TY - JOUR

T1 - Prespeech motor learning in a neural network using reinforcement

AU - Warlaumont, Anne

AU - Westermann, Gert

AU - Buder, Eugene

AU - Oller, D. Kimbrough

PY - 2013

Y1 - 2013

N2 - Vocal motor development in infancy provides a crucial foundation for languagedevelopment. Some significant early accomplishments include learningto control the process of phonation (the production of sound at the larynx)and learning to produce the sounds of one's language. Previous work hasshown that social reinforcement shapes the kinds of vocalizations infantsproduce. We present a neural network model that provides an account ofhow vocal learning may be guided by reinforcement. The model consistsof a self-organizing map that outputs to muscles of a realistic vocalizationsynthesizer. Vocalizations are spontaneously produced by the network. If avocalization meets certain acoustic criteria, it is reinforced, and the weightsare updated to make similar muscle activations increasingly likely to recur.We ran simulations of the model under various reinforcement criteria andtested the types of vocalizations it produced after learning in the differentconditions. When reinforcement was contingent on the production of phonated (i.e. voiced) sounds, the network's post learning productions were almost always phonated, whereas when reinforcement was not contingent on phonation, the network's post-learning productions were almost always not phonated. When reinforcement was contingent on both phonation and proximity to English vowels as opposed to Korean vowels, the model's postlearning productions were more likely to resemble the English vowels and vice versa.

AB - Vocal motor development in infancy provides a crucial foundation for languagedevelopment. Some significant early accomplishments include learningto control the process of phonation (the production of sound at the larynx)and learning to produce the sounds of one's language. Previous work hasshown that social reinforcement shapes the kinds of vocalizations infantsproduce. We present a neural network model that provides an account ofhow vocal learning may be guided by reinforcement. The model consistsof a self-organizing map that outputs to muscles of a realistic vocalizationsynthesizer. Vocalizations are spontaneously produced by the network. If avocalization meets certain acoustic criteria, it is reinforced, and the weightsare updated to make similar muscle activations increasingly likely to recur.We ran simulations of the model under various reinforcement criteria andtested the types of vocalizations it produced after learning in the differentconditions. When reinforcement was contingent on the production of phonated (i.e. voiced) sounds, the network's post learning productions were almost always phonated, whereas when reinforcement was not contingent on phonation, the network's post-learning productions were almost always not phonated. When reinforcement was contingent on both phonation and proximity to English vowels as opposed to Korean vowels, the model's postlearning productions were more likely to resemble the English vowels and vice versa.

KW - Infant vocalization

KW - Motor development

KW - Neural network

KW - Reinforcement

KW - Neuromuscular control

KW - Articulatory speech synthesis

U2 - 10.1016/j.neunet.2012.11.012

DO - 10.1016/j.neunet.2012.11.012

M3 - Journal article

VL - 38

SP - 64

EP - 75

JO - Neural Networks

JF - Neural Networks

SN - 0893-6080

ER -