Active and equal participation is important for the regulation of collaboration (Isohätälä et al., 2017). An approach for improving such regulation in collaborative learning is providing prompts or scripts from software or an agent to regulate or guide the group members’ activity (e.g., Borge et al., 2018; Vogel, et al., 2016, for a review). Conversational agents are widely used for prompting and scripting computer-mediated discussions, and prior studies showed a range of their effectiveness in supporting collaborative learning (e.g., Dyke et al., 2013).
Social agents such as robots and virtual agents are typically applied to regulate face-to-face collaborative learning (Okita & Clarke, 2021). Such social agents for collaborative learning are expected to support interactive learning through dialogues among learners by providing appropriate feedback with less authority than that when teachers do so, thus the learners can be aware of and internalize such scripts in their learning processes. As such, a social agent is expected to work with students as “a learning peer” who helps them enjoy their classroom or small group settings, and thus the students remain engaged in the learning process and maintain apt participation in the discussion (Miyake & Okita, 2012).
Ochibi is a three-dimensional holographic embodied conversational agent that can work as a facilitator for a student group discussion, providing suitable collaboration prompts (Strauß & Rummel, 2021) with some verbal and non-verbal reactions (nodding, pointing, turning its face sideways, etc.). The agent system collects all voice data of each member from a headset fitted with a directional microphone and calculates a variety of variables such as how long each student utters within every 20 seconds, a group’s ratio of silence durations for every 20 seconds, each student’s frequency of utterances, the turn-taking frequency and directions, etc. The system calculates such variables in real-time, and the agent provides prompts as configured when the threshold matches the corresponding pattern.
Ochibi itself does not make sophisticated judgments like an AI or speak up on its own as it deems appropriate. However, the research revealed that an equilibrium in the level of participation among participants is created when supported by the agent’s prompts (Nishimura, et al., 2004). We have also analyzed how relationships are built between agents and participants (Mochizuki, et al. 2023). The relationship development between a social agent and learners has rarely been considered in the CSCL, even though the literature on peer learning has identified that the quality of the relationship impacts the quality of peer learning (Riese et al., 2011). Even though users may consider conversational agents as more goal-oriented tools rather than social actors for establishing communication (Rehu et al., 2020), there is a need to investigate how learners perceive an agent and how they establish a relationship with the agent so that they have a quality socio-emotional state with the agent that leads to improve and regulate their collaborative discussion.
This project is in collaboration with Dr. Hiro Egi (University of Electro-Communications) and Dr. Yutaka Ishii (Okayama Prefectural University)’s teams.
Recent Publications
Hisatomi, A., Ishii, Y., Mochizuki, T., Egi, H., Kubota, Y., & Kato, H. (2017). Development of a Prototype of Face-to-Face Conversational Holographic Agent for Encouraging Co-regulation of Learning. Proceedings of the 7th International Conference on Human-Agent Interaction (HAI2017), pp.308-310【Best Poster Award】
Mochizuki, T., Egi, H., Ishii, Y., Yuki, N., Kubota, Y., Kato, H., & Takeuchi, M. A. (2023) Investigating Relationship Development Processes between 3D Conversational Agents and Learners in Collaborative Discussions. In Damșa, C., Borge, M., Koh, E., & Worsley, M. (Eds.), Proceedings of the 16th International Conference on Computer-Supported Collaborative Learning – CSCL 2023 (pp. 201-204). International Society of the Learning Sciences.