Predictive coding in typical speech perception and dyslexia

点击翻译按钮获取中文摘要

PROJECT SUMMARY/ABSTRACT The most common and best understood cause of the reading difficulty that defines dyslexia is an alteration in the processing of spoken language. While speech perception deficits in dyslexia have been reported for decades, a burgeoning literature describes behavioral deficits that rely on the exploitation of regularities in the sensory environment, as well as reduced neural adaptation to consistent stimulation. Because efficient speech processing relies on rapid plasticity for acoustic features characteristic of particular voices, coupled to semantic predictions constrained by context, a rapid plasticity impairment in the auditory cortical hierarchy is a candidate core deficit in dyslexia. Here we explore whether reduced plasticity due to short-term experience and/or top- down expectation characterizes speech perception in dyslexia. By recording magnetoencephalography (MEG) while individuals listen to pairs of words, we will determine how predictability differentially modulates neural responses in dyslexia. Aim 1 is to characterize the spatiotemporal patterns of auditory repetition suppression deficits in dyslexia. It is not known whether reduced neural adaptation is due to bottom-up or top-down mechanisms. We will assess bottom-up repetition suppression by measuring responses to pairs of speech stimuli in which the word, voice, or both are repeated unexpectedly, revealing with high spatiotemporal detail how neural populations encode these features. Attenuated repetition suppression suggests that the auditory system changes less due to short-term experience with word forms and voices, which may be a core neurobiological difference in dyslexia. Aim 2 is to characterize expectation suppression and prediction error deficits for speech in dyslexia. We will assess top-down expectation suppression by measuring responses to pairs of speech stimuli in which listeners have high expectation that stimuli will repeat. Consistent with a predictive coding account, we expect that fulfilled expectations will generate little response, while violated expectations will evoke large prediction error responses, signaling a need to update the prediction. In dyslexia, abnormalities in these phenomena suggest inadequate prediction of voice phonetics and/or word phonology, implicating a higher-order deficit. Aim 3 is an exploratory quantification of the emergence of expected stimulus feature encoding in neural signals. We will train a neural pattern classifier to distinguish words and voices from the MEG data, investigating whether features emerge earlier and more robustly when they are predicted vs. unpredicted, as would be explained by top-down influences. We will investigate whether individual differences in classifier accuracy correlate with the magnitude of neural prediction error and with language abilities. These aims advance a mechanistic understanding of speech processing differences that can lead to dyslexia. Reduced plasticity due to short-term experience and/or inadequate prediction of speech features may prevent the brain from building and updating models of phonetic-phonological relationships that underlie children's phonological awareness, sound-to-print mapping, and, ultimately, reading.