Generative AI and Second Language Writing

This article provides a critical synthesis and analysis of the research on the application of generative AI (GenAI) in second language (L2) writing. It conceptualizes GenAI literacy, synthesizes the research on written feedback, establishes a framework for prompt engineering, critiques the research examining the validity of GenAI ratings in writing assessment, and summarizes empirical evidence on the differences between GenAI and human writing. Specifically, the following findings and arguments are presented and discussed. GenAI literacy consists of four components pertaining to users’ competence and knowledge of GenAI basics, effective use, output evaluation, and ethics. The research on written feedback shows that teacher feedback focuses more on content, while GenAI feedback focuses more on organization. This research also suggests a need for criteria-based feedback and feedback evaluation. Prompt engineering is discussed along three dimensions: input, task, and output, followed by snapshots of prompts used in feedback research. The studies on writing assessment reveal that GenAI ratings are more consistent with human ratings when GenAI is trained using a large number of scored essays and when the rating criteria are well-defined. Comparisons of GenAI and human writing demonstrate that GenAI writing is more formal, academic, and impersonal, while human writing is more personal, creative, and linguistically accessible. This article concludes by making sense of the research findings, identifying future directions, and proposing three principles that may guide the research, practice, and theory construction for GenAI: individualization, domain-specificity, and writer agency.