Scholarlink

<p>Students’ reading comprehension depends on their reading skill, the readability of the text, and the interaction of those two factors. For instance, developing readers benefit from easier texts – texts that use simpler words and are more cohesive. However, instructors may be unable to personalize texts for students due to time and knowledge constraints. Generative Large Language Models (LLMs) such as ChatGPT may afford instructors the ability to adapt texts to students’ individual skill more easily and at a larger scale compared to manual editing. This study evaluates the ability of LLMs to ease text readability by assessing LLM-altered texts using natural language processing (NLP) techniques. In Experiment 1, 30 texts were rewritten by ChatGPT 4 to be more readable. In Experiment 2, the same 30 texts were rewritten by ChatGPT 4o and MagicSchool.ai to be readable for students at a particular grade. Both experiments used NLP tools to derive measures of linguistic sophistication, text cohesion, and holistic readability. Only the models and prompts tested in Experiment 2 demonstrated evidence of successfully easing text readability.</p>

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

ChatGPT, Can You Make this Text Easier?: Investigating the Utility of LLMs in Easing Text Readability

  • 期刊论文

  • 发表日期：

    2025年10月27日

查看全文
  • Micah Watanabe

摘要

Students’ reading comprehension depends on their reading skill, the readability of the text, and the interaction of those two factors. For instance, developing readers benefit from easier texts – texts that use simpler words and are more cohesive. However, instructors may be unable to personalize texts for students due to time and knowledge constraints. Generative Large Language Models (LLMs) such as ChatGPT may afford instructors the ability to adapt texts to students’ individual skill more easily and at a larger scale compared to manual editing. This study evaluates the ability of LLMs to ease text readability by assessing LLM-altered texts using natural language processing (NLP) techniques. In Experiment 1, 30 texts were rewritten by ChatGPT 4 to be more readable. In Experiment 2, the same 30 texts were rewritten by ChatGPT 4o and MagicSchool.ai to be readable for students at a particular grade. Both experiments used NLP tools to derive measures of linguistic sophistication, text cohesion, and holistic readability. Only the models and prompts tested in Experiment 2 demonstrated evidence of successfully easing text readability.