Exploring How Agent Voice Accents Shape Human-AI Collaboration in K-12 Group Learning

Synthesis: Ravi et al. investigate how the voice accent of a Generative AI conversational peer agent shapes learners' perceptions, trust, and interactional dynamics in K-12 group learning. While prior work examined agent accent effects in one-to-one settings, little is known about how these effects manifest in multi-party group contexts — a critical gap as AI peer agents enter collaborative classrooms.

The between-subjects mixed-methods study involved 33 teachers interacting with a GenAI voice agent in three accent conditions — British, Indian, and African American. Key findings:

1. The British-accented agent was largely treated as a tool and engaged with in detached, utility-based ways — less anthropomorphized, more like an external resource.
2. Indian- and African American-accented agents were more readily anthropomorphized and integrated as peers — building stronger trust, engagement, and reliance over time.
3. These role expectations influenced collaboration dynamics: turn-taking, questioning patterns, and perceived social presence all shifted based on accent condition.

The findings advance understanding of how GenAI's sociolinguistic design features shape group dynamics in CSCL (computer-supported Collaborative Learning), with implications for designing culturally inclusive AI partners. The work connects to Human AI Collaboration debates about teacher and AI roles in collaborative classrooms, and raises important questions about how accent may inadvertently reinforce or disrupt power dynamics in educational AI.

What this means for practice

• Instructors. Name the agent's role explicitly before learners meet it and use discussion-based value prompts rather than information-focused icebreakers: the authors found their own icebreaker failed to scaffold Phoenix's role, so groups defaulted to familiar voice-assistant mental models and evaluated the system instead of partnering with it.
• Instructors. Choose the agent's voice against your pedagogical goal rather than as neutral polish: an agent perceived as a detached tool is used transactionally, one perceived as an authority invites uncritical deference, and one perceived as a peer invites dialogic engagement — with accent subtly steering which of these orientations appears.
• Instructors. Set engagement norms for how the agent's contributions are evaluated and design activities around them, since over-humanization can foster unrealistic expectations, discomfort, or distraction from the task.
• Learners. Push back on the agent rather than deferring to it: the authors warn that group members may defer uncritically to an agent they read as an authority, while a peer-perceived partner supports dialogic, collaborative engagement.
• Designers. Treat a group-facing agent's voice as part of the activity design, not an after-the-fact styling choice: sociolinguistic cues shape authority and participation in the classroom, so the role framing and the voice need to be designed together.

Limitations

• The convenience sample is 33 teachers working in 11 small groups, recruited as educators predisposed to educational technology; the authors state this constrains generalizability, and the agent was fielded with teachers rather than students.
• Interactions were one-off and lab-based rather than longitudinal classroom use, so the study captures only a snapshot of how agent roles and group dynamics vary with accents.
• A technical oversight meant two Black-accent groups received no audible agent output, so the condition structure became unbalanced — Black (n = 6), voiceless (n = 6), Indian (n = 11), British (n = 10) — and only 31 of 33 participants returned post-surveys.
• Technical constraints including latency and limited model transparency affected interaction flow and trust, and the quantitative check found no significant accent differences on any CASUX subscale (Proficiency F(3, 27) = 0.232, p = .873; Etiquette & Mannerism F(3, 27) = 1.393, p = .266; Personality F(3, 27) = 0.776, p = .517), leaving the sample possibly underpowered for small effects.

Citation

Ravi, P., Stevens, C., Hurt, B., Hanks, B., Lin, G., & Anderson, E. (2026). Exploring How Agent Voice Accents Shape Human-AI Collaboration in K-12 Group Learning.