Apprendimento degli idiomatismi e modelli del linguaggio : una valutazione di GPT

Large language models, which exploit neural architectures to generate text, are now easily accessible even to people not skilled in such technologies due to the possibility of interacting through a chatbot or conversational agent. For this reason, they can be a source of linguistic knowledge, particularly for students or language learners, who take advantage of the generative capabilities of the models to ask for information or to perform text-writing tasks in different languages. In this contribution, we propose an evaluation of GPT as a tool for the acquisition of idioms. We use a corpus consisting of 900 replies of GPT-3.5-turbo referring to 300 verbal locutions extracted from the GRADIT dictionary (De Mauro 1999) to evaluate the model capability to provide correct meanings of the selected locutions, together with the contextual examples and support in this way the learning of Italian idioms.