ABSTRACT
What are the neural mechanisms underlying vocal learning in speaking humans and in singing birds? Young infants are capable of statistically learning serial order in speech sounds with great ease (Saffran, 1996). However, the neural mechanisms involved in this process are far from understood due to their experimental inaccessibility. By contrast, the songbird has been adopted as a model system in which the neural mechanisms subserving song learning from a tutor are highly accessible (Hahnloser and Kotowicz, 2010; Mooney, 2009; Ölveczky and Gardner, 2010). A great deal of experimental effort has provided us with a wealth of data about the songbird system across multiple layers of biological organization ranging from molecular, and cellular levels, to the circuit level. However, despite this wealth of data, the basic computational principles underlying vocal development remain elusive. Here, we take a rst principles approach to the problem of vocal learning by formulating it as a simple mathematical problem that we then solve using a minimal set of neurobiological assumptions. This approach yields a new Hebbian theory of motor learning with delayed sensory feedback. Using three simple assumptions involving the existence of certain plasticity rules, gating of auditory feedback, and sources of motor
28.1 Introduction .......................................................................................................................... 547 28.2 Vocal Learning Problem .......................................................................................................548
