Inference-based form selection in native and non-native speakers: Evidence from subject-verb agreement error detection in French

This study examines native and non-native retrieval of the third person number agreement morphology for the French future tense (e.g. mangera (EAT-FUT.3PS.SG) vs. mangeront (EAT-FUT.3PS.PL), 'will eat'), in the light of subset and superset based late-insertion models. We present evidence for the coactivation of subset vocabulary items where selecting the more specified item-ont (3PS.PL) in third-person plural agreement requires eliminating the less specified item -a (3PS.SG). The lack of plural specification for -a is interpreted as [-Plural] by a real-time inference on a <-ont,-a > information scale defined by [Number: o/Plural]. Using picture classification time differences during a grammaticality judgment task, we argue that i.) subset vocabulary items are activated and that ii.) domain-general inferencing enables form selection in multiple activations. Hence, picture-probes related to the ongoing interpretation of the sentence presented immediately after a mismatching singular verb form (subset item error) took longer to classify in L1-English L2-French learners and in French natives tested with a speeded task. This interaction of form with meaning construction justifies subset access triggering domain-general inference-based selection in multiple activations over a superset-based model of vocabulary item activation or information-load and frequency-based approaches. This model provides an understanding of temporary irretrievability in L2 acquisition. (c) 2021 Elsevier B.V. All rights reserved.