How vocabulary knowledge influences word associations: applications of lexical metrics and latent space network models

Abstract

This study examines how word association networks differ as a function of vocabulary knowledge using two methods: lexical metrics and latent space modeling. College students ( N = 44) completed a standardized assessment of receptive vocabulary knowledge and a repeated word association task, where they responded to cue words with the first word that came to mind over three list repetitions. Word associations were coded for cue-response similarity (word embedding, taxonomic, phonological) and word-level features (concreteness, age of acquisition, frequency). Participants with higher vocabulary knowledge more often produced lower frequency words with a later age of acquisition than their counterparts with lower vocabulary knowledge. Over list repetitions, cue-response similarity decreased and responses more often utilized lower frequency words with a later age of acquisition. We pooled word associations to construct a latent space model, and used lexical metrics and vocabulary knowledge (above-average vs. below-average) to predict edge weights (i.e., word association strength). Both word embedding similarity and word frequency predicted stronger edge weights. Over list repetitions, edge weights decreased with a larger effect in the below-average vocabulary network. The above-average vocabulary network exhibited more clusters with shorter average distances between nodes, suggesting greater differentiation within the lexicon. Taken together, the results indicate minimal differences in cue-response similarities of word associations of adults varying in their vocabulary knowledge, but more diverse word associations among those with above-average vocabularies. Growing one’s vocabulary over the lifespan may influence the organization of the mental lexicon by altering proximities between neighboring words.