Our lab combines a computational approach to describing learning (by modeling the relationship between sensory inputs and motor outputs) with a neurophysiological approach to investigating motor control (by studying the relationship between neural activity and song). In the Bengalese finch, young birds learn to sing by imitating the song of an adult “tutor,” providing one of the few examples of nonhuman vocal learning. Once song acquisition is complete, song is extraordinarily stable and remains highly stereotyped for the rest of the bird’s life. In a series of behavioral experiments (Sober and Brainard, Nature Neuroscience, 2009), we demonstrated that the stability of adult birdsong reflects a lifelong process of error correction and described the computational rules governing how sensory experience drives behavioral plasticity. We fit adult Bengalese finches with (Hoffmann et al., J. Vis. Exp. 2012) designed to provide online perturbations of auditory feedback. When the pitch (fundamental frequency) of auditory feedback was shifted, adult birds adaptively modified their vocalizations to reduce the experienced auditory error (for example, an upward shift in the pitch of auditory feedback resulted in a downward shift in the pitch of song). These results are the first demonstration of error-corrective learning in adult songbirds, and suggest that the stability of adult song reflects the stability of a sensory target used throughout the bird’s life to correct perceived vocal errors. Surprisingly, we also found that smaller shifts in the pitch of auditory feedback drove faster changes in behavior than did larger shifts (Sober and Brainard, PNAS, 2012). More recently, we have used this technique to explore the computations used to transform sensory errors into changes in behavior (Hoffmann and Sober, J. Neuroscience, 2014; Kelly and Sober, Frontiers in Integrative Neuroscience, 2014; Wyatt et al, eNeuro, 2017).
In parallel with these behavioral studies, we have used neural recordings to describe how pitch and other acoustic properties of song are controlled by the brain and vocal muscles. By recording from single neurons in premotor nucleus RA in singing birds, we demonstrate for the first time that trial-by-trial variations in RA activity predict variations in behavioral output (Sober et al., J. Neuroscience, 2008; Wohlgemuth, Sober, and Brainard, J. Neuroscience, 2010) and how variations in behavior are controlled by the precise timing of action potentials in the motor system (Tang et al., PLoS Biology, 2014; Srivastava et al., PNAS, 2017). Other current studies in the lab examine how vocal muscles transform neural signals into acoustic output (Srivastava, Elemans, and Sober, J. Neuroscience, 2015; Elemans et al., Nature Communications 2015), evaluate the role of the neurotransmitter dopamine in regulating skill learning (Hoffmann et al., J. Neuroscience 2016), and explore changes in neural activity that underlie adult song learning and answer fundamental questions about how behavior is controlled by a hierarchy of neural circuits. These studies emphasize the interplay between psychophysical descriptions of the rules governing song plasticity, physiological measures of learning-related changes in neural and muscular activity, and mathematical models of learning.