Effects of neural noise on predictive model updating across the adult lifespan

Abstract

AbstractIn the perceptual and sensorimotor domains, ageing is accompanied by a stronger reliance on top-down predictive model information and reduced sensory learning, thus promoting simpler, more efficient internal models in older adults. Here, we demonstrate analogous effects in higher-order language processing. One-hundred and twenty adults ranging in age from 18 to 83 years listened to short auditory passages containing manipulations of adjective order, with order probabilities varying between two speakers. As a measure of model adaptation, we examined attunement of the N400 event-related potential, a measure of precision-weighted prediction errors in language, to a trial-by-trial measure of speaker-based adjective order expectedness (“speaker-based surprisal”) across the course of the experiment. Adaptation was strongest for young adults, weaker for middle-aged adults, and absent for older adults. Over and above age-related differences, we observed individual differences in model adaptation, with aperiodic (1/f) slope and intercept metrics derived from resting-state EEG showing the most pronounced modulations. We suggest that age-related changes in aperiodic slope, which have been linked to neural noise, may be associated with individual differences in the magnitude of stimulus-related prediction error signals. By contrast, changes in aperiodic intercept, which reflects aggregate population spiking, may relate to an individual’s updating of inferences regarding stimulus precision. These two mechanisms jointly contribute to age-related changes in the precision-weighting of prediction errors and the degree of sensory learning.