Analyzing Conversational Responsivity with AI in Education

This lecture delves into the computational analysis of conversation dynamics, focusing on participant responsivity. We discuss how responsivity distinguishes 'true dialogue' from 'parallel monologue,' examining failures of semantic similarity baselines versus the effectiveness of LLM-based methods like GPT-4. Results show LLMs achieving high alignment with human evaluations (91.9%), revealing insights from civic conversation case studies, e.g., personal stories fostering greater responsivity....