Scholarlink

Technology-Enhanced Learning powered by Artificial Intelligence in Education (AIED) systems offers substantial benefits, including automated assessments and personalized feedback. Yet, deploying these systems in resource-constrained classrooms is challenging due to limited internet connectivity and insufficient access to digital devices. The AIED Unplugged (AIEDU) paradigm addresses these issues by emphasizing offline operation, teacher-mediated proxy interactions, multi-user capabilities, and minimal digital skill requirements. However, scaling AIEDU applications remains problematic. This study investigates integrating state-of-the-art multimodal Large Language Models (LLMs) into an Automatic Essay Scoring system specifically designed for low-resource educational settings. This study uses datasets of handwritten Brazilian Portuguese essays to assess LLM effectiveness in transcription accuracy and essay evaluation according to official Brazilian educational criteria. Results highlight Gemini 1.5 Pro’s superior transcription accuracy and Gemini 2.0 Flash’s outstanding essay evaluation performance, with Mean Absolute Errors of approximately 110 points overall on a scale of 0–1,000.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Empowering Equitable Learning with LLMs: Enhancing Writing Skills in Low-Resource Contexts

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Rafael Ferreira Mello,
  • Lenon Anthony,
  • Jamilla Lobo,
  • Fabíola Gonçalves Coelho Ribeiro,
  • Cleon Xavier,
  • Newarney Torrezão da Costa,
  • Dragan Gasevic,
  • Luiz Rodrigues

摘要

Technology-Enhanced Learning powered by Artificial Intelligence in Education (AIED) systems offers substantial benefits, including automated assessments and personalized feedback. Yet, deploying these systems in resource-constrained classrooms is challenging due to limited internet connectivity and insufficient access to digital devices. The AIED Unplugged (AIEDU) paradigm addresses these issues by emphasizing offline operation, teacher-mediated proxy interactions, multi-user capabilities, and minimal digital skill requirements. However, scaling AIEDU applications remains problematic. This study investigates integrating state-of-the-art multimodal Large Language Models (LLMs) into an Automatic Essay Scoring system specifically designed for low-resource educational settings. This study uses datasets of handwritten Brazilian Portuguese essays to assess LLM effectiveness in transcription accuracy and essay evaluation according to official Brazilian educational criteria. Results highlight Gemini 1.5 Pro’s superior transcription accuracy and Gemini 2.0 Flash’s outstanding essay evaluation performance, with Mean Absolute Errors of approximately 110 points overall on a scale of 0–1,000.