Convergent neural signatures of speech prediction error are a biological marker for spoken word recognition

Abstract

We used MEG and fMRI to determine how predictions are combined with speech input in superior temporal cortex. We compared neural responses to bisyllabic spoken words in which the first syllable strongly or weakly predicts the form of the second syllable (e.g. "bingo" versus "tango"). We further compared neural responses to the same second syllables when heard in an unfamiliar pseudoword and therefore in a situation in which predictions mismatch with sensory input (e.g. "snigo" and "meago"). Across multiple imaging modalities and analysis approaches (including information-based pattern analysis methods), we show that neural representations of second syllables are suppressed by strong predictions when predictions match sensory input. However, neural representations of the same second syllables show the opposite effect (i.e. enhanced representations following strongly than weakly-predicting syllables) when predictions mismatch with sensory input. Computational simulations show that this interaction between prediction strength and (mis)match is consistent with prediction error but not alternative (sharpened signal) computations. Neural signatures of prediction error are observed early (beginning 200 ms after the onset of the second syllable), localise to early auditory regions (in fMRI, bilateral Heschls gyrus and STG) and are expressed as changes in low-frequency (theta and alpha) power. Our study therefore provides convergent neural evidence that speech perception is supported by the computation of prediction errors in auditory brain regions. These prediction error computations play a central role in the identification of familiar spoken words and perception of unfamiliar pseudowords.