The Effect of Previously Encountered Sensory Information on Neural Representations of Predictability: Evidence from Human EEG

Abstract

Accumulating evidence suggests that the brain continuously monitors the predictability of rapidly evolving sound sequences, even when they are not behaviorally relevant. An increasing body of empirical evidence links sustained tonic M/EEG activity to evidence accumulation and tracking the predictability, or inferred precision, of the auditory stimulus. However, it remains unclear whether, and how, this process depends on auditory contextual memory. In the present EEG study, we examined neural responses to sound sequences across two experiments, and compared them to predictions from ideal observer models with varying memory spans. Stimuli were sequences of 50 ms long tone-pips. In Experiment 1 (N=26; both sexes), a regularly repeating sequence of 10 tones (REG) transitioned directly to a different regular sequence (REGxREGy). In Experiment 2 (N=28; both sexes), the same regular sequence was repeated after an intervening random segment (REGxINTREGx). Results from Experiment 2 revealed that the inferred predictability of the resumed REGx pattern was influenced by the preceding INT tones, even several seconds after they ended, indicating that the brain retains contextual memory over time. In contrast, neural responses in Experiment 1 were best explained by models with minimal memory. This dissociation implies that the brain can dynamically adjust its strategy based on inferred environmental structure--resetting context when interruptions signal change, and preserving context when patterns are likely to resume.