L2ers’ predictions of syntactic structure and reaction times during sentence processing

This paper investigates how Korean L2 learners of English predict upcoming syntactic structure based on a newly received word during sentence processing. Studies like Linzen and Jaeger (2016) suggest that readers use their probabilistic inference developed by their experience of the language to which they have been exposed to predict the most appropriate syntactic structure. This study replicates the experiment for L2ers following Linzen and Jaeger (ibid.), which investigates the way of predicting syntactic structure by using the subcategorization frame of a verb to understand L1 language processing. We employ the information-complexity metrics such as surprisal, entropy, and entropy reduction to quantify the uncertainty/unexpectedness of a given word that reflects the processing difficulty during sentence processing. The results show that L2ers’ tendency to read different regions of a sentence varies. Reading times are longer in the verb and the ambiguous regions of the structurally ambiguous than of the structurally unambiguous sentences. Likewise, reading times are longer in the disambiguating region of the unambiguous than of the ambiguous sentences. Reading times are also longer when the surprisal increases in the disambiguating region. Overall, the findings reveal that such information-complexity metrics as entropy reduction and surprisal play an instrumental role in accounting for the aspects of sentence processing by Korean L2 learners of English.