Contextual and neural representations of sequentially complex animal vocalizations

Anonymous

Keywords: sequence learning, birdsong, auditory neuroscience, generative models, context

TL;DR: We compare perceptual, neural, and modeled representations of animal communication using machine learning, behavior, and physiology.

Abstract: Holistically exploring the perceptual and neural representations underlying animal communication has traditionally been very difficult because of the complexity of the underlying signal. We present here a novel set of techniques to project entire communicative repertoires into low dimensional spaces that can be systematically sampled from, exploring the relationship between perceptual representations, neural representations, and the latent representational spaces learned by machine learning algorithms. We showcase this method in one ongoing experiment studying sequential and temporal maintenance of context in songbird neural and perceptual representations of syllables. We further discuss how studying the neural mechanisms underlying the maintenance of the long-range information content present in birdsong can inform and be informed by machine sequence modeling.