How An Associative Learning Account Of Language Exposure Predicts Vocabulary Growth By Word Length, Word Frequency, And Neighbourhood Density

Abstract

Children who hear lots of language have larger vocabularies. The words within the language also affect learning: short words, high frequency words, and words from dense neighbourhoods are all more likely to be learned quickly. However, little consideration has been given to the actual learning that occurs based on language exposure and how this subsequently influences word-level effects. We present a computational model of associative learning that extracts sublexical and lexical information based on exposure to the maternal utterances in twelve 2-3 year old mother- child dyads. By analysing the vocabulary of the mothers, children, and model, we show that both model and children are qualitatively different to the maternal input, showing: (a) superior learning for monosyllabic over bisyllabic words that reduces over time; (b) a decline in the proportion of high frequency words learned over time with marginal increases for mid and low frequency words; (c) greater likelihood of learning words from dense neighbourhoods; and (d) as word frequency declines, the influence of neighbourhood density dramatically increases. Associative learning operating on the linguistic environment of the child captures a range of effects seen in vocabulary learning over time, suggesting that word-level effects are largely determined by language exposure.