|Home | About | Journals | Submit | Contact Us | Français|
Neuropsychological and neurophysiological evidence point to a role for the left fusiform gyrus in visual word recognition, but the specific nature of this role remains a topic of debate. The aim of this study was to measure the sensitivity of this region to sublexical orthographic structure. We measured blood oxygenation (BOLD) changes in the brain with functional magnetic resonance imaging while fluent readers of English viewed meaningless letter strings. The stimuli varied systematically in their approximation to English orthography, as measured by the probability of occurrence of letters and sequential letter pairs (bigrams) comprising the string. A whole-brain analysis showed a single region in the lateral left fusiform gyrus where BOLD signal increased with letter sequence probability; no other brain region showed this response pattern. The results suggest tuning of this cortical area to letter probabilities as a result of perceptual experience and provide a possible neural correlate for the ‘word superiority effect’ observed in letter perception research.
Much evidence supports the idea that perceptual systems become selectively efficient at processing inputs that are encountered frequently. In learning a written language, for example, human brains appear to become tuned to the recurring visual patterns of the language represented in its orthographic structure. This is illustrated by the fact that letters embedded in words (such as S in the English word FLASH) or in word-like letter strings (S in FRISH) are more efficiently recognized than letters embedded in unusual letter strings (S in RFHSL) (McClelland and Rumelhart, 1981; Reicher, 1969). Such evidence suggests that normal readers use information about frequently recurring letter combinations, encoded as a result of experience with a specific written language, to more efficiently perceive and identify letters and letter strings.
Neuropsychological evidence for an orthographic processor in the brain comes from patients who exhibit ‘letter-by-letter reading’ after left occipitotemporal brain injury (Binder and Mohr, 1992; Cohen et al., 2003; Leff et al., 2001; Sakurai et al., 2000). Such patients have normal language functions, including good recognition of single letters, but show profoundly impaired processing of letter strings, suggesting focal damage to systems responsible for storing or using orthographic information (Behrmann et al., 1998; Patterson and Kay, 1982; Warrington and Shallice, 1980). This localization is supported by neuroimaging experiments in normal readers, which have identified a region in the lateral left fusiform (occipitotemporal) gyrus that responds more strongly to words and word-like nonwords than to consonant letter strings or nonsense characters (Cohen et al., 2002; Dehaene et al., 2001; Polk and Farah, 2002; Tarkiainen et al., 1999). Several elegant studies showed that this orthographic system employs an abstract code that is unaffected by changes in letter case (Dehaene et al., 2001, 2004; Polk and Farah, 2002).
Although activation in this brain region appears to be related to the presence of orthographic structure in the input stimuli, the details of this relationship have not yet been quantified. Most studies on this topic have demonstrated effects of orthographic structure by comparing words or word-like nonwords (‘pseudowords’) to extremely unfamiliar stimuli such as consonant strings or nonsense characters. Yet letter strings can vary continuously in their degree of orthographic approximation to a familiar language (Miller et al., 1954), and this relative degree of orthographic structure can be defined quantitatively in terms of the frequency of occurrence in the language of letters and short letter sequences (e.g., bigrams and trigrams) within the string. The argument that the left fusiform gyrus processes ‘word forms’ would be strengthened if this region were to show a graded response correlated with the degree of orthographic structure in a string, and particularly if this type of graded response was unique to this brain region.
We addressed this issue by recording blood oxygen-level-dependent (BOLD) signals with functional magnetic resonance imaging (fMRI) while English-language readers processed letter strings varying in sublexical orthographic familiarity. Items were nonwords ranging gradually from very unfamiliar consonant letter strings to strings composed of letter sequences occurring relatively frequently in English (Table 1). During fMRI, participants were asked to perform a nonlinguistic feature detection task to divert attention from the linguistic properties of the stimuli and to reduce overt attempts at pronunciation. Our aim in employing this task was to focus on automatic orthographic processes that might provide a neural correlate for increased perceptual efficiency of familiar letter sequences.
Participants were 30 healthy, literate adults (15 women), aged 18–49 years, with no history of neurological disease or learning disability. Years of education ranged from 12 to 23 (mean, median and mode=16). All were right-handed on the Edinburgh handedness inventory (Oldfield, 1971), spoke English as a first language, and had normal or corrected-to-normal vision. All participants provided written informed consent and were paid an hourly stipend.
The stimuli were 5-letter nonwords varying in sublexical orthographic familiarity (Miller et al., 1954). There were four conditions, each consisting of 144 items. Half of the items in each condition contained an ‘ascender’ feature in one of the letters. This term denotes a vertical line rising above the midpoint of the letter space, as occurs in the lowercase letters b, d, f, h, k, l, and t. Dots above the letters i and j were not counted as ascenders. Ascenders appeared with equal probability at all letter positions in all conditions.
Stimuli were constructed by random sequential letter selection. In the least familiar condition, labeled CON, strings were created by random sequential selection of consonant letters. Items containing medium- or high-frequency bigrams or trigrams (calculated from English word frequencies in the CELEX database; Baayen et al., 1995) were excluded in an effort to keep orthographic familiarity as low as possible in this condition. In the other conditions, strings were created by random sequential selection of consonants or vowels. A large pool of these strings was generated, and a measure of sublexical orthographic familiarity, the mean positional bigram frequency (MPBF), was computed for each item. The position-specific frequency of a bigram B in any given string of length N is the frequency with which B occurs at the same sequential position among all words of length N. This quantity, the positional bigram frequency (PBF), is computed by adding the word frequencies of all words of length N containing B at the index position. The average of the PBFs of all bigrams in the string is the MPBF. Three familiarity conditions, labeled CV1, CV2, and CV3, were created from the larger pool of randomly generated strings by randomly selecting stimuli in three logarithmically related MPBF ranges. MPBF values for each condition are given in Table 1. For comparison, the mean MPBF of all 5-letter English words in the CELEX database (n=4703) is 1145 (99% confidence intervals=1080–1210). That is, the orthographic familiarity (as measured by MPBF) of even the most familiar items used in the experiment was far less than a typical real word.
Although we used MPBF to operationally define orthographic familiarity, single letter, bigram, and trigram probabilities tend to be highly correlated, and these quantities were somewhat correlated in the overall item list used in this study (MPLF vs. MPBF, R2=0.344; MPBF vs. MPTF, R2=0.187). Thus, we have not attempted to parcel out brain responses as a function of sequence fragment length. Conclusions regarding effects of the stimulus manipulation therefore apply to sublexical orthographic familiarity as a general factor and not to bigram frequency per se.
Stimuli were computer generated using Psyscope software (Cohen et al., 1993), which also recorded accuracy and response time (RT) data. A liquid crystal display projector was used to rear-project the stimuli onto a screen located near the participant's feet, which was viewed through prism lenses. Letter strings were presented in white lowercase Geneva font on a black background and subtended an average horizontal visual angle of 2.7°. A letter string was presented every 2 s at the center of the field of view, with a duration of 1 s. A fixation cross appeared during each 1-s interstimulus interval. Stimuli were blocked by condition, with 12 trials per block. Two trials occurred during each 4-s image volume acquisition (see below). There were 12 blocks per condition, with condition order randomized across four scanning runs. In addition, 12 blocks of baseline were randomly interspersed among the task blocks. During the baseline, a string of five dots () was presented at the same duration and frequency as the letter strings, and participants were instructed simply to look at the dots.
Participants were asked to press a key with the index finger of the left hand, as quickly and accurately as possible, if the string contained an ascender letter (Price et al., 1996). No response was required for items without an ascender. This task was identical across all conditions.
MRI data were acquired on a GE Signa 1.5 Tesla scanner (GE Medical Systems, Milwaukee, WI) using a 3-axis, local gradient coil with a built-in transmit-receive RF coil (Medical Advances, Inc., Milwaukee, WI). High-resolution, T1-weighted anatomical reference images were acquired as a set of 124 contiguous sagittal slices (0.9375×0.9375×1.2 mm) using a spoiled-gradient-echo sequence (“SPGR”, GE Medical Systems, Milwaukee, WI). Functional imaging used a gradient-echo echoplanar sequence with the following parameters: 40 ms echo time, 4 s repetition time, 24 cm field of view, 64×64 pixel matrix, in-plane voxel dimensions 3.75×3.75 mm, and slice thickness 6 mm. There were 21–23 contiguous sagittal slice locations covering the entire brain, the number depending on brain width. Four sets of time series echoplanar imaging runs were acquired, each composed of 94 whole-brain image volumes.
All image analysis was done with the AFNI software package (available at http://afni.nimh.nih.gov/afni)(Cox, 1996). Motion artifacts were minimized by within-participant registration of echoplanar image volumes. Estimates of the three translation and three rotation movements at each point in each time series were computed during registration and saved. The first four images of each series, during which spin relaxation reaches an equilibrium state, were discarded, and the mean, linear, and second-order polynomial trends across time were removed on a voxel-wise basis from the remaining 90 image volumes of each series.
Multiple regression analysis of the combined image time series (360 volumes) in each participant was performed using an event-related approach. The reason for adopting an event-related analysis was to increase sensitivity to orthographic familiarity effects by modeling the variation in orthographic familiarity within, as well as between, conditions. Orthographic familiarity was modeled by convolving the demeaned MPBF of each stimulus with a canonical hemodynamic response (the sum of two gamma functions with delay, dispersion, and amplitude parameters of 6, 0.8, 1 and 16, 1, −0.1, respectively). Letter-string events were coded separately from fixation events to obtain a map of areas activated in common by all letter strings. Translation and rotation movement parameters estimated during image registration were included in the model to remove residual variance associated with motion-related changes in BOLD signal. The resulting parametric maps included contrast coefficient (magnitude), variance and t statistic parameters (relative to the baseline) showing activation changes due to any letter-string event and to variation in MPBF.
The resulting contrast coefficient maps from each participant were linearly resampled in standard stereotaxic space to a voxel size of 1 mm3 and spatially smoothed with a 6-mm full-width-half-maximum Gaussian kernel to compensate for variance in anatomical structure. The smoothed coefficient maps were then subject to a random effects analysis comparing the coefficient values to a null hypothesis mean of zero across participants. The resulting group activation maps were thresholded at a voxel-wise, uncorrected, 2-tailed probability of p<0.00001 (|z-deviate|≥4.42). Finally, Monte Carlo simulation was used to estimate the chance probability of spatially contiguous clusters of voxels passing this threshold. Clusters smaller than 200 μl in the group maps were removed, resulting in a corrected 2-tailed probability threshold of p<0.05 for each group map.
To control attentional state and minimize explicit attempts to pronounce the letter strings, participants were asked to press a key whenever a stimulus contained an ‘ascender’ letter (b, d, f, h, k, l, or t) (Price et al., 1996). Mean accuracy on this task across all conditions was 99.1% correct (SD=1.5). Mean response time was 726 ms (SD=100). Repeated-measures ANOVA demonstrated a small effect of condition on accuracy (F(3, 87)=3.00; p=0.035), though none of the post hoc pair-wise comparisons showed significant differences. There was a robust effect of condition on RT (F(3, 87)=20.17; p<0.0001). RT increased monotonically by 30 ms across the four conditions (Table 2), indicating that combinatorial letter information was processed automatically and interfered slightly with efficiency on the feature detection task.
Collapsed across all conditions, the feature detection task activated extensive and approximately symmetric regions in visual association cortex, intraparietal sulcus, anterior cingulate gyrus, supplementary motor area and midline cerebellum relative to baseline (Fig. 1). Right-lateralized activations occurred in the caudate nucleus, putamen, internal capsule and primary sensorimotor cortex, consistent with execution of the motor response with the left hand. Left-lateralized responses occurred in the lateral fusiform gyrus and premotor cortex. Stereotaxic coordinates for all activation peaks are provided in Appendix A.
The main analysis searched for voxels in which BOLD activation increased as a function of orthographic familiarity, operationally defined by MPBF. As shown in Fig. 2A, four activation clusters were modulated by MPBF. The largest of these was located in the left lateral fusiform gyrus, with smaller foci in the intraparietal sulcus bilaterally and the left precentral sulcus (see Appendix B for coordinates of peak activations). The left fusiform and bilateral intraparietal sulcus areas were also activated across all conditions relative to fixation (see corresponding axial slices at z=−12 and z=+40 in Fig. 1), whereas the left precentral sulcus cluster modulated by MPBF was slightly inferior and anterior to the left premotor activation observed in the overall map.
Considering the fact that increasing orthographic familiarity also led to longer processing times in the ascender detection task, it is likely that some or all of the activation associated with increasing MPBF actually reflects processes associated with increased task difficulty, such as increased demands on attention and response selection systems. To tease apart these potentially confounded effects, we conducted a second analysis in which each participant's block-wise, demeaned response time (i.e., average RT for each 12-trial block), convolved with the canonical hemodynamic function, was included in the analysis to model effects related to task difficulty and attention. This second analysis thus identifies brain areas where variation in BOLD signal is uniquely accounted for by orthographic structure. A single activated cluster (volume=1137 μl) with two activation peaks was identified in the lateral left fusiform gyrus (Fig. 2B, Table 3). This cluster was nearly identical in size and location to the fusiform gyrus cluster observed in the analysis without RT. Thus, although the variation in BOLD signal in intraparietal and precentral regions can be accounted for largely by variation in response time, activation in the left lateral fusiform gyrus correlates more closely with MPBF and is largely independent from RT. Notably, the coordinates of the main peak in this cluster match almost perfectly the average peak coordinates commonly given for the ‘visual word form area’ (−43, −57, −12) (Cohen and Dehaene, 2004).
Finally, a region-of-interest (ROI) analysis was conducted to determine more precisely the relationship between BOLD activation and orthographic familiarity, as operationally defined by MPBF. The voxel cluster in Fig. 2B was used as an ROI applied to the individual raw BOLD signal time series, and mean BOLD values were determined for each condition relative to baseline. These mean activation values in each condition were then averaged across participants. A plot of these grand means against MPBF for each condition revealed a monotonically increasing BOLD response with increasing MPBF in the lateral left fusiform gyrus (Fig. 3). Interestingly, the function decelerates, such that the incremental increase in BOLD signal lessens at higher levels of familiarity.
The results contribute in several ways to our understanding of brain mechanisms underlying orthographic processing. First, unlike many previous experiments that focused on processing of words and word-like nonwords, the stimuli used in this experiment were not word-like or easily pronounceable. The focus here was on letter probabilities rather than on whole-word lexical forms. The results demonstrate sensitivity of the left lateral fusiform gyrus to familiar over unfamiliar letter sequences, even when these sequences do not resemble words. The results thus suggest perceptual tuning of the left fusiform gyrus to these letter probabilities as a result of sensory experience. Similar suggestions have been made regarding sensitivity of nearby regions in the ventral extrastriate cortex to faces, environmental scenes, colors and other visual stimuli (Aguirre et al., 1998; Epstein and Kanwisher, 1998; Gauthier et al., 1999; Kanwisher et al., 1997; McKeefry and Zeki, 1997) (for reviews on the relative locations of these putative specialized areas, see Hasson et al., 2003; Kanwisher et al., 2001; McCarthy, 2000).
A second point is that perceptual tuning to letter probabilities is not necessarily related to categorization of single letters. Single letters must be recognized despite wide variations in size, font, case and retinal location. Several researchers have documented insensitivity of the ‘visual word form area’ to these variables and proposed that one of the functions of this region is to extract an invariant, abstract representation of letter identity (Dehaene et al., 2001, 2004; Polk and Farah, 2002). In the present study, we focused on the probability of letter combinations rather than individual letters and found sensitivity to letter sequence familiarity in a region that appears to be identical in location to the visual word form area (Cohen and Dehaene, 2004). Thus, in addition to its proposed role in computing invariant letter representations, we suggest that this region processes language-specific orthographic structure represented at the level of letter combinations, as has been proposed by several previous authors (Dehaene et al., 2005; Whitney, 2001). Whether these two putative functions are carried out by the same neurons, by different but spatially overlapping neurons, or by spatially distinct but neighboring neuron clusters must be addressed in future studies by comparing the two processes directly using carefully constructed task manipulations and higher resolution imaging techniques.
A third point is that the orthographic familiarity effect arose even under task conditions requiring participants to attend to nonlinguistic visual features of the stimuli. That is, the explicit task during fMRI did not depend on processing of letter probabilities and was identical across all conditions, suggesting that the processing of sublexical orthographic structure occurs automatically, without the need for directed attention to this information. This task, used successfully by other researchers to demonstrate automatic processing of linguistic stimuli (Paulesu et al., 2000; Price et al., 1996), was used here in a dual effort to control attention across conditions and to inhibit as much as possible any high-level processing of the stimuli, such as conscious attempts at pronunciation. This method is in contrast to many prior studies of orthographic processing, in which participants were encouraged to silently read the stimuli, presumably resulting in extensive activation of phonological and semantic codes in addition to orthographic processes.
Given the amount of experience normal readers gain with letter sequences and the striking level of reading fluency that typically results, it is not surprising that the visual recognition system would develop a degree of perceptual expertise for familiar letter combinations (Gauthier et al., 1999). Tuning the system to high-probability sequences allows the perceptual mechanism to represent the input in larger ‘chunks’, conferring a greater degree of parallel processing and increasing perceptual efficiency. Chunking of letter sequences is presumably also useful for solving problems posed by irregularities of grapheme-to-phoneme conversion, such as the fact that ‘tion’ in English should usually be pronounced as ‘shun’, that some bigrams such as ‘ch’ and ‘th’ correspond to single phonemes, and that a word-initial ‘g’ should usually be pronounced /j/ when followed by ‘e’ but /g/ when followed by ‘a’, ‘o’, or ‘u’ (Seidenberg and McClelland, 1989).
As mentioned earlier, there is considerable behavioral evidence of experience-dependent expertise with letter strings. Facilitation of letter recognition by orthographic familiarity, known as the ‘word superiority effect’, is one of the more extensively documented phenomena in cognitive science (Carr, 1986; McClelland and Rumelhart, 1981; Reicher, 1969). Letters embedded in orthographically familiar strings (words or pseudowords) are processed more efficiently during very brief presentations than are letters embedded in unfamiliar strings or even single letters. Moreover, carefully controlled experiments suggest that this phenomenon occurs at a relatively early perceptual stage and is not an artifact of post-perceptual inference or phonological memory. Our results suggest the intriguing possibility that the neural basis for this phenomenon may be coding of orthographic structure in the lateral left fusiform gyrus.
This hypothesis would be strengthened if a similarity could be demonstrated between the pattern of BOLD responses we observed in the left fusiform gyrus and actual performance on a letter perception task. We sought such evidence in a follow-up behavioral study performed outside the scanner, using the letter identification task developed by Reicher (1969) and Wheeler (1970) and the same set of stimuli as in the fMRI study. Ten healthy, fluent readers of English (all right-handed adults, mean education 18 years) viewed each stimulus briefly, preceded and followed by a pattern mask (#####). Participants were seated in front of a computer screen with their viewing distance fixed via a chin rest. A fixation cross at the center of the screen indicated the beginning of each trial. Following a random interval of 1000–1500 ms, the pattern mask was presented for 250 ms, followed by brief presentation of the letter string stimulus. The pattern mask then appeared again, this time with two letter choices presented above and below one of the character positions in the mask. Participants were required to indicate, by a 2-alternative forced choice button response, which of the two choices matched the letter that had just been presented at the indicated position. Only letters at positions 2, 3 and 4 were assessed. The possibility of ‘guessing’ correctly based on a priori knowledge of letter probabilities was eliminated by matching the incorrect (foil) letter choice on each trial to the correct choice on both single letter probability (e.g., the target ‘e’ was never paired with the foil ‘q’) and the positional probability of flanking bigrams that would result from substitution of the foil for the target. The duration of stimulus exposure was determined individually for each participant using 120 training stimuli (5-letter random consonant strings different from the experimental strings). Exposure duration during training was systematically increased or decreased (in 16-ms increments) to produce an overall accuracy of approximately 70% with the consonant strings. The mean of the durations selected for the 10 participants was 228 ms (range 156 to 306).
Repeated-measures ANOVA showed a strong effect of condition on accuracy (F(3,36)=4.97, p=0.006) (Fig. 4). These results confirm that the orthographic familiarity manipulation resulted in more efficient letter perception across conditions. Notably, the increase in performance decelerates at higher levels of familiarity, closely mirroring the pattern of BOLD signal increase in the lateral left fusiform gyrus. Indeed, these mean accuracy values are highly correlated with the mean BOLD signals measured across the four conditions (R2=0.936, p=0.032).
Although these correlations between letter perception and BOLD signal cannot be considered proof of a causal link, they are consistent with such a link, suggesting that activation of neural ensembles in the lateral left fusiform gyrus that are tuned to familiar letters and letter combinations may underlie the word superiority effect. Also consistent with this hypothesis is the location of these BOLD effects in the ventral extrastriate visual pathway, which would be expected given the early perceptual locus of the effect. Finally, it is noteworthy that the lateral left fusiform gyrus was the only brain area to show this pattern of BOLD changes, suggesting that it may play a relatively unique role in the processing of orthographic structure.
Although we propose here that perceptual expertise for familiar letter sequences could underlie the word superiority effect, an alternative and widely held view is that this effect is due to positive feedback from lexical representations. In the well-known interactive activation model of McClelland and Rumelhart, single letter representations exchange activation directly with localist whole-word nodes (McClelland and Rumelhart, 1981). There is ample evidence, however, that a wide range of orthographic processes can be accounted for equally well by models that use sublexical representations (Massaro and Cohen, 1994; Seidenberg and McClelland, 1989; Van Orden et al., 1990). In the interactive activation model, for example, the word superiority effect could be explained by probabilistic letter-sequence nodes (rather than word nodes) providing feedback to letter nodes. Recent priming experiments provide independent support for such letter-sequence representations (Grainger and Whitney, 2004; Perea and Lupker, 2003; Whitney, 2001). Although it may be possible to account for the perceptual effects observed in the present study by a mechanism involving partial activation of a cohort of lexical codes, we suggest that the stimuli were sufficiently unlike words that any such activation would have been very minimal. On the other hand, the strong manipulation of sublexical familiarity provides what seems to us a simpler and more parsimonious explanation of the observed facilitation effects.
The proposal that a part of the ventral visual stream shows expertise for processing letter sequences does not mean that this is the only function of this region, nor does it mean that other regions do not also participate in the complex behavior of reading. Although this region of the lateral fusiform gyrus has been called a ‘visual word form area’, it appears to respond to a range of other visual stimuli as well, including object pictures, colors and nonobject shapes (Price and Devlin, 2003). Thus, this region may participate in the recognition of many or even most visual inputs and yet also display perceptual expertise for letter strings. Several authors have reported stronger activation in this general region (and surrounding extrastriate cortex) for reading word-like pseudowords compared to words (Binder et al., 2005a; Kronbichler et al., 2004; Mechelli et al., 2003; Price et al., 1996; Xu et al., 2001). It seems very likely that this difference is related to the longer processing time and visual attention required for reading pseudowords, as the same region showed activation correlated with RT during overt word and pseudoword naming (peak at −42, −55, −10) (Binder et al., 2005a) and during a visual lexical decision task (peak at −42, −52, −17) (Binder et al., 2005b) and was activated by covert shifts of visual attention in an experiment using meaningless geometric shapes to cue the spatial location of targets (peak at −45, −69, −6) (Gitelman et al., 1999). Thus, this general region of visual extrastriate cortex appears to be sensitive to task difficulty and attentional modulation. Several studies have even shown responses in this region to spoken speech stimuli, though this activation seems to occur with specific linguistic tasks and not with passive listening (Binder et al., 1997, 2000; Booth et al., 2002; Dehaene et al., 2002), prompting some investigators to propose ‘top-down’ activation of orthographic codes in this region as an explanation for responses to speech input (Cohen et al., 2004; Dehaene et al., 2002).
Another alternative account of the current results is that the fusiform gyrus activity represents an early stage of phonological processing of the stimuli, such as orthographic-to-phonological conversion. Phonotactic familiarity (ease of pronunciation) and orthographic familiarity are inevitably correlated; items that are more orthographically familiar also tend to be more easily pronounceable. Although the feature detection task used during fMRI did not encourage phonological access and should have acted to inhibit such processing, it is likely that the letter strings were processed at higher linguistic levels to some degree. The small increase in RT across conditions is consistent with this claim, and suggests that as the letter strings became orthographically more familiar, more neural resources may have been diverted to task-irrelevant processes that interfered with the feature detection task. Although this interference effect was somewhat surprising, a prior study by Price et al. (1996) using the same task and a similar orthographic manipulation produced similar findings. In that study, the ascender detection task took significantly longer for words than for false font, and both words (542 ms) and pseudowords (569 ms) produced longer RTs than consonant strings (530 ms), though these latter differences were not statistically reliable for the small sample sizes used (4–10 subjects).
We propose that these RT effects are due to more extensive linguistic (orthographic and phonological) processing of the orthographically familiar items, which interferes with explicit perceptual analysis of individual letter features and formulation of a response based on this analysis, much like phonological access from print interferes with color naming in the Stroop task (Macleod, 1991). Based on much prior work linking the left premotor region to phonological processing (Fiez et al., 1999; Mechelli et al., 2003; Xu et al., 2001), it is likely that the orthographic familiarity effect observed in the left precentral sulcus (Fig. 2A) reflects a greater degree of phonological processing in this region as orthographic familiarity increases. Similarly, we hypothesize that the bilateral intraparietal sulcus activation correlated with orthographic familiarity represents additional demands on visual attention as the increasing phonological processing leads to interference with the feature detection task (Corbetta and Shulman, 2002; Gitelman et al., 1999; Kastner and Ungerleider, 2000). Whereas activation of these regions was correlated with orthographic structure in the stimuli, this activation was even more closely related to individual variation in RT, as demonstrated by the fact that these regions showed no significant modulation by MPBF after adding RT as an explanatory variable (Fig. 2B).
Although there was evidence to suggest additional phonological processing with increasing orthographic familiarity in some brain regions, it appears unlikely that modulation of the left fusiform gyrus activation was phonologically mediated. After accounting for variance due to RT, the left fusiform gyrus still showed robust effects of orthographic structure. As mentioned previously, these effects are occurring in the ventral left visual processing pathway—precisely where they would be expected to occur on the basis of anatomical and human lesion data. Attributing these effects to phonological access is less consistent with available lesion data, which link phonological impairments – including phoneme recognition deficits, phonological dyslexias, phonemic paraphasia and phonological dysgraphia – to lesions in the superior temporal lobe and temporoparietal junction (Alexander et al., 1992; Beauvois and Derouesne, 1979; Caplan et al., 1995; Damasio and Damasio, 1980; Marin, 1980) rather than to lesions in the ventral visual pathway.
Notably, the letter strings employed here did not produce activation (Fig. 1) or modulation of activation (Fig. 2) in the posterior superior temporal lobe, a region often activated by words and pronounceable nonwords in comparison to unpronounceable strings (Booth et al., 2002; Herbster et al., 1997; Howard et al., 1992; Indefrey et al., 1997; Price et al., 1996; Tagamets et al., 2000). We propose that this posterior superior temporal region (centered on the posterior superior temporal sulcus) is sensitive to the pronounceability of letter strings, consistent with its hypothesized role in phonological access and visual–auditory integration (Callan et al., 2005; Hickok et al., 2000; Indefrey and Levelt, 2000; Van Atteveldt et al., 2004; Wise et al., 2001). The modulation here of the left fusiform gyrus independently from this more dorsal temporal region suggests at least a partial independence of orthographic and phonological processes, even if these systems normally operate in a highly interactive way during visual word recognition.
This research was supported by the National Institute of Neurological Diseases and Stroke grant R01 NS33576, the National Institute of Mental Health grant P01 MH51358 and the National Institutes of Health General Clinical Research Center grant M01 RR00058. We thank B. D'Angelo, E.T. Possing, J.A. Frost, T.E. Prieto and B.D. Ward for technical assistance.
|Region||Brodmann area||Left hemisphere||Right hemisphere|
|x||y||z||Z score||x||y||z||Z score|
|Inf. occipital g.||18||−30||−93||−12||6.15||26||−96||−10||6.56|
|Inf. occipital g.||18||−40||−85||−13||5.72||36||−89||−8||6.97|
|Ant. cingulate g.||24/32||−10||10||37||5.11||14||8||30||5.65|
|Region||Brodmann area||Left hemisphere||Right hemisphere|
|x||y||z||Z score||x||y||z||Z score|