Using large language models to enhance data-driven learning in academic writing

Data-Driven Learning (DDL) has been widely validated for its effectiveness in language teaching across educational levels (O'keeffe, 2021). However, challenges persist, such as the need for advanced linguistic skills and time-consuming manual analysis (Ädel, 2010). The emergence of large language models (LLMs) offers promising solutions. Recent studies, including Liu et al. (2021) and Fathi & Rahimi (2024), demonstrate LLMs' efficacy in enhancing language learning, particularly in writing instruction. This study presents a two-year classroom practice integrating LLMs (e.g., DeepSeek and ChatGPT) into academic writing instruction for 673 undergraduate and graduate students in science and engineering. The tools automated the identification and correction of writing issues in synthesis tasks, enabling students to analyze errors and receive immediate feedback during and after class. This approach significantly reduces the limitations of traditional DDL. The results show that the combined DDL and LLMs approach improves performance in synthesis writing and error correction, with students achieving higher scores on quizzes and exams. By combining DDL with LLMs, we offer a more accessible and efficient learning experience, especially for students with varying proficiency levels. This integration represents a significant advancement in academic writing instruction, providing a scalable solution to traditional DDL challenges.