![[2602.22524] Iterative Prompt Refinement for Dyslexia-Friendly Text Summarization Using GPT-4o](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.22524] Iterative Prompt Refinement for Dyslexia-Friendly Text Summarization Using GPT-4o
Summary
This paper explores an iterative prompt refinement method for creating dyslexia-friendly text summaries using GPT-4o, demonstrating improved readability and comprehension for dyslexic readers.
Why It Matters
Dyslexia affects a significant portion of the population, and traditional assistive technologies often fall short in addressing linguistic complexity. This research provides a novel approach to enhance accessibility in text summarization, paving the way for more inclusive AI applications.
Key Takeaways
- The study focuses on dyslexia-friendly text summarization using GPT-4o.
- Iterative prompt refinement significantly improves readability, achieving a Flesch Reading Ease score of 90 or higher.
- Most summaries meet readability targets within four attempts, with many succeeding on the first try.
- A composite score shows stable performance in readability and semantic fidelity.
- The findings establish a baseline for future accessibility-driven NLP research.
Computer Science > Computation and Language arXiv:2602.22524 (cs) [Submitted on 26 Feb 2026] Title:Iterative Prompt Refinement for Dyslexia-Friendly Text Summarization Using GPT-4o Authors:Samay Bhojwani, Swarnima Kain, Lisong Xu View a PDF of the paper titled Iterative Prompt Refinement for Dyslexia-Friendly Text Summarization Using GPT-4o, by Samay Bhojwani and 2 other authors View PDF Abstract:Dyslexia affects approximately 10% of the global population and presents persistent challenges in reading fluency and text comprehension. While existing assistive technologies address visual presentation, linguistic complexity remains a substantial barrier to equitable access. This paper presents an empirical study on dyslexia-friendly text summarization using an iterative prompt-based refinement pipeline built on GPT-4o. We evaluate the pipeline on approximately 2,000 news article samples, applying a readability target of Flesch Reading Ease >= 90. Results show that the majority of summaries meet the readability threshold within four attempts, with many succeeding on the first try. A composite score combining readability and semantic fidelity shows stable performance across the dataset, ranging from 0.13 to 0.73 with a typical value near 0.55. These findings establish an empirical baseline for accessibility-driven NLP summarization and motivate further human-centered evaluation with dyslexic readers. Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI) Cite...
