The late frontal positivity reflects incremental mental model updating: Graded predictability effects during authentic text reading

• Word predictability is measured using surprisal values from a transformer model. • Participants read two-sentence passages extracted from naturalistic texts. • The predictability of each word shows a graded effect on late frontal positivity. • The late frontal positivity effect reflects mental model updating. High word predictability facilitates the access and integration of word meaning, indicated by a reduced N400 effect. However, whether word predictability affects later phases of reading comprehension (mental model updating), as assessed by a late positivity, remains unsettled. Some studies suggest that unexpected but plausible words increase positivity in frontal regions, while others do not. To gain clarity on this issue, we used two-sentence passages from The New York Times articles that do not contain implausible words. Using a language model with a transformer architecture, we assessed the predictability (surprisal) of each word in these texts on a continuous scale. Linear mixed-effects modeling of the EEG dataset (Chen et al., 2025) showed that higher word predictability reduced N400 in central-parietal regions, whereas lower predictability increased late positivity in frontal regions. These findings suggest that word predictability has a graded effect on late frontal positivity, reflecting mental model updating.