Effects of a higher-level distributional statistic on second language learners’ restriction of linguistic generalization in L2 construction development

A central issue in SLA is how, without negative evidence, L2 learners come to know particular new word-construction combinations are acceptable while other new combinations are not allowed in the target language. Recently, L2 researchers have attempted to extend the statistical approach proposed for L1 to address this L2 learnability problem, with a focus on verb-specific statistics (i.e. how likely a particular verb is to co-occur with a particular construction). This study investigates whether a higher-level verb distribution (i.e. the distribution of one-construction only verbs and alternating verbs in the training input) influences Chinese EFL learners’ generalization of the English prenominal past participle (PPP) via novel verbs. Ninety first-year high-school students in China were recruited and randomly assigned to three groups: a lexicalist group, a generalist group, and a mixed group. During the treatment, the Lexicalist group received input involving one-construction only verbs exclusively; the Generalist group, alternating verbs only; the Mixed group, both one-construction only and alternating verbs. Results from an acceptability judgment task indicated that, although this higher-level verb distribution did not impact on the generalization of real verbs, it influenced the generalization of both extensive exposure novel verbs and minimal exposure novel verbs. Meanwhile, this higher-level verb distribution also affected the regularization (operationalized as the number of participants inclined to accept each verb in only one of two competing constructions) of no exposure novel verbs. The findings suggest that higher-level distributional statistics could affect the restriction of L2 construction overgeneralization.