Self-production facilitates and adult input interferes in a neural network model of infant vowel imitation


It is well known that greater amounts of adult input facilitate a child's language development. Thus, one might expect that increased amounts of adult input would help an infant learn to accurately imitate the vowels of his/her native language. In addition, an infant's own production of sounds during cooing, babbling, etc. is known to be important to the development of speech abilities. We simulate infant vowel development using a neural network that contains a layer of auditory neurons, a layer of motor neurons, and bidirectional connections linking these perceptual and motor layers. During an initial babbling phase, the system produces random motor activations, hears the acoustic consequences of these motor activations, and adjusts the weights between its auditory and motor layers in a Hebbian fashion. In simulations, passive auditory input from an external "caregiver" is also included during the babbling phase, and is used to update existing auditory-motor connections. In a testing phase, the model is given adult vowels as auditory input and asked to imitate them. Results indicate that self-productions do promote the development of the ability to imitate, but, somewhat counterintuitively, the more adult input this model receives during babbling, the less accurate its imitations are during test. Explanations and implications of this finding are discussed.12.

Publication Title

AISB 2011: Computational Models of Cognitive Development

This document is currently not available here.