Proceedings of the Annual Meeting of the Cognitive Science Society

Comprehenders routinely generate predictions during language comprehension in real time, yet the mechanisms underlying prediction generation remain debated. The prediction-by-production hypothesis proposes that comprehenders use their own production system to anticipate upcoming linguistic input. While an initial body of work supports production-based accounts of prediction, it remains unclear whether increased engagement of the production system benefits all predictable words – i.e., both high- and low-cloze continuations – or only the most expected ones in a "winner-takes-all" fashion. In the present work, we investigate this question through a German self-paced reading and an event-related potentials (ERP) experiment employing a novel within-participants task paradigm manipulating engagement of the production system during comprehension. Reading times and ERPs show that increased engagement of the production system facilitiates processing for both high- and low-cloze words, suggesting that prediction-by- production supports parallel pre-activation of multiple plausible continuations, with ERP results showing strongest facilitation for highly predictable words.