Connectionist Modeling and the Microstructure of Phonological Development: A Progress Report

摘要: Children learning to pronounce the sounds of their languages exhibit individual differences and a varying but often high degree regularity in rendering adult words. The mappings from child words show great phonetic context dependency considerable stability over time, much lexical irregularity other ‘unruly behavior’. Standard phonology models, derived adult-based phonological theory, ignore such unruly phenomena as non-rule-governed template matching, crosstalk between rules, fuzzy boundaries rule domains. Connectionist models are principle well adapted simulating these properties; model progress, GEYKO, is sketched. have access several feedback loops pronounce, both internal (auditory, motor, proprioceptive) external (parental social material reinforcement). GEYKO’s speech gesture planning module intended learn production with aid feedback, its auditory will perceptual categorization spectral data by unsupervised methods.