Valutare la naturalezza dell’output prodotto dall’IA generativa: uno studio empirico sull’uso dei suffissi alterativi in un corpus di favole sintetiche e scritte da esseri umani

This study empirically assesses the naturalness of AI-generated texts in comparison with human-written ones, using a corpus of fables authored by humans and generated by four LLMs (GPT-3.5 Turbo, GPT-4o mini, Teuken-7B, and Minerva-7B) trained on different datasets. To evaluate the naturalness of the fables, the study focuses on alterative suffixes, a distinctive feature of Italian that commonly occurs in child-oriented text genres. The results reveal that synthetic fables exhibit a lower usage of these suffixes, suggesting an indirect Anglicisation of these texts.