Open books. Image credit: Patrick Tomasso (CC0)
Understanding language requires more than simply recognizing individual words. Instead, readers and listeners continuously generate predictions about what might come next. For instance, after reading “She walked her…”, most individuals would probably anticipate the word “dog.”
Such predictive processes facilitate comprehension, making it faster and more efficient. Yet, a long-standing debate concerns whether these predictions emerge effortlessly and automatically, or whether they depend on domain-general executive resources, including working memory, attention, inhibitory control and goal-directed behaviour. Importantly, these resources are finite: they can be temporarily depleted when performing additional non-linguistic tasks and generally decline with age.
Schuckart et al. investigated whether language prediction during reading relies on executive resources and how language prediction might change with age as executive functioning declines. Employing a dual task reading paradigm, they examined how imposing a cognitively demanding secondary task influenced the effect of word predictability on reading speed in adults aged 18 to 85. This approach addressed a fundamental question in psycholinguistics – whether prediction is automatic or resource-dependent – and shed light on why language comprehension remains robust despite age-related cognitive decline.
The study involved 175 participants, with an independent replication sample of 96, who read texts word by word while sometimes performing a concurrent working-memory task. Word predictability was quantified using ‘surprisal’, a lexical score derived from the large language model GPT-2, which provides information about how unpredictable a word is.
The results revealed that less predictable words were read more slowly, confirming that readers actively generated linguistic predictions. The effect of predictability diminished under increased cognitive load, demonstrating that language prediction draws on executive resources. Older adults exhibited stronger predictability effects overall, but these were also more susceptible to reduction when executive resources were taxed. Together, the results show that language prediction is not cost-free and changes systematically across the adult lifespan.
These findings advance our understanding of how predictive language processing relies on executive resources and may have implications for interventions in conditions where these resources are compromised, such as following a stroke.
