Generative Artificial Intelligence-Assisted Language Learning: Cognitive Mechanisms, Theoretical Tensions, and Ethical Governance

Generative artificial intelligence (AI) is changing the environment of language learning, but its theoretical basis and ethical boundaries have not been fully explored. This paper builds a multi-level analytical system for AI-assisted language learning based on cognitive load theory, the zone of proximal development, generative linguistics and the technology acceptance model. AI can help students learn better by reducing extraneous cognitive load and providing easy-to-access support and instant correction; however, due to technological limitations, it may also reduce students' sense of ability and self-management. Generative linguistics and generative AI are not in opposition; rather, the former advocates for data frugality and innate mechanisms, while the latter is characterized by data maximalism and statistical learning; they are in harmony at the level of third factors and the Minimalist Program. According to the above research, chatbots have shown a modest positive effect on language learning; this degree varies depending on the learner's level, interface Design, interaction characteristics, etc. Based on the above, this paper introduces a "Smart Chinese" collaborative learning plugin that integrates ChatGPT with special scaffolding tools for assessment, planning, practice, feedback and community support. Finally, it presents the ethical problems of algorithmic bias, data privacy, over-reliance, and educational gaps, and proposes a multi-level governance system at the institutional, technological and pedagogical levels.