AI

Artificial intelligence in ELT covers the cluster of computational technologies that produce, evaluate, or interact with natural language: large language models (GPT, Claude, Gemini), automated speech recognition, automated writing evaluation, neural machine translation, and adaptive learning platforms built on top of these. The field shifted from experimental to mainstream between 2022 and 2025, driven primarily by the public release of conversational LLMs and their absorption into classroom and self-study workflows (Kohnke, Moorhouse, & Zou, 2023).

What AI Actually Does for Language Learning

The headline applications cluster into four pedagogically distinct functions. Generation produces texts at controlled levels (reading passages, listening scripts, exercise items) and is now standard in materials production pipelines. Feedback evaluates learner output, with automated writing evaluation tools (Grammarly, Write & Improve, ETS e-rater) and pronunciation feedback engines (ELSA Speak, SpeechAce) offering targeted form-focused commentary. Conversation uses chatbots to provide low-stakes practice, with strong evidence for anxiety reduction and mixed evidence for fluency gains (Lyu, 2025). Personalisation drives adaptive platforms that adjust content difficulty and recycling based on learner performance.

The research base is still maturing. Lyu's (2025) meta-analysis reports a medium positive effect for chatbots on L2 learning (g = 0.608), but most published studies are short, treatment is heterogeneous, and the comparison conditions vary widely. Findings most consistent across studies: AI feedback works better in combination with teacher feedback than as a replacement, and task design determines outcomes more than the choice of tool (Godwin-Jones, 2022).

Limits

Three limitations have proved structural rather than transient. AI feedback degrades on pragmatics, discourse coherence, register, and culturally specific language: precisely the areas where advanced learners most need attention. Automatic speech recognition fails most often on the learner errors that need diagnosis, because the model trained on standard speech cannot decode the deviation. And LLM hallucination, especially in metalinguistic explanations, makes critical AI literacy a precondition for productive learner use rather than a nice-to-have.

See Also

For the full pedagogical treatment, applications by skill, integration patterns, and design principles, see AI in Language Teaching. For chatbot-specific evidence and tool comparisons, see Chatbots in ELT.

References

• Godwin-Jones, R. (2022). Partnering with AI: Intelligent writing assistance and instructed language learning. Language Learning & Technology, 26(2), 5–24.
• Kohnke, L., Moorhouse, B. L., & Zou, D. (2023). ChatGPT for language teaching and learning. RELC Journal, 54(2), 537–550.
• Lyu, B. (2025). Effectiveness of chatbots in improving language learning: A meta-analysis. International Journal of Applied Linguistics, 35(1).