The effects of corpus model predictors on L2 constructional use: the case of <i>let</i> and <i>permit</i>

Abstract Usage-based theories posit that first language (L1) and second language (L2) speakers recruit constructional usage patterns in language input as part of their linguistic knowledge. This study focused on L2 English speakers’ knowledge of two near-synonymous constructions, let +V 2 and permit +V 2 , which typically express permission. The study has two goals. First, we examined the extent to which linguistic factors predicting the use of these constructions in the British National Corpus (BNC) affected L2 speakers’ constructional choices in a forced-choice task, in which they selected a preferred construction or indicated equal preference for both constructions to complete contextual sentences. Second, we explored how L2 proficiency mediated these effects. We used Bayesian multilevel logistic regression to determine key predictors of constructional choices in the BNC data and Bayesian multilevel ordinal regression to analyze forced-choice data. Participants were 46 L1 English speakers in the UK and 42 L1 Thai-L2 English graduate students in Thailand. The results showed that the corpus predictors impacted responses from both groups. L2 proficiency did not mediate these effects but was still influential. The findings support the cognitive plausibility of corpus model predictors and align with usage-based theories.