Listening to disfluent speech: Robust effect at processing may not extend to learning

Fillers, such as uh and um in English, are reliable predictors of upcoming novelty, and listeners harness these statistics to process disfluent speech efficiently. Predictive processing begets prediction error, and therefore it is possible that fillers impact word learning when they precede unexpected referents. However, the effect of fillers on word learning is unknown, especially over different time scales. Across five experiments, we sought to investigate the impact of fillers on language processing and learning, focusing on the relationship between fillers and predictive processing. Following exposure to novel words under fluent and disfluent learning conditions, we measured participants' word recognition at immediate vs. delayed testing. Participants consistently used fillers to predict upcoming novelty at exposure, but they tended to demonstrate similar retention of words taught in fluent and disfluent conditions, whether or not they made prediction errors. We frame our findings in the context of predictive processing and disfluency, concluding that disfluency shapes listeners' predictions but this effect does not carry over to word retention.