The Complexity of the Visual Word Form Area’s Function
Over the course of history, developments in human culture have dramatically changed our environment and required our brains to adapt. A prime example is a region in the left ventral occipitotemporal cortex that, in literate individuals, becomes specialized for processing written words – the “visual word form area” (VWFA) (Nobre et al., 1994; McCandliss et al., 2003; Dehaene and Cohen, 2011).
While the VWFA’s plasticity during literacy acquisition is well established (Dehaene et al., 2015; Kubota et al., 2019), less is known about how its stimulus selectivity changes in the short term to meet task demands. The nature of the VWFA’s functional specialization remains a topic of debate. Some researchers characterize it primarily as a visual region tuned for a particular category of stimulus (McCandliss et al., 2003; Dehaene et al., 2005), potentially serving as the “orthographic lexicon” that identifies familiar letter strings (Glezer et al., 2009; Dehaene and Cohen, 2011; Yeatman and White, 2021). Others emphasize its integration of linguistic information across sensory modalities (Price and Devlin, 2011; Qin et al., 2021; Dȩbska et al., 2023).
In this study, we contribute to this lively discussion by examining how task demands shape the VWFA’s stimulus selectivity, as well as its communication with the canonical language network. By fully crossing task demands and stimulus types, we aim to provide some resolution to the ongoing debate about this important brain region.
The VWFA’s Selectivity Depends on Top-Down Factors
Like other areas in the ventral visual stream, the VWFA is sensitive to various visual stimulus properties. While it responds above baseline to many types of images, it consistently shows a preference for text over other visual stimuli (Ben-Shachar et al., 2007; Muayqil et al., 2015; White et al., 2023). Its responses are modulated by visual features of words, such as their length (Woolnough et al., 2021), position (Rauschecker et al., 2012), and contrast (Kay and Yeatman, 2017).
Beyond these bottom-up visual factors, the VWFA’s activity is also influenced by attentional allocation and task demands (Mano et al., 2013). Words evoke larger responses when they are attended than when they are ignored (Kay and Yeatman, 2017; White et al., 2019, 2023). There is even evidence for top-down language influences without any visual input, such as during spoken language comprehension (Planton et al., 2019) or reading Braille (Reich et al., 2011; Striem-Amit et al., 2012).
To better understand the interplay between bottom-up and top-down factors that shape the VWFA’s function, we used functional magnetic resonance imaging (fMRI) to examine how its stimulus selectivity changes with different task demands. On each trial, participants viewed a briefly flashed character string that was either a real English word, a pronounceable pseudoword, or a string of unfamiliar “false font” characters (Fig. 1C). The false fonts were matched to the Latin fonts in visual features (size, number of strokes, perimetric complexity, etc.; Vidal et al., 2017; Vildavski et al., 2022).
Participants performed one of three different tasks while viewing these stimuli: detecting real words, detecting color in the characters, or detecting color in the fixation mark. This allowed us to assess the VWFA’s responses under conditions where participants were explicitly trying to read the words, attending to the stimuli without a linguistic task, or ignoring the stimuli altogether.
Key Findings About the VWFA’s Modulation
Our results revealed three primary findings about the VWFA’s response:
-
Inherent Selectivity for Familiar Letter Strings: The VWFA preferred letter strings over unfamiliar characters even when the stimuli were ignored during the fixation task. This is consistent with the idea that the VWFA is a visual region tuned for a particular category of stimulus – familiar orthography.
-
Enhancing and Suppressing Responses Based on Task Demands: Compared to the baseline responses in the fixation task, engaging in the word reading task enhanced the VWFA’s response to words and pronounceable pseudowords (with an even greater increase for pseudowords). However, it suppressed the response to visually matched unfamiliar characters (false fonts).
-
Negligible Effects of Visual Attention Alone: Attending to either type of character string to judge its color had little impact on the VWFA’s response, compared to when the stimuli were ignored. This suggests that strong VWFA activation requires specific stimuli (strings of familiar letters) presented during specific tasks (explicitly trying to read words), and not just visual attention to the stimuli.
Taken together, these findings rule out several hypotheses about what determines the magnitude of VWFA responses:
-
Word-likeness of Visual Stimuli: The VWFA’s selectivity for letter strings over false fonts was greatly magnified during the lexical task, suggesting its sensitivity to lexical attributes is contingent on top-down cognitive influences.
-
Top-Down Boost for Attended Stimuli: The VWFA responded similarly to attended words and ignored words, violating the prediction that attention alone would boost responses to task-relevant text.
-
Automaticity of Word Recognition: The VWFA did not respond more strongly when participants attended to words to judge their color, compared to when words were ignored. Strong VWFA activation requires voluntary effort to read the words.
-
Task Difficulty: While there was a correlation between reaction times and VWFA responses during the lexical decision task, this relationship was much weaker during the other tasks. VWFA responses were best predicted by a model including task and stimulus conditions in addition to reaction times.
The VWFA’s Function Depends on Interaction with Language Networks
Our findings suggest that the VWFA’s activity is determined by an interaction between its inherent stimulus selectivity and top-down input from frontal language regions. The VWFA is maximally activated when letter strings are presented and the observer is voluntarily trying to read them. However, engagement in a lexical task does not always enhance the VWFA’s response; in fact, it suppresses the response to unfamiliar characters.
Functional connectivity analyses revealed that communication between the VWFA and a left frontal “Broca’s area” region increased when participants engaged in the linguistic task, compared to the other tasks. Interestingly, this increased connectivity was observed both when the VWFA’s response magnitudes were enhanced (for text) and when they were suppressed (for false fonts).
One interpretation is that the frontal “Broca’s area” region functions as a control area, applying positive or negative modulations to other parts of the reading network based on the task demands. When letter strings are presented during the lexical task, this region becomes engaged and in turn excites the VWFA and other word-responsive areas. But when false fonts are detected during the lexical task, the frontal region suppresses activity in these regions.
Implications and Future Directions
Our results demonstrate that the VWFA’s function cannot be fully explained by either bottom-up stimulus selectivity or top-down modulation alone. Rather, its responses reflect an interaction between these factors, which is shaped by communication with the broader language network.
These findings open the door to future work investigating the VWFA’s role in more naturalistic reading tasks, as well as comparisons to other specialized parts of the visual system. There is also more to learn about functional subdivisions within the VWFA and surrounding regions, as some previous studies have suggested (Lerma-Usabiaga et al., 2018; Hagoort, 2014).
Ultimately, our study provides an important step toward resolving the ongoing debate about the VWFA’s functional specialization. By fully crossing task demands and stimulus types, we have shown that the VWFA is inherently selective for familiar orthography, but this selectivity is further reshaped by voluntary language processing and interactive feedback from the broader reading network.
Availability of Data and Code
Deidentified raw and processed data have been deposited in an OpenNeuro repository. Processed summary data files, statistical outputs, and analysis code have been shared on Open Science Framework. Additional information can be provided by the corresponding author upon request.
