Is a Native Speaker Editor Still Necessary? Investigating L2 Learners' Responses to ChatGPT's Revisions

This study explored how English as a Second Language (L2) learners respond to outputs generated by ChatGPT. Four Japanese learners were tasked with creating original documents using mechanical assistance, after which they were presented with multiple corrected outputs from ChatGPT versions 3.5 and 4.0. They then selected the corrections they deemed most effective. These documents were subsequently assessed by native English-speaking university faculty members to evaluate the disparity between the learners' perceptions of writing quality and those of native speakers. The findings revealed that learners unconditionally accepted the corrections provided by version 4.0 when they were comprehensible. However, the experts often preferred simpler outputs, citing issues such as unnatural register in the more complex corrections. Additionally, corrections that were difficult to understand frequently coincided with errors originally made by the learners. Overall, the study indicated that, beyond grammatical accuracy, the learners’ ability to effectively utilize ChatGPT's outputs reflected their proficiency.