The present study investigates the automatization of second language (L2) sentence processing. It compares the extent to which a mere speedup (faster execution) and restructuring (more stable execution) of sentence processing contribute to L2 oral performance. The maze task is used to measure the speed (reaction time, RT) and processing stability (coefficient of variance, CV) of sentence processing. The elicited imitation (EI) task measures L2 oral proficiency (repetition accuracy and accuracy in plural and third person s). These tasks were performed by 110 English-as-a-foreign-language learners with Japanese as their L1. The results show that only RT, not CV, significantly predicts L2 oral proficiency. Even though a subgroup of learners, who previously stayed in an English-speaking country, demonstrated some indications of automatization, RT was a better predictor of L2 oral proficiency than CV, irrespective of immersion experience. These findings suggest that CV has little practical value in predicting L2 oral proficiency.

This study measures whether number and type of morphemes in an elicited imitation string result in a greater number of modifications with L2 experience. Rationale is drawn from L2 working memory processing limitations at distinct levels of proficiency. 38 subjects (L2 Spanish university students) comprise three proficiency groups: beginning, undergraduate majors and graduate students. Number of morphemes was varied within each syllable count; and responses were either correct or modified (lexemically/inflectionally as deletions or substitutions). One or two way ANOVAs determined significance between mean proportions for each group. Findings indicate that increases in number of morphemes yielded significant differences, and that while the lowest proficiency group produced higher proportions of lexical deletions, the modifications made by more advanced groups were inflectional substitutions.