Abstract:
Communication is the cornerstone of civilization. The ability to learn and produce complex vocalizations has helped human societies to flourish. On the tree of life, humans and birds belong to a subgroup that shows sensorimotor vocal learning. Songbirds learn their songs through vocal imitation, and this behavior is controlled by a series of interconnected brain regions called the song system.Bengalese finches are a species of domesticated songbirds that are known for their sequentially variable songs, with probabilistic sequencing. Bengalese finch songs contain sequence structures such as chunks, which are syllable sequences that always appear in the same order, repeat phrases, in which a single syllable type can repeat a variable number of times and INs, which are made of introductory notes and inter-notes. Introductory notes are 1-3 syllable types which represent premotor activity and are repeated at the beginning of song bouts, while inter-notes have the same syllable types, but occur between syllables within a bout. In this thesis, I first characterize the development of higher-order sequencing in Bengalese finch song. I establish pipelines for segmentation and annotation of developmental songs of Bengalese finches. I also establish a method to identify chunks methodically. I show repeats as well as INs have monotonically decreasing repeat number distributions early in development, but repeat phrase distributions are refined over development, and become characteristically peaked. I show that sequence structures like chunks are formed due to the acquisition of syllable-to-syllable
bigram transitions. Chunks can also be repeated variably in song, forming higher-order sequence structures called metarepeats. Meta-repeat number distributions also follow a developmental trajectory similar to repeat number distributions, and are also refined over development. I also compare the developmental trajectories of these phrases across birds with a common tutor and show that juveniles can copy acoustic structure and global transition patterns, while maintaining individual variability in local transition statistics. Next, I investigate the neural mechanisms behind higher-order sequencing in Bengalese finches through lesion studies. I show that the pallial nucleus mMAN (medial magnocellular nucleus of the anterior nidopallium) affects sequence entropy in song, and affects the variability of repeat phrases. I show evidence that mMAN is necessary for learned changes to flexible sequencing in song. My preliminary work shows that lesions of nucleus Nif (nucleus interfacialis of the nidopallium) suggest that Nif may also be involved in the control of sequencing variability. My work provides insights into the contributions of inputs to premotor nucleus HVC (proper name) in the control of variable sequencing in song. It also provides an important data basis on which statistical models of song syntax can be developed and refined. Bengalese finch songs display long-range dependencies between song elements, which are not well captured by simple Markov models. In the outlook, I present ongoing work on computational modeling of song syntax, and show that a hierarchical Bayesian sequence model can capture the long-range dependencies in Bengalese finch song.
birdsong