Fig. 1.

Motor sequence learning: hypotheses and testing method. a Sensorimotor learning may require adapting the structure of a motor gesture to a desired target (left, different colors indicate a structural mismatch); adapting the temporal order of gestures to a target sequence (middle); or adapting a sequence of unformed gestures to a target sequence (right). In the latter case, the number of possible combinations of structural and temporal adjustments (vertical and horizontal arrows) increases exponentially with sequence length. b Song learning in zebra finches: a juvenile male gradually matches its own unformed vocalizations (bottom sonogram) to a memorized song of a tutor (top sonogram). Letters represent consecutive syllables of the bird’s song (S₁, S₂…) and the target (T₁, T₂…) c Hypothetical strategies of motor sequence learning: left, motor gestures are matched to temporally corresponding target gestures; middle, gestures are matched to the most structurally similar targets, minimizing structural changes, but possibly requiring considerable sequential rearrangements; right, error is computed across ‘chunks’ of gestures attempting to achieve local (non-optimal) tradeoffs between structural and temporal adjustments. d Experimental setup for testing vocal learning strategies. Left, a young bird hears a playback of an artificial song (source); once he learns it (middle), training is switched to a playback of a second song (target). The mismatch between the two songs is designed by the experimenter. The bird’s vocal practice trajectory as it corrects the mismatch is continuously recorded