Traduzione automatica neurale in ambito letterario dall'inglese all'italiano: un caso di studio sulla saga Hunger Games

In the Italian publishing industry, literary translation plays a key role in the market dynamics due to the strong interest of Italian readers in foreign fiction. This highlights the need to address market trends quickly and efficiently. In recent years, with the success of Machine Translation (MT), and the rise of Neural Machine Translation (NMT), the possibility of applying such technologies to translate literary texts has become an increasingly viable option. This dissertation aims to assess the quality of Machine Translation in the literary field, focusing on the English-to-Italian language combination and offering a comprehensive overview of its current limitations and potential. It focuses on dystopian young adult literature, and specifically on the Hunger Games saga by Suzanne Collins. Three extracts were chosen as representative of three distinct narrative modes, i.e., dialogue, description, and action. Translations were generated using two models of the ModernMT system, a general-purpose one, and one specifically tailored for literary texts. The evaluation was carried out using three approaches: automatic evaluation, error analysis, and quality analysis. These methods allowed me to determine whether domain adaptation improves the quality of MT output and to identify the main errors and challenges for Machine Translation. The results revealed that the adapted model has slightly outperformed the generic model, especially with regards to terminology. However, its overall quality remains insufficient for publication due to many stylistic and semantic errors. This suggests the need to further explore literary Machine Translation from English into Italian in future studies.