Alladi Ramakrishnan Hall

Inference of Syntax from Songbird Songs and Implications for Neural Mechanisms

Sumithra Surendralal

Symbiosis International (Deemed University), Pune

The neural and peripheral mechanisms underlying sequence generation in vocal learners still need to be fully understood. One of the approaches used to gain insight into this behaviour involves the quantitative characterization of the structure of vocal sequences, also known as the syntax, based on statistical regularities. I will discuss the inference of syntax models for learned animal vocalisations using the example of songbird songs. These songs consist of variable sequences of acoustic units called syllables that follow probabilistic rules. The probability of producing a particular syllable in a song can depend on the syllables that came before it. Similar context dependencies are widely observed in animal behaviours. Using Bengalese finch song, I will detail how context dependencies can be modelled using Partially Observable Markov Models (POMMs), a special case of Hidden Markov Models (HMMs). POMMs consist of states and probabilistic transitions between them. Each state is associated with a syllable, and one syllable can be associated with multiple states. This multiplicity of syllable-to-states-association distinguishes a POMM from a simple Markov model and captures context dependencies. Previous studies have shown that disrupting auditory feedback by deafening adult Bengalese finches can cause changes in the structure and sequencing of syllables in their songs. This suggests that auditory feedback could play a role in shaping the context dependencies in songbird song syntax. Comparing the song syntax of Bengalese finches before and shortly after deafening reveals that auditory feedback plays an important but not exclusive role in creating context dependencies, positing the need to investigate the role of intrinsic neural circuitry as well.