# Automated Reading Passage Generation with OpenAI's Large Language Model

**Authors:** Ummugul Bezirhan, Matthias von Davier  
**arXiv:** [2304.04616](https://arxiv.org/abs/2304.04616)  
**Date:** 2023-04-10  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2304.04616v1/

## Key points

- Automated item generation (AIG) is a process that uses computer technology to generate new items for assessments, reducing reliance on human subject experts.
- Machine learning algorithms have the potential to greatly improve the efficiency and effectiveness of AIG.
- The paper presents an approach that utilizes OpenAI's latest transformer-based language model, GPT-3, to generate reading passages automatically.
- The AI-generated text had similar content and structure to a fourth-grade reading passage.
- Multiple passages were generated using GPT-3, and the final passage was selected based on its Lexile score agreement with the original passage.
- A human editor revised the selected passage to ensure it was free of any grammatical or factual errors.
- All AI-generated passages along with original passages were evaluated by human judges according to their coherence, appropriateness for fourth graders, and readability.
- The results showed that GPT-3 could generate coherent and appropriate reading passages that were comparable in quality to those written by humans.
- This study demonstrates how machine learning algorithms can be used effectively in AIG processes such as generating reading passages for educational assessments.

## Summary

In recent years, the widespread use of computer-based assessments and individualized learning platforms has led to an increased demand for the rapid production of high-quality items. To address this need, automated item generation (AIG) was proposed as a process that uses item models to generate new items with the help of computer technology, reducing reliance on human subject experts at each step of the process. While AIG has been used in test development for some time, machine learning algorithms have introduced the potential to greatly improve its efficiency and effectiveness. This paper presents an approach that utilizes OpenAI's latest transformer-based language model, GPT-3, to generate reading passages automatically. The authors carefully engineered prompts using existing reading passages to ensure that the AI-generated text had similar content and structure to a fourth-grade reading passage. For each prompt, multiple passages were generated using GPT-3, and the final passage was selected based on its Lexile score agreement with the original passage. In the final round, a human editor revised the selected passage to ensure it was free of any grammatical or factual errors. To evaluate their approach, all AI-generated passages along with original passages were evaluated by human judges according to their coherence, appropriateness for fourth graders, and readability. The results showed that GPT-3 could generate coherent and appropriate reading passages that were comparable in quality to those written by humans. Overall, this study demonstrates how machine learning algorithms can be used effectively in AIG processes such as generating reading passages for educational assessments. By leveraging advanced language models like GPT-3 and carefully engineering prompts and evaluation criteria, researchers can produce high-quality items more efficiently than ever before while maintaining consistency and accuracy across different versions of tests.