Why Phonological Constraints Are so Granular

Speech perception models with a fast phonological preprocessor (an FPP) require a uniform phonology throughout the speech community, because chunking errors would otherwise impede lexical access. The phonological constraints encapsulated in an FPP arise as regularities over the lexicon. However, different people know different words. A Monte Carlo simulation of vocabulary acquisition using the Celex database reveals that coarse-grained phonological constraints can be acquired reliably from 30% or less of the monomorphemic words (e.g. by early adolescence). An hypothetical fine-grained constraint combining a trisyllabic stress template and a nasal-obstruent restriction cannot be reliably learned without the large vocabulary of an educated adult. Thus, the nexus of speech perception, phonological well-formedness, and individual variation in vocabulary means that phonological constraints must be coarse-grained.