Faces and Word Forms Can Coexist in the Right Ventro-occipital Temporal Cortex—Evidence from Left-Hemisphere Perinatal Stroke

Abstract

In most healthy adult humans, the right ventro-occipital temporal cortex (vOTC) is more involved in face perception than its left-hemisphere counterpart. Several hypotheses link this right-lateralization for face processing to left-lateralization of language and visual word form processing. In its strongest form, this account assumes an encroachment by word form processing into left vOTC territory that would otherwise be dedicated to face processing, causing face processing that would otherwise be bilateral to rely relatively more on right vOTC. By this logic, if visual word recognition came to rely predominantly on the right hemisphere, face processing should become left-lateralized. Alternatively, it would have to share neural territory with words in right vOTC, which might negatively impact behavioral performance due to crowding. Here, we used functional MRI and the Cambridge Face Memory Test in 15 male and female adolescents and young adults with a history of left-hemisphere perinatal stroke (LHPS) who are right-lateralized for language and word form processing and whose vOTC is anatomically and functionally intact in both hemispheres. Comparison with 14 neurotypical controls revealed no significant group differences in strength or lateralization of face activation and no significant differences in face recognition. These results demonstrate that after LHPS, face and word form processing can both be right-lateralized without significant detriment to either. This is reminiscent of the observation that emotional prosody and sentence processing—typically lateralized to the right and left hemisphere, respectively—are both successfully supported by the right hemisphere in the same participant group.

Introduction

The fusiform face area (FFA) is a region of the ventro-occipital temporal cortex (vOTC) that responds more to faces than to other objects or scenes. It can be identified using functional neuroimaging (Kanwisher et al., 1997), and while its exact location shows considerable interindividual variability (Glezer and Riesenhuber, 2013), it consistently falls within the same general territory (Zhen et al., 2015). Functional neuroimaging studies (Sergent et al., 1992; Kanwisher et al., 1997; McCarthy, 1997; Yovel et al., 2008; Rossion et al., 2012; Zhen et al., 2015; Pitcher et al., 2023; Zhang et al., 2025), visual half-field studies (Leehey and Cahn, 1979; Levine and Koch-Weser, 1982; Young et al., 1985), lesion studies (Landis et al., 1986; De Renzi et al., 1994; Barton et al., 2002; Tranel et al., 2009), and neurostimulation studies (Jonas and Rossion, 2021) consistently show that face processing in healthy adults relies more on right than left vOTC.

The neuronal recycling hypothesis (Dehaene and Cohen, 2007) and the theory of graded hemispheric specialization (Behrmann and Plaut, 2015) link this right-lateralization of face processing to reading (Dehaene et al., 2010; Plaut and Behrmann, 2011; Behrmann and Plaut, 2020)—an idea supported by correlations between literacy and lateralization of face processing (Dundas et al., 2013; Gabay et al., 2017), EEG measures (Dundas et al., 2014), and neuroimaging (Dehaene et al., 2010; Monzalvo et al., 2012; Pinel et al., 2015; Dehaene-Lambertz et al., 2018). The theories posit that face processing is initially bilateral (Golarai et al., 2007; Dundas et al., 2013; Hildesheim et al., 2020; Kosakowski et al., 2024) but becomes right-lateralized when, during reading acquisition, a visual word form area (VWFA; Dehaene et al., 2002) emerges in left vOTC, presumably due to its colateralization with the (usually) left-dominant frontotemporal language system (Cai et al., 2008; Gerrits et al., 2019). This then induces “a small competition with faces at this location” and “a significant right-hemispheric shift of face responses” (Dehaene et al., 2010), and “due to competition from emerging word representations, face representations that were initially bilateral become lateralized to the right” (Behrmann and Plaut, 2015).

However, while evidence for complementary lateralization of face and word form processing at the population level is strong (Badzakova-Trajkov et al., 2010; Gerrits et al., 2020), evidence for direct competition between the two is mixed (Rossion and Lochy, 2022). Some studies show correlations between left-lateralization for word forms and right-lateralization for faces (Dundas et al., 2014; Behrmann and Plaut, 2020; Brederoo et al., 2020), but others do not (Pinel et al., 2015; Davies-Thompson et al., 2016; Gerrits et al., 2019; Canário et al., 2020). Studies investigating a trade-off between the left vOTC response to word forms and faces did not find it (Hervais-Adelman et al., 2019; Yeatman et al., 2024), and it has been suggested that what is recycled for word form processing was previously specialized for limbs, not faces (Nordt et al., 2021). Moreover, evidence for the behavioral impairment one might expect due to crowding (Teuber, 1974) if both functions lateralize to the same hemisphere is weak. Even childhood hemispherectomies that leave only one vOTC to house both face and word form processing are associated with only moderate performance decrements (Granovetter et al., 2022), and it is unclear whether those are attributable to crowding or to diffuse damage from the intractable seizures that necessitated surgery.

To approach the question of competition between word forms and faces from a novel angle, we examined adolescents and young adults who sustained a large perinatal stroke to the left hemisphere’s middle cerebral artery (MCA) territory and consequently developed a right-lateralized language system and a right-lateralized VWFA (Seydell-Greenwald et al., 2025) to ask: (1) Is right lateralization of the VWFA associated with stronger reliance of faces on left vOTC (which is spared by the lesion)? (2) Does competition in right vOTC lead to neural and/or behavioral consequences for face processing, such as a reduced fMRI response and/or impaired face recognition?

Materials and Methods

This study was conducted in accordance with the Declaration of Helsinki and with the approval of the Georgetown University Institutional Review Board. Written informed consent (or, for participants younger than 12, assent) was obtained from all participants. For participants under the age of 18, written parental consent was also obtained.

Participants.

Data presented here are from the same participants as in a previous report on the effect of a large left-hemisphere perinatal stroke (LHPS) to the MCA territory on lateralization of visual word form processing (Seydell-Greenwald et al., 2025). These participants were adolescents and young adults with a history of LHPS (N = 15, 9 males, 6 females, mean age at time of testing, 19.42; SD = 4.69 years; see Table 1 for individual ages and other demographic information). Their strokes involved at least one-third of the cortical territory fed by the left MCA. They were selected for this study because they showed clear right-hemisphere dominance for language processing as assessed using an fMRI sentence processing task. Importantly, the lesions of all participants spared the vOTC region in which the VWFA is typically found. Nonetheless, they were right- rather than left-lateralized for visual word form processing (Seydell-Greenwald et al., 2025), which makes them an ideal group in which to test whether face processing lateralizes to whichever hemisphere is nondominant for word form processing (here, the left hemisphere) or whether face processing lateralizes to the right hemisphere even if this hemisphere is also dominant for visual word form processing.

Table 1.

Participant overview

LHPS	L2	L3	L4	L5	L6	L7	L8	L9	L10	L11	L12	L13	L15	L16	L17
Age	19.7	16.5	25.3	21.1	24.2	14.7	26.5	18.0	15.1	17.9	10.0	19.8	16.8	17.8	26.7
Sex	m	m	m	m	m	f	m	f	m	f	f	m	f	m	f
Race	w	w	b	w	w	w	w	b	w	w	w	w	w	b	w
Ethnicity	nh	nh	nh	nh	nh	nh	nh	nh	nh	nh	nh	nh	nh	nh	nh
CTRL	C1	C2	C3	C4	C5	C6	C7	C8	C10	C11	C12	C13	C14	C15
Age	14.9	11.5	13.8	14.5	16.5	16.8	15.3	9.8	17.3	29.5	17.8	22.2	14.8	11.2
Sex	m	m	m	m	f	m	f	m	m	f	m	f	f	f
Race	w	w	w	b	w	w	w	w	w	w	w	w	b	o
Ethnicity	nh	nh	nh	nh	nh	nh	nh	nh	nh	nh	nh	nh	nh	h

Participant IDs correspond to those in Newport et al. (2022) and Seydell-Greenwald et al. (2025). m, Male; f, Female; w, White; b, Black or African American; o, Other; nh, not Hispanic/Latino; h, Hispanic/Latino.

A group of controls (CTRL; N = 14, 8 males, 6 females; age mean, 16.20; SD, 4.93 years; see Table 1 for individual ages and other demographic information) who did not have a history of perinatal stroke was recruited from the same families and thus matched the LHPS group in key demographics. All controls showed clear left-hemisphere dominance for language processing. As reported previously (Seydell-Greenwald et al., 2025), all participants (LHPS and CTRL) were fluent readers, and multiple assessments such as the Clinical Evaluation of Language Fundamentals (CELF-5; Wiig et al., 2013) and the Test for Reception of Grammar (TROG-2; Bishop, 2003) confirmed that the LHPS group’s language skills were not impaired despite the atypical lateralization of their language system.

Functional MRI task.

Each participant contributed two 4 min runs of a visual localizer task that interleaved blocks of images from the categories of interest (Faces, Places, and Word Forms, in separate blocks) with periods of Rest. Each block lasted 18 s, and block order was counterbalanced within and across the two runs. During Rest blocks, participants were asked to rest their eyes on a fixation cross displayed at screen center. During stimulation blocks, images from the category of interest were presented at screen center in rapid succession, with 24 images per block shown for 500 ms each and separated by a 250 ms fixation period. Participants were asked to push a button held in their dominant hand if the same image was shown twice in a row, which happened two or three times per block (at random intervals between 4 and 12 stimuli apart). Other than those immediate repetitions, images were not repeated within or across runs. The task was designed to be easy and served to ensure that participants remained attentive to the stimuli throughout. Indeed, as reported previously (Seydell-Greenwald et al., 2025), accuracy was high, and reaction times were fast in all conditions, and neither measure differed between groups.

A detailed description of the stimuli has been provided previously (Seydell-Greenwald et al., 2025). In brief, all images were shown in gray-scale at screen center and subtended roughly 5° of the visual angle. Images of faces were cropped to only show eyes, nose, and mouth and did not include jewelry or distinctive facial hair. Images of places showed outside views of various buildings. Pseudowords were short (5–7 letter) strings that followed the phonological structure of English but were not real English words. To enhance visual similarity between the stimuli in the different categories, pseudowords were shown in front of a background of phase-scrambled place stimuli, and to encourage similar looking patterns between pseudoword and face stimuli, pseudowords were shown at roughly the same height as the eyes [which tend to attract the most eye movements during visual exploration of a face (Hernandez et al., 2009)] in the face stimuli. Example stimuli and an illustration of the time course of a functional run are shown in Figure 1. Stimulus presentation in synchronization with the image acquisition and collection of button push responses were coordinated by E-Prime 2.0.

Figure 1.

Functional MRI task. A, Excerpt of the stimulus streams shown during the three experimental conditions; Faces (top, blue outline), Places (middle, yellow outline), and Word Forms (bottom, red outline). Each functional block contained 24 stimuli, for a block duration of 18 s, and each functional run contained three blocks of each condition, shown in counterbalanced order and interleaved with Rest blocks (fixation only). B, Timing of an example functional run. Each participant contributed two runs of data. Adapted from Seydell-Greenwald et al. (2025) with permission.

MRI data acquisition.

Functional magnetic resonance imaging was performed at Georgetown’s Center for Functional and Molecular Imaging on a research-dedicated 3 Tesla Siemens MRI scanner. This scanner received an upgrade from Trio to Prisma during the data collection period. Of the 15 participants in the LHPS group, 5 were scanned after the upgrade; of the 14 participants in the CTRL group, 2 were scanned after the upgrade. While the upgrade necessitated switching from a 12-channel to a 20-channel head coil, imaging parameters were held as constant as possible, and a direct comparison of pre- and postupgrade fMRI results of Georgetown volunteers showed no systematic changes in the resulting activation maps.

During the scan, participants wore foam ear plugs underneath noise-cancelling headphones and, if needed, MRI safe glasses roughly matching their prescription lenses. Participants lay in a supine position, with the head stabilized by a neck pillow and inflatable pads inserted between the headphones and the head coil, and they viewed the visual stimuli projected onto a screen at the back of the scanner bore through a slanted mirror mounted on the head coil. To indicate that they had seen a stimulus repetition, they pushed a Cedrus fiber-optic button box held in their dominant hand.

From each participant, we acquired a high-resolution T1-weighted whole-brain anatomical scan (MPRAGE; 176 sagittal slices, effective voxel size 1 × 1 × 1 mm, 256 × 256 matrix, TR of 2,350 ms, TE of 3.5 ms, TI of 1,100 ms, flip angle of 7°) and at least two runs of T2*-weighted functional scans (EPI; 50 horizontal slices, effective voxel size 3 × 3 × 3 mm, 64 × 64 matrix, TR of 3,000 ms, TE of 30 ms, flip angle of 90°). Runs were repeated if participant motion, as assessed in real time, exceeded 1 mm in either direction, or if the participant failed to respond to repetitions frequently enough to raise concerns that they might have fallen asleep. The analyses include two good runs for each participant.

MRI data analysis.

Preprocessing and modeling of MRI data were performed using SPM 12 (7,771) for MATLAB (R2023b). After conversion from dicom to nii format, each functional run’s individual images were aligned first to the run’s mean functional image and then to the corresponding participant’s native-space MPRAGE. The MPRAGE was warped to standard MNI space using SPM’s built-in unified segmentation and normalization approach. If necessary to achieve good alignment of the warped MPRAGE to the MNI space template, a manually drawn lesion mask was used in combination with enantiomorphic lesion healing (Nachev et al., 2008) to prevent the presence of a large lesion from distorting the warp. Once acceptable warping results were achieved, the same transformation was applied to the participant’s functional images. Standard-space images were then smoothed with a 6 mm FWHM Gaussian kernel.

The general linear model used to fit voxel time courses at the individual subject level included boxcar predictors to capture the block timing for each of the three conditions (convolved with the default hemodynamic response function), the six motion predictors estimated during realignment of the run’s images to their mean, and a constant for each of the runs. Linear drift was removed from the voxel time courses by applying a high-pass filter with a 400 s cutoff. Activation maps were generated comparing each condition (Faces, Places, and Word Forms) against the resting baseline (contrasts F > Rest, P > Rest, and W > Rest). These maps capture the response to faces, places, and word forms in their entirety, including low-level visual responses as well as more specific visual object recognition responses. To narrow the analysis down to the visual response specifically to faces and word forms as opposed to visual stimulation more generally, we then created maps comparing the response to faces and word forms with the response to places (contrasts F > P and W > P).

To constrain the analysis to areas in which other studies have found FFA and VWFA, we used the meta-analytic tool Neurosynth (https://neurosynth.org/). The association test map for search term “ffa,” as retrieved in May 2025, was based on 99 studies and, when thresholded at z = 11 as shown in Figure 2A, revealed two clusters in the vOTC of each hemisphere: a large, anterior cluster (peaking around an MNI y-coordinate of −52 and larger in the right than in the left hemisphere), and a smaller, more posterior cluster (peaking around y = −80). Because the location of the posterior cluster is more consistent with that of the occipital face area, we utilized only the anterior cluster as our region of interest (ROI) for the FFA. Consistent with the notion that visual processing of faces is right-lateralized, the FFA cluster was larger in the right than in the left hemisphere. The association test map for search term “word form” was based on 93 studies and, when thresholded at z = 11 as shown in Figure 2B, revealed a single cluster in the left vOTC, which we utilized as our region of interest for the VWFA.

Figure 2.

ROI and nonoverlap with lesions, visualized on the ICBM template brain using the Mango MRI Viewer. A, Results of an association test for term “ffa” across 99 studies on the Neurosynth platform. B, Results of an association test for term “word form” across 93 studies. C, Union ROI generated from the above maps to capture all voxels that might include FFA (blue), VWFA (yellow), or both (green) in either hemisphere. D, Lesion overlap map for the N = 15 participants in the LHPS group, showing that there was no overlap between the lesion and the ROI in any participant.

To ensure that our analysis included voxels that in other studies were identified as FFA, VWFA, or both, as well as their counterparts in the opposite hemisphere, we first flipped the FFA and VWFA ROI maps across the midline and took the union of the original and flipped map to generate bilaterally symmetric FFA and VWFA ROIs. We then took the union of these symmetrized FFA and VWFA maps to generate a Union ROI (Fig. 2C). The resulting Union ROI had a total size of 5,376 mm³ per hemisphere, of which 1,256 mm³ belonged to both the symmetrized FFA and the symmetrized VWFA ROIs, 2,208 mm³ were unique to the FFA ROI and located anterior and inferior to the overlap region, and 1,912 mm³ were unique to the VWFA ROI and located posterior and superior to the overlap region. While the exact locations of FFA and VWFA can vary considerably across individuals (Glezer and Riesenhuber, 2013; Zhen et al., 2015), each individual participant’s FFA and VWFA should fall (and potentially compete for territory) within this large Union ROI. Importantly, as illustrated in Figure 2D, the ROI did not overlap with the lesion in any of the participants.

Lateralization of activation inside the ROI was quantified by computinga lateralization index (LI) for each participant. In general, LIs contrast activation in the left and right hemisphere and then normalize the result by the total amount of activation [LI = (left − right)/(left + right)] such that LIs near 1 indicate left-lateralization, LIs near −1 indicate right-lateralization, and LIs near zero indicate roughly equal activation on both sides. Specifically, we computed LIs with the LI toolbox (Wilke and Lidzba, 2007) for SPM, which utilizes an outlier-resistant bootstrapping approach to estimate LIs at a range of different thresholds and then computes a weighted mean across those threshold-dependent LI estimates such that more weight is given to LI estimates from stricter thresholds but LI estimates from lenient thresholds still figure into the final estimate (Wilke and Schmithorst, 2006).

To additionally examine measures of activation separately for the two hemispheres, we extracted the following summary statistics from the Union ROI of each hemisphere for each participant using in-house MATLAB (The MathWorks Inc., 2023) scripts: The average t value of all voxels inside the ROI, as a measure of the overall response across the ROI; the number of voxels with t values significant at p < 0.01 uncorrected, as a measure of the spatial extent of significant activation within the ROI; and the average t value of the eight voxels with the highest t values inside the ROI, as a measure of the peak strength of the response. (We used the average of the top eight rather than just the peak voxel to make this measure more resilient against the influence of single-voxel outliers.) With regard to our research questions, these measures provide complementary evidence and are considered in conjunction. The average t value measure ensures that all vOTC voxels in which face and/or word form processing might occur are included in the analysis; however, because the individual FFA and VWFA are likely much smaller than the ROI, it does so at the expense of including a large number of voxels whose signal is not informative regarding our question and might dilute lateralization effects. The spatial extent measure overcomes this difficulty by including only voxels with suprathreshold activation for the contrast of interest and can reveal whether the amount of cortical territory dedicated to face versus word form processing differs between groups and/or hemispheres. Lastly, the top eight measure allows us to compare the peak response with the stimulus of interest, as an indicator of each hemisphere’s best response to it. While these measures are not independent and can be expected to be correlated, it is still conceivable that one might show an effect while the others do not. For example, LHPS participants might show a more focal response in right vOTC to avoid overlap between face and word form processing. Relative to CTRL participants, this could result in comparable average t values across the entire ROI but lower spatial extent and higher peak responses. For this reason, we test and report results for each measure separately without correcting for multiple comparisons, and our conclusions consider the results in conjunction. The overall absence of a significant group difference is inferred only if none of the measures show one, and the general presence of lateralization is concluded—separately for each participant group—only if all measures show it.

Magnetic resonance imaging data were visualized using Mango (4.1) and BrainNet Viewer [1.7 (Xia et al., 2013)]. Graphs and t tests were completed using Excel (16.16.27), ANOVAs and planned comparisons in JASP (0.95.4, ANOVAs), and regression analyses in R (4.5.2).

Behavioral assessment of face processing.

To determine whether face processing might be impaired in the LHPS group, we used the Cambridge Face Memory Test (CFMT; Duchaine and Nakayama, 2006) and, for five participants (four CTRL, one LHPS) who were younger than 14.5 years at the time of testing, a version with faces of children rather than of adults (Dalrymple et al., 2014). This test assesses recognition of unfamiliar (newly learned) faces using pictures taken from different perspectives and in different lighting conditions, with all but facial features (e.g., hair, ears) cropped from the image. In the first part of the test, participants are introduced to six faces, one at a time, by being shown the same face sequentially from three angles: left one-third profile, straight frontal view facing the camera, and right one-third profile. Immediately afterward, one of the views is shown along with two distractors (views of faces not previously shown), and the participant is asked to identify which of the three faces shown is the one they just saw. This procedure is repeated three times with different views and different distractors, for all six faces (18 trials total). In the second part of the test, frontal views of all six newly learned faces are shown together as a reminder for 20 s. After this, participants once again have to pick which of three simultaneously shown faces was among the six they just saw; however, in this part of the test, lighting and pose of familiar face stimuli are not identical to the ones shown during the exposure phase (30 trials). The third part of the test follows the same procedures as the second one, but images are degraded by addition of Gaussian noise (24 trials). Our outcome measure was accuracy across the entire test (with chance performance at 33%). Both versions of the test have been shown to result in roughly the same average accuracy for neurologically healthy participants of the target age range relevant to our study [i.e., neurologically healthy young adults performed about as well on the adult version (Duchaine and Nakayama, 2006) as children ages 10 and up performed on the child version (Dalrymple et al., 2014)].

Results

Inspection of individual activation maps

Face-selective responses, defined as significantly (p < 0.001) stronger activation in the Face condition than in the Place condition (contrast F > P; Fig. 3A), could be detected in at least one hemisphere’s vOTC in 13 of the 14 CTRL participants and 13 of the 15 LHPS participants. As expected, there was considerable variation in cluster location and cluster extent, highlighting the importance of basing analyses on individual activation maps rather than group-level maps, but in general, activation clusters were in locations captured by our ROI.

Figure 3.

Individual activation maps contrasting the response to faces with that to places (A) and showing the response to faces (B) and places (C) relative to rest. Example maps from CTRL participants are shown in the first row of each subpanel, example maps from LHPS participants in the second row. For reference, the location of the Union ROI to which statistical analyses were limited is indicated by means of a semitransparent green overlay. Activation maps are projected onto the smoothed ICBM152 T1 template in the BrainNet Viewer software using nearest voxel mapping. Individual activation maps for the word form condition have already been published previously (Seydell-Greenwald et al., 2025). The complete set of individual activation maps can be found in Figure S1.

Somewhat surprisingly, the face-selective response did not appear particularly right-lateralized even in the individual activation maps of the CTRL participants, where we had expected to identify face-selective regions in both hemispheres but more strongly on the right. Inspection of individual activation maps for contrast P > Rest (Fig. 3C) revealed that the response to place stimuli was itself stronger in right than left vOTC. This finding may seem at odds with studies showing that the fMRI response to place, house, or scene stimuli is typically bilateral. However, bilaterality does not imply symmetry, and indeed several other studies have reported results consistent with a right-hemisphere bias for visual processing of places/houses/scenes (Epstein and Kanwisher, 1998; Dehaene et al., 2010; Stevens et al., 2012).

While in direct comparison to a right-lateralized place response, the face response may no longer look right-lateralized, this does not mean that the face response itself (relative to baseline) is not right-lateralized. In fact, activation maps for contrast F > Rest (Fig. 3B) did appear more right-lateralized than those for F > P in many participants even as local activation maxima were evident in locations quite similar to those of the F > P maps.

Lateralization of face and word responses in the bilateral FFA/VWFA Union ROI

Because the LI toolbox aborts its bootstrapping-based LI computation unless a sufficient number of suprathreshold voxels can be found in both hemispheres and because activation maps for several subjects did not show a face-selective response in both hemispheres, LI analyses could not be conducted for contrast F > P. (The ROI ANOVAs reported in the next sections still allow an investigation of hemispheric differences for this contrast.) We thus conducted LI analyses only for the fMRI response to faces and word forms relative to fixation-only periods (contrasts F > Rest and W > Rest).

As shown previously (Seydell-Greenwald et al., 2025), the ROI response to word forms (contrast W > Rest) was left-lateralized for most participants in the CTRL group and right-lateralized for most participants in the LHPS group (Fig. 4A). At the group level, lateralization was significantly above zero (i.e., left-lateralized) for the CTRL group (mean = 0.22; SD = 0.31; t₍₁₃₎ = 2.60; p = 0.011) and significantly below zero (i.e., right-lateralized) for the LHPS group (mean = −0.42; SD = 0.21; t₍₁₄₎ = −7.82; p < 0.001). Average LIs for both groups exceeded an absolute value of 0.2, a commonly used cutoff for what constitutes a meaningful deviation from bilateral activation (Seghier, 2008). The group difference was sizeable and significant (t₍₂₇₎ = 6.48; p < 0.001).

Figure 4.

Lateralization of activation for word forms (A) and faces (B) along with their correlation (C). Each dot represents an individual subject. In panels A and B, blue and orange horizontal lines indicate the group means for the CTRL and LHPS group, respectively, and the vertical lines crossing them represent the standard error of the mean. In panel C, the blue and orange dashed lines indicate the best fit linear trend for the CTRL and LHPS data. As expected based on prior findings, the CTRL group showed left-lateralization for word form processing and right-lateralization for face processing. In contrast, the LHPS group was right-lateralized for both word form and face processing. Both groups showed positive correlations between LIs for faces and word forms.

In contrast, lateralization of the response to faces (Fig. 4B) did not differ significantly between groups (t₍₂₇₎ = 1.02; p = 0.315), and lateralization was significantly below zero (i.e., right-lateralized) for both groups (CTRL, mean = −0.34; SD = 0.23; t₍₁₃₎ = −5.60; p < 0.001; LHPS, mean = −0.25; SD = 0.26; t₍₁₄₎ = −3.67; p = 0.001) and exceeded the conventional threshold of |LI| > 0.2 for meaningful lateralization. Both groups had positive correlations between LIs for word form and face processing (Fig. 4C): Stronger right-lateralization for faces was associated with stronger right-lateralization for word forms in the LHPS group (r = 0.44) and with weaker left-lateralization for word forms in the CTRL group (r = 0.67).

Taken together, these LI results for contrasts F > Rest and W > Rest suggest that in our LHPS group, both faces and word forms were processed predominantly in right vOTC, and the unusual right-lateralization of word form processing was not associated with a shift of face processing toward left-lateralization.

Selectivity to faces (contrast F > P) in the left and right FFA/VWFA Union ROI

Because LIs represent the ratio of left and right responses, the absence of a significant group difference in face processing LIs does not necessarily imply the absence of group differences in the face response of each individual hemisphere. For example, activation in the LHPS group could be significantly lower than that of the CTRL group in both hemispheres and still result in comparable lateralization estimates. Therefore, we next explored activation separately for each hemisphere. For this analysis, we used repeated-measures ANOVAs to test main effects of group, hemisphere, and interactions of group and hemisphere, with planned comparisons for effects of hemisphere in each group and for group differences within each hemisphere. Nonsignificant group differences were followed by Bayesian analyses to determine the strength of evidence in favor of a true null effect.

As evident in Figure 5, the study groups did not differ regarding their face-selective ROI response (contrast F > P) in either hemisphere. For the average t value across the ROI (Fig. 5A), the repeated-measures ANOVA showed no significant main effect of group (F_(1,27) = 0.49; p = 0.488), and planned comparisons revealed no group difference for either the left hemisphere (CTRL mean = 0.74; SD = 0.89; LHPS mean = 0.53; SD = 0.93; t₍₂₇₎ = 0.65; p = 0.524) or the right hemisphere (CTRL mean = 1.00; SD = 1.07, LHPS mean = 0.75; SD = 0.98; t₍₂₇₎ = 0.65; p = 0.524). There was no significant main effect of hemisphere (F_(1,27) = 2.78; p = 0.107), and planned comparisons showed that face selectivity was not significantly stronger across the right than across the left ROI for either the CTRL group (LH mean = 0.74; SD = 0.89; RH mean = 1.00; SD = 1.07; t₍₂₇₎ = 1.23; p = 0.231) or the LHPS group (LH mean = 0.53; SD = 0.93; RH mean = 0.75; SD = 0.98; t₍₂₇₎ = 1.13; p = 0.069). There was also no significant interaction of group and hemisphere (F_(1,27) = 0.10; p = 0.923). Follow-up Bayesian t tests on the group differences separately for each hemisphere provided weak evidence for the null hypothesis (no group difference) in both the left ROI (BF₀₁ = 2.45) and the right ROI (BF₀₁ = 2.45).

Figure 5.

ROI summary statistics for contrast F > P. Conventions as in Figure 4. Neither the average t value across the vOTC ROI (A) nor the number of voxels with t values exceeding a significance threshold of p < 0.01 for the contrast (B) nor the average t value of the eight voxels with the highest t values inside the ROI (C) showed significant differences between the CTRL group (blue) and the LHPS group (orange). While the face-selective response was slightly higher in the right than in the left hemisphere, this difference reached significance only for the top-voxel measure (panel C) and only for the LHPS group.

For the number of suprathreshold t values inside the ROI (Fig. 5B), the repeated-measures ANOVA again showed no significant main effect of group (F_(1,27) = 0.003; p = 0.954), and planned comparisons revealed no group difference for either the left hemisphere (CTRL mean = 87; SD = 73; LHPS mean = 87; SD = 117; t₍₂₇₎ = −0.03; p = 0.981) or the right hemisphere (CTRL mean = 122; SD = 129; LHPS mean = 117; SD = 120; t₍₂₇₎ = 0.12; p = 0.906). There was a significant main effect of hemisphere (F_(1,27) = 5.20; p = 0.031), but planned comparisons of LH and RH face selectivity did not reach significance in either the CTRL group (LH mean = 87; SD = 73; RH mean = 122; SD = 129; t₍₂₇₎ = 1.74; p = 0.093) or the LHPS group (LH mean = 87; SD = 117; RH mean = 117; SD = 120; t₍₂₇₎ = 1.48; p = 0.151). There was no significant interaction of group and hemisphere (F_(1,27) = 0.50; p = 0.825). Follow-up Bayesian t tests on the group differences separately for each hemisphere provided weak evidence for the null hypothesis (no group difference) in both the left ROI (BF₀₁ = 2.87) and the right ROI (BF₀₁ = 2.85).

For the average t value of the eight most strongly activated voxels inside the ROI (Fig. 5C), the repeated-measures ANOVA again showed no significant main effect of group (F_(1,27) = 0.14; p = 0.714), and planned comparisons revealed no group difference for either the left hemisphere (CTRL mean = 3.90; SD = 1.53; LHPS mean = 3.38; SD = 1.29; t₍₂₇₎ = 1.00; p = 0.327) or the right hemisphere (CTRL mean = 4.09; SD = 1.48; LHPS mean = 4.23; SD = 1.60; t₍₂₇₎ = −0.25; p = 0.807). There was a significant main effect of hemisphere on face selectivity (F_(1,27) = 7.86; p = 0.009). Planned comparisons showed that face selectivity was stronger in the right than in the left hemisphere for the LHPS group (LH mean = 3.38; SD = 1.29; RH mean = 4.23; SD = 1.60; t₍₂₇₎ = 3.31; p = 0.003) but not for the CTRL group (LH mean = 3.90; SD = 1.53; RH mean = 4.09; SD = 1.48; t(₂₇) = 0.70; p = 0.491). There was no significant interaction of the group and hemisphere (F_(1,27) = 3.24; p = 0.83). Follow-up Bayesian t tests on the group differences separately for each hemisphere provided weak evidence for the null hypothesis (no group difference) in both the left ROI (BF₀₁ = 1.97) and the right ROI (BF₀₁ = 2.80).

In summary, there was no evidence in either of the three measures that face selectivity (contrast F > P) or its lateralization differed between groups. Evidence for a right-lateralization of face selectivity was somewhat inconsistent across measures and groups, reaching significance as a main effect for those measures focusing on voxels with suprathreshold activation but not for the average across all voxels in the ROI and reaching significance in only one planned contrast (for the most strongly activated voxels in the LHPS group). Importantly, the absence of strong and consistent right-lateralization in the face-selective response should not be interpreted as an absence of lateralization in the face response. Rather, as discussed above in the section about individual activation maps, the face-selective response, as assessed with contrast F > P, does not look right-lateralized because the response to places (contrast P > Rest) was itself right-lateralized.

By the same logic, the absence of a group difference for contrast F > P does not necessarily imply the absence of a group difference for contrast F > Rest. If the LHPS group showed a significantly stronger LH response to both faces and places, an analysis of contrast F > P alone would not capture this. Thus, we also tested the response to faces relative to that during fixation-only periods (contrast F > Rest).

Activation to faces (contrast F > Rest) separately for left and right FFA/VWFA Union ROI

As evident in Figure 6, the study groups also did not differ regarding their response to faces relative to fixation-only baseline. For the average t value across the ROI (Fig. 6A), the repeated-measures ANOVA revealed no significant main effect of group (F_(1,27) = 0.43; p = 0.519), and planned comparisons revealed no group difference for either the left hemisphere (CTRL mean = 2.02; SD = 1.26; LHPS mean = 2.00; SD = 1.53; t₍₂₇₎ = 0.01; p = 0.991) or the right hemisphere (CTRL mean = 3.07; SD = 0.87; LHPS mean = 2.80; SD = 1.17; t₍₂₇₎ = 0.71; p = 0.483). There was a significant main effect of hemisphere (F_(1,27) = 20.93; p < 0.001), and planned comparisons showed that activation was stronger in the right than in the left hemisphere for both the CTRL group (LH mean = 2.02; SD = 1.26; RH mean = 3.07; SD = 0.87; t₍₂₇₎ = 3.64; p = 0.001) and the LHPS group (LH mean = 2.00; SD = 1.53; RH mean = 2.80; SD = 1.17; t₍₂₇₎ = 2.82; p = 0.009). There was no significant interaction of group and hemisphere (F_(1,27) = 0.43; p = 0.519). Follow-up Bayesian t tests on the group differences separately for each hemisphere provided weak evidence for the null hypothesis (no group difference) in both the left ROI (BF₀₁ = 2.87) and the right ROI (BF₀₁ = 2.37).

Figure 6.

ROI summary statistics for contrast F > Rest. Conventions as in Figure 4. In agreement with the LI results, both groups showed significantly stronger activation in the right compared with the left ROI for all ROI summary statistics: The average t value across the entire ROI (A), the number of voxels with t values exceeding a significance threshold of p < 0.01 for the contrast (B), and the average t value of the eight voxels with the highest t values inside the ROI (C). None of the measures showed significant differences between the CTRL group (blue) and the LHPS group (orange).

For the number of suprathreshold t values inside the ROI (Fig. 6B), the repeated-measures ANOVA again showed no significant main effect of group (F(_1,27) = 0.16; p=0.697), and planned comparisons revealed no group difference for either the left hemisphere (CTRL mean = 288; SD= 163; LHPS mean = 281; SD = 188; t₍₂₇₎ = 0.10; p = 0.923) or the right hemisphere (CTRL mean = 422; SD = 120; LHPS mean = 388; SD = 127; t₍₂₇₎ = 0.73; p = 0.470). There was a significant main effect of hemisphere (F(_1,27) = 23.67; p < 0.001), and planned comparisons showed that activation was stronger in the right than in the left hemisphere for both the CTRL group (LH mean = 288; SD = 163; RH mean = 422; SD = 120; t₍₂₇₎ = 3.77; p < 0.001) and the LHPS group (LH mean = 281; SD = 188; RH mean = 388; SD = 127; t₍₂₇₎ = 3.10; p = 0.004). There was no significant interaction of group and hemisphere (F(_1,27) = 0.30; p = 0.586). Follow-up Bayesian t tests on the group differences separately for each hemisphere provided weak evidence for the null hypothesis (no group difference) in both the left ROI (BF₀₁ = 2.86) and the right ROI (BF₀₁=2.34).

For the average t value of the eight most strongly activated voxels inside the ROI (Fig. 6C), the repeated-measures ANOVA again found no significant main effect of group (F(_1,27) = 0.31; p = 0.583), and planned comparisons revealed no group difference for either the left hemisphere (CTRL mean = 6.27; SD = 1.97; LHPS mean = 5.90; SD = 2.55; t₍₂₇₎ = 0.43; p = 0.673) or the right hemisphere (CTRL mean = 7.62; SD = 1.48; LHPS mean = 7.27; SD = 1.55; t(₂₇) = 0.63; p = 0.535). There was a significant main effect of the hemisphere (F(_1,27) = 17.58; p < 0.001), and planned comparisons showed that activation was stronger in the right than in the left hemisphere for both the CTRL group (LH mean = 6.27; SD = 1.97; RH mean = 7.62; SD = 1.48; t₍₂₇₎ = 2.91; p = 0.007) and the LHPS group (LH mean = 5.90; SD = 2.55; RH mean = 7.27; SD = 1.55; t₍₂₇₎ = 3.03; p = 0.005). There was no significant interaction of the group and hemisphere (F(_1,27) < 0.01; p = 0.990). Follow-up Bayesian t tests on the group differences separately for each hemisphere provided weak evidence for the null hypothesis (no group difference) in both the left ROI (BF₀₁ = 2.68) and the right ROI (BF₀₁ = 2.47).

In summary, for all three measures of activation, the results showed right-lateralization of activation for contrast F > Rest in both groups, with no group differences in lateralization and no group differences in activation in either hemisphere. There was no evidence that in the LHPS group relative to the CTRL group, the LH response was stronger in any way (in fact, across the measures, it was numerically slightly lower), nor was the RH response significantly weaker on average or more focal (lower spatial extent, higher peak).

Taken together, the results for contrasts F > P and F > Rest converge in showing no group differences between the LHPS and the CTRL group regarding the response to faces or its lateralization. Compared with the response to places (which is itself right-lateralized), the face response is not consistently right-lateralized, but this is true for the CTRL group as well as for the LHPS group, and there are no significant differences between the groups. Compared with baseline, the response is consistently right-lateralized for both groups on all measures (peak, extent, and average ROI activation) and again not significantly different between groups. Bayesian analyses provide additional support for the absence of any group difference.

Correlations between the fMRI response to faces and word forms in right FFA/VWFA Union ROI

If visual processing of faces and word forms competed for cortical territory in right vOTC, one would expect a negative correlation between the ROI responses to faces and word forms. However, as is evident in Figure 7, we observed no such negative correlation in either of our participant groups, neither for the response to faces relative to places (Fig. 7A–C) nor for the response to faces when compared with that during fixation-only periods (Fig. 7D–F). In fact, for the latter comparison (F > Rest), we found positive correlations for both participant groups across all measures. These positive correlations are likely driven largely by the fact that participants who respond more strongly to one kind of visual stimulus (faces) also respond more strongly to any other kind of visual stimulus (word forms). Nonetheless, the absence of negative correlations even when subtracting the response to places to control for the strength of the response to visual stimuli more generally (Fig. 7A–C) argues against any kind of direct competition for territory within the ROI.

Figure 7.

Correlations between the visual response to faces and word forms in right vOTC. Neither the average t value across the vOTC ROI (A, D) nor the number of voxels with t values exceeding a significance threshold of p < 0.01 for the contrast (B, E) nor the average t value of the eight voxels with the highest t values inside the ROI (C, F) showed the significant negative correlations in the ROI response that one might expect if visual processing of faces and word forms competed inside the ROI. This held both for the response relative to places (A–C) and relative to fixation-only baseline (D–F).

Association between performance on the CFMT and fMRI results

As mentioned in the introduction, the hemispheric crowding hypothesis (Teuber, 1974) posits that after early injury to the left hemisphere, language functions can be supported by the right hemisphere but that “this resiliency is purchased at the expense of” functions normally supported by the right hemisphere, which are crowded out. If this logic applies to visual object processing, the development of a VWFA in right vOTC after LHPS, especially without a compensatory increase in left vOTC face processing activation, should occur to the detriment of face processing. To test this, we examined performance on the CFMT (Duchaine and Nakayama, 2006; Dalrymple et al., 2014), which was completed by all participants in the CTRL group and all but one participant in the LHPS group. Because the CTRL group was on average slightly (although not significantly) younger than the LHPS group and had a higher proportion of participants (4/14 in the CTRL group vs 1/14 in the LHPS group) who completed the child version of the CFMT instead of the adult version, group differences were tested for significance using a linear regression model that adjusted for age and test version.

While the LHPS group, on average, indeed performed slightly worse than the CTRL group on the CFMT (Fig. 8A), the adjusted group difference was not significant (beta_group = −8.08; t(₂₆) = −1.63; p = 0.117). A Freedman–Lane permutation test with 5,000 permutations determined the 95% confidence interval to be [−9.53, 9.48], confirming the absence of a significant group difference. To determine whether the absence of a significance group difference should be interpreted as evidence of absence or might just be reflective of a difference that did not reach significance due to small sample size and low power, we next conducted a Bayesian linear regression (R’s bayesglm function) with weakly normal informative priors. Regardless of which standard deviation was chosen for the prior (5, 7, 10, or even 20), the posterior mean-adjusted group difference estimate was around −8 (lower CFMT accuracy in the LHPS group than the CTRL group), and while the 95% confidence interval around this estimate always included zero, the posterior distribution was centered clearly below zero, and only ~27% of its mass fell within the region of −5 to 5 (group differences we would consider as negligible or equivalent). Thus, while the LHPS group did not perform significantly worse on the CFMT than the CTRL group, this should not be taken as indicating that they performed equally well. A more likely interpretation of the results is a small reduction in CFMT performance in the LHPS group that did not reach significance due to the small sample size and large variability.

Figure 8.

Performance on the CFMT and correlation with fMRI results. A, Individual results for the adult version of the CFMT are shown as solid dots; individual results from the child version are shown as open circles. The horizontal and vertical lines represent the mean and standard error of the CTRL (blue) and LHPS (orange) groups. While slightly lower on average, the performance of the LHPS group on the CFMT was not significantly worse than that of the CTRL group, and both groups performed significantly above chance (dashed gray line), suggesting no significant behavioral impact of right-lateralized word form processing on face processing. In both groups, the ROI response to faces was positively correlated with performance on the CFMT in both the left and right hemisphere, both for contrast F > P (B, C) and for contrast F > Rest (D, E).

In comparison to published CFMT data, the performance of both of our test groups (CTRL mean = 72% correct; SD = 12.6; LHPS mean = 66% correct; SD = 11.8) is slightly worse than the roughly 80% correct reported in the papers introducing the adult version of the CFMT (Duchaine and Nakayama, 2006) and the child version of the CFMT (Dalrymple et al., 2014). However, other studies employing the CFMT in neurologically healthy participants reported accuracy levels comparable to ours [e.g., 74% correct in a large sample of adolescents and young adults tested with the adult version (Wilmer et al., 2012), 63.2% correct in a sample of neurologically healthy young adults participating in a tractography study (Unger et al., 2016), and between 65% and 70% correct in an unselected sample of primary school children in Grades 5 and 6 performing the child version (Bennetts et al., 2017)]. In contrast, children and adults with prosopagnosia have been reported to score well below 60% correct on the CFMT even in those studies where the control groups scored ~80% correct (Duchaine and Nakayama, 2006; Dalrymple et al., 2014). Thus, like the group comparison within our own data, this review of our data in the context of the literature indicates that the performance of our LHPS group falls within the normal range and shows, at most, a modest reduction quite unlike that observed in individuals with diagnosed face memory impairments.

It also appears less pronounced than the significant reduction in face recognition performance reported for patients with hemispherectomies relative to matched controls (Granovetter et al., 2022; Simmons et al., 2024). In the first study (Granovetter et al., 2022), face recognition performance was assessed using a one-back same/different judgment task. Controls performed very close to ceiling (above 95% correct on average), whereas patients showed a large range of performance, with some close to chance (50% correct) and the group average falling ~10 percentage points below that of controls. In contrast, in our study, controls were very clearly off ceiling, all patients scored clearly above chance, and the average group difference was smaller even though the plausible range of accuracy scores (33–100%) was larger. The second study (Simmons et al., 2024) used a child-friendly version of the CFMT (Croydon et al., 2014) in which the response screens offer only two rather than three choices (thus, chance performance is at 50%), and participants have to memorize only five rather than six faces. Patients were on average significantly worse (~75% correct) regardless of the side of hemispherectomy compared with controls (who were at 91% correct). Thus, while direct comparisons of performance between these hemispherectomy studies and our perinatal stroke study are hampered by the different assessments used, it seems that performance is affected more clearly after hemispherectomy, when only one hemisphere is available to support both face and word form processing, than after LH MCA territory perinatal stroke, where both functions lateralize to the RH, but the LH vOTC is nonetheless available to contribute to face processing. It is also possible that the intractable seizures that necessitated the hemispherectomy surgery caused diffuse damage that contributed to lower performance in that patient group. However, face processing was also found to be significantly impaired after early life unilateral lesions to vOTC in the absence of seizures (De Schonen et al., 2005). Thus, we think that the absence of a significant impairment in our study group is owed to the fact that vOTC was intact in both hemispheres.

Corroborating the role of both left and right vOTC in face processing, both participant groups displayed strong positive correlations between performance on the CFMT and the fMRI response to faces (both relative to places and relative to fixation-only baseline) in both left and right vOTC. The correlations shown in Figure 8B–F are with the average response of the eight most strongly activated voxels of each participant inside the ROI, but positive correlations were also found with the average response of all voxels inside the ROI and with the number of voxels inside the ROI whose activation passed a significance threshold of p < 0.01.

Discussion

This study investigated the neural and behavioral profile for representation of faces in individuals who had sustained a LHPS and had developed a right-lateralized language system and VWFA. It was inspired by the neuronal recycling hypothesis (Dehaene and Cohen, 2007) and the graded hemispheric specialization hypothesis (Behrmann and Plaut, 2015), which posit that in neurologically healthy literate adults, face processing is right-lateralized because learning to read leads to development of a VWFA in left vOTC. While the original formulation of the neuronal recycling hypothesis (Dehaene and Cohen, 2007) posits that the VWFA occupies territory previously specialized for face processing, the theory of graded hemispheric specialization (Behrmann and Plaut, 2015) and a revised version of the recycling hypothesis (Dehaene-Lambertz et al., 2018) suggest that the VWFA emerges in previously unspecialized territory, thereby blocking an expansion of face processing in left vOTC that would otherwise occur (and does occur in right vOTC) as visual expertise for faces matures. In either case, the result is that less of left vOTC is dedicated to face processing than if the VWFA had not developed there and that face processing relies more on right vOTC.

The notion of competition for cortical territory evokes the hemispheric crowding hypothesis (Teuber, 1974), according to which resiliency of language function after early brain injury “is purchased at the expense of nonspeech functions as if one had to admit a factor of competition in the developing brain for terminal space” (p. 73). While it was formulated to explain why left and right-hemisphere lesions alike result in “only slight difficulties with linguistic development” but “marked deficits on visuospatial tasks,” we speculate that Teuber would likewise predict impairments of face processing when visual word form processing utilizes space that would otherwise be dedicated to face processing. Indeed, early lesions to either hemisphere’s vOTC result in impairments of face processing (De Schonen et al., 2005), as do childhood hemispherectomies involving either hemisphere’s vOTC (Granovetter et al., 2022; Simmons et al., 2024). However, it is unclear whether this is caused by crowding of face and word form processing in the one remaining vOTC or simply by loss of part of the bilateral cortical territory normally involved in face processing.

Our LHPS group provides a unique opportunity to investigate competition and crowding because vOTC is intact bilaterally, but the VWFA, along with the rest of the frontotemporal language system, is right-lateralized. Thus, if competition with visual word form processing drives face lateralization toward the other hemisphere, one might expect left-lateralization for face processing in this group. Alternatively, if faces remained right-lateralized, one might expect behavioral and neural evidence for crowding, such as impaired face processing and/or a weaker or spatially more confined fMRI response to faces in right vOTC.

Our results provide no evidence for a change in the neural representation of faces due to competition by visual word forms: Even though word form processing was right-lateralized in our LHPS participants, their face processing was no less right-lateralized than that of the CTRL participants. They showed no significant reduction in the average strength, spatial extent, or peak strength of the face response in right vOTC, neither relative to baseline nor relative to the response to place stimuli. Inconsistent with the assumption of competition, there were no negative (but rather, positive) correlations between the fMRI response to word forms and faces or its lateralization. Colateralization of face and word form processing in the LHPS group was also not associated with behavioral evidence for crowding: CFMT performance was not significantly worse in the LHPS group compared with the CTRL group, and although our data are consistent with a small performance decrement, it appears to be less pronounced than that observed in patients with early life unilateral lesions to vOTC (De Schonen et al., 2005), patients with a history of childhood hemispherectomy (Granovetter et al., 2022; Simmons et al., 2024), or developmental prosopagnosia (Duchaine and Nakayama, 2006; Dalrymple et al., 2014). Lastly, there were positive correlations between face memory performance and the vOTC response to faces in both participant groups in both hemispheres. Taken together, these results suggest bilateral (if right-lateralized) face processing in both groups. They imply that if word form processing is established in right vOTC, face processing can nonetheless remain right-lateralized without inducing crowding. Thus, the impairments observed after lesions/resections involving vOTC are likely due to loss of face processing territory in the damaged hemisphere rather than to crowding in the remaining hemisphere.

If face and word form processing can colateralize to right vOTC without negative consequences, why do they lateralize to opposite hemispheres in most adults? Computational simulations (Jacobs et al., 1991; Plaut and Behrmann, 2011) demonstrate that maximum efficiency is achieved when distinct resources are allocated to closely related tasks that each have idiosyncratic processing features. Thus, if word forms lateralize to left vOTC, it may increase efficiency if face processing relies more on right vOTC. This idea underlies the graded hemispheric specialization hypothesis (Behrmann and Plaut, 2015), whose supporting evidence includes the observation that in participants with lesions to left or right vOTC, left lateralization for word processing is correlated with right lateralization for face processing (Behrmann and Plaut, 2020). However, several studies in neurologically healthy participants instead found positive rather than negative correlations between activation strength and lateralization for face and word processing (Pinel et al., 2015; Davies-Thompson et al., 2016; Hervais-Adelman et al., 2019; Blauch et al., 2025), as also observed in this study. Thus, the competitive relationship between faces and word forms is not strong enough to emerge consistently across studies.

If there is an efficiency advantage to opposite lateralization of face and word form processing, why did we not observe opposite lateralization in our LHPS group? While left vOTC itself was intact, lesions overlapped with frontal and temporal regions known to encode behaviorally relevant aspects of faces not only in humans but also nonhuman primates (Perrett et al., 1988; Rolls, 2007) and whose connectivity with vOTC has been found to predict the vOTC response to faces (Saygin et al., 2012). Lesions also overlapped with major left-hemisphere fiber tracts whose right-hemisphere counterparts may play a role in face recognition, as reduced face recognition performance in developmental prosopagnosia may be associated with reduced integrity of these tracts (Thomas et al., 2009; although see Gomez et al., 2015 and Song et al., 2015 for suggestions that local connectivity within vOTC rather than long-range connectivity to remote areas may be a key predictor of behavioral performance). Thus, the damaged LH of our LHPS group may have been insufficiently connected to take on the dominant role for face processing, even though left vOTC was intact.

There are also many reasons to assume that face processing would have remained right-lateralized even if the entire LH had been intact. Areas responsive to faces emerge in both hemispheres early in infancy (Kosakowski et al., 2024), and there is a predominance of right-lateralization for faces even in preliterate (De Schonen and Mathivet, 1990; de Haan and Nelson, 1997; Cantlon et al., 2011; De Heering and Rossion, 2015) and illiterate humans (Dehaene et al., 2010; Hervais-Adelman et al., 2019). This suggests influences on face lateralization that are present early in life and prior to reading acquisition. For example, some have suggested that the right hemisphere prefers low-frequency input and tends to integrate across larger/longer spatial and temporal windows (Sergent, 1982; Ivry and Robertson, 1998), rendering it better suited for global (holistic) processing, a hallmark of face perception (Ramon and Rossion, 2011; Richler et al., 2011; Rossion, 2013).

In summary, in the unusual case of LHPS, where visual word form processing is established in the right vOTC, faces can share neural territory with words without compromising neural or behavioral responses. Such unusual functional sharing of right vOTC territory is consistent with other findings from our LHPS participants showing that two aspects of language (sentence processing and emotional prosody—usually lateralized to opposite hemispheres) can share cortical territory in the right (Newport et al., 2022; Martin et al., 2024). Regarding the question of whether the typical vOTC lateralization pattern (faces right, word forms left) is driven by competition between word forms and faces, we conclude that other factors, such as connectivity with ipsilateral brain areas and hemispheric preferences for integrating information at different scales, contribute to right-lateralization of face processing. In our LHPS group, their influence outweighed any pressure that competition might have exerted on face processing to become left-lateralized when visual word form processing was established in right vOTC. The absence of significant impairments in reading and face memory in our stroke participants (which contrast with significant impairments after childhood hemispherectomy) and the positive correlations between behavioral performance and the fMRI response in both left and right vOTC suggest that the cost of having word and face processing lateralized to the same hemisphere is minor as long as both hemispheres can contribute to face processing.

Supplementary Material

This paper contains supplemental material available at: https://doi.org/10.1523/JNEUROSCI.1854-25.2026

Significance Statement.

This study informs our understanding of the causes and consequences of lateralization for visual object processing in ventro-occipital temporal cortex (vOTC) by demonstrating that after a large perinatal stroke to the left hemisphere, faces and words can share right vOTC territory without compromise of neural or behavioral outcomes. This suggests that the lateralization pattern typically observed in neurologically healthy adults (left-lateralization of word forms and right-lateralization of faces) is unlikely to be driven solely by competition for left vOTC territory. It highlights the influence of other factors, such as intrahemispheric connectivity and asymmetrical hemispheric processing preferences, in determining hemispheric lateralization for face processing.