Area (VWFA) (Cohen et al., 2000). We have shown recently that this area can be identi®ed in any single individual as a region showing a response to visual words, independently of their location on the retina. Its peak is consistently found at the same location in Talairach space (approximately ±43 ±54 ±12), with a standard deviation of only ~0.5 cm (Cohen et al., 2000). It is also thought to be the source of electrical and magnetic ®elds that are recorded over the left ventral occipito-temporal region, with a latency of ~150±200 ms, whenever subjects see words (Nobre et al., 1994; Salmelin et al., 1996; Tarkiainen et al., 1999; Cohen et al., 2000). For the sake of simplicity, we will refer to this region as the VWFA, although it should be clear that this denomination rests on functional hypotheses that are still open to debate. The VWFA hypothesis gains support from both brain imaging and neuropsychological data suggesting that the representation subtended by the VWFA is invariant for the spatial location and the speci®c case or font used to present the words. The VWFA is activated by words irrespective of the visual hemi®eld in which words are presented, and thus represents stimuli in a location-invariant format (Cohen et al., 2000). Furthermore, a recent study using masked repetition priming demonstrates invariance for typographic case (Dehaene et al., 2001): the VWFA shows a reduced activation to conscious words whenever the target is primed by the subliminal presentation of the same word. Crucially, this repetition suppression effect is identical irrespective of the case in which the prime is printed. Observations of brain damaged patients provide a set of parallel data. The VWFA seems to be the critical lesion site for pure alexia, a unimodal de®cit of word reading, with sparing of writing and of auditory word comprehension (Dejerine, 1892; Damasio and Damasio, 1983; Binder and Mohr, 1992; Beversdorf et al., 1997; Leff et al., 2001). Although the lesion affects the left inferotemporal cortex, the de®cit typically extends to the entire visual ®eld of those patients who are not hemianopic (Dejerine, 1892). Finally, patients with severe pure alexia may be unable to reach a case-invariant representation of letters, as evidenced for instance by their inability to decide whether `a' and `A' represent the same abstract grapheme (e.g. Miozzo and Caramazza, 1998). Less clear at present is whether the middle portion of the left fusiform gyrus houses a strictly visual representation of words, as our VWF hypothesis predicts. Certainly, many studies report no activation of this region during spoken word processing (for reviews see Binder et al., 2000; Giraud and Price, 2001). In a recent study, using an identical same± different judgement task with pairs of visual or auditory stimuli, we have demonstrated that while the VWFA is strongly activated by strings of letters, it shows no activation to auditory words or pseudowords (Dehaene et al., 2002). It therefore appears as a unimodal area, at least as far as word perception is concerned. However, in some studies, words in the auditory or tactile modality have been shown to induce 1055 left fusiform activations, which should not be expected if this region was purely devoted to visual processing (DeÂmonet et al., 1992; DeÂmonet et al., 1994; Binder et al., 1996; Vandenberghe et al., 1996; D'Esposito et al., 1997; BuÈchel et al., 1998; Perani et al., 1998; Chee et al., 1999; Buckner et al., 2000; PihlajamaÈki et al., 2000; Wise et al., 2000; Giraud and Price, 2001). Such evidence has been taken as an indication that the fusiform gyrus could embody the phonological output lexicon (Brunswick et al., 1999) or support lexical access from semantics (Foundas et al., 1998). This objection to the VWFA hypothesis, however, seems less than conclusive. First, the mid-fusiform gyrus may encompass several distinct areas involved in language processing. The VWFA proper would occupy the middle portion of the left fusiform gyrus, with Talairach coordinates (TC) close to y = ±60, while increasingly more abstract and supramodal representations would occupy its more anterior sectors. This idea is compatible with an overview of 20 recent imaging studies reporting left fusiform activations that fell within the following bounding coordinates: ±50 < x < ±30; ±80 < y < ±30; z < 0. Fourteen fusiform activation peaks were elicited by non-visual verbal stimuli, or were common to visual and non-visual stimuli, thus suggesting an abstract representation beyond the VWF system. Those peaks had an average anteroposterior coordinate of y = ±43 (range ±54 to ±32), clearly anterior to our postulated VWFA (DeÂmonet et al., 1992; DeÂmonet et al., 1994; Binder et al., 1996; Price et al., 1996b; D'Esposito et al., 1997; BuÈchel et al., 1998; Perani et al., 1998; Chee et al., 1999; Buckner et al., 2000; Wise et al., 2000). Interestingly, all these activations correspond to contrasts between conditions with higher versus lower semantic processing demands, which is in good agreement with the idea that this region is the siege of an abstract supramodal representation of words. Conversely, we compiled 25 activation peaks observed when contrasting alphabetic strings with non-alphabetic stimuli such as false fonts or ®xation, which might correspond to the real VWFA. These had a signi®cantly more posterior average coordinate of y = ±60 (range ±43 to ±70) [t(37) = 5.9; P = 10±6] (Price et al., 1996a; Puce et al., 1996; Beauregard et al., 1997; Wagner et al., 1998; Brunswick et al., 1999; Fiez et al., 1999; Kiehl et al., 1999; Buckner et al., 2000; Paulesu et al., 2000; Tagamets et al., 2000; Xu et al., 2001). Thus, this analysis is suggestive of an anterior-to-posterior functional differentiation in the fusiform gyrus, with a supramodal sector anterior to the VWFA proper. However, such evidence should obviously be interpreted with the greatest caution given the potential pitfalls of comparisons that merge various tasks, imaging and statistical methods. Secondly, in interpreting fusiform activations during nonvisual word processing, one should consider that visual regions, including primary cortex, can be activated from the top down in the absence of any visual stimulation, depending on the cognitive strategy induced by the task (Kosslyn et al., 1999; Tomita et al., 1999). Top-down activations have been evidenced in high-level visual cortex comparable to the 1056 L. Cohen et al. VWFA, using mental imagery tasks for faces or places (Ishai et al., 2000). Buckner and colleagues proposed that top-down in¯uences from frontal regions accounted for the fusiform responses to auditory verbal stimuli in a stem completion task (Buckner et al., 2000). This may also explain the activation of the VWFA when Japanese subjects were asked to write complex kanji ideograms or to imagine doing so, a task that may have an important visual component (Tokunaga et al., 1999; Nakamura et al., 2000). Thirdly, as a further indication that the VWFA supports an essentially visual representation of words, lesions affecting the mid-fusiform gyrus give rise to pure alexia (Binder and Mohr, 1992; Leff et al., 2001), while lesions that induce de®cits of lexical retrieval seem to impinge on the polar (Grabowski et al., 2001) or lateral temporal cortex (Foundas et al., 1998). Therefore, as a conclusion of this brief overview, and although the issue cannot be considered as fully settled at present, the bulk of the evidence seems consistent with our hypothesis that the left mid-fusiform region activated during reading is the cerebral substrate of an abstract visual representation of letter strings akin to the VWF. Language-speci®c tuning in the fusiform gyrus? In the present paper, we examine whether these properties depend, at least in part, on a process of ®ne tuning of the VWFA to the process of visual word recognition in the course of learning to read. Are the properties of the VWFA in¯uenced by the acquisition of a speci®c language and of a speci®c writing system? Or is this cortical sector merely responding spontaneously to the kind of features that compose letters and words? This issue can be broken down into three distinct questions. First, is the VWFA tuned to speci®c letter shapes, as opposed to visually equivalent pseudoletters? Few studies have compared directly the activation of the VWFA induced by letters and pseudoletters. In early PET experiments, Petersen et al. (1990) showed stronger activations to words than to strings of pseudocharacters in mesial extrastriate cortex. This region, however, was far more mesial and superior than the VWFA as de®ned here. More recently, Price and colleagues have obtained stronger activations to strings of consonants than to strings of pseudoletters at coordinates close to the VWFA (Price et al., 1996a). Secondly, is the VWFA sensitive to the lexical status of letter strings? Several studies using PET scanning reveal left fusiform activations stronger for pseudowords than for real words (Brunswick et al., 1999; Fiez et al., 1999; Xu et al., 2001). We have recently observed a similar trend in an event-related fMRI study (Dehaene et al., 2002). Thirdly, one may ask whether this area is tuned to the orthographic rules that constrain which combinations of letters are legal in a given writing system. This question can be addressed by comparing activations induced by legal versus illegal strings of real letters. Beauregard and colleagues mention stronger basal temporal activations for words than for random letters (Beauregard et al., 1997). However, they report neither a direct test of this contrast, nor the exact TCs. Price and colleagues, using a visual feature detection task, observed stronger activations for pseudowords than for consonant strings at coordinates compatible with the VWFA (TC ±46 ±52 ±12) (Price et al., 1996a). However, in this blockeddesign PET study, the activation of the VWFA might have been contaminated by attentional ¯uctuations between blocks of consonant strings and blocks of pseudowords, as only the latter are likely to engage subjects in spelling-to-sound translation and in lexical access to orthographically similar real words. Similar methodological remarks apply to a recent PET study by Xu and colleagues (Xu et al., 2001), as well as to block-design fMRI studies (BuÈchel et al., 1998; Rees et al., 1999). In the present study, our goal is to determine whether the functional properties of the VWFA have been attuned to the orthographic regularities of the writing system. We do not address the general issue of the parcellation of ventral temporal cortex into multiple domain-speci®c areas (see, for example, Kanwisher et al., 1999; Gauthier, 2000; Haxby et al., 2000). We are primarily concerned with the clari®cation of the role that the VWFA plays in reading. Nevertheless, our results may prove relevant to some aspects of the specialization debate. If indeed the VWFA turns out to be more activated by words than by consonant strings, it would imply that the activation of inferotemporal areas does not depend exclusively on the geometric features of the incoming stimuli. More speci®cally, it would shed light on the role of experience in the development of the functional tuning of the VWFA, and it would imply that this area has become partially specialized for reading. We studied the activation of the VWFA by real words, compared with consonant strings. Consonant strings are well matched to words in their visual features, yet they radically violate the language-dependent letter combination rules. If the VWFA has become attuned to the processing of words in the subjects' script, it should respond more to words than to consonant strings. If, on the contrary, the VWFA is tuned only to the visual features of letter strings, it should be activated equally for both types of stimuli. In order to distinguish the VWFA from other ventral temporal regions responsive to visual words, we recorded fMRI activations during the presentation of stimuli in either the left or the right hemi®eld. We reasoned that most of the reading network, starting from the VWFA, should be activated identically for words presented in the left and right halves of the visual ®eld (Fig. 1). Note that beyond the isolation of position-invariant activations, the split-®eld display technique can also reveal distinct patterns of activation for stimuli presented in the left and right visual hemi®elds (LVF and RVF, respectively). A secondary goal of our experiments, therefore, was to use such hemi®eld-dependent activations to shed some light on the mechanisms that underlie the well known advantage for verbal stimuli presented in the RVF (GruÈsser and Landis, Language-speci®c tuning of visual cortex 1057 and had normal anatomical MRIs. All gave their written informed consent. The experiment was approved by the Ethical Committee of the HoÃpital de BiceÃtre. Stimuli Word stimuli consisted of 168 frequent and highly imageable nouns (mean log10 of frequency/million = 1.75, range 1.0± 3.3; mean imageability rating = 4.8, range 4±6) (Content et al., 1990), 3±6 letters and 1±3 syllables in length. Non-words were derived from words by creating 168 consonant strings matched one-to-one in number of letters. The frequency distribution of consonants was the same in consonant strings as in real words. All words and non-words were presented in lower case letters, once in the LVF and once in the RVF. Stimulus presentation was controlled by the Expe6 software (Pallier et al., 1997). Experimental tasks Fig. 1 Illustration of the six types of stimuli used in Experiments 1 and 2. The stimuli were words, consonant strings and chequerboards displayed in either the left or the right hemi®eld. The main contrasts that were used in the analysis of data are summarized in coloured rectangles, with arrows pointing to the corresponding cerebral activations in group analyses (voxelwise P < 0.01; corrected P < 0.05 for cluster extent). Red and green: posterior hemi®eld-dependent activations induced by alphabetic stimuli and chequerboards. Magenta: activation of the VWFA induced by alphabetic stimuli relative to chequerboards in either hemi®eld. Activations were highly similar in the block-design and the event-related experiments. 1991). In the ®rst experiment, we compared words and consonant strings using a blocked-design fMRI paradigm. In the second experiment, in order to prevent any attentional ¯uctuation between blocks, we adapted the same paradigm to an event-related fMRI design. Experiment 1 Methods Subjects Seven subjects (six females, one male), all aged 20±30 years, with a university education, and fully right-handed according to the Edinburgh Inventory, participated in the study. All were drug free, had no neurological or psychiatric history, Subjects were asked to ®xate a permanent central ®xation point, while stimuli were ¯ashed in their right or left visual hemi®eld (Fig. 1). Words and non-words extended from 2° to a maximum of 6° away from ®xation. Chequerboards consisted of 12 3 4 small rectangles, and extended from 2° to 6° away from ®xation, with approximately the same vertical size as letter strings. Each trial consisted of a 550 ms ®xation period followed by a 200 ms presentation of the target. Subjects were asked to pay attention to all stimuli, words, non-words and chequerboards. Procedure fMRI experiment Subjects received three fMRI sequences, each comprising 12 experimental blocks, i.e. two blocks for each condition. Each block comprised 28 trials. Blocks were presented in pseudo-random order within sequences, so as to maximize the variety of transitions between conditions while avoiding any repetition of the same condition in successive blocks. Trials were presented in random order within blocks. The order of blocks within sequences and the order of trials within blocks differed across the three sequences, but were identical for all subjects. The order of sequences was randomized across subjects. Behavioural control In seven subjects, including ®ve subjects from the fMRI experiment, eye movements and word identi®cation were studied outside of the MRI scanner. The angular parameters of the stimuli were the same as in the fMRI experiment. Eye movements were monitored using an Iris infrared-light eyetracker (Skalar Medical, Delft, The Netherlands). First, 1058 L. Cohen et al. Statistical analysis of fMRI data Fig. 2 Cortical network activated by alphabetic stimuli relative to checkerboards, including the left fusiform VWFA (white square) (group analyses; voxelwise P < 0.01, corrected P < 0.05 for cluster extent). subjects were presented with a complete sequence from the fMRI experiment, with the original timing parameters. Eye movements were monitored but no overt response was required. Secondly, subjects were presented with 56 real words drawn from the fMRI material. Words were presented once in each hemi®eld, arranged in four blocks of 28 words each (RVF, LVF, RVF, LVF). We used a longer stimulus onset asynchrony (SOA) than in the fMRI experiment (2000 ms), in order to allow for overt naming. Naming errors and eye movements were monitored. Imaging parameters Each sequence consisted of 12 s of initial ®xation, followed by 12 stimulation blocks of 21 s each, as described before, followed by 12 s of ®nal ®xation. In each sequence, 92 functional volumes sensitive to blood oxygen level dependent (BOLD) contrast were acquired with a T2-weighted gradient echo, echo planar imaging sequence on a 1.5 tesla Signa Imager [General Electric, TR (relaxation time) = 3000 ms, µ = 90°, TE (echo time) = 60 ms, ®eld of view = 240 3 240 mm, in-plane resolution = 3.75 3 3.75 mm2]. Each volume comprised 20 axial slices of 5 mm thickness covering most of the brain. The ®rst four volumes were discarded to reach signal equilibrium. High-resolution images (3D fast gradientecho inversion-recovery sequence, TI (inversion time) = 600 ms, TR = 1100 ms, TE = 2 ms, µ = 20°, ®eld of view = 240 3 180 mm, slice thickness = 1.5 mm, in-plane resolution = 0.94 3 0.94 mm2) were also acquired for anatomical localization. Functional images were analysed with the Statistical Parametric Mapping software (SPM99b). To correct for motion, functional scans were realigned using the last image as a reference. The anatomical image was transformed stereotactically to TCs using the standard template of the Montreal Neurological Institute. The functional scans were then normalized using the same transformation. Functional images were smoothed with a Gaussian spatial ®lter to a ®nal smoothness of ~ 8 mm. The resulting images had cubic voxels of 3 3 3 3 3 mm3. For single-subject analyses, activation on each of the six types of trial was modelled by a combination of the standard SPM haemodynamic function and its temporal derivative. Only the former function was used for statistical contrasts. Three additional variables of non-interest modelled constant differences across the four sequences. Long-term signal variations were eliminated with a high-pass ®lter set at 120 s. Low-pass ®ltering was achieved by convolution with the haemodynamic response function. We also performed a random-effect group analysis with subjects as random variable. Results Subjects made signi®cantly more errors when reading aloud LVF words than RVF words (29.8% versus 14.5% errors; Wilcoxon one-tailed test P = 0.014). The reading performance thus displayed the classical RVF advantage for verbal material, replicating the results obtained in a previous fMRI study with split-®eld reading (Cohen et al., 2000). The relatively high error rate in this behavioural test is imputable to the simultaneous recording of eye movements, which required subjects to inhibit eye blinks while wearing cumbersome equipment close to the eyes. The mean number of saccades was 1.7 (0.5% of trials) during a complete fMRI sequence with covert naming instructions. Thus, the stimulation paradigm used in this study allowed for a satisfactory identi®cation of target words, while keeping lateral saccades at a minimum. Group analyses In group analyses, we used a voxelwise threshold of P < 0.01, with a corrected P < 0.05 for cluster extent. We tested the contrast of alphabetic stimuli (words or consonant strings) versus chequerboards. This revealed a bilateral cortical network, including bilateral mesial frontal and rolandic cortex, right intraparietal and prefrontal cortex, Broca's area, and a left inferotemporal focus (Fig. 2 and Table 1). The highly signi®cant left fusiform focus closely matched the previously reported coordinates of the VWFA [TC ±39 ±57 ±9; t(6) = 10.01]. No activation was observed at the homologous location in the right hemisphere. As the reason for using lateralized stimuli was to pick out activations invariant for spatial location, we checked that this region was Language-speci®c tuning of visual cortex 1059 Table 1 Experiment 1: group analysis of the reading network activated by alphabetic stimuli relative to chequerboards Alphabetic stimuli versus chequerboards Main effect Side Left Left Left Left Right Right Right Bilateral Area Fusiform (VWFA) Rolandic Rolandic Broca's area Prefrontal Rolandic Intraparietal Mesial frontal TC Cluster Peak voxel x y z P corrected No. of voxels t ±39 ±21 ±48 ±30 18 51 33 12 ±57 ±3 3 21 45 9 ±39 21 ±9 60 45 9 33 30 51 60 <0.001 0.005 0.034 0.003 <0.001 0.016 <0.001 <0.001 349 80 59 88 446 67 135 364 10.01* 11.44* 7.75* 7.60* 10.46* 9.39* 9.24* 8.76* LVF RVF t t Interaction with hemi®eld t 8.05* 3.48** 2.30*** 1.54 1.41 3.18** 0.69 1.69 6.17* 7.64* 2.48*** 3.36** 1.74 3.34** 5.07* 2.60*** 2.94*** 3.24*** 0.68 0.92 0.15 0.01 3.15*** 0.84 Words versus consonants t 3.10*** 5.06* 3.69** 2.35*** 4.45** 3.95** 1.94 5.70* *P < 0.001; **P < 0.01; ***P < 0.05 (two-tailed P for the interaction column, one-tailed P otherwise). Fig. 3 Sagittal cut at coordinate x = ±40, showing the activation of the VWFA by alphabetic stimuli relative to chequerboards (magenta; voxelwise P < 0.01 for both the right and the left hemi®eld). The major part of the VWFA shows a stronger activation for words than for consonant strings (yellow; voxelwise P < 0.05). also activated in analyses restricted either to LVF stimuli [TC ±39 ±57 ±9; t(6) = 8.05; 121 voxels] or to RVF stimuli [TC ±39 ±60 ±12; t(6) = 7.80; 272 voxels]. The left fusiform activation, de®ned as the voxels responsive to alphabetic stimuli in both the LVF and the RVF (voxelwise P < 0.01 each), extended from y = ±75 to y = ±35 along the sagittal axis, stopping short of the temporal magnetic susceptibility artifact. The difference between alphabetic stimuli and chequerboards was stronger for RVF than for LVF stimuli at the peak of the VWFA [interaction with hemi®eld, t(6) = 2.94; P < 0.05], as well as at the left rolandic and right intraparietal peaks [t(6) = 3.24 and t(6) = 3.15; both Ps < 0.05] (see Table 1). This interaction might be related to the behavioural RVF advantage in reading tasks. The VWFA activation was signi®cantly stronger for words than for consonant strings at the peak voxel [t(6) = 3.10, P = 0.011]. Due to the intrinsic resolution of the fMRI method and to the smoothing of functional images, this statistical test performed at the peak voxel actually re¯ects the activity of a larger surrounding cortical volume. Nevertheless, in order to check further that the effect was not restricted to this single voxel, we looked for voxels activated more strongly by words than by consonant strings (voxelwise P < 0.05) within the reading network, de®ned as the voxels responsive to alphabetic stimuli in both the LVF and the RVF (voxelwise P < 0.01 each). We found a cluster of 56 voxels coextensive with the VWFA [TC ±33 ±69 ±6; t(6) = 5.69; voxelwise P = 0.002; P < 0.001 for cluster extent corrected for multiple comparisons within the search volume]. As shown in Fig. 3, this lexicality effect was characteristic of a large set of left fusiform voxels in the vicinity of the peak of the VWFA. The effect was of the same size, whether the stimuli were presented in the RVF or the LVF [interaction with hemi®eld, t(6) = 0.32]. Stronger activation to words was also observed at all other peaks of the reading network, except for the right intraparietal focus (see Table 1), again without interaction with the visual hemi®eld (all interaction Ps > 0.16). The VWFA was distinct from more posterior hemi®elddependent occipito-temporal regions that showed greater activation to contralateral than to ipsilateral stimulation [LVF: TC 33 ±81 15, t(6) = 15.97, 518 voxels; RVF: TC ±18 ±84 ±6, t(6) = 10.90, 490 voxels], whether the stimuli consisted of alphabetic strings or chequerboards (all Ps < 0.0005 at the peak voxel; Fig. 1). We tested whether the nature of the stimuli in¯uenced the activation of those regions, de®ned as the sets of voxels activated by contralateral versus ipsilateral alphabetic strings, and by contalateral versus ipsilateral checkerboards (for each test, voxel-level P < 0.01, corrected P < 0.05 for cluster extent). 1060 L. Cohen et al. Fig. 4 (Left panel) peak of the VWFA identi®ed in individual subjects (green squares) and in the group analyses (yellow circles) projected onto the inferior surface of a normalized brain. (Right panel) percentage change in BOLD signal for words and consonant strings versus chequerboards in the left and right hemi®elds at the peak of the group VWFA, averaged across subjects (bars represent the intersubject standard error). In both the block-design and the event-related experiments, alphabetic stimuli yielded signi®cant activations relative to chequerboards, and real words relative to consonant strings. To increase sensitivity, these regions were searched for voxels showing differential activation to contralateral alphabetic versus chequerboard stimuli, using a statistical test corrected for multiple comparisons within the small volume thus de®ned. In the right hemisphere, no difference was observed. In the left hemisphere, however, a small hemi®eld-dependent region posterior to the VWFA was more responsive to contralateral alphabetic stimuli than to chequerboards [TC ±24 ±78 ±12; t(6) = 5.60; P < 0.001; corrected P = 0.003 for cluster extent]. No difference between words and consonant strings was observed at this peak location (P > 0.05). Individual analyses In individual analyses, we used a voxelwise threshold of P < 0.001, with a corrected P < 0.05 for cluster extent. We looked for the VWFA in each individual subject, using the contrast between alphabetic stimuli and checkerboards, masked by the same contrast restricted to LVF and to RVF stimuli (voxelwise P < 0.01 each). As shown in Fig. 4 and Table 2, all seven subjects showed a signi®cant activation in the immediate vicinity of the peak isolated in the group analysis. The location of this focus was remarkably stable across subjects, with a standard deviation of 5 mm, 5 mm and 4 mm Language-speci®c tuning of visual cortex 1061 Table 2 Experiment 1: activations of the left fusiform VWFA in individual subjects Subject Alphabetic stimuli versus chequerboards TC Cluster Peak voxel x y z P corrected No. of voxels t 1 2 3 4 5 6 7 ±42 ±42 ±51 ±36 ±36 ±48 ±42 ±60 ±63 ±63 ±48 ±54 ±57 ±51 ±15 ±21 ±12 ±18 ±18 ±9 ±15 0.005 <0.001 <0.001 0.002 0.008 0.005 <0.001 30 415 94 32 20 13 59 9.93* 12.94* 11.29* 7.25* 6.79* 6.11* 9.03* Mean Standard deviation ±42 5 ±57 5 ±15 4 Words versus consonants t 6.46* 3.15* 3.6* 1.9*** 5.64* 2.97** 1.05 *P < 0.001; **P < 0.05; ***P < 0.01. Due to the small size of the VWFA cluster in Subjects 5 and 6, the voxelwise threshold was raised to 10±4 in order to reach clusterwise signi®cance. In Subject 7 there was a signi®cant difference between words and consonant strings 6 mm away from the peak of the VWFA (t = 3.42; P = 0.024 corrected for multiple comparisons in a 6 mm radius sphere). Table 3 Experiment 1: activations of the right hemispheric counterpart of the VWFA in individual subjects Subject No. Alphabetic stimuli versus checkerboards TC 3 7 Cluster Peak voxel x y z P corrected No. of voxels t 51 42 ±60 ±45 ±15 ±15 0.001 0.001 18 18 8.12* 6.13* Words versus consonants t 3.52* 1.34 *P < 0.001. Due to the relatively small size of the cluster, the voxelwise threshold was raised to 10±4 in order to reach clusterwise signi®cance. along the x-, y- and z-axes, respectively. In only two subjects was the peak activation signi®cantly stronger for RVF than for LVF alphabetic stimuli. The contrast of words versus consonant strings was signi®cant in six out of seven subjects at the peak of the individually identi®ed VWFA (Table 2). In the remaining subject, a signi®cant cluster was found within a few millimetres of the expected coordinates [TC ±42 ±57 ±15; t(90) = 3.42; voxelwise P = 0.024 corrected for multiple comparisons in a 12 mm sphere centred on the expected location]. As in the group analysis, the effect of lexical status did not interact with visual hemi®eld (P > 0.05 in all subjects). In addition to the VWFA proper, two out of seven subjects showed a right fusiform activation cluster for alphabetic stimuli relative to checkerboards and invariant for spatial location (Table 3). In one of these two subjects (Subject 3), this peak showed a signi®cant effect of lexicality, with no interaction with hemi®eld. Figure 5 shows a summary of the functional properties of the left and right fusiform foci in this individual subject. The posterior hemi®eld-dependent activations could be identi®ed in all subjects at coordinates close to the peaks of the group analysis, using the contrast between contralateral and ipsilateral stimuli, masked by the same contrast restricted to alphabetic stimuli and to chequerboards (voxelwise P < 0.01 each). We looked, within those regions, for increased activations for contralateral alphabetic strings relative to chequerboards (voxelwise P < 0.01; P < 0.05 for cluster extent corrected for multiple comparisons within the small volume). In agreement with the group analysis, left-hemispheric activations were found in all subjects. A similar region was also found in the right hemisphere of four out of seven subjects, although it was always smaller and less signi®cant than in the left hemisphere. In three out of seven subjects, the left-hemispheric peak activation was stronger for words than for consonant strings (P < 0.05). Such was also the case for one out of four subjects with a right-hemispheric activation (P < 0.05). Correlates of the RVF advantage We attempted further to identify the neural correlates of the behavioural RVF advantage in word reading. Behaviourally, we observed a better performance for real words displayed in the RVF than for similar words displayed in the LVF. We 1062 L. Cohen et al. Fig. 5 Fusiform activations, with percentage change in BOLD signal, in Subject 3 (Experiment 1). In addition to the real VWFA, this subject showed a symmetrical but smaller right-hemispheric activation for alphabetic stimuli. Both foci were more strongly activated by real words than by consonant strings. therefore analysed fMRI data using the very same contrast, with the same statistical thresholds as in previous group analyses. Several regions were more strongly activated by RVF than LVF words. They naturally included the left occipital hemi®eld-dependent region described earlier (TC ±30 ±78 ±3), but also three additional regions: the left precuneus (TC ±12 ±51 45), the left thalamus (TC ±18 ±24 3) and a small right prerolandic focus (TC 48 ±15 36). Note that these three regions did not show a stronger activation to RVF than to LVF chequerboards (P > 0.30 for all three). The opposite contrast (LVF versus RVF words) only revealed right occipital regions (TC 30 ±81 18). processing might have decreased relative to the blocks in which real words were presented. Such a non-speci®c attentional in¯uence might explain at least in part why the VWFA, among other regions, was in¯uenced by orthographic well-formedness. In order to clarify this point, we performed a second experiment, very similar to Experiment 1 except that we resorted to using an event-related design, with the six types of trial randomly intermixed. We expected that this design would largely avoid the spurious effects of task-related attentional ¯uctuations. Experiment 2 Nine subjects (seven female, two male), ful®lling the same criteria as in Experiment 1, participated in the study. In Experiment 1, the VWFA was reliably identi®ed on the basis of positional invariance, and was shown to be more strongly activated by real words than by consonant strings. However, this difference does not necessarily re¯ect the intrinsic processing properties of the VWFA. Rather, it could be a consequence of the blocked structure of the experiment. During the blocks with consonant strings, the subjects' general level of arousal or their attention to alphabetic Methods Subjects Stimuli, tasks and procedure We used a subset of the verbal material of Experiment 1, comprising 100 words and the corresponding consonant strings. All words and non-words were presented in lower case letters, once in the LVF and once in the RVF. The Language-speci®c tuning of visual cortex 1063 Table 4 Experiment 2: group analysis of the reading network activated by alphabetic stimuli relative to checkerboards Side Area Alphabetic stimuli versus checkerboards Main effect TC Left Left Left Right Right Right VWFA Rolandic Broca's area Prefrontal Rolandic Mesial frontal Cluster Peak voxel x y z P corrected No. of voxels t ±42 ±42 ±45 45 39 ±6 ±57 ±3 12 18 6 ±3 ±15 42 12 0 51 48 0.031 0.001 <0.001 101 184 426 <0.001 1663 4.61* 8.57* 6.38* 9.02* 7.54* 9.46* LVF RVF t t Interaction with hemi®eld t 4.70* 4.57* 2.32*** 9.33* 3.10** 3.21** 3.57** 2.44*** 3.81** 4.77* 4.33** 1.24 1.14 1.41 0.66 0.68 0.24 0.82 Words vs consonants t 2.31*** 0.85 1.35 4.13** 0.78 0.05 *P < 0.001; **P < 0.01; ***P < 0.05 (two-tailed P for the interaction column, one-tailed P otherwise). Since the right hemispheric rolandic and prefrontal activations belonged to the cluster whose main peak was mesial frontal, these two peaks are reported without clusterwise signi®cance level. experimental task was the same as in Experiment 1, except that each trial consisted of a 2200 ms ®xation period followed by a 200 ms presentation of the target. Subjects received four fMRI sequences, each comprising a total of 150 trials (25 words, 25 consonant strings and 25 checkerboards in each hemi®eld). Trials were presented in a random order within sequences. As a behavioural control, seven other righthanded subjects were presented with the same material (except the checkerboards) outside of the MRI device. They were asked to name real words and to utter the word `consonants' when presented with consonant strings. Errors and naming latencies were monitored. Imaging parameters and statistical analysis of fMRI data Each sequence consisted of 12 s of initial ®xation, followed by 150 stimulation trials. In each sequence, 155 functional volumes sensitive to blood oxygen level dependent contrast were acquired using the same parameters as in Experiment 1, except for the TR (2400 ms) and the number of functional axial slices (n = 16). The ®rst ®ve volumes were discarded to reach signal equilibrium. Anatomical images were acquired as in Experiment 1. Functional images were corrected for slice acquisition delays by Fourier interpolation, and then realigned, normalized and smoothed as in Experiment 1. For statistical modelling, we applied the same procedures and parameters as in Experiment 1. Results Subjects tested outside the MRI magnet again showed a signi®cant advantage when reading aloud RVF versus LVF words, both on error rates (3.1% versus 9.4% errors; Wilcoxon one-tailed P = 0.012) and on latencies [mean correct latency: 707 ms versus 758 ms; t(6) = 4.73; one-tailed P = 0.0016]. Group analyses In group analyses, we used a voxelwise threshold of P < 0.01, with a corrected P < 0.05 for cluster extent. We tested the global contrast between alphabetic stimuli and chequerboards. A bilateral cortical network was observed, similar to the pattern reported in Experiment 1 (Fig. 2 and Table 4). It included the left fusiform VWFA, located within a few millimetres of the coordinates observed in Experiment 1 (TC ±42 ±57 ±15). The same contrast restricted to RVF or LVF stimuli was signi®cant at the peak voxel of the VWFA (both Ps < 0.004), with no signi®cant interaction, thus achieving invariance for spatial location. The left fusiform activation, de®ned as in Experiment 1, extended from y = ±60 to y = ±41 along the sagittal axis, stopping at ~ 6 mm from the temporal magnetic susceptibility artifact. As in Experiment 1, the VWFA was activated more strongly by real words than by consonant strings [t(8) = 2.31; P < 0.025]. Applying the same statistical procedure as in Experiment 1, we found that the peak of the VWFA belonged to a cluster of 12 voxels coextensive with the VWFA [TC ±42 ±54 ±12; t(8) = 2.71; voxelwise P = 0.013; P < 0.001 for cluster extent]. As shown in Fig. 3, this lexicality effect concerned the major part of the VWFA. Again, there was no interaction between lexicality and hemi®eld (P > 0.15). Replicating Experiment 1, we found posterior hemi®elddependent activations to contralateral stimuli [LVF: TC 36 ±72 ±3, t(8) = 20.49, 1093 voxels; RVF: TC ±24 ±81 ±3, t(8) = 7.07, 509 voxels], responding to both alphabetic strings and checkerboards (all Ps < 0.005 at the peak voxel; Fig. 1). However, contrary to Experiment 1, no difference between contralateral alphabetic stimuli versus chequerboards was found within these hemi®eld-dependent activations. Individual analyses This event-related experiment was less sensitive than its blocked-design counterpart, and individual analyses often 1064 L. Cohen et al. Table 5 Experiment 2: activations of the left fusiform VWFA in individual subjects Subject Alphabetic stimuli versus checkerboards TC 1 2 3 4 5 6 7 8 9 Mean SD Cluster Peak voxel Words versus consonants t x y z P corrected No. of voxels t ±45 ±42 ±36 ±45 ±45 ±51 ±60 ±54 ±54 ±45 ±21 ±12 ±24 ±18 ±18 <0.001 <0.001 >0.05 0.001 <0.001 6479 257 9 202 196 5.95* 7.2* 3.08* 6.47* 4.98* 0.56 1.72 0.08 0.37 4.12* ±48 ±48 ±42 ±44 4 ±54 ±60 ±60 ±55 5 ±12 ±15 ±12 ±17 4 >0.05 >0.05 0.007 14 70 143 2.99** 3.98* 7.03* 1.78 0.59 0.56 *P < 0.001; **P < 0.01. yielded marginal results. In an attempt to identify the VWFA in individual subjects, we used the global contrast of alphabetic stimuli versus chequerboards, at a lowered threshold (voxelwise P < 0.01). In eight out of nine subjects, an activation cluster was observed in the immediate vicinity of the peak determined in the group analysis (see Table 5 and Fig. 4). This cluster, however, reached a corrected P < 0.05 for cluster extent in only ®ve out of eight subjects. Still, the location of this focus was remarkably stable across subjects, with standard deviations of 4, 5 and 4 mm along the x-, y- and z-axes, respectively. At the peak of the VWFA, the contrast of real words versus consonant strings was signi®cant in only one subject. No right-hemispheric equivalent of the VWFA was found in any of the subjects. Correlates of the RVF advantage As in Experiment 1, we compared activations induced by real words displayed in the RVF versus in the LVF. As before, the left occipital hemi®eld-dependent region was activated (TC ±21 ±84 0), as well as the precuneus (TC ±9 ±54 30). Again, the precuneus did not show a stronger activation to RVF than to LVF chequerboards (P > 0.50). The opposite contrast (LVF versus RVF words) only revealed right occipital regions (TC 33 ±69 ±6 and 36 ±78 12, respectively). Discrepancies between Experiments 1 and 2 Before turning to the general discussion, we will brie¯y consider some apparent discrepancies between the general pictures of the reading network that emerged from the two experiments. If we consider Fig. 2 and Tables 1 and 4, in which the contrast of alphabetic stimuli versus chequerboards is reported, some discrepancies appear in the pattern of frontal and parietal activations, beyond the overall similarity of the activation patterns. First, in Experiment 2, there was a stronger bilateral inferior rolandic activation than in Experiment 1. This region corresponds to the sensorimotor cortex devoted to the control of facial and articulatory movements. In Experiment 2, the display of each stimulus was followed by a 2200 ms ®xation period, while this period lasted only 550 ms in Experiment 1. Thus, if an overt vocal response had been required, subjects would have had suf®cient time to utter it only in Experiment 2. It is therefore plausible that in Experiment 2 subjects could covertly read stimuli up to a full-¯edged motor plan, which might be re¯ected in the strong lower rolandic activations. Secondly, the activation of the anterior portion of Broca's area which was apparent in Experiment 1 was apparently lacking in Experiment 2. Actually, this region was activated in Experiment 2 also [t(8) = 3.06; P < 0.01 at the peak voxel TC ±30 21 9], but it was included in the large left inferior frontal cluster of 426 voxels with a rolandic main peak. This cluster included a secondary peak within a few millimetres of the activation observed in Experiment 1 [TC ±33 27 6; t(8) = 3.78; P = 0.003]. Thirdly, the right parietal activation observed in Experiment 1 was lacking in Experiment 2. This lack of activation persisted even when lowering the statistical threshold, and should be considered meaningful. This discrepancy should be related to the major difference in task structure between the two experiments. In Experiment 1, subjects could focus their attention in advance towards the region of space where the next stimulus was expected, while in Experiment 2 the location of the next stimulus was always unpredictable. Considering its role in the orientation of spatial attention (see review in Mesulam, 1999), it is not surprising that the right parietal cortex was more strongly involved in the block-design than in the event-related experiment. As an illustration of converging data, Gitelman and colleagues elicited right parietal activations very similar to the present ones by inducing lateralized expectancy for subsequent target Language-speci®c tuning of visual cortex appearance, irrespective of the direction of the attentional shift (Gitelman et al., 1999). Discussion In this study, normal subjects were presented with real words, consonant strings and chequerboards in their left or right visual hemi®eld. The ®rst experiment followed a blockeddesign paradigm, while in the second experiment the six types of stimuli were randomly mixed, following an event-related design. The analysis of activations concentrated on fusiform regions, and particularly on the putative VWFA. Localization of the VWFA The VWFA was identi®ed easily in both experiments as a left-hemispheric inferotemporal area displaying two key features. First, it shows a stronger activation to alphabetic stimuli than to checkerboards. Secondly, it is invariant for spatial location, as revealed by comparable activation patterns irrespective of the stimulated hemi®eld. Over the two experiments, this area could be detected unambiguously in 15 out of 16 subjects. It was located in the mid-portion of the fusiform gyrus, generally within the occipitotemporal sulcus. The TCs of the VWFA were remarkably stable across subjects and across experiments, invariably falling within a radius of a few millimetres. This location is in good agreement with a number of previous activation studies (for reviews see Cohen et al., 2000; Dehaene et al., 2001; Dehaene et al., 2002). The VWFA was shown to be activated by meaningless visual letter strings (either pronounceable pseudowords or consonant strings) relative to non-alphabetic visual stimuli such as false fonts or faces (Price et al., 1996a; Puce et al., 1996), or by similar contrasts using real words (Beauregard et al., 1997; Wagner et al., 1998; Fiez et al., 1999; Kiehl et al., 1999; Buckner et al., 2000; Paulesu et al., 2000). The VWFA also appears to be strongly lateralized. A symmetrical right-hemispheric homologue could be identi®ed in only two out of 16 subjects, and it was always much more weakly activated than its left-hemispheric counterpart. Differences between words and consonant strings Our main goal was to establish whether the VWFA was tuned to language-dependent parameters, and particularly to the orthographic regularities that constrain the combination of letters into pronounceable strings. In both experiments, we observed stronger activations to words than to strings of consonants, in good agreement with previous studies using PET or block-design fMRI (Price et al., 1996a; BuÈchel et al., 1998; Rees et al., 1999; Xu et al., 2001). It may be objected that in Experiment 1 this effect could re¯ect attentional ¯uctuations between blocks: with real words subjects could engage in attention-demanding phonological and semantic 1065 operations, therefore increasing the general level of activation of the entire word processing network. This ampli®cation would include the VWFA and be confounded with any intrinsic preference of this region for words over consonant strings. In Experiment 2, however, all types of stimuli alternated randomly at a rapid rate, making it unlikely that there would be any systematic difference in attentional engagement between words and consonant strings. One may note in retrospect that the amplitude of the lexicality effect in terms of the percentage change in BOLD signal was comparable in both experiments (Fig. 4), suggesting that attentional factors actually had little in¯uence even on the lexicality effect observed in Experiment 1. Still, different tasks and experimental designs are likely to induce different levels of involvement and of top-down activation of the VWFA. Thus, Rees and colleaugues found a difference between words and consonant strings only whenever subjects paid attention to the stream of alphabetic stimuli (as opposed to a concurrent stream of object pictures) (Rees et al., 1999). However, once stimuli enter the ®eld of attention, a difference between words and consonants can be elicited using a variety of tasks [e.g. passive viewing as in the present study, visual feature detection as in Price et al. (1996a), etc.]. Task-related parameters may also explain why, in contrast with a number of previous studies, Tagamets and colleagues showed no change of activation in the VWFA across words, pseudowords, consonant strings and even false fonts (Tagamets et al., 2000). At the same time, they showed progressively increasing activations in parietal and frontal regions presumably involved in attentional control, and an ampli®cation of posterior visual regions. It is possible that the demanding repetition detection task grew more dif®cult with more unfamiliar stimuli, progressively amplifying the activity of visual regions, including the VWFA. This effect could have masked the intrinsic preference of this region for words over consonant strings and false fonts. Our results clearly show that the operation of the VWFA is in¯uenced by language-dependent parameters, and does not depend exclusively on the visual features of stimuli. However, words and consonant strings differ not only in their obedience to letter combination rules, but also in their lexical status. Only words possess stable semantic and phonological representations in memory. Therefore, critical to any detailed interpretation of the present data is the status of the VWFA relative to pseudowords, which are comparable to real words inasmuch as they obey orthographic rules, and comparable to consonant strings inasmuch as they lack a lexical representation. We capitalized on previous data showing, in comparable experimental settings, that pseudowords induced an activation of the VWFA at least as strongly as did real words (Brunswick et al., 1999; Fiez et al., 1999; Xu et al., 2001; Dehaene et al., 2002). This result is congruent with the postulated function of the VWFA, namely the representation of any well-formed letter string at a prelexical level to serve as input to subsequent language-related processes. We may therefore conclude that the weaker 1066 L. Cohen et al. activation of the VWFA by consonant strings is indeed due to their violation of letter combination rules, and not to their lack of a lexical entry per se. Note that the superiority of words over consonant strings could in principle re¯ect the simple presence of vowels, rather than the obedience to orthographic rules. If this hypothesis were true, one would predict identical activation levels for non-words such as BRSTEAIU and for real words or pseudowords. This possibility remains to be tested in future experiments. Cerebral bases of the RVF advantage in reading The primary reason for using split-®eld presentation of stimuli was to isolate areas invariant for spatial position. However, it also provided an opportunity to observe behavioural asymmetries, and to try to clarify some of their underlying mechanisms. In both experiments we observed the classical behavioural advantage for RVF over LVF words. Schematically, two types of theories have been proposed to account for this RVF advantage, although these two approaches are by no means mutually exclusive. According to `structural' accounts, RVF verbal stimuli bene®t from a more effective processing by virtue of a direct access to the left-hemisphere language areas, while LVF stimuli must follow a longer and `noisier' transcallosal pathway. It is also possible that the left-hemispheric retinotopic areas that process RVF stimuli are more adequately tuned to visual word recognition than their right-sided counterparts. According to `attentional' accounts, the RVF advantage results from an attentional bias toward the right side of space during reading. This bias could result from the intrinsically verbal nature of the task, yielding a global left-hemispheric activation (Kinsbourne, 1970), or from various strategies adopted by the reader, some of which may depend on the direction of reading. For instance, bilingual subjects tend to report arrays of Hebrew letters starting with the rightmost letter, and arrays of English letters starting with the leftmost letter (Lubow et al., 1994). Similarly, the preferred point of initial eye ®xation is left of word centre in English (with most of the word falling in the RVF) and right of word centre in Hebrew (Deutsch and Rayner, 1999; Aghababian and Nazir, 2000), suggesting that attentional biases contribute to the RVF advantage in left-to-right reading systems. Our brain-imaging results suggest that both mechanisms are at work during split-®eld word reading. First, when examining the activations induced by RVF or LVF words relative to words presented in the opposite hemi®eld, we found the expected contralateral occipital activations. These activations probably re¯ect relatively low-level processing of contralateral stimuli, and may be thought to be essentially equivalent in the left and right hemisphere. Nevertheless, as mentioned earlier as one possible `structural' component of the RVF advantage, these perceptual regions may be better tuned to reading in the left hemisphere than in the right. Indeed, in Experiment 1, a left extrastriate region that was only responsive to contralateral hemi®eld stimulation was signi®cantly more activated by words than by checkerboards, while no such difference was observed in the right-hemispheric extrastriate areas. Interpretation should be cautious, since this difference was not replicated in Experiment 2 with an event-related design (and, therefore, a decreased statistical power). One interesting possibility, which should be subjected to further testing, is that this region has developed greater perceptual learning for words than its right-hemispheric counterpart, due to normal readers' bias for ®xating words left of their geometrical centre (Nazir, 2000). This region may be the equivalent, in the domain of word perception, of the so-called Occipital Face Area, a region of similar location responding to faces more than to common objects or letters (Kanwisher et al., 1997; Gauthier et al., 2000). Secondly, when comparing RVF versus LVF words, we observed in both experiments a predominantly left-sided activation of the precuneus, while the LVF versus RVF contrast never yielded any activation beyond the hemi®elddependent occipital areas. Furthermore, the precuneus was not activated when comparing RVF versus LVF checkerboards. This con®rms that the pattern of activation was not related to some non-speci®c visual processing asymmetry, but probably to the RVF advantage speci®c to word processing. In Experiment 1, a stronger activation for RVF than for LVF words was also observed in the left thalamus. Both the precuneus and the thalamus can plausibly be related to the orientation of attention towards the opposite side of space: spatial attentional shifts are correlated with activations in the thalamus and precuneus (Gitelman et al., 1999), and thalamic lesions can result in contralateral spatial neglect (for reviews see Mesulam, 1999; Vallar, 2001). The emergence of the VWFA In conclusion, is it possible to propose a broader view of the properties of the VWFA that are relevant to its role in reading? Writing systems are a recent cultural invention, and it is clear that in children, prior to the acquisition of reading, the patch of cortex that will eventually become the VWFA obviously cannot possess any predisposition for a speci®c spelling system. Why then is this particular region selected, within the extent of the ventral visual cortex, to play a central role in the processing of letter strings? The fact that the location of the VWFA is highly reproducible across subjects suggests that some initial properties intrinsic to this region and to its pattern of connectivity are the cause of its subsequent specialization for reading. Recently, Levy and colleagues have shown that high-level inferotemporal visual areas are localized according to a central versus peripheral visual ®eld bias (Levy et al., 2001). They show that regions sensitive to faces such as the so-called Fusiform Face Area (FFA) (Kanwisher et al., 1997) fall within an antero-posterior Language-speci®c tuning of visual cortex strip of cortex that is biased toward representing foveal stimuli, while regions more reactive to pictures of buildings fall in a more lateral periphery-biased strip. Reading typically requires predominantly foveal processing, and it may be expected that the VWFA should be located close to the FFA in terms of excentricity bias. This appears to be the case, as the coordinates of the VWFA are essentially symmetrical to those of the FFA (Kanwisher et al., 1997; Tarr and Gauthier, 2000). Lerner and colleagues have also described another gradient of perceptual predisposition, orthogonal to the eccentricity gradient, corresponding to a shift from sensitivity to local object features to a more holistic mode of representation as one goes from posterior to more anterior occipitotemporal cortex (Lerner et al., 2001). One may speculate that, even prior to the acquisition of reading, the VWFA shows a particular combination of eccentricity tuning, local versus global processing, and other parameters such as invariance for position and size (Ito et al., 1995; Grill-Spector et al., 1999), which makes it particularly suitable for representing strings of letters. This approach is related to the object-form topology hypothesis proposed by Haxby and collaborators (Ishai et al., 1999; Haxby et al., 2000), according to which regions of the ventral visual cortex differ in their perceptual predisposition for various categories of objects such as chairs, faces or buildings. Interhemispheric differences in those basic perceptual abilities may play a role in the genesis of the clearcut asymmetry of the VWFA. Hemispheric differences in the processing of parts versus whole objects, or in low versus high spatial frequencies, may contribute to the emergence of this lateralization (see, for example, Kitterle and Selig, 1991; Robertson and Lamb, 1991). In addition to its intrinsic perceptual predispositions, the anatomical and functional shaping of the mature VWFA may also result from its connections to other components of the language system. Anatomical links between the visual system and left-hemispheric language areas may have a causal role in the left-lateralization of the VWFA, possibly in combination with the aforementioned perceptual asymmetries. Indeed, in developmental dyslexics, phonological impairments may lead to an abnormal development of the VWFA. Although the VWFA is probably not the primary origin of dyslexia (Habib, 2000), it is one of the brain regions that do not activate normally in dyslexic subjects during reading (Helenius et al., 1999; Paulesu et al., 2001). One may speculate that this defect is a consequence of a phonological impairment on the ®ne tuning of the VWFA for reading. We are now in a position to summarize the impact of the acquisition of reading on the functional properties of the VWFA. Learning to read entails learning to recognize letter shapes, learning that different shapes may correspond to a single abstract letter identity (e.g. `a' and `A'), and learning that only some combinations of those letters are legal. All three types of knowledge are re¯ected in the functional properties of the VWFA. First, the VWFA is activated more strongly by real characters than by pseudofonts of equal 1067 visual complexity (Price et al., 1996a; Tarkiainen et al., 1999). Secondly, the VWFA represents words in an abstract case-invariant format, as shown by the recent ®nding of a case-independent word repetition priming in this region (Dehaene et al., 2001). Thirdly, we have demonstrated in the present study that the orthographic regularities that distinguish legal from illegal strings are also incorporated in the functional properties of the VWFA. All of these properties imply that the VWFA is plastic and becomes progressively attuned to details of the reading process during the acquisition of literacy. Acknowledgements This work was supported by a `Cognitique' grant from the French Ministry of Research. References Aghababian V, Nazir TA. Developing normal reading skills: aspects of the visual processes underlying word recognition. J Exp Child Psychol 2000; 76: 123±50. Beauregard M, Chertkow H, Bub D, Murtha S, Dixon R, Evans A. The neural substrate for concrete, abstract, and emotional word lexica: a positron emission tomography study. J Cogn Neurosci 1997; 9: 441±61. Beversdorf DQ, Ratcliffe NR, Rhodes CH, Reeves AG. Pure alexia: clinical-pathologic evidence for a lateralized visual language association cortex. Clin Neuropathol 1997; 16: 328±31. Binder JR, Mohr JP. The topography of callosal reading pathways. A case-control analysis. Brain 1992; 115: 1807±26. Binder JR, Frost JA, Hammeke TA, Rao SM, Cox RW. Function of the left planum temporale in auditory and linguistic processing. Brain 1996; 119: 1239±47. Binder JR, Frost JA, Hammeke TA, Bellgowan PS, Springer JA, Kaufman JN, et al. Human temporal lobe activation by speech and nonspeech sounds. Cereb Cortex 2000; 10: 512±28. Brunswick N, McCrory E, Price CJ, Frith CD, Frith U. Explicit and implicit processing of words and pseudowords by adult developmental dyslexics: a search for Wernicke's Wortschatz? Brain 1999; 122: 1901±17. BuÈchel C, Price C, Friston K. A multimodal language region in the ventral visual pathway. Nature 1998; 394: 274±7. Buckner RL, Koutstaal W, Schacter DL, Rosen BR. Functional MRI evidence for a role of frontal and inferior temporal cortex in amodal components of priming. Brain 2000; 123: 620±40. Chee MWL, O'Craven KM, Bergida R, Rosen BR, Savoy RL. Auditory and visual word processing studied with fMRI. Hum Brain Mapp 1999; 7: 15±28. Cohen L, Dehaene S, Naccache L, LeheÂricy S, Dehaene-Lambertz G, HeÂnaff MA, et al. The visual word form area: spatial and temporal characterization of an initial stage of reading in normal subjects and posterior split-brain patients. Brain 2000; 123: 291± 307. 1068 L. Cohen et al. Content A, Mousty P, Radeau M. Brulex, une base de donneÂes lexicale informatiseÂe pour le francËais eÂcrit et parleÂ. Ann Psychol 1990; 90: 551±66. GruÈsser OJ, Landis T. Visual agnosias and other disturbances of visual perception and cognition. London: Macmillan Press; 1991. Damasio AR, Damasio H. The anatomic basis of pure alexia. Neurology 1983; 33: 1573±83. Habib M. The neurological basis of developmental dyslexia: an overview and working hypothesis. [Review]. Brain 2000; 123: 2373±99. Dehaene S, Le Clec'H G, Poline JB, Le Bihan D, Cohen L. The visual word form area: a prelexical representation of visual words in the left fusiform gyrus. 2002; In press. Dehaene S, Naccache L, Cohen L, Bihan DL, Mangin JF, Poline JB, et al. Cerebral mechanisms of word masking and unconscious repetition priming. Nat Neurosci 2001; 4: 752±8. Dejerine J. Contribution aÁ l'eÂtude anatomo-pathologique et clinique des diffeÂrentes varieÂteÂs de ceÂcite verbale. Mem Soc Biol 1892; 4: 61±90. DeÂmonet J-F, Chollet F, Ramsay S, Cardebat D, Nespoulous J-L, Wise R, et al. The anatomy of phonological and semantic processing in normal subjects. Brain 1992; 115: 1753±68. Haxby JV, Ishai II, Chao LL, Ungerleider LG, Martin II. Objectform topology in the ventral temporal lobe. Response to I. Gauthier (2000). Trends Cogn Sci 2000; 4: 3±4. Helenius P, Tarkiainen A, Cornelissen P, Hansen PC, Salmelin R. Dissociation of normal feature analysis and de®cient processing of letter-strings in dyslexic adults. Cereb Cortex 1999; 9: 476±83. Ishai A, Ungerleider LG, Martin A, Schouten JL, Haxby JV. Distributed representation of objects in the human ventral visual pathway. Proc Natl Acad Sci USA 1999; 96: 9379±84. Ishai A, Ungerleider LG, Haxby JV. Distributed neural systems for the generation of visual images. Neuron 2000; 28: 979±90. DeÂmonet J-F, Price C, Wise R, Frackowiak RS. A PET study of cognitive strategies in normal subjects during language tasks: in¯uence of phonetic ambiguity and sequence processing on phoneme monitoring. Brain 1994; 117: 671±82. Ito M, Tamura H, Fujita I, Tanaka K. Size and position invariance of neuronal responses in monkey inferotemporal cortex. J Neurophysiol 1995; 73: 218±26. D'Esposito M, Detre JA, Aguirre GK, Stallcup M, Alsop DC, Tippet LJ, et al. A functional MRI study of mental image generation. Neuropsychologia 1997; 35: 725±30. Kanwisher N, McDermott J, Chun MM. The fusiform face area: a module in human extrastriate cortex specialized for face perception. J Neurosci 1997; 17: 4302±11. Deutsch A, Rayner K. Initial ®xation location effects in reading Hebrew words. Lang Cogn Proc 1999; 14: 393±421. Kanwisher N, Stanley D, Harris A. The fusiform face area is selective for faces not animals. Neuroreport 1999; 10: 183±7. Fiez JA, Petersen SE. Neuroimaging studies of word reading. [Review]. Proc Natl Acad Sci USA 1998; 95: 914±21. Kiehl KA, Liddle PF, Smith AM, Mendrek A, Forster BB, Hare RD. Neural pathways involved in the processing of concrete and abstract words. Hum Brain Mapp 1999; 7: 225±33. Fiez JA, Balota DA, Raichle ME, Petersen SE. Effects of lexicality, frequency, and spelling-to-sound consistency on the functional anatomy of reading. Neuron 1999; 24: 205±18. Kinsbourne M. The cerebral basis of lateral asymmetries in attention. Acta Psychol (Amst) 1970; 33: 193±201. Foundas AL, Daniels SK, Vasterling JJ. Anomia: case studies with lesion localization. Neurocase 1998; 4: 35±43. Kitterle FL, Selig LM. Visual ®eld effects in the discrimination of sine-wave gratings. Percept Psychophys 1991; 50: 15±8. Gauthier II. What constrains the organization of the ventral temporal cortex? Trends Cogn Sci 2000; 4: 1±2. Kosslyn SM, Pascual-Leone A, Felician O, Camposano S, Keenan JP, Thompson WL, et al. The role of area 17 in visual imagery: convergent evidence from PET and rTMS. Science 1999; 284: 167± 70. Gauthier I, Tarr MJ, Moylan J, Skudlarski P, Gore JC, Anderson AW. The fusiform `face area' is part of a network that processes faces at the individual level. J Cogn Neurosci 2000; 12: 495±504. Giraud AL, Price CJ. The constraints functional neuroimaging places on classical models of auditory word processing. J Cogn Neurosci 2001; 13: 754±65. Gitelman DR, Nobre AC, Parrish TB, LaBar KS, Kim YH, Meyer JR, et al. A large-scale distributed network for covert spatial attention: further anatomical delineation based on stringent behavioural and cognitive controls. Brain 1999; 122: 1093±106. Grabowski TJ, Damasio H, Tranel D, Ponto LL, Hichwa RD, Damasio AR. A role for left temporal pole in the retrieval of words for unique entities. Hum Brain Mapp 2001; 13: 199±212. Grill-Spector K, Kushnir T, Edelman S, Avidan G, Itzchak Y, Malach R. Differential processing of objects under various viewing conditions in the human lateral occipital complex. Neuron 1999; 24: 187±203. Leff AP, Crewes H, Plant GT, Scott SK, Kennard C, Wise RJ. The functional anatomy of single-word reading in patients with hemianopic and pure alexia. Brain 2001; 124: 510±21. Lerner Y, Hendler T, Ben-Bashat D, Harel M, Malach R. A hierarchical axis of object processing stages in the human visual cortex. Cereb Cortex 2001; 11: 287±97. Levy I, Hasson U, Avidan G, Hendler T, Malach R. Centerperiphery organization of human object areas. Nat Neurosci 2001; 4: 533±9. Lubow RE, Tsal Y, Mirkin A, Mazliah G. English and Hebrew letter report by English- and Hebrew-reading subjects: evidence for stimulus control, not hemispheric asymmetry. Brain Cogn 1994; 25: 34±51. Mesulam MM. Spatial attention and neglect: parietal, frontal and cingulate contributions to the mental representation and attentional Language-speci®c tuning of visual cortex 1069 targeting of salient extrapersonal events. [Review]. Philos Trans R Soc Lond B Biol Sci 1999; 354: 1325±46. versus inattentional amnesia for ®xated but ignored words. Science 1999; 286: 2504±7. Miozzo M, Caramazza A. Varieties of pure alexia: the case of failure to access graphemic representations. Cogn Neuropsychol 1998; 15: 203±38. Riesenhuber M, Poggio T. Hierarchical models of object recognition in cortex. Nat Neurosci 1999; 2: 1019±25. Nakamura K, Honda M, Okada T, Hanakawa T, Toma K, Fukuyama H, et al. Participation of the left posterior inferior temporal cortex in writing and mental recall of kanji orthography: a functional MRI study. Brain 2000; 123: 954±67. Nazir TA. Traces of print along the visual pathway. In: Kennedy A, Radach R, Heller D, Pynte J, editors. Reading as a perceptual process. Amsterdam: Elsevier; 2000. p. 3±22. Nobre AC, Allison T, McCarthy G. Word recognition in the human inferior temporal lobe. Nature 1994; 372: 260±3. Pallier C, Dupoux E, Jeannin X. EXPE: an expandable programming language for on-line psychological experiments. Behav Res Methods Instrum Comput 1997; 29: 322±7. Paulesu E, McCrory E, Fazio F, Menoncello L, Brunswick N, Cappa SF, et al. A cultural effect on brain function. Nat Neurosci 2000; 3: 91±6. Paulesu E, Demonet JF, Fazio F, McCrory E, Chanoine V, Brunswick N, et al. Dyslexia: cultural diversity and biological unity. Science 2001; 291: 2165±7. Perani D, Paulesu E, Galles NS, Dupoux E, Dehaene S, Bettinardi V, et al. The bilingual brain: pro®ciency and age of acquisition of the second language. Brain 1998; 121: 1841±52. Petersen SE, Fox PT, Snyder AZ, Raichle ME. Activation of extrastriate and frontal cortical areas by visual words and word-like stimuli. Science 1990; 249: 1041±4. PihlajamaÈki M, Tanila H, HaÈnninen T, KoÈnoÈnen M, Laakso M, Partanen K, et al. Verbal ¯uency activates the left medial temporal lobe: a functional magnetic resonance imaging study. Ann Neurol 2000; 47: 470±6. Price CJ, Functional anatomy of reading. In: Frackowiak RSJ, Friston KJ, Frith CD, Dolan RJ, Mazziotta JC, editors. Human brain function. San Diego: Academic Press; 1997. p. 301±28. Price CJ, Wise RJ, Frackowiak RS. Demonstrating the implicit processing of visually presented words and pseudowords. Cereb Cortex 1996a; 6: 62±70. Price CJ, Wise RJ, Warburton EA, Moore CJ, Howard D, Patterson K, et al. Hearing and saying. The functional neuro-anatomy of auditory word processing. Brain 1996b; 119: 919±31. Puce A, Allison T, Asgari M, Gore JC, McCarthy G. Differential sensitivity of human visual cortex to faces, letterstrings, and textures: a functional magnetic resonance imaging study. J Neurosci 1996; 16: 5205±15. Rees G, Russell C, Frith CD, Driver J. Inattentional blindness Robertson LC, Lamb MR. Neuropsychological contributions to theories of part/whole organization. [Review]. Cognit Psychol 1991; 23: 299±330. Salmelin R, Service E, KiesilaÈ P, Uutela K, Salonen O. Impaired visual word processing in dyslexia revealed with magnetoencephalography. Ann Neurol 1996; 40: 157±62. Tagamets MA, Novick JM, Chalmers ML, Friedman RB. A parametric approach to orthographic processing in the brain: an fMRI study. J Cogn Neurosci 2000; 12: 281±97. Tarkiainen A, Helenius P, Hansen PC, Cornelissen PL, Salmelin R. Dynamics of letter string perception in the human occipitotemporal cortex. Brain 1999; 122: 2119±32. Tarr MJ, Gauthier I. FFA: a ¯exible fusiform area for subordinatelevel visual processing automatized by expertise. [Review]. Nat Neurosci 2000; 3: 764±9. Tokunaga H, Nishikawa T, Ikejiri Y, Nakagawa Y, Yasuno F, Hashikawa K, et al. Different neural substrates for Kanji and Kana writing: a PET study. Neuroreport 1999; 10: 3315±9. Tomita H, Ohbayashi M, Nakahara K, Hasegawa I, Miyashita Y. Top-down signal from prefrontal cortex in executive control of memory retrieval. Nature 1999; 401: 699±703. Vallar G. Extrapersonal visual unilateral spatial neglect and its neuroanatomy. Neuroimage 2001; 14 (1 Pt 2): S52±8. Vandenberghe R, Price C, Wise R, Josephs O, Frackowiak RS. Functional anatomy of a common semantic system for words and pictures. Nature 1996; 383: 254±6. Wagner AD, Schacter DL, Rotte M, Koutstaal W, Maril A, Dale AM, et al. Building memories: remembering and forgetting of verbal experiences as predicted by brain activity. Science 1998; 281: 1188±91. Warrington EK, Shallice T. Word-form dyslexia. Brain 1980; 103: 99±112. Wise RJ, Howard D, Mummery CJ, Fletcher P, Leff A, BuÈchel C, et al. Noun imageability and the temporal lobes. Neuropsychologia 2000; 38: 985±94. Xu B, Grafman J, Gaillard WD, Ishii K, Vega-Bermudez F, Pietrini P, et al. Conjoint and extended neural networks for the computation of speech codes: the neural basis of selective impairment in reading words and pseudowords. Cereb Cortex 2001; 11: 267±77. Received June 8, 2001. Revised October 12, 2001. Accepted November 30, 2001