Elad Yacobson, Ofra Amir, Ayelet Baram-Tsabari E107 Proceedings of the 21st Chais Conference for the Study of Innovation and Learning Technologies: Learning in the Digital Era I. Blau, A. Caspi, Y. Eshet-Alkalai, N. Geri, Y. Kalman, D. Olenik-Shemesh, Y. Sidi, & N. Brandel (Eds.), Ra'anana, Israel: The Open University of Israel Advancing Dialogic Communication Skills through LLM-Based Simulation (Short paper) Elad Yacobson Technion – Israel Institute of Technology eladyacobson@campus.technio n.ac.il Ofra Amir Technion – Israel Institute of Technology oamir@technion.ac.il Ayelet Baram-Tsabari Technion – Israel Institute of Technology ayelet@technion.ac.il קידום מיומנויות תקשורת דיאלוגית באמצעות סימולציה מבוססת מודלי שפה גדולים (LLM) )מאמר קצר( אילת ברעם - צברי הטכניון – מכון טכנולוגי לישראל ayelet@technion.ac.il עפרה עמיר הטכניון – מכון טכנולוגי לישראל oamir@technion.ac.il אלעד יעקבסון הטכניון – מכון טכנולוגי לישראל eladyacobson@campus.techni on.ac.il Abstract Communication skills are increasingly recognized as essential for science students. However, existing communication training programs face scalability challenges due to limited resources, time constraints, and the need for expert instructors. Consequently, these programs rarely provide scientists with sufficient opportunities to practice dialogic communication. This study explores the use of large language models (LLMs), specifically ChatGPT-4o, as a novel tool for scalable and personalized science communication training. We developed an LLM-based dialogue simulator grounded in Prodigy, a new theoretical framework that defines productive dialogue across four dimensions: Content, Interpersonal Rapport, Perspective Taking & Listening, and Integrity & Humility. In an exploratory study with 37 science students, participants conducted two voice conversations with ChatGPT 4o, receiving AI-generated feedback between sessions. Analysis of the conversations revealed significant improvements in dialogic performance across three of the four dimensions, along with high levels of user satisfaction. These results suggest that LLMs hold promise as effective tools for scalable communication training. Keywords: large language models, science communication, soft skills training, computer-human interaction, AI in education.
RkJQdWJsaXNoZXIy Mjk0MjAwOQ==