Surprisal in reading: language models predict the N400 for L2 readers

Research on L1 reading has demonstrated context-dependent predictive/integrative processes during reading a word in context. This study examined these processes in proficient L2 speakers by measuring surprisal effects during reading English texts and compared the findings with those of native English speakers. We used ERPs to examine whether surprisal derived from different probabilistic language models affects the N400 response in L2 readers. Overall, the results showed that surprisal is a strong predictor of the N400 in L2 reading, suggesting that predictive/integrative processes play an important role during L2 reading. Further analyses suggested that both syntactic and lexical predictive/integrative processes are functional in L2 reading. Crucially, L2 predictive/integrative processes incorporate hierarchical syntactic information. Additionally, the language models that best predicted N400 differed between L2 and L1 readers, potentially reflecting differences in cognitive processing between native and non-native speakers of English.