Search Abstracts | Symposia | Slide Sessions | Poster Sessions
The neural architecture of predictive language processing: Prefrontal and parietal contributions
Poster Session C, Thursday, October 1, 10:45 am - 12:45 pm, Wangari Maathai
Sandra Martin1,2, Anna-Lisa Schuler1, Ole Numssen1, Merle Schuckart3, Jonas Obleser3, Gesa Hartwigsen1,2; 1Max Planck Institute for Human Cognitive and Brain Sciences, 2Leipzig University, 3University of Lübeck
Efficient language comprehension relies on the cognitive ability to generate probabilistic predictions about upcoming input (Ryskin & Nieuwland, 2023). Predictive coding frameworks propose a hierarchical neural architecture integrating domain-specific linguistic processes and domain-general executive resources (Caucheteux et al., 2023). The Angular Gyrus (AG) supports semantic integration, whereas the Dorsolateral Prefrontal Cortex (DLPFC) exerts cognitive control to maintain context under uncertainty. Here, we used transcranial magnetic stimulation (TMS) to test their causal roles during naturalistic reading combined with working memory demands, hypothesizing that DLPFC disruption would impair prediction generation, whereas AG modulation would interfere with semantic integration. In a preregistered within-subject repeated-measures design, 25 right-handed participants (M = 31.3, SD = 6.6, range = 21–41 years; 13 female) completed three sessions: offline repetitive TMS (rTMS) over the AG (1 Hz), offline+online rTMS (1 Hz over AG + 10 Hz over DLPFC during tasks), and sham stimulation. Participants performed self-paced reading of texts alone, an n-back task (1- and 2-back) alone, or a dual task, which combined self-paced reading with a concurrent 2-back task on the words’ font color. Reading times, n-back reaction times, text comprehension accuracy, and n-back performance were analyzed using linear and generalized additive mixed-effects models. Cognitive modeling using Bayesian hierarchical drift-diffusion models characterized latent decision components (e.g., drift rate and boundary separation) in the single n-back and dual tasks. Models included stimulation condition and lexical surprisal as key predictors. Surprisal robustly slowed reading across tasks. Stimulation effects emerged in a task-dependent manner, with the most pronounced modulation under high cognitive load. In the dual-task condition, combined offline AG + online DLPFC stimulation reduced the processing cost associated with high-surprisal words relative to sham, indicating altered predictive processing dynamics when executive resources were taxed. Text comprehension accuracy was preserved across stimulation conditions. For the n-back task, combined stimulation produced faster reaction times in the demanding 2-back condition compared to sham. This speed gain was accompanied by reduced discriminability (d-prime), whereas offline AG stimulation alone yielded an improvement in n-back accuracy relative to sham, pointing to dissociable contributions of the two targeted regions. Hierarchical Bayesian modeling provided mechanistic insight into these patterns. During the dual task, offline AG stimulation showed a consistent negative effect on drift rate relative to sham indicating a decrease in the n-back performance, which was not observed after offline+online stimulation. For the n-back task alone, combined offline AG + online DLPFC stimulation enhanced drift rates while also increasing decision thresholds during the 2-back task relative to sham, suggesting that participants accumulated evidence more efficiently under high cognitive load despite adopting conservative decision boundaries. These findings provide evidence that DLPFC and AG jointly regulate cognitive performance when predictive language processing occurs under executive load. The dissociation between single-task and dual-task effects indicates that frontoparietal contributions to language prediction become critical when domain-general control demands compete with domain-specific linguistic processing. The differential modulation of decision thresholds and drift rates suggests complementary roles: AG contributes to information integration efficiency, while DLPFC supports performance when executive and linguistic processes compete.
Topic Areas: Control, Selection, and Executive Processes, Meaning: Lexical Semantics