Åpen forelesning: Hvordan kan teknologi gi barn med språkvansker en bedre læringsopplevelse?

Simplification of Word Problems using Large Language Models

Av: Bernard Veldkamp and Emma Tas, University of Twente, The Netherlands

Interactive digital technology offers personalized learning experiences to meet the individual needs and preferences. While such technology has proven to be effective in primary education, it falls short in supporting children with language development disorders. For these children, the language of an assignment is often too difficult, while the content of the assignment is appropriate for their skill level. However, due to their language deficiencies, they cannot demonstrate their true ability. Currently, no digital tools can adjust a task's reading level without altering its content.      

Therefore, we aim to answer the following research question: To what extent can Large Language Models be used to adapt word problems to the needs of children with a language development disorder?

To answer the question, a dataset of 200 Dutch math problems for primary school children aged 11-12 was collected.  We created five prompts for simplification, including age-based and Dutch reading level-based prompts, as well as a prompt aimed to rewrite the problem as a bulleted list. The outcomes of the generated simplified word problems were calculated and compared to the answers of the original word problem, to test if information was lost during simplification. To quantify the impact of the different prompting methods on the text complexity, a linguistic complexity analysis was done. Preliminary findings using ChatGPT3.5 show that the bulleted list prompt was most effective in preserving content. The linguistic complexity analysis shows that there are some features that are consistently changed while simplifying the word problems. Overall, these outcomes suggest that a Large Language Model such as ChatGPT can be used to create adaptive digital technology that can benefit children with language development disorders.