Abstract
Learning new sequential movements is a fundamental skill for many animals. Motor sequence learning may arise from three distinct processes: (1) improved execution of individual movements independent of their sequential context; (2) enhanced anticipation of "what" movement should be executed next, enabling faster initiation; and (3) the development of motoric sequence-specific representations that encode "how" movements should be optimally performed within a sequence. However, many existing paradigms conflate the "what" and "how" components of learning, as participants often acquire both the sequence content (what to do) and its execution (how to do it). This overlap obscures the distinct contributions of each mechanism to motor sequence learning. In this study, we disentangled these mechanisms in a continuous reaching task by varying how many upcoming targets were visible. When participants (n = 14, 8F) could only see one future target, improvements were mostly due to them learning which target would come next. When they could see four future targets, participants immediately demonstrated faster movement times and increased movement smoothness, surpassing late-stage performance in the one-target condition. Crucially, even with full visibility of future targets, participants showed further sequence-specific learning driven by a continuous optimization of movement trajectories. Follow-up experiments (n = 42, 21F) revealed that the learned sequence representations did not generalize in extrinsic coordinates across limbs and encoded contextual information of four movements or longer. Our paradigm dissociates between the "what" and "how" components of motor sequence learning and provides evidence for the development of motoric sequence representations that guide optimal movement execution.