Rethinking inner speech through linguistic active inference

Yao, Bo (2025) Rethinking inner speech through linguistic active inference. Psychological Review. ISSN 0033-295X

Abstract

This article introduces the linguistic active inference theory, which proposes that inner speech augments the brain's predictive processes by transforming prior expectations and sensorimotor predictions to help reduce prediction error and uncertainty. By leveraging language's unique properties-its efficiency in representing sensorimotor information, its ability to extend across time and space, and its generativity in constructing novel predictions-inner speech enables predictive processes to transcend immediate experience, encoding complex sensory experiences into linguistic forms for perceptual inference while decoding abstract goals into situated actions for active control. Linguistic active inference theory provides a unifying framework explaining how inner speech's diverse functions, varied phenomenology, and neurocognitive developmental mechanisms all emerge from its augmentation of perceptual inference and active control. It posits that inner speech dynamically adapts its form and function in response to computational demands and ongoing prediction errors to reduce the imprecision in the brain's generative model. This synthesis advances foundational theories and provides a roadmap for future research: generating novel testable hypotheses, motivating a shift toward dynamic and integrative methodologies, and opening new perspectives on related mental phenomena and the broader role of symbolic systems in cognition.