The use of conditional inference to reduce prediction error—A mismatch negativity (MMN) study

Abstract

The brain uses regularities in the sound environment to build inference models predicting the most likely attributes of subsequent sounds. When the inference model fails, a prediction-error signal (the mismatch negativity or MMN) is generated. This study is designed to explore the capacity to use information about when a deviant sound will occur to switch between inference models in memory. We measured MMN generated to rare frequency, duration, intensity and spatial deviant sounds randomly occurring in a stream of identical repeating “standard” sounds. We then measured MMN to the same deviants in a linked sequence where deviants were paired—duration deviants followed an intensity change and spatial deviants followed a frequency change. To minimise prediction error, the brain should use the occurrence of the intensity and frequency deviant to prompt a change in the dominant inference (“expect-the-standard”) to anticipate the characteristics of the linked deviant. Anticipation was quantified as the proportion decline in duration and spatial MMN in the linked versus random sequence. We report three main outcomes on a sample of 23 healthy adults: (1) a significant reduction in duration MMN amplitude in linked versus random sequence; (2) a subgroup of participants exhibited significant reduction in spatial MMN amplitude in linked versus random sequence; and (3) the capacity to anticipate a linked deviant (reduce MMN) was a related to performance on the Continuous Performance Task-Identical Pairs. The results are discussed with respect to a possible co-reliance of CPT-IP and inference models on the inferior frontal gyrus.