Student Theses and Dissertations

Date of Award


Document Type


RU Laboratory

Reeke Laboratory


Variability is a defining feature of the oscine song learning process, reflected in song and in the neural pathways involved in song learning. For the zebra finch, juveniles learning to sing typically exhibit a high degree of vocal variability, and this variability appears to be driven by a key brain nucleus. It has been suggested that this variability is a necessary part of a trial-­‐and-­‐error learning process in which the bird must search for possible improvements to its song. Our work examines the role this variability plays in learning in two ways: through behavioral experiments with juvenile zebra finches, and through a computational model of parts of the oscine brain. Previous studies have shown that some finches exhibit less variability during the learning process than others by producing repetitive vocalizations. A constantly changing song model was played to juvenile zebra finches to determine whether auditory stimuli can affect this behavior. This stimulus was shown to cause an overall increase in repetitiveness; furthermore, there was a correlation between repetitiveness at an early stage in the learning process and the length of time a bird is repetitive overall, and birds that were repetitive tended to repeat the same thing over an extended period of time. The role of a key brain nucleus involved in song learning was examined through computational modeling. Previous studies have shown that this nucleus produces variability in song, but can also bias the song of a bird in such a way as to reduce errors while singing. Activity within this nucleus during singing is predominantly uncorrelated with the timing of the song, however a portion of this activity is correlated in such a manner. The modeling experiments consider the possibility that this persistent signal is part of a trial-­‐and-­‐error search and contrast this with the possibility that the persistent signal is the product of some mechanism to directly improve song. Simulation results show that a mixture of timing-­‐dependent and timing-­‐independent activity in this nucleus produces optimal learning results for the case where the persistent signal is a key component of a trial-­‐and-­‐error search, but not in the case where this signal will directly improve song. Although a mixture of timing-­‐locked and timing-­‐independent activity produces optimal results, the ratio found to be optimal within the model differs from what has been observed in vivo. Finally, novel methods for the analysis of birdsong, motivated by the high variability of juvenile song, are presented. These methods are designed to work with sets of song samples rather than through pairwise comparison. The utility of these methods is demonstrated, as well as results illustrating how such methods can be used as the basis for aggregate measures of song such as repertoire complexity.


A thesis presented to the faculty of The Rockefeller University in partial fulfillment of the requirements for the degree of Doctor of Philosophy.

Permanent URL

Included in

Life Sciences Commons