Executive and language systems competitively shape predictive processing during reading
When we read, our brains constantly generate predictions about upcoming words, which is thought to make comprehension faster and more efficient. Yet it remains unclear whether this predictive mechanism relies on the same domain-general executive resources the brain uses for other demanding cognitive tasks, or operates largely independently within the language system. Here we tested this question using transcranial magnetic stimulation (TMS) in twenty-five healthy adults, who performed self-paced naturalistic reading, an n-back task, and a dual task of concurrent reading and n-back, allowing us to place competing demands on cognitive resources. In three sessions, we stimulated the left angular gyrus alone, the angular gyrus together with dorsolateral prefrontal cortex, a hub of executive control, or delivered sham stimulation, and used computational modeling of decision-making to isolate how each brain state altered the underlying cognitive processes. Combined stimulation of the angular gyrus and prefrontal cortex selectively weakened the reading-time costs of unexpected words, but only when reading competed with the working-memory task, revealing that linguistic prediction draws on executive resources specifically under cognitive load. Computational models showed that the relationship was bidirectional: Unpredictable words also disrupted concurrent working-memory decisions, slowing the accumulation of evidence and increasing response caution. Together, these findings show that predictive language processing is not a fixed, encapsulated function but is dynamically coupled to executive control when cognitive demands compete. Our results provide evidence that prediction during language comprehension is embedded within a broader, demand-sensitive cognitive architecture that flexibly coordinates linguistic and executive resources.