An Associational Hypothesis for Sensorimotor Learning of Birdsong

Songbirds learn to sing in a two stage process. First, the bird forms a sensory template by listening to and memorizing its father's song. Second, the bird practices singing and learns to match its song to the memorized template. Two experimentally well-supported hypotheses locate song sequence generation in nucleus HVc, and the sensory template in the anterior forebrain pathway (AFP). However, there is no known connection from the AFP to HVc. Furthermore, due to auditory feedback delay, reinforcement signals from vocalization are likely to reach the motor pathway after the burst of motor activity responsible for the vocalization. Thus, it is unclear how template information and auditory feedback could be used to guide learning of song. A model of sensorimotor learning of syllables and song sequences is outlined that resolves these problems, using only simple associational and reinforcement learning. First, HVc uses auditory feedback to learn an eeerence copy, i.e. a prediction of the auditory vocalization that will result from premotor activity. Second, the AFP uses this eeerence copy, rather than auditory feedback, to compare to the template, and uses the result to reinforce appropriate syllable representations within RA. Sequence production results from an interplay of premotor activity and an auditory feedback-induced eeerence copy within HVc. Finally, by innuencing the statistics of the RA patternss resulting from HVc premotor activity, the AFP ensures that sequences generated in HVc are mapped onto the appropriate sequence of states within RA.