Possibilities and limitations of generative artificial intelligence for the personalization of ELE/L2/LH teaching
The presentation began by outlining the general limitations of AI and reviewing recent data on the market penetration of major chatbots such as ChatGPT, Claude, Gemini, DeepSeek, Copilot, and Grok. It then examined how university students are using these tools for academic work. Shifting to the use of generative AI in second-language classrooms, it compared human language with the synthetic language produced by large language models to evaluate how effectively chatbots can personalize the learning of a foreign, second, or heritage language, in this case, Spanish.
The presentation is grounded in the insight that both pragmatic and humanistic approaches to personalized learning emphasize the need to consider students’ voices and needs when defining goals, learning pathways, and forms of assessment. It acknowledges that contemporary chatbots allow for a limited degree of customization through prompt engineering, retrieval-augmented generation (RAG), and the development of local LLMs. Nevertheless, it argues that inherent limitations of generative AI—such as its inability to admit ignorance, reliance on biased or incomplete training data, post-training incentives, and the fundamentally synthetic nature of its language—constrain personalization to largely remedial functions, practice activities, and supplementary explanations. Even within these domains, LLM errors can create challenges. Furthermore, the linear and often monotonous structure of chatbot-generated content can result in boredom and disengagement. Finally, the lack of genuine communicative intent and the absence of a true hermeneutic contract significantly limit the value of chatbots as tools for fully personalized learning, particularly for autonomous use without teacher supervision.
More information about the conference here.
This summary in English was refined with ChatGPT-5
November 28, 2025