Applications of Large Language Models in Automatic Error Correction

Abstract

A system that can detect and correct errors in a text is highly desirable. Such a tool, also known as a grammar checker, may be the holy grail of intelligent computer-assisted writing systems. It has several potential pedagogical benefits for second language writing, such as reducing learners' errors and improving the quality of their writing. If properly designed, it can increase learners' awareness of their writing problems and develop learner autonomy. For instructors, such a tool can reduce their workload of error correction, allowing them to focus more on the content and organization of their students' writing. With recent advances in large language models, AI-powered systems have revealed unprecedented capabilities. In this colloquium, I will show how large language models such as ChatGPT and Llama 2 can support teaching and research in second language writing in ways that were not possible even two years ago. I will demonstrate how they can help researchers address important and difficult research questions such as common grammatical errors for a given level and the development of an error-tagged learner corpus without human intervention.