Isolating action-dependent transfer in motor skill learning

Motor skill training can improve performance in similar skills, yet it has been controversial which components of training contribute to such transfer. A main reason for this long-standing controversy is that movements comprising a skill (actions) are tightly intertwined with the sensory information that guides training (cues), obscuring the contribution of trained actions to skill transfer. Here, we provide evidence for action-dependent transfer in a sequential finger movement skill. Participants learned the skill with one hand in either action-based (sequential typing guided by cues) or cue-only (memorizing numerical cues without typing) training. In an intermanual transfer test with the other hand, we scrambled trained cues via numerical mirroring (swapping 1↔5 and 2↔4; e.g., 152→514), such that the mirrored sequences required the same trained actions in the absence of their trained cues. In Experiment 1, the action-based training group executed the mirrored sequences faster than novel sequences, revealing action-dependent transfer. In contrast, cue-only training showed no performance advantage for mirrored over novel sequences, confirming the elimination of cue-dependent transfer. Notably, action-dependent transfer occurred in both intrinsic (body-centric) and extrinsic (world-centric) coordinates, suggesting that trained actions are encoded in an abstract format or multiple coordinates. In Experiment 2, we quantified the additional benefit of original trained cues and tested distinctions between cue- and action-dependent transfer components. Inter-press-interval analyses showed a qualitative distinction: the action-dependent transfer effect was distributed across the sequences, whereas the cue-dependent transfer effect was concentrated toward sequence endings. Together, these findings establish that trained action constitutes a distinct component of motor skill transfer.