How adaptable is the ChatGPT large language model for translating different text types? An Empirical Study

Abstract

Recent advancements in Artificial Intelligence and Large Language Models, such as ChatGPT, have improved machine translation capabilities. However, little research explores their adaptability across different text types. This study evaluates ChatGPT's translation quality using Reiss' text typology theory, focusing on three text types: informative, expressive, and operative. Through a corpus-based approach, the study combines automated and human evaluations, with textual features analyzed using Coh-Metrix 3.0. The results show significant variation in ChatGPT’s translation quality, with the best performance in informative texts and lower quality in expressive and operative ones. The study also shows that while ChatGPT captures general variations in text types, it struggles to replicate the nuanced characteristics of specific genres as accurately as human translators.

Ähnliche Arbeiten