Exploring Generative AI for Personalized Readability
The ongoing research of the TRC and beyond shows content and format impact reading for readers of all ages. The TRC is investigating how GenAI can personalize reading materials to support more optimal fluency and comprehension.
For content personalization, our research explores how Large Language Models (LLMs) can automatically adjust the reading (complexity) level of text. Research has long suggested that readers may struggle to comprehend text when it is written far above an individual’s reading skill level. In education settings, matching reading materials to individual student reading abilities is fundamental for effective learning. However, creating differentiated content presents a significant challenge for educators. This capability could revolutionize how instructors prepare materials, saving valuable time and allowing for greater differentiation in the classroom. For students, AI-powered text adaptation promises a more equitable learning environment where content complexity is no longer a barrier to understanding.
Our current work focuses on understanding the effectiveness and trade-offs of using LLMs for automatic text simplification. We are evaluating performance across various models and prompting techniques to simplify texts. Key questions guiding this research include:
- How accurately can LLMs adjust text to a specific target reading level?
- How well are crucial keywords, concepts, and details preserved during simplification?
- How significantly does the text change in length and structure?
While preliminary results show variability across our evaluation metrics, the potential is promising and we are continuing to investigate how we can improve automatic text simplification with GenAI.
More details about this work can be found in this paper.
For format personalization, we are evaluating how GenAI can augment the visual presentation of text for individuals. As informed by our research and the work of the broader readability community, we are developing tools to reflow the visual presentation of text based on reader profiles (such as age, dyslexia, etc.) and preferences.
FlexibilityAI
Our vision is an AI-assisted approach to empower readers with more accessible formats, adjusted reading levels, and beyond to foster more engaged and fluent reading.
Our tool, FlexibilityAI, combines GenAI for format and content in one text personalization tool. This technology can benefit educators, students, and experts across various domains. FlexibilityAI can assist in tailoring complex information for diverse audiences, creating personalized learning experiences, and ultimately enhance knowledge dissemination and learning for all.
“This tool could assist me in providing more personalized content for my students matched to their individual learning needs and reading skill level.” –Educator
“Grounding typographic choices in empirical readability research enhances accessibility across diverse populations.” –Researcher
“Including vetted typographic research data allows LLMs to generate contextually correct advice and content.” –Engineer
“This system helps me ensure I understand tradeoffs in my designs, and allowing my work to be both beautiful and effective.” –Typographer
“These tools unlock new markets and enhance global communication, reducing the barrier to culturally-aware localized typographic assets for the next billion users.” –MBA
This tool is in early development. To see re-leveled passage samples and the FlexibilityAI demo, click here.
To learn more or to join us in this work, please send inquiries to team@thereadabilityconsortium.org