LLM-based personalization of learning materials for children with a developmental language disorder

Children with developmental language disorder would benefit from personalization of the language complexity of the learning materials. This research focuses on whether Large Language Model can adapt the language usage of word problems while preserving their content to the needs of the children. Developmental language disorder can affect children's speaking, listening, readind and writing. In this project, the focus is on reading. We first conducted a pilot study to test the technical feasibility of our idea. A dataset of 200 Dutch word problems for grade 5/6 students was collected and five prompts were designed for adaptation. These prompts were aimed at adapting test complexity in general, based on age, reading level, or text structure. The results show that all prompts perform similarly in terms of preserving content, with the Bulleted List prompt altering or removing the least amount of crucial information. Complexity analysis indicates that a prompt based on reading levels lowers the text complexity of the word problems the most, compared to the original word problem. Overall, the outcomes of this pilot suggest that a Large Language Model, such as ChatGPT, can be used to create digital technology that can benefit children with a developmental language disorder. The next step is to design a real-world validation study involving children with DLD and educators. In this work in progress session, we aim to discuss the design of our validation study.