Salience-based reinforcement of a spiking neural network leads to increased syllable production

Abstract

Canonical babbling is vocal babbling that contains syllabic patterning like that in adult speech. Its emergence during the first year of human infancy is one of the most significant pre-speech vocal motor milestones. This paper focuses on a spiking neural network model that controls the lip and jaw muscles of an articulatory speech synthesizer and learns to produce canonical babbling. The model was adapted to receive reinforcement when it produced a sound with high auditory salience. Salience-reinforced versions of the model increased their rates of canonical babbling over the course of learning more than their yoked controls. This supports the idea that both intrinsic reinforcement and social reinforcement both contribute to human acquisition of canonical babbling.