Master’s thesis · UMBC
Evaluating Large Language Models as Generators and Judges of Dialogic Reading Questions for Early Literacy
Investigates large language models as both generators and evaluators of dialogic reading questions for early literacy. The research examines question quality and instructional alignment for kindergarten and first-grade learners.
Aryan Dhiren Jagani · University of Maryland, Baltimore County · ProQuest Dissertations & Theses, 2026 · Publication no. 32792284.
Read the research approach
The thesis uses the CROWD framework: Completion, Recall, Open-ended, Wh-, and Distancing questions. The rubric examines instructional alignment, answerability, text relevance, engagement, age-appropriate vocabulary, and grammar. It compares zero-shot and few-shot prompting, AI-generated and human-authored questions, and the use of LLMs as evaluators.