The influence of children’s exposure to language from two to six years: The case of nonword repetition

Nonword repetition (NWR) is highly predictive of vocabulary size, has strong links to language and reading ability, and is a clinical marker of language impairment. However, it is unclear what processes provide major contributions to NWR performance. This paper presents a computational model of NWR based on Chunking Lexical and Sub-lexical Sequences in Children (CLASSIC) that focuses on the child's exposure to language when learning lexical phonological knowledge. Based on language input aimed at 2-6year old children, CLASSIC shows a substantial fit to children's NWR performance for 6 different types of NWR test across 6 different NWR studies that use children of various ages from 2;1 to 6;1. Furthermore, CLASSIC's repetitions of individual nonwords correlate significantly with children's repetitions of the same nonwords, NWR performance shows strong correlations to vocabulary size, and interaction effects seen in the model are consistent with those found in children. Such a fit to the data is achieved without any need for developmental parameters, suggesting that between the ages of two and six years, NWR performance measures the child's current level of linguistic knowledge that arises from their exposure to language over time and their ability to extract lexical phonological knowledge from that exposure.