Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Neural predictive coding during naturalistic listening comprehension in dyslexia
Poster Session E, Sunday, September 14, 11:00 am - 12:30 pm, Field House
Lauren K. Salig1, Rachel L. Eggleston1, Aleksandra Witkowska1, Ziqian Shen1, Chi-Lin Yu2, Xiaosu Hu1, Ioulia Kovelman1, Jonathan R. Brennan1; 1University of Michigan, 2Oklahoma State University
It is well established that learners with dyslexia typically have difficulties with phonology and sound-to-letter mappings that affect their reading ability, but deficits in listening comprehension are also common in individuals with dyslexia. While those with dyslexia effectively communicate verbally, complex linguistic tasks often reveal processing challenges beyond phonology, extending to listening comprehension. However, the sources of variability in listening comprehension deficits in individuals with dyslexia remain unclear. Guided by the phonological deficit hypothesis of dyslexia, we predicted that even during naturalistic listening comprehension, children with dyslexia may exhibit differences in the engagement of brain regions associated with phonological processing. Prior research with traditional single-word phonological assessments (e.g., rhyme judgment) has consistently shown hypoactivation in left temporo-parietal regions in children with dyslexia, relative to typical readers. Here, we explored if these traditional phonological task-based findings extend to naturalistic listening contexts. Children with dyslexia (n=73) and typically developing children (n=95), ages 7–12 (all native English speakers) listened to a naturalistic audiobook during fNIRS neuroimaging. The large, pre-trained artificial neural network GPT-2 was used to quantify each word’s predictability within the audiobook context (surprisal), and we correlated the model-predicted brain activity based on surprisal with children’s fNIRS-recorded brain activity to identify brain regions that were sensitive to surprisal—one reasonable way to pinpoint the complex, expectation-based processes involved in language comprehension. Using this approach, we evaluate differences between the brain activity of typically developing children vs. children with dyslexia during naturalistic listening comprehension. Results from our novel, participant-friendly and ecologically valid naturalistic listening protocol are consistent with decades of prior neuroimaging work on dyslexia that used isolated word tasks (e.g., rhyme judgment). Typically developing learners exhibited a stronger left temporo-parietal response, whereas those with dyslexia exhibited a stronger bilateral frontal and right temporo-parietal response. The findings are in line with past work and the phonological perspectives on language comprehension in dyslexia, with hypoactivation in phonology-related regions and stronger activation in frontal and right-hemisphere regions that may be compensatory. Additional insights were afforded through a forecast analysis to determine how far ahead brain regions were making linguistic predictions. Both groups’ brains showed short-range language predictions (~12 words ahead), but this was detected in the left STG of typical learners and in the left IFG of those with dyslexia. Typical learners also exhibited long-range language predictions (~18 words ahead) in the left IFG, mirroring neurotypical adults (Caucheteux et al., 2022). In contrast, children with dyslexia only demonstrated short-range prediction. Previous research has linked dyslexia to functional and anatomical deficits in left temporal regions. Despite appearing fluent in conversation, our findings suggest that children with dyslexia rely more, though not entirely, on frontal regions to compensate for inefficiencies in temporal lobe function. The study of variation in individuals’ ability to predict words from verbal context at the neural level may provide new insights into the language comprehension challenges associated with dyslexia.
Topic Areas: Disorders: Developmental, Language Development/Acquisition