Constructing an AI Tutor for Argumentative Writing Feedback (Poster 37)

• In Event: AERA Poster Session Sunday 1:30 p.m.
  In Poster Session: Artificial Intelligence in Education and Learning Technologies

Abstract

Composition and education scholars have called for more process-oriented pedagogy that allows for repeated, deliberate practice in long-form writing assignments, but these calls are complicated by the high degree of human labor required to create substantive formative feedback. We address this challenge by developing a student-facing argument mining tool to enhance the writing process with detailed, context-aware feedback. Utilizing LLaMA3-8B and GPT-3.5, we fine-tuned models to generate holistic feedback and employed a student simulator for iterative optimization. Our method outperformed baseline models in intrinsic and extrinsic evaluations, showing significant improvements in pedagogical alignment. The empirical results highlight the potential of integrating language models into educational tools, offering a scalable solution for improving argumentative writing skills.

Authors

• Jason Godfrey, Harvard University

• Nair Inderjeet, University of Michigan

• Lu Wang, University of Michigan

• Xu Wang, University of Michigan

• Anne R. Gere, University of Michigan