Master Thesis: Predicting Pedagogical Intent through LLM Tool Calling

Summary

Manually annotating educational dialogues with dialogic acts (the pedagogical intent behind utterances) is costly, and LLMs still struggle to predict them. This thesis frames dialogic acts as tools to test whether tool-calling LLMs can transfer their tool-selection skills to this task. You will extend an educational dialogue benchmark, compare against conventional approaches, explore fine-tuning, and test generalization across contexts and taxonomies. Requires Python and LLM/NLP interest.