Bird Vocalization - Auditory Feedback in Birdsong Learning

Auditory Feedback in Birdsong Learning

Early experiments by Thorpe in 1954 showed the importance of a bird being able to hear a tutor's song. When birds are raised in isolation, away from the influence of conspecific males, they still sing. While the song they produce, called "isolate song", resembles the song of a wild bird, it shows distinctly different characteristics from the wild song and lacks its complexity. The importance of the bird being able to hear itself sing in the sensorimotor period was later discovered by Konishi. Birds deafened before the song-crystallization period went on to produce songs that were distinctly different from the wild type and isolate song. Since the emergence of these findings, investigators have been searching for the neural pathways that facilitate sensory/sensorimotor learning and mediating the matching of the bird's own song with the memorized song template.

Several studies over recent decades have looked at the neural mechanisms underlying birdsong learning by performing lesions to relevant brain structures involved in the production or maintenance of song or by deafening birds before and/or after song crystallization. Another recent experimental approach was recording the bird's song and then playing it back while the bird is singing, causing perturbed auditory feedback (the bird hears the superposition of its own song and a fragmented portion of a previous song syllable). After Nordeen & Nordeen made a landmark discovery as they demonstrated that auditory feedback was necessary for the maintenance of song in adult birds with crystallized song, Leonardo & Konishi (1999) designed an auditory feedback perturbation protocol in order to explore the role of auditory feedback in adult song maintenance further, to investigate how adult songs deteriorate after extended exposure to perturbed auditory feedback, and to examine the degree to which adult birds could recover crystallized song over time after being removed from perturbed feedback exposure. This study offered further support for role of auditory feedback in maintaining adult song stability and demonstrated how adult maintenance of crystallized birdsong is dynamic rather than static.

Brainard & Doupe (2000) posit a model in which LMAN (of the anterior forebrain) plays a primary role in error correction, as it detects differences between the song produced by the bird and its memorized song template and then sends an instructive error signal to structures in the vocal production pathway in order to correct or modify the motor program for song production. In their study, Brainard & Doupe (2000) showed that while deafening adult birds led to the loss of song stereotypy due to altered auditory feedback and non-adaptive modification of the motor program, lesioning LMAN in the anterior forebrain pathway of adult birds that had been deafened led to the stabilization of song (LMAN lesions in deafened birds prevented any further deterioration in syllable production and song structure).

Currently, there are two competing models that elucidate the role of LMAN in generating an instructive error signal and projecting it to the motor production pathway:

• Bird’s Own Song (BOS)-tuned Error Correction Model

During singing, the activation of LMAN neurons will depend on the match between auditory feedback from the song produced by the bird and the stored song template. If this is true, then the firing rates of LMAN neurons will be sensitive to changes in auditory feedback.

• Efference Copy Model of Error Correction

An efference copy of the motor command for song production is the basis of the real-time error-correction signal. During singing, activation of LMAN neurons will depend on the motor signal used to generate the song, and the learned prediction of expected auditory feedback based on that motor command. Error correction would occur more rapidly in this model.

Leonardo tested these models directly by recording spike rates in single LMAN neurons of adult Zebra Finches during singing in conditions with normal and perturbed auditory feedback. His results did not support the BOS-tuned error correction model, as the firing rates of LMAN neurons were unaffected by changes in auditory feedback and therefore, the error signal generated by LMAN appeared unrelated to auditory feedback. Moreover, the results from this study supported the predictions of the efference copy model, in which LMAN neurons are activated during singing by the efference copy of the motor signal (and its predictions of expected auditory feedback), allowing the neurons to be more precisely time-locked to changes in auditory feedback.