Evidence for highly selective neuronal tuning to whole words in the “Visual Word Form Area”

Summary

Theories of reading have posited the existence of a neural representation coding for whole real words (i.e. an orthographic lexicon), but experimental support for such a representation has proved elusive. Using fMRI rapid adaptation techniques, we provide evidence that the human left ventral occipitotemporal cortex (specifically the “visual word form area”, VWFA) contains a representation based on neurons highly selective for individual real words, in contrast to current theories that posit a sublexical representation in the VWFA.

Introduction

Reading written words is an important cognitive skill that, given the cultural recency of reading and the variability of lexica across languages, arguably needs to depend on neural representations that are acquired through experience with written words (Binder et al., 2006; Dehaene et al., 2005; Vinckier et al., 2007). Recent neuroimaging studies (Dehaene et al., 2004; Vinckier et al., 2007) have provided evidence for a hierarchical organization in the ventral visual pathway (Ungerleider and Haxby, 1994) for the visual word form, leading to the proposal that running posterior to anterior along this pathway, neurons are tuned to increasingly complex word features, viz. from oriented bars, to letters, bigrams, and finally quadragrams (Dehaene et al., 2005; Dehaene et al., 2004; Vinckier et al., 2007). However, while some theories of reading as well as neuropsychological and experimental data (Coltheart, 2004) have argued for the existence of a neural representation for whole real words (i.e., an orthographic lexicon), experimental evidence for such a representation has been elusive: Numerous human neuroimaging studies have identified an area in the left occipitotemporal cortex that appears to play a key role in whole word reading (Baker et al., 2007; Dehaene et al., 2004; Dehaene et al., 2001; Gaillard, 2006; Kronbichler et al., 2004; Vinckier et al., 2007), termed the “visual word form area” (VWFA) (Cohen et al., 2002; Dehaene et al., 2004; Kronbichler et al., 2004; Vinckier et al., 2007). However, these studies have failed to find a preference in the VWFA for real words (RW) over pseudowords (i.e., pronounceable legal letter strings; PW) (Binder et al., 2006; Devlin et al., 2006; Kronbichler et al., 2004; Vinckier et al., 2007), leading to the hypothesis that the VWFA is tuned to sublexical orthographic structure (Binder et al., 2006; Dehaene et al., 2005; Vinckier et al., 2007).

One reason for this failure to find selectivity for real words over pseudowords might be that most prior fMRI studies investigating VWFA selectivity have examined the average BOLD-contrast response in the VWFA to the stimuli of interest (but see (Dehaene et al., 2004; Dehaene et al., 2001; Devlin et al., 2006)). However, inferring neuronal selectivity based on average BOLD-contrast signal change is complicated by the fact that both the density of selective neurons as well as the broadness of their tuning contribute to the average activity level in a voxel, making it difficult to infer properties of the underlying neural representation from the total response to the stimuli of interest (Jiang et al., 2006): A given BOLD-contrast signal change in a voxel could arise from a small number of strongly activated units with high selectivity, or from a large number of broadly-tuned neurons with low activity. By contrast, it has been suggested that fMRI rapid adaptation (fMRI-RA) techniques – in which two stimuli are presented sequentially in each trial, and the BOLD-contrast response to the pair is taken to reflect similarity of the neuronal activation patterns corresponding to the two individual stimuli, with the lowest response for two stimuli activating identical neuronal populations, and maximum signal if the two stimuli activate disjoint groups of neurons (Jiang et al., 2006) – are capable of probing neuronal tuning more directly and selectively (for a recent review, see (Grill-Spector et al., 2006)). Several studies have used fMRI-RA to examine single-word reading by varying properties of the word form (Dehaene et al., 2004; Dehaene et al., 2001), providing evidence that the VWFA contains an abstract representation of the visual word form that is invariant to case, font, size, and location. Here, we used an fMRI-RA paradigm to examine neuronal tuning specificity and the nature of the representation in the VWFA by systematically altering the visual word form and lexicality to examine the effect of similarity on the hemodynamic response. We hypothesized, in contrast to the current theory of a sublexical representation in the VWFA, that the VWFA contains neurons tightly tuned to whole real words, akin to neurons highly selective for individual training objects found in monkey inferotemporal cortex after training the animals to discriminate between similar novel objects (Logothetis et al., 1995). This selective tuning for specific real words would be the result of visual experience with real words and the behavioral requirement to discriminate highly visually similar real words (such as “farm” and “form”), producing RW-tuned neurons that show high selectivity to a specific RW, but show little response to other real words, even if they are orthographically similar. In contrast, lack of experience with pseudowords and the lack of a need to discriminate real words from pseudowords (such as “farm” from “tarm”) during normal reading would lead to a lower degree of experience-driven response differentiation of specific RW-tuned neurons to orthographically similar pseudowords, analogous to previous training studies in the monkey using an orientation discrimination task, which found increased neuronal selectivity in primary visual cortex only around the trained orientation but not for untrained orientations (Schoups et al., 2001). These RW-tuned neurons then not only respond strongly to their preferred RW, but also respond to a lesser degree to similar pseudowords, thus accounting for VWFA responsiveness also to pseudowords (Fiez et al., 1999; Mechelli et al., 2003; Paulesu et al., 2000). For example, neurons tuned to the word “farm” would respond strongly to the word “farm” but very little to the real words “form”, “firm”, “harm”, etc. In contrast, the pseudoword “tarm” would cause low-level activation of neurons tuned to orthographically similar real words (“farm”, “harm”, “term”, “tarp”, etc.), leading to a total signal for PWs that might be equal to or greater than that evoked by RWs.

To test these hypotheses, we conducted three sets of fMRI experiments. Experiments 1 and 2 involved 24 subjects total (12 in each experiment) who performed a semantic oddball detection task in the scanner on visually presented words (requiring subjects to press a button whenever they saw a fruit or vegetable word; see Experimental Procedures). Experiment 3 involved 10 subjects who performed an orthographic oddball detection task on visually presented words (requiring subjects to press a button whenever they saw a particular sequence of letters embedded in a word; see Experimental Procedures). In all three experiments, words were presented in prime/target pairs, and we examined three conditions: 1) “same”, in which the same stimulus was presented twice (as prime and target) in each trial, 2) “1L”, in which the prime and target differed by one letter and, 3) “different”, in which the target shared no letters with the prime (Fig. 1). For these three conditions, one group of 12 subjects was presented with RW pairs (RW group) in Experiment 1, and the other group of 12 subjects was presented with PW pairs (PW group) in Experiment 2. The data from Experiment 1 and 2 were analyzed together using a mixed-design ANOVA. In the third experiment (Orthographic group, or O group), subjects were presented with both RW pairs and PW pairs, and a within-subject ANOVA was conducted. For the RW pairs (in both the RW and the O groups), 47 high frequency target words were chosen from the CELEX Lexical Database (Baayen, 1995). The “1L” condition was created by changing one letter of the target word to make another RW of comparable frequency. In the “different” condition, words of equal frequency were chosen so that no letters were repeated in any position. For the PW pairs in the PW group (Experiment 2), the target words were created by altering one letter of the RW target words while maintaining length, bigram and trigram (where applicable) frequency, and orthographic neighborhood (using the N-Watch program (Davis, 2005)). The “1L” condition was created by altering one letter of the PW target, again matched for bigram and trigram frequency. Finally in the “different” condition, PWs were generated using the ARC Nonword Database (Rastle et al., 2002) with no letters repeated in any position. For the PW pairs in the O group (Experiment 3), a new set of target PWs were generated to avoid overlap with the RW pairs (using the ARC Nonword Database (Rastle et al., 2002)). These PW targets were matched to the RW targets for length, bigram and trigram (where applicable) frequency, and neighborhood size. The “1L” and “different” stimuli were created as described above for the PW group. For all three experiments, all grapheme changes resulted in only one phoneme change and prime/target sets were matched for length, part of speech (where applicable), bigram and word frequency, and orthographic neighborhood (using the N-Watch program (Davis, 2005)).

Figure 1.

Experimental paradigm. For all three experiments, (A) shows the different stimulus conditions for the real word (RW) pairs, and (B) shows the conditions for the pseudoword (PW) pairs.

For all three groups we predicted the lowest signal for the “same” condition, as the two stimuli were identical and would therefore repeatedly activate the same neural populations, causing maximum adaptation (Grill-Spector et al., 2006). Likewise, we predicted the least amount of adaptation for the “different” condition because two stimuli that share no letters should not activate overlapping populations of neurons, regardless of whether neurons in the VWFA are tuned to whole words or to prelexical letter combinations. The crucial prediction that differentiates a whole-word representation from a part-word or sublexical representation is the response of the “1L” condition when comparing the PW pairs and the RW pairs (in all three groups): If neurons in the VWFA are tuned to sublexical features, one would predict a gradual response increase with increasing dissimilarity for both the RW pairs (in the RW and O groups) and the PW pairs (in the PW and O groups), as the “same” condition has maximal sublexical unit overlap, the “1L” condition has partial sublexical unit overlap and “different” has no sublexical overlap. In contrast, if neurons in the VWFA are tightly tuned to whole real words, then two similar but non-identical real words should have minimal neural overlap and therefore no adaptation should occur. This would lead to a total activation in the RW pairs for the “1L” condition similar to that caused by the two completely different words in the “different” condition (in the RW and O groups). In addition, assuming that VWFA responsiveness to pseudowords is caused by low-level activation of RW-tuned neurons to orthographically similar PWs, we should see a gradual BOLD response increase with increasing dissimilarity (from “same” to “1L” to “different”) in the PW pairs (in the PW and O groups), due to the lower degree of selectivity of the RW-tuned neurons for pseudowords.

Results and Discussion

The VWFA regions were identified for each individual subject independently through localizer scans (see Experimental Procedures), using a contrast of word versus fixation masked by the contrast of word versus scrambled word, selecting ROIs closest to the published location of the VWFA, approximate Talairach coordinates −43 −54 −12 ±5 (Cohen et al., 2002; Kronbichler et al., 2004) and within the individual subject’s occipitotemporal sulcus/fusiform gyrus region (See Fig. 2A for the left VWFA from a representative subject); see Experimental Procedures for further information. The average locations of the thus-defined VWFA ROI for the RW, PW, and O groups were Talairach coordinates (−37±4 −54±8 −11±6), (−39±4 −57±9 −11±6), and (−40±3 −58±7 −11±6), respectively. For each subject, we then analyzed the activity in their individually defined ROI during the separate event-related scans that used the rapid adaptation paradigm.

Figure 2.

Sample VWFA ROI and mean percent signal change in relation to orthographic similarity in the fMRI-RA scans of Experiments 1 and 2. (A) Activation from a representative subject during the localizer scans showing the localization of the VWFA. The VWFA was defined by words versus fixation (p<10⁻⁵, uncorrected) masked by words versus scrambled (p<0.05, uncorrected). Only clusters that were significant at the corrected cluster-level of p<0.05 were selected. This representative subject had a cluster size of 42 voxels (cluster-level, p<10⁻⁴). This figure also shows some right hemisphere activation, which is common in some subjects (Baker et al., 2007; Cohen et al., 2002) and has been shown to be less selective (Vinckier et al., 2007). In our study 14/34 subjects showed (typically smaller and less significant) homologous right hemisphere activation in or near the corresponding location as their individually defined (left hemisphere) VWFA ROI. (B,C) Mean percent signal change in the VWFA in the event-related fMRI-RA scans in RW (B) and PW (C) groups (Experiments 1 and 2, respectively). Error bars represent within-subject SEM. Significance levels: **** = < 0.0001; *** = < 0.001, ** = < 0.01; * = <0.05, n.s. = not significant.

Experiments 1 and 2

For Experiments 1 and 2 (PW and RW groups) (Fig. 2B and C, resp.), an ANOVA, with within-subject factor condition (“same”, “1L”, “different”) and between-subject factor group (RW vs. PW), revealed a significant effect of the three experimental conditions (F(2,44)=40.113, p<0.001), and a significant interaction between the three experimental conditions and the two groups (F(2,44)=4.833, p=0.019), but no significant difference between the two groups (F(1,22)=2.004, p=0.171). Post-hoc analyses indicated that the significant interaction was due to the difference in the 1L condition between the two groups, as there was a significant interaction with the RW vs. PW experiments from the comparison of “same” vs. “1L” (F(1,22)=5.577, p=0.027), and of “different” vs. “1L” (F(1,22)=14.249,p=0.001), but not of “same” vs. “different” (F(1,22)=0.011, p=0.919). We then conducted repeated-measures ANOVAs on the data from the two groups separately. Significant differences were found for both groups (for RW, F(2,22)=48.591, p<0.001; for PW, F(2,22)=12.014, p=0.001). Additionally, we conducted paired t-tests. For the RW group, we found significant adaptation in the VWFA ROIs for “same” when compared with “different” and “1L” (p<0.0001). However, response levels did not differ significantly between the “1L” and “different” conditions (p=0.09). For the PW group, there was also significant adaptation for “same” when compared to both “different” and “1L” (p=0.001 and p=0.03, respectively). However, crucially, in this group, “1L” and “different” also differed significantly (p=0.008). Similar results were obtained when we used a fixed-size ROI for all subjects (see Supplementary Material).

The results of Experiments 1 and 2 are in concert with the hypothesis that neurons in the VWFA are highly selective for whole real words: In the RW group, the finding that “1L” and “different” were not reliably different even though the two words in “1L” shared sublexical units suggest very tight tuning in the VWFA to whole real words. In contrast, the gradual BOLD response increase with increasing dissimilarity (from “same” to “1L” to “different”) for the PW group suggests broader tuning to pseudowords, compatible with an experience-driven refinement of tuning of neurons in the VWFA as a result of extensive visual experience with real words (and the requirement to discriminate these words) but not with pseudowords.

Experiment 3

We then conducted an additional experiment, Experiment 3, to address two possible concerns from the first two experiments: First, results in Experiments 1 and 2 may have been influenced by the use of a semantic task which could possibly result in differential processing for the RW versus the PW pairs. Second, it would be desirable to show the key effect of differential selectivity within subjects rather than through a between-subject design (even though the main effect and interactions were significant when comparing the results of Experiments 1 and 2). Experiment 3 addressed both of these issues. First, by using an orthographic task, we eliminated any differential effect a semantic task might have on processing RWs vs. PWs. Second, we included both RW and PW pairs in the same scanning session, allowing a complete within-subject analysis. For the O group (Fig. 3), an 2×3 repeated-measures ANOVA with within-subjects factor lexical group (RW vs. PW), and experimental condition (“same”, “1L”, and “different”) revealed a significant effect of the three adaptation conditions (F(2,18)=24.393, p<0.001), and a significant difference between the two lexical groups (F(1,9)=6.186, p=0.035), but no significant interaction between the two factors (F(2,18)=1.993, p=0.183). Post-hoc analyses indicated that there was a significant interaction between the lexical group (RW vs. PW) and experimental condition (“1L” vs. “Diff”), F(1,9)=6.148, p=0.035, in line with the data from Experiment 1 and 2. We then conducted repeated-measures ANOVAs on the data from the RW pairs and the PW pairs separately. Significant differences were found for both lexical groups (for PW, F(2,18)=14.488, p=0.002; for RW, F(2,18)=18.193, p<0.001). Additionally, we conducted paired t-tests. For the RW pairs we found significant adaptation in the VWFA ROIs for “same” when compared with “different” and “1L” (p=0.001 and p<0.001, respectively). However, response levels did not differ significantly between the “1L” and “different” conditions (p=0.622). For the PW pairs, there was also significant adaptation for “same” when compared to both “different” and “1L” (p=0.003 and p=0.002, respectively). However, “1L” and “different” also differed significantly (p=0.016). As in Experiments 1 and 2, we obtained similar results when we used a fixed size ROI for all subjects (see Supplementary Material). The results from Experiment 3 thus confirm our findings from Experiments 1 and 2, and show that the difference in selectivity in the VWFA for real words vs. pseudowords can be found independent of task.

Figure 3.

Plots of mean percent signal change in relation to orthographic similarity in the fMRI-RA scans of Experiment 3 (O group). Results from RW pairs (real words) and PW pairs (pseudowords). Error bars represent within-subject SEM. Significance levels: *** = < 0.001, ** = < 0.01; * = <0.05, n.s. = not significant.

To address the potential concern that the release from adaptation for RWs might have been driven by differences in word meaning between the prime and target despite the use of an orthographic task in Experiment 3, we conducted an additional control experiment that included a semantically-related condition (“semanticR”), in which the prime/target words in each trial were semantically related (e.g., “boat” and “ship”, see Figure S4). We hypothesized that if the release in adaptation for RWs were driven by differences in meaning between the two words in a pair, we should find a difference in activation for semantically related (“semanticR” condition) and unrelated pairs (“1L” and “different” conditions). We found no such difference (Figure S5), further supporting that the results of Experiments 1 and 3 were not due to semantic effects. Additionally, we conducted a whole-brain group analysis. The only area showing consistent adaptation effects across all experiments was the left VWFA (see Table S1), suggesting that the observed high selectivity in the VWFA was indeed a property of the stimulus representation and not a reflection of real-word or semantic selectivity in another part of the brain modulating activation in the VWFA.

In summary, our results therefore provide strong support for theories of experience-driven plasticity of the neural representations underlying reading (Baker et al., 2007; Binder et al., 2006; Dehaene et al., 2005; Vinckier et al., 2007), establishing that this learning does not just apply to lower level representations, for characters (Baker et al., 2007) and sublexical letter combinations (Binder et al., 2006; Dehaene et al., 2005; Vinckier et al., 2007) but also to whole words, compatible with the concept of an “orthographic lexicon” postulated by some models of reading (Coltheart, 2004). This “simple-to-complex” hierarchy of single-word reading fits well with general theories of object recognition in cortex (Riesenhuber and Poggio, 2002), and also provides a powerful framework to study reading errors: Corrupting neural activity at lower levels, e.g., the letter level would predict errors on similar words (e.g., misreading “farm” as “form”), whereas noise at higher levels could lead to errors on increasingly larger units such that eventually the whole word is difficult to access (e.g. pure alexia, Gaillard et al., 2006).

These results are not just relevant for theories of reading and reading acquisition but also for our understanding of the mechanisms underlying experience-driven cortical plasticity in general. It will be interesting in future studies to investigate how the specificity of the representation in the VWFA changes during development and how it might differ in individuals with reading disorders.

Experimental Procedures

Participants

A total of forty-one (26 female) right-handed normal adults who were native English speakers (aged 18–32) took part in the Experiment. Participants were excluded from further analysis if they had excessive head motion or if they were not able to stay awake for the experimental scanning sessions. The fMRI data from 6 participants were excluded for these reasons. Additionally, one participant was excluded when review of structural images demonstrated a 5–8 mm vascular anomaly in his right subcortical frontal lobe. Following exclusion of subjects for the aforementioned reasons, a total of 12 subjects (8 female) took part in Experiment 1, 12 subjects (8 female) in Experiment 2, and 10 (6 female) subjects in Experiment 3. Experimental procedures were approved by Georgetown University’s Institutional Review Board, and written informed consent was obtained from all subjects prior to the experiment.

Stimuli

Experiment 1 (RW group): Real word stimuli were chosen using the CELEX database (Baayen, 1995). Forty-seven high frequency (>50 per million) 3–6 letter target words were chosen. The target words were matched to the three different types of primes discussed above (“same”, “1L”, and “different”) plus one additional prime (“PW”) beyond the scope of the main paper, but discussed in the Supplementary Material for completeness. Experiment 2 (PW group): Target PW stimuli were created by altering the target words from the RW experiment by one grapheme to create an orthographically legal PW, and three prime types (“same”, “1L”, “different”) were created for each target word as described in the Introduction above. Experiment 3 (O group): RW pairs from Experiment 1 were used. To avoid overlap of the PW pairs with the RW pairs, a new set of target PWs were generated using the ARC Nonword Database (Rastle et al., 2002). These PWs were matched to the RWs for length, bigram and trigram (where applicable) frequency, and neighborhood size. The “1L” and “different” stimuli were created as described above for the PW group.

Tasks

To engage subjects’ attention yet avoid potential task-related confounds of the BOLD-contrast response to the conditions of interest (Grady et al., 1996; Sunaert et al., 2000), subjects were asked to perform an “oddball” detection task (Jiang et al., 2006; Jiang et al., 2000) in the scanner. For Experiments 1 and 2 (RW and PW groups), subjects were asked to press a button whenever a word appeared that referred to either a fruit or vegetable. These “oddball” stimuli were created by selecting common fruit and vegetable words that were 3–6 letters in length. In the RW group, these “oddball” words were randomly inserted into the prime/target word pairs either in the prime word position or in the target word position. In the PW group, to equate for task difficulty, the “oddball” words were matched to RW “foils” of equal length. The “oddball” and “foil” words were inserted into trials in the same manner as in the RW experiment. In Experiment 3 (O group), subjects were asked to press a button every time they saw the sequential letters “xyz” or “abc”. These “oddball” stimuli were created by randomly replacing 3 sequential letters at the beginning, middle, or end of either the prime or the target stimuli in both the RW and PW pairs. We conducted behavioral testing to verify there was no difference in reaction time or accuracy between RWs and PWs (see Supplementary Materials).

All stimuli were rendered in Courier font (36 point size, 100 ppi), average letter size ¼ × ¼ inch (25 × 25 pixels), for an approximate size of 0.5 degrees of visual angle per letter in the scanner.

Functional localizer scans

Separate localizer scans were conducted to identify the VWFA in each subject individually. Using a block design, EPI images from two functional localizer scans were collected, one at the beginning and one at the end of the scanning session for Experiments 1 and 2 and one at the beginning and one in the middle of the scanning session for Experiment 3. Participants passively viewed blocks of images of written words (high frequency nouns, >50 per million, different from those used in the event-related scans), scrambled words, faces, and objects. Each block lasted 20400ms (stimuli were displayed for 500ms and were separated by a 100ms blank interval), and stimulus blocks were separated by a 10200 ms fixation block. Each run consisted of 2 blocks of each group (words, scrambled words, faces, objects) and 8 fixation blocks. The face and object images used in the localizer scans were purchased from http://www.hemera.com and post-processed using programs written in MATLAB (The Mathworks, MA) to eliminate background variations and to adjust image size, luminance, and contrast. The final size of all images was scaled to 200 × 200 pixels. Word stimuli were chosen using the CELEX database (Baayen, 1995). Scrambled images of words were generated by scrambling the word images with a tile size of 4 pixels.

Rapid event-related (ER) scans

To probe the effects of varying orthographic similarity and lexicality on BOLD-contrast response in the VWFA, MRI images from four (Experiment 1 and 2) and six (Experiment 3) ER scans were collected. For Experiments 1 and 2, each run lasted 448.8s and had two 10.2s fixation periods, one at the beginning and the other at the end. Between the two fixation periods, a total of 110 trials were presented to participants at a rate of one every 4080ms. During each trial, the two stimuli (prime/target) were displayed sequentially (timing: fixation for 1000ms, prime for 33ms, blank for 337ms, target for 500ms, and blank for 2210ms). For Experiment 3, each run lasted 473.28s and had two 10.2s fixation periods one at the beginning and the other at the end. Between the two fixation periods, a total of 116 trials were presented to participants at a rate of one every 4080ms. During each trial, the two stimuli (prime/target) were displayed sequentially (timing: prime for 300ms, blank for 400ms, target for 300ms, and blank for 3080ms). For all three experiments the number of repetitions of each word stimulus was counterbalanced across all conditions to control for long-lag priming effects (Henson et al., 2000). Trial order and timing was adjusted using M-sequences (Buracas and Boynton, 2002). Participants were asked to watch all stimuli and respond to instances of the “oddball” stimuli by pressing a button with the right hand. The stimuli of both localizer and ER scans were presented using E-Prime software (http://www.pstnet.com/products/e-prime/), back-projected on a translucent screen located at the rear of the scanner, and viewed by participants through a mirror mounted on the head coil.

MRI acquisition

All MRI data were acquired at Georgetown University’s Center for Functional and Molecular Imaging using an echo-planar imaging (EPI) sequence on a 3 Tesla Siemens Trio scanner. For Experiments 1 and 2, an 8-channel head coil was used (Flip angle = 90°, TR = 2040ms, TE = 30ms, FOV = 205 mm, 64×64 matrix). For Experiment 3, a 12-channel head coil was used (Flip angle = 90°, TR = 2040ms, TE = 29ms, FOV = 205 mm, 64×64 matrix). For all experiments, 35 interleaved axial slices (thickness = 4.0 mm, no gap; in-plane resolution = 3.2×3.2 mm²) were acquired. At the end, three-dimensional T1-weighted MPRAGE images (resolution 1×1×1 mm³) were acquired from each subject.

MRI data analysis

All preprocessing and most statistical analyses were done using the SPM2 software package (http://www.fil.ion.ucl.ac.uk/spm/software/spm2/). After discarding the first five acquisitions of each run, the EPI images were temporally corrected to the middle slice (for event-related scans only), spatially realigned, resliced to 2 × 2 × 2 mm³ and normalized to a standard MNI reference brain in Talairach space. Images were then smoothed with an isotropic 6.4mm Gaussian kernel. The VWFA regions were identified for each individual subject independently with the data from the localizer scans. We first modeled the hemodynamic activity for each condition (word, scrambled word, face, object, and fixation) in the localizer scans with the standard canonical hemodynamic response function, then identified a word-selective ROI with the contrast of word versus fixation (at least p<0.00001, uncorrected) masked by the contrast of word versus scrambled word (at least p<0.05, uncorrected). This contrast typically resulted in only 1–2 foci in the left ventral occipitotemporal cortex (p<0.05, corrected). ROIs were selected by identifying in each subject the most anterior cluster that was significant at the corrected cluster-level of at least p<0.05 in the ventral occipitotemporal cortex (specifically, the occipitotemporal sulcus//fusiform gyrus region) with at least 10 but not more than 100 contiguous voxels in a location closest to the published location of the VWFA, approximate Talairach coordinates −43 −54 −12 ±5 (Cohen et al., 2002; Kronbichler et al., 2004) (see Fig. 2A for the results from one representative subject). Additionally, another set of VWFA ROIs using a fixed size for all subjects was extracted to assess the robustness of the results. There, VWFA ROIs were defined by a sphere of fixed size (radius = 4mm) centered at the peak of each individual’s aforementioned VWFA region, which resulted in a set of ROIs of approximately 34 voxels for each subject. Results using the fixed-size ROIs were similar to those presented in the main paper (see Supplementary Material). In the ER scans, after applying global scaling following verification that there were no correlations of the global signal with the experimental conditions (Aguirre et al., 1998) and removing low frequency temporal noise from the EPI runs with a high pass filter (1/128Hz), we modeled fMRI responses with a design matrix comprising the onset of trial types and movement parameters as regressors using a standard canonical hemodynamic response function (HRF), then extracted the mean percent signal change of the VWFA ROIs for each subject with the MarsBar toolbox (Brett, 2002) and conducted statistical analyses (mixed-design ANOVA (Experiments 1 and 2) and within-subject repeated measures ANOVA (Experiment 3) with Greenhouse-Geisser correction, followed by planned t-tests, a=0.05, two-tailed) on the percent signal change.