On the dimensional structure of vocabulary and grammar in early language development

The relationship between lexical and grammatical knowledge in young children is impressively strong. Indeed, the correlation between productive vocabulary and grammar (r = .84) is larger than that between productive and receptive vocabulary (r = .63) when measured with the commonly used Communicative Development Inventories (CDIs). This correlation fits cleanly with usage-based theories of language, which assume no clear distinction between the lexicon and grammar (Tomasello, 2003). However, it could also reflect separate systems that are mutually causally related (mutualism); initially uncorrelated domains can gradually become so correlated as to be statistically indistinguishable when they are mutually causally related (Van der Maas et al 2006). Disentangling these accounts is complicated by the non-linear relationship between true and measured grammatical/lexical knowledge, which is not accounted for in traditional regression-based approaches. Here we present a new approach to disentangling these accounts which overcomes these measurement challenges. We examined the dimensional structure of item-level data from CDI data on Wordbank (Frank et al. 2017) using item-response theory and the DETECT method (Stout et al. 1996). We first considered all non-longitudinal data from the American English subsample of Wordbank. A DETECT analysis found evidence of moderate multidimensionality with vocabulary and grammar items clustering separately, contra some usage-based accounts which assume no distinction between grammatical and lexical knowledge. Given that mutualism predicts that two domains become increasingly correlated with age, we next ran a similar analysis in separate sets of younger (~18 months) and older (~28 months) children. These data were unidimensional at 18 months and multidimensional at 28 months. In sum, our results did not strongly support either account described above and are most consistent with an initially integrated lexico-grammatical system that becomes decoupled between the second and third year