Evolving Connectionist Models to Capture Population Variability across Language Development: Modeling Children's Past...

Area of Science:

• Computational Neuroscience
• Developmental Psychology
• Behavioral Genetics

Background:

• Children's English past tense acquisition is a key area for language development theories.
• Existing computational models often overlook individual differences in learning.
• English past tense involves regular and irregular verb forms, presenting a complex learning challenge.

Purpose of the Study:

• To develop computational models of past tense acquisition that account for individual differences.
• To simulate genetic and environmental influences on language learning variability.
• To explore the application of evolutionary principles to cognitive development.

Main Methods:

• Utilized populations of artificial neural networks (ANNs) that evolve based on behavioral genetics principles.
• Simulated genetic influences via parameter variations in ANNs and environmental influences through a training data filter.
• Employed a novel twin model based on meiosis and fertilization to disentangle genetic and environmental effects.

Main Results:

• The model captured population variability in children's past tense acquisition.
• Identified divergence in selection favoring regular, irregular, or both verb types.
• Observed canalization and the limiting effects of stochastic selection and variable environments on evolutionary power.
• Found an inverse relationship between heritability and optimization; selected traits showed lower heritability with reduced genetic variation.

Conclusions:

• Linking individual differences, cognitive development, and generational selection is viable within a computational framework.
• The model successfully simulates both individual differences and developmental trajectories.
• Findings highlight the interplay of genetic and environmental factors in shaping language learning outcomes.