Fundamental predictive computations in upper-limb and speech adaptation

Project Summary Motor learning is a critical function of the vertebrate nervous system. Foundational theoretical work has described how motor learning may proceed at the algorithmic level, where an internal model predicts the sensory outcome of descending motor commands and adapts future movements to reduce prediction errors. At present, the internal model framework does not specify the necessary and sufficient neural signals that are required for error-based motor learning. This project is designed to fill this gap. In our framing, motor learning is made possible by two signals – a motor plan and a subsequent sensory prediction error. In light of this hypothesis, we predict that motor learning can proceed even in the absence of movement. Here, we test this hypothesis in two distinct domains – upper-limb and speech adaptation. This project thus proposes a unique hypothesis concerning multiple forms of motor learning and is poised to increase our understanding of this critical function of the human motor system. The research plan centers on two Specific Aims. In Aim 1 we use behavioral studies, electromyography, and functional neuroimaging (fMRI) to ask if motor planning and error processing are necessary and sufficient for upper limb motor learning, even in the absence of movement execution. We test the idea that motor plans need not reach the muscles to act as effective input to the brain’s internal models for movement. We also posit several constraints on motor learning without movement, where motor errors must be spatially localized near the goal of preceding motor plans to induce learning, and that the two signals must also occur close in time. Finally, using a novel fMRI experimental design, we attempt to map the neural correlates of the planning and error signals that constitute the key ingredients of motor learning. Aim 2 takes a parallel tack, but in the domain of speech motor learning. Thus, Aim 2 provides a stringent generalization test of our framework, asking whether predictive processing is a fundamental and universal computation of the motor system. Moreover, our novel behavioral design helps us to, for the first time, directly test competing theories of speech adaptation. Aim 2 also utilizes a novel fMRI protocol for mapping neural circuits underlying planning and prediction error processes in speech adaptation. It is critical to better understand how motor learning works at a basic algorithmic level – the knowledge gleaned by this proposal will support the development of unique mechanistic and clinical insights into future rehabilitation strategies, the underlying nature of BCI learning, and the power of human predictive processing.