Published in the Proceedings of the Workshop on Orchestrating Human and AI Agency for Proactive and Reflective Learning (HAI-Agency 2026), co-located with the 27th International Conference on Artificial Intelligence in Education (AIED 2026), Seoul, Republic of Korea, CEUR Workshop Proceedings Vol-4248.
The paper builds a pipeline in which human annotators set quality rubrics and AI agents generate, critique and refine Socratic hints, producing MultiHint-9K: 9,987 verified question-answer-hint tuples in English, Italian and Farsi. Fine-tuning five open models (7B to 24B) on the data improved ROUGE-L by about 67% over the base models, with the largest gains in Farsi.
Authors: Kamyar Zeinalipour, Amir Sadeghi, Giovanni Angelini, Leonardo Rigutini, Marco Gori, Marco Maggini.
Read the paper | Code | Models and data
All the lab's language releases are listed in Language models and datasets.