Note: Symposia have not yet been scheduled.
From vision to vocabulary: a multi-disciplinary investigation of reading
Chairs: Dr Ian Chris Baker1, Joseph Salvo1; 1National Institute of Mental Health
Presenters: Oscar Woolnough, Elizabeth Schotter, Gina Kuperberg, Joseph Salvo
Reading is a complex process that requires successful integration of several difficult abilities. Precise eye movements flood the visual system with written information, while rapid visual and linguistic processing provide real-time interpretations that can in turn influence eye movements. Despite the prevalence of reading ability, there are open questions regarding how fluent readers successfully combine these elements, and how each stage of reading may or may not influence the others. Answers may be found through understanding how different processing components contribute to reading, from the link between eye behavior and comprehension, to the functional connections that share information between sensory and linguistic areas, to the timing of each stage and influence of context. This symposium will discuss research using tools including fMRI, M/EEG, and eye tracking to explore the neural building blocks of reading on a range of temporal and spatial scales.
Presentations
A causal, spatiotemporal map of the reading network
Oscar Woolnough1; 1University of Texas Health Science Center at Houston
Fluent literacy is a crucial skill in modern civilisation and forms the basis of our broader education. This skill is critically dependent on the dynamic interactions of a distributed network across the brain, with disruption of any of these network hubs resulting in reading deficits. To investigate the neural basis of reading we use large population intracranial recordings and direct cortical stimulation, causally probing brain-behaviour associations across the literacy network at high spatiotemporal resolution. Mid-fusiform cortex is a critical early hub for reading, showing early sensitivity to word frequency and semantics of written words before propagating this information back to earlier visual cortex and to higher-order language regions. Mid-fusiform shows dynamic network interactions with hubs across frontoparietal cortex to form a distributed literacy network. This network facilitates both comprehension and production of written words, allowing access to their meanings, sounds, and spellings. Direct cortical stimulation of mid-fusiform results in selective reading disruptions, dissociable from other visual language functions. Behavioural disruptions resulting from mid-fusiform stimulation include full reading arrests, reading delays, semantic errors, letter intrusions and transpositions, and modulations of eye movements. This work demonstrates the importance of mid-fusiform cortex as a critical, early, causal reading hub that shows dynamic interactions across the distributed literacy network.
The eye-brain link in reading: Coordination between slow language processes and rapid eye movements
Elizabeth Schotter1, Brennan Payne2, Sara Milligan1, Allyson Copeland2, Jack Silcox1; 1University of South Florida, 2University of Utah
Humans read incredibly quickly, fixating a word for only a quarter of a second before moving their eyes. But neural data suggests it takes 2-4 times longer than a single fixation to fully recognize and interpret a word in context, raising the question, “how can the eyes be faster than the brain” (Rayner & Clifton, 2009)? Co-registration of neural and behavioral measures is key in studying neural processes during naturalistic reading, but progress is limited if we treat the two data streams independently, rather than examining the coupling between temporally extended neural effects and eye movement decisions. In this talk, I will discuss prior work from my lab using both univariate analysis of co-registration data (i.e., treating eye movement behavior and neural data as independent outcomes), as well as emerging work using behavior-contingent analysis (BCA) to reveal how language-related brain processes (e.g., N400, LPC) are only observed when the reader exhibits specific behaviors (e.g., skipping, regressions). I will also discuss the theoretical implications of these findings, which suggest that saccade decisions are triggered at an intermediate point of lexical processing, followed by complete recognition and integration of a word into its context. This account can resolve the apparent paradox that the eyes can move through text faster than the brain understands it.
Reading in time: EEG, MEG, and computational modeling reveal recurrent feedforward and feedback dynamics during word-by-word comprehension
Gina Kuperberg1,2,3, Thomas Hansen1, Aileen Guo1, Samer Nour Eddine1,4, Anthony Yacovone1,5; 1Tufts University, 2Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, 3Massachusetts General Hospital, 4University of Maryland, College Park, 5Boston University
Reading comprehension unfolds word-by-word. With each word, the brain must compare incoming lexical information with predictions from preceding context and update an evolving higher-level representation. In this talk, I show how EEG, MEG, and computational modeling can be combined to characterize feedforward and feedback dynamics supporting this process. I first discuss EEG during self-paced reading of multi-sentence discourse, with responses to every content word and contextual probability estimated from human cloze norms and GPT-2 XL. Between 300–500 ms, N400 amplitude scaled with contextual surprisal. In the same window, semantic relatedness to the preceding content word provided additional facilitation, especially for less predictable words, revealing complementary contributions of top-down prediction and bottom-up lexical-semantic priming. Contextual probability also modulated a later frontal positivity (600–1000 ms), linked to successful shifts in higher-level interpretation. I then discuss simultaneous MEG/EEG to characterize the sources and representational content of these effects. Between 300–500ms, expectancy effects localized to left temporal and inferior frontal cortices. Representational similarity analyses showed item-specific information in temporal cortex, but primarily magnitude differences in inferior frontal cortex. Between 600–1000 ms, many regions were re-engaged with reversed dipole polarity. Finally, I show how a hierarchical predictive-coding model reproduces these dynamics: predictions propagate downward, incoming input drives feedforward processing, and subsequent feedback realigns representations across the hierarchy. Together, these findings suggest that word-by-word comprehension emerges through recurrent cycles in which lexical information is carried forward to update an evolving higher-level interpretation, which in turn feeds back to refine lower-level representations.
Language network and ventral visual regions interface at the Visual Word Form Area
Joseph Salvo1,2, Nathan Anderson1,3, Rodrigo Braga1,4; 1Northwestern University Feinberg School of Medicine, 2National Institute of Mental Health, 3Brigham Young University, 4Northwestern University
Successful reading involves both visual and linguistic processing. A possible bridge between the two systems is a Visual Word-Form Area (VWFA), which responds to words and pseudowords written in a familiar script. This VWFA is typically found in mid-inferotemporal cortex, as part of the ventral visual pathway, and in the same general area as a region of the language network. To investigate the relationship between the VWFA and the visual and language systems we used multi-echo functional magnetic resonance imaging to analyze task-based activations and resting-state functional connectivity from 8 neurotypical participants. We identified possible language regions using functional connectivity, in the absence of a language task, and tested whether these estimated areas aligned with regions active during language tasks. In the same participants, we used object and text localizers to map ventral visual regions and a possible VWFA. As expected, visual regions responded to pseudowords and characters written in an unfamiliar script, while language regions did not. Conversely, language regions responded to speech while visual regions did not. The possible VWFA shared features with each system, responding to both written pseudowords and speech, but not to the unfamiliar script. Further, this VWFA showed functional connections with language network regions. We replicated our results using the Natural Scenes Dataset. These findings suggest that the VWFA is uniquely poised to serve as a transition point between sensory-specific processing in visual regions and sensory-non-specific processing in language regions.
CNS Account Login
