Teachers have long relied on information gaps, mingles & role-plays to get learners talking. These activities create movement, unpredictability & the kind of communicative pressure that gap-fills can’t. Role-play in particular has become a classroom staple: Ladousse popularised the term in ELT in the 1980s (with a book which was to be found on the shelves of countless staffrooms, back when we had books, shelves & staffrooms), though dramatic techniques had been used in language teaching long before that. The basic idea hasn’t changed -simulate a communicative situation, assign roles, rehearse the pragmatics- but the tools available to us certainly have.
It also sits neatly beside a core goal of contemporary ELT: intercultural communicative competence (ICC). ICC is the ability to navigate meaning, relationships & expectations across cultures while communicating in another language. Traditionally, teachers have tried to build this through peer role-play, but as Widdowson pointed out, monolingual classrooms often create a “blind leading the blind” dynamic. Learners share the same cultural assumptions, so the “otherness” of the interlocutor is hard to simulate.
Which brings us to a new study by Zhanbo Qu & Liyun Chen (Northwest Minzu University & Longnan Normal University), published in Frontiers in Psychology. It asks a timely question: what happens when learners practise intercultural role-play not with peers, but with a voice-based GenAI interlocutor designed to embody a specific cultural persona?
The study
Two classes of Chinese undergraduates (N=59) completed a 10‑week ICC course. One group practised weekly intercultural role-plays with Doubao, a voice-based GenAI agent configured through system prompts to adopt culturally coherent identities (e.g., a high‑context professor, where much meaning is conveyed implicitly through shared assumptions & indirectness, as opposed to a low‑context professor who prefers explicitness & direct verbal clarity). The other group did parallel role-plays with peers using role cards. Both groups received identical ICC instruction.
Learners in the AI group interacted individually for at least 15 turns per task, with the agent signalling misunderstanding (“I beg your pardon?”) rather than correcting errors mid-flow. After each dialogue, learners received AI-generated feedback & submitted chat logs. The peer group followed similar themes but worked in dyads [Ed. ‘Pairs’]. They assessed learners’ intercultural speaking performance, their perceived ICC, their speaking anxiety, patterns in their role-play interactions & insights from post-course interviews.
The findings
The GenAI group outperformed peers in intercultural speaking performance, barely edging ahead but still statistically significant. They also reported significantly higher perceived ICC. In addition, speaking anxiety dropped sharply in the AI group.
Interaction patterns were revealing. Social politeness dominated (33%), followed by topic extension (22%) & robust negotiation-of-meaning routines. Learners treated the AI as a legitimate social partner, not a drill machine. A socio-cognitive triangle emerged -politeness, repair, extension- showing learners sustained socially meaningful dialogue even when facing pragmatic or semantic challenges.
Interviews highlighted two major affordances:
- a psychologically safe space free from peer judgement
- persona fidelity & just-in-time cultural feedback that helped learners adjust register & tone
Constraints included the absence of paralinguistic cues such as tone, facial expression & gesture, which sometimes made pragmatic interpretation harder, as well as occasional AI logic loops when the agent struggled to maintain persona coherence.
Context
In monolingual classrooms, peer role-play often struggles to simulate genuine cultural difference. GenAI changes that by offering interlocutors who behave, respond & signal meaning in culturally distinct ways — something peers can rarely provide.
This aligns with Byram’s ICC framework, the interaction hypothesis & the CASA paradigm, which shows learners often treat computers as social actors. It also echoes work on willingness to communicate & affective filter theory: remove fear of negative evaluation & learners speak more, experiment more & adapt more.
A simple example: imagine a learner negotiating a complaint with a simulated British hotel manager. If they say, “You must fix this now”, the AI might respond with cool formality, prompting a shift to “I’m wondering if it might be possible to…”. That kind of situated pragmatic adjustment is hard to achieve with peers who share the same cultural background & lack the linguistic repertoire to model it.
Teacher takeaways?
- Use persona-based prompts to create richer pragmatic challenges than generic chat. These let you simulate culturally distinct communication styles -indirectness, formality, mitigation, turn‑taking norms- that peers can’t always model spontaneously. This gives learners a chance to practise adjusting tone, register & politeness strategies in response to a partner who behaves differently from them.
- Frame breakdowns as cultural moments, not linguistic failures. When misunderstandings occur, treat them as opportunities to explore how cultural expectations shape meaning. A repair sequence isn’t just “fixing language”; it’s a moment to notice how assumptions, politeness norms or indirect cues differ across cultures, & how learners can navigate them.
- Treat AI role-play as rehearsal before peer or real-world interaction, especially for anxious learners. AI interlocutors offer a psychologically safe space where learners can rehearse tricky interactions -complaints, refusals, negotiations- without fear of judgement. This lowers the affective filter, builds confidence & prepares them to transfer those pragmatic moves into peer work or real-world encounters.
How do you approach role-play in your classroom?
Leave a Reply