VietTESOL International Convention 2026

The current study examines subject–verb agreement (SVA) errors in L2 English using a large-scale spoken learner corpus. Drawing on data from 238 learners across three L1 backgrounds and four proficiency levels, the study analyzes 6,762 clauses extracted from 2,142 task-based transcripts. Each clause in the dataset was annotated for a wide range of linguistic and extralinguistic predictors including L1 background, proficiency, task type, verb type, subject number, subject animacy, noun phrase modification, subject–verb distance, verb frequency, and lexical reliability. A Bayesian mixed-effects logistic regression model was used to examine the effects of the selected variables.

Results indicate that SVA accuracy is primarily shaped by lexical and syntactic factors rather than learner background. Verb type emerged as a key predictor, with copula verbs associated with higher error rates and lexical verbs generally reducing errors relative to auxiliaries. However, these effects were strongly moderated by subject number. Greater subject–verb distance and noun phrase premodification also increased error likelihood. By contrast, L1 background and proficiency level showed limited reliable effects. The findings highlight the complexity of agreement processing in L2 English and demonstrate the value of multifactorial approaches in learner corpus research.

Overall, the current project represents the first comprehensive investigation of how syntactic, semantic, and morphological factors jointly influence SVA accuracy within a multifactorial Bayesian framework.