fMRI correlates of cortical specialization and generalization for letter processing

Abstract

The present study used functional magnetic resonance imaging to examine cortical specialization for letter processing. We assessed whether brain regions that were involved in letter processing exhibited domain-specific and/or mandatory responses, following Fodor's definition of properties of modular systems (Fodor, J.A., 1983. The Modularity of Mind. The MIT Press, Cambridge, MA.). Domain-specificity was operationalized as selective, or exclusive, activation for letters relative to object and visual noise processing and a baseline fixation task. Mandatory processing was operationalized as selective activation for letters during both a silent naming and a perceptual matching task. In addition to these operational definitions, other operational definitions of selectivity for letter processing discussed by [Pernet, C., Celsis, P., Demonet, J., 2005. Selective response to letter categorization within the left fusiform gyrus. NeuroImage 28, 738–744] were applied to the data. Although the left fusiform gyrus showed a specialized response to letters using the definition of selectivity put forth by [Pernet, C., Celsis, P., Demonet, J., 2005. Selective response to letter categorization within the left fusiform gyrus. NeuroImage 28, 738–744], this region did not exhibit specialization for letters according to our more conservative definition of selectivity. Instead, this region showed equivalent activation by letters and objects in both the naming and matching tasks. Hence, the left fusiform gyrus does not exhibit domain-specific or mandatory processing but may reflect a shared input system for both stimulus types. The left insula and some portions of the left inferior parietal lobule, however, did show a domain-specific response for letter naming but not for letter matching. These regions likely subserve some linguistically oriented cognitive process that is unique to letters, such as grapheme-to-phoneme translation or retrieval of phonological codes for letter names. Hence, cortical specialization for letters emerged in the naming task in some peri-sylvian language related cortices, but not in occipito-temporal cortex. Given that the domain-specific response for letters in left peri-sylvian regions was only present in the naming task, these regions do not process letters in a mandatory fashion, but are instead modulated by the linguistic nature of the task.

Introduction

Alphabetic systems are a relatively recent development from an evolutionary perspective, having first emerged around 1500 BC (Driver, 1976). Because cortical specializations for reading and other modern-day cognitive capacities have not had enough time to evolve (Tooby and Cosmides, 2000), the capacity for recognizing letter forms may have exploited neural circuitry that was already in place for recognizing object form (Joseph et al., 2003). Moreover, given that alphabetic systems emerged from pictographic systems (Driver, 1976) the same neural substrates that recognize objects (e.g., the fusiform gyrus) may extend to recognizing highly stylized symbols that now compose alphabetic systems. However, it is entirely possible that a brain region (or a cortical network) becomes specialized for letter processing due to experience with the domain (Polk and Farah, 1998, Polk et al., 2002).

The neuroimaging evidence to date seems to support a specialized letter processing region in the left fusiform gyrus or in left extrastriate cortex. This general region is consistently activated for single letters or non-pronounceable letter strings as compared with other categories (Cohen et al., 2002, Dehaene et al., 2002, Flowers et al., 2004, James et al., 2005, Jessen et al., 1999, Longcamp et al., 2003, Pernet et al., 2005, Polk and Farah, 1998, Polk et al., 2002, Price et al., 1996, Puce et al., 1996, Sergent et al., 1992, Tagamets et al., 2000). Cohen and colleagues (Cohen et al., 2002, Dehaene et al., 2002) have referred to a left fusiform region as the visual word form area (VWFA) based on the finding that alphabetic strings activate the region more strongly than do checkerboard patterns. They have demonstrated that activation in this region is not driven by low-level visual information nor is it retinotopically organized. In fact, this region does not appear to be sensitive to letter case, implying that it processes abstract letter form (but see Gauthier et al., 2000). In addition, the response in this region is modulated by task demands. As an example, Gros et al. (2001) showed that the left fusiform gyrus showed an adaptation response to an ambiguous stimulus (e.g., a shape that could be perceived as either a circle or the letter “O”) only when it was primed by letters but not when primed by shapes. Hence, the left fusiform gyrus responded to an abstract representation of letter form that was driven by top-down information about the visual category rather than driven by low-level visual information in a bottom-up fashion. Importantly, this body of work has outlined factors that may modulate VWFA responses, but the question still remains as to whether VWFA processing is unique to letters or whether the processing in this region is shared with other visual categories.

Joseph et al. (2003) showed no letter-selective activation in the fusiform gyrus during passive viewing and silent naming of individual letters in an fMRI study. Instead, the left fusiform gyrus was equally activated by letter and object naming, as revealed by a type of conjunction analysis (Joseph et al., 2002, Nichols et al., 2005). Joseph et al. (2003) suggested that this region does indeed process abstract representations of form as suggested by others (Cohen et al., 2002, Dehaene et al., 2002, Gros et al., 2001, Pernet et al., 2005), but such representations are not unique to letters. Additional support for this conclusion is suggested by studies showing no differential activation when words are compared with objects (Jessen et al., 1999, Joseph et al., 2003, Price et al., 1996, Sergent et al., 1992, Tagamets et al., 2000).

The goal of the present study is to examine whether brain regions that appear to respond specifically to visual letters encapsulate processing that is unique to letters. The present analysis will use two properties of modularity proposed by Fodor (1983) as a starting point to shape an operational definition of cortical specialization. The property of domain specificity emerges when a particular type of input (e.g., human speech or face recognition) makes special processing demands that cannot be accommodated by existing input systems. Domain specific systems only represent a narrow range of properties from the environment and they require specialized computations to process that narrow range of input. The property of mandatory processing implies that a module will automatically process the information to which it is specifically tuned in a bottom-up or data driven fashion. In other words, the module will process the specific domain regardless of other task demands. These two properties are considered by some (Garfield, 1987) to be essential to the concept of modularity.

In the present study, we isolate a number of different response profiles using a conservative hypothesis testing approach (Joseph et al., 2002) and use these profiles as operational definitions for domain specificity and mandatory processing (Fig. 1). We suggest that selective responses (Fig. 1A) provide the strongest evidence for domain-specificity. Selective activation for letter processing is defined as a statistically significant response for letters relative to all other conditions (objects, visual noise and a baseline task of visual fixation), but no statistically significant response among objects, noise and baseline. Hence, selective activation reflects processing that is unique to letters and not shared by other visual categories manipulated in a given study. With a selective response, some of the comparisons are required to be statistically equivalent and others are required to be statistically non-equivalent. However, statistical equivalence of certain conditions must co-occur with statistical non-equivalence of another subset of conditions in the same voxel. Hence, a single test of no differences is not sufficient in the logical combination tests, but is only part of a larger prediction that includes positive results. Specifically, these individual comparisons are only valid in the context of a significant main effect or interaction, much like conducting post hoc comparisons.

A selective response can be contrasted with a preferential response, in which the experimental condition of interest yields a greater response than at least one other control condition and/or baseline in a given brain region. Graded responses (described in Appendix) are similar to preferential responses, but slightly more stringent. With both preferential and graded responses, the control condition(s) could yield a greater-than-baseline signal, which does not occur with selective responses (see Fig. 1B); consequently, the region subserves some type of processing that is shared by both conditions, but to varying degrees.

We also isolate a conjoined response for letters and objects in which letters and objects produce a statistically greater response than fixation and noise, but objects and letters are statistically equivalent (Fig. 1C). The statistical equivalence of two conditions implies that the region is equally recruited in both conditions and likely subserves some form of processing that is shared by both conditions (see Friston et al., 2005, Joseph et al., 2002, Nichols et al., 2005, Price and Friston, 1997 for discussions of conjunction analysis). Again, statistical equivalence is only valid in the context of a main effect of condition, as discussed above. A conjoined response provides evidence for cortical generalization rather than specialization.

The property of mandatory processing for letters has also been addressed previously. Pernet et al. (2005) and Pernet et al. (2004) describe a response in which a brain region responds to the preferred category in two different tasks (Fig. 1D), which we refer to as a task-independent response. In this example, letters produce a greater response than noise and baseline in both a naming (gray bars) and a matching task (white bars). A task-independent response embodies both domain-specificity and mandatory processing in that the response is specific to the domain of interest (preferential) and is automatic whenever the information is presented regardless of the task (mandatory). The present analysis will also modify the definition of task-independence to use selective rather than preferential responses as a more restrictive test of domain-specificity. This selective + task-independent response is shown in Fig. 1E.

Pernet et al. (2005) defined another response profile that isolates domain-specific responses that are not mandatory. Here, the preferred category induces a response in only one task (Fig. 1F). Pernet et al. (2005) termed this activation pattern as “selective,” but to avoid confusion with Joseph et al.'s, (2002) use of the term, we adopt the term task-dependent. A task-dependent response does not guarantee that the region would be specialized for letter processing because the definition implies that other categories can activate the region greater than baseline. In addition, even if a letter-preferential response does not emerge in the matching task, it is possible that the matching task significantly activates the region above baseline in a non-preferential manner. Hence, those regions in which matching activation is significantly greater than baseline would need to be excluded before concluding that a region is involved in letter naming specifically. Fig. 1G illustrates a selective + task-dependent response, in which the response to letters in the naming task is selective and the matching task does not activate the region greater than the baseline condition.

The present study combines the data of Joseph et al. (2003) with a new dataset that uses the same stimuli (i.e., letters, objects, visual noise and a visual fixation baseline), but employs a perceptual matching task instead of passive viewing and silent naming. The analysis will apply the various definitions of cortical specialization (Figs. 1D–G) and generalization (Fig. 1C) to this combined data set. Although the focus is on the left fusiform gyrus throughout this analysis given its proposed role as a letter processing region, we also report on any other regions that satisfy the different definitions of cortical specialization or generalization.