Limited cognitive resources reduce the language predictability benefit across the adult lifespan

In everyday communication, humans predict upcoming language seemingly effortlessly. However, it remains unclear to what extent the formation of such predictions taxes executive resources. Our study set out to investigate how a limitation of executive resources impacts natural language prediction on multiple timescales in a novel dual-task paradigm, and how this impact is modulated by age. Participants (N = 175; 18-85 years) read short newspaper articles, presented word by word in varying font colours. This self-paced reading task was either performed in isolation or paired with a competing n-back task (1-back or 2-back) on the words' font colour. We measured word-level reading time and block-level reading comprehension as well as n-back performance. To quantify word predictability, we estimated word surprisal on four distinct timescales (i.e., context lengths ranging from words to paragraphs) using a large language model. Under high cognitive load, adults aged 60 and over benefited most from high word predictability. Our results show that independent of timescale, higher cognitive load diminishes the benefits of high word predictability on reading time, suggesting language predictions draw on executive resources.