In this project, we will develop statistical and neural network models to describe the sequential structure of song in Japanese zebra finches (Lonchura striata domestica) and test these models with behavioral and neurophysiological experiments. The song of this species is characterized by a combination of systematic and random elements across multiple timescales: single syllables; deterministic chunks containing multiple syllables in the same sequence; stochastic bifurcations where the following syllable may depend on preceding syllables or chunks; and longer sequential dependencies that extend over the entire song. These elements influence the plasticity of different parts of the song in training experiments; the sequential structure of the song, derived from these elements, also allows us to predict the functional organization of the neural circuits responsible for song control, including the song nucleus HVC (proper name). The sequential structure of song, however, is not yet fully characterized, and the neural realization of the various organizational levels has also been scarcely investigated. We will create a large database of automatically annotated songs and apply modern machine learning methods to the songs of individual birds to test the hierarchical dependencies that emerge from the models through behavioral experiments and neurophysiology. At the behavioral level, we will manipulate auditory feedback at statistically significant points in the song and measure the resulting sequence changes. At the neural level, we will manipulate the same points using electrical microstimulation and measure the activity of HVC projection neurons. This same approach can ultimately be used to model the vocal sequences of other species in the research group in order to characterize their complex vocalizations at both the behavioral and neural levels.