Modeling English vocabulary acquisition through the biomechanics of speech and Large Language Models
Jingya Shang · Molecular & cellular biomechanics · 2025
This study investigates the relationship between biomechanical constraints in speech production and English vocabulary acquisition by integrating Large Language Models (LLMs). Using a sample of 51 Mandarin Chinese speakers in Shenzhen, China, divided into three age groups (children: 8–12 years, adolescents: 13–17 years, and adults: 18–25 years), we conducted a 12-week longitudinal study combining articulatory measurements with computational analysis. The research employed electromagnetic articulography, surface electromyography, and advanced language modeling to examine speech patterns and learning outcomes. Results reveal significant age-related differences in articulatory kinematics, with children showing larger tongue displacements (14.3 ± 2.1 mm) and higher muscle activation levels than adults. Integrating biomechanical constraints into LLM analysis improved prediction accuracy by 18.7% for children and 14.2% for adults, though at the cost of increased computational resources. Strong negative correlations were found between articulatory effort and learning success (r = −0.824 for children, p < 0.001), with retention rates significantly influenced by motor complexity. These findings suggest that biomechanical factors play a crucial role in vocabulary acquisition, particularly in younger learners, and that incorporating these constraints into computational models can enhance our understanding of language learning processes. This integrated approach offers new insights for developing age-appropriate language teaching methodologies and improving predictive models for learning outcomes.