Neural Evidence for Tonal Prediction: Multivariate Decoding of Predicted Tone Categories Using Functional Magnetic Resonance Imaging Data

Predictive processing plays a central role in language comprehension, allowing listeners to generate predictions about upcoming linguistic input. Although considerable evidence supports segmental prediction, less is known about whether listeners can form predictions about suprasegmental features such as lexical tone. This study investigates whether listeners can generate and neurally represent predicted tonal information in the absence of auditory input. Using a Mandarin Chinese tone sandhi paradigm, we established tonal predictions based on sentence and visual context, recording brain activity with functional magnetic resonance imaging. Multivariate pattern analysis showed that predicted tonal categories could be decoded from brain activity even without tonal stimuli present. These representations were localized in auditory areas, articulatory motor regions, and the right cerebellum. We also found that predicted tone representations had distinct neural substrates compared to perceived tone representations. The study provides direct neural evidence that listeners can form representations of lexical tone in predictions of auditory input. The findings expand our understanding of suprasegmental prediction in speech and highlight the cerebellum's role in linguistic prediction.