Executive resources shape the impact of language predictability across the adult lifespan.

Less predictable words (higher surprisal) were read more slowly, confirming that readers actively generate linguistic predictions during word-by-word reading.

• Word predictability was indexed by surprisal derived from a pre-trained large language model (GPT-2).
• Higher surprisal values indicate less predictable words, and these were associated with longer reading times.
• This effect was observed in both the main sample (N = 175) and the independent replication sample (N = 96).
• The finding replicates the well-established surprisal effect in reading time research.

Experimentally limiting executive resources through a concurrent n-back task reduced the effect of language predictability on reading time.

• Participants completed a secondary font colour n-back task while reading short articles presented word-by-word, varying cognitive demand.
• The reduction in the surprisal effect under increased cognitive load indicates that language prediction draws on executive control resources.
• This dual-task manipulation addresses whether prediction is automatic or resource-dependent.
• The finding was replicated across two independent samples (N = 175 and N = 96).

Older adults exhibited stronger predictability effects overall compared to younger adults.

• The study spanned a broad adult age range of 18–85 years (N = 175, replication N = 96).
• Older adults showed larger surprisal effects on reading time, indicating greater reliance on or benefit from linguistic predictions.
• This age-related increase in predictability effects is consistent with the idea that older adults compensate for declining processing speed by leaning more heavily on predictions.

The dependency of language prediction on executive resources became more pronounced with age.

• The reduction in the predictability effect under high cognitive load was larger for older adults than for younger adults.
• This age-by-cognitive-load interaction indicates that older adults' language predictions are more susceptible to disruption when executive resources are taxed.
• The pattern was consistent across both the main sample and the independent replication sample.
• Executive resources, including working memory, attention, and inhibitory control, generally decline with age.

The findings were replicated in an independent sample, supporting the robustness of the resource-dependent predictability effect.

• An independent replication sample of N = 96 participants was used.
• Both the main effect of surprisal on reading time and its modulation by cognitive load were replicated.
• The age-related modulation of the cognitive load effect on predictability was also replicated in the second sample.

The study employed a dual-task reading paradigm combining word-by-word self-paced reading with a concurrent font colour n-back task to manipulate cognitive demand.

• Participants read short articles presented word-by-word (self-paced reading time paradigm).
• The secondary task was a font colour n-back task, designed to vary working memory and executive control load.
• GPT-2-derived surprisal was used as a continuous measure of word predictability.
• The total sample included 175 participants (18–85 years) in the main study and 96 in the replication.