Plan- versus motion-referenced generalization of fast and slow processes in reach adaptation

Generalization in motor learning refers to the transfer of a learned compensation to other relevant contexts. The generalization function is typically assumed to be of Gaussian shape, centered on the planned motion, although more recent studies associate generalization with the actual motion. Because motor learning is thought to involve multiple adaptive processes with different time constants, we hypothesized that these processes have different time-dependent contributions to the generalization. Guided by a model-based approach, the objective of the present study was to experimentally examine these contributions. We first reformulated a validated two-state adaptation model as a combination of weighted motor primitives, each specified as a Gaussian-shaped tuning function. Adaptation in this model is achieved by updating individual weights of the primitives of the fast and slow adaptive process separately. Depending on whether updating occurred in a plan-referenced or a motion-referenced manner, the model predicted distinct contributions to the overall generalization by the slow and fast process. We tested 23 participants in a reach adaptation task, using a spontaneous recovery paradigm consisting of five successive blocks of a long adaptation phase to a viscous force field, a short adaptation phase with the opposite force, and an error-clamp phase. Generalization was assessed in 11 movement directions relative to the trained target direction. Results of our participant population fell within a continuum of evidence for plan-referenced to evidence for motion-referenced updating. This mixture may reflect the differential weighting of explicit and implicit compensation strategies among participants.NEW & NOTEWORTHY Error-based reach adaptation can be modeled by fast and slow adaptive processes. Using a spontaneous recovery paradigm and model-based analyses, we tested how these processes generalize during force-field reach adaptation. Depending on whether the fast and slow adaptive processes operate by crediting the planned or actual motion, the model predicts distinct contributions of them to the overall generalization function. We show that human participants fall within a continuum of evidence for plan-referenced to motion-referenced updating.