Lingue generate: i large language model tra riflessione linguistica e implicazioni didattiche

The rapid diffusion of Generative Artificial Intelligence tools, particularly Large Language Models (LLMs), challenges linguists and educators to reassess key linguistic notions. Drawing on Coseriu’s tripartite model, this contribution examines whether and how its categories can be applied to LLMs. It is argued that while LLMs do not contain a linguistic system in the traditional sense, they simulate multiple systems probabilistically based on large training datasets. Similarly, although they do not encode a socially shared norm, they reflect statistical regularities of language use. Finally, LLMs cannot be said to produce acts of parole, as they lack intentionality, subjectivity, and communicative purpose. Nonetheless, several pedagogical implications emerge: the probabilistic nature of LLMs, their ambiguous treatment of norms, and their simulation of dialogic interaction suggest valuable opportunities for developing learners’ metalinguistic awareness and for critically addressing language variation, usage, and acceptability in educational contexts.