Oral practice in language teaching usually takes place in classroom settings, offering limited opportunities to learners for practice. The development of GenAI and the increased possibility of conversing in digital safe spaces with a chatbot brings about new hope for more inclusive approaches to the learning and teaching of foreign languages.
Oral practice is a cornerstone of language learning, yet in traditional classroom settings, opportunities to speak can be limited - especially for learners who experience anxiety or lack confidence. The ‘Voice Chatbot for Language Learning’ project set out to address this by developing an AI-powered voice chatbot that offers low-pressure, personalised speaking practice. Learners would then receive a transcription of both their own speech and the chatbot’s responses, allowing them to review and reflect after each conversation.
Focusing initially on French, the team - supported by Dominik Lukes, Lead Business Technologist, and Edward Fauchon-Jones, Senior Research Software Engineer, from the AI Competency Centre - first tested GPTs within ChatGPT Edu and other voice chatbot platforms. The project then progressed to a standalone application building on the work done with the GPTs, resulting in a first test rollout in week 4 of Trinity term.
In week 9, Dominik Lukes ran workshop at the Language Centre which brought together language teachers to explore the tool’s potential in Russian, Spanish, and Italian. Teachers responded enthusiastically, noting the chatbot’s adaptability and potential for integration into their courses.
Results were encouraging: the chatbot demonstrated strong comprehension of learner input, with some students sustaining extended exchanges even when gaps in understanding arose. Requests for continued access over the long vacation reflected learner engagement and enthusiasm.
However, the project also surfaced important considerations. Not all students were keen to engage, citing time constraints, environmental concerns about AI, or uncertainty over institutional support. The team noted that discipline-specific use would require careful prompt design to avoid overly general or verbose AI responses. Key lessons included ensuring that students have the choice not to use AI, embedding tools in meaningful scenarios, and involving learners as active partners in development.
AI-Driven Feedback Loops
AI in Education