Automated Translation and the Performance of BLEU and METEOR: Economic Translation as a Case Study[Traduction automatique et performance des métriques BLEU et METEOR: Cas de la traduction économique]

This article analyzes the performance of the BLEU and METEOR metrics in terms of linguistic aspects of translated texts. In fact, style, terminology, and sentence structure are all factors that influence the quality of these scores. This is particularly true in the context of economic discourse. Commonly used in the field of machine translation (MT), these metrics spark debates regarding their effectiveness and relevance, particularly due to their limitations in assessing the qualitative and contextual aspects of the translations produced. Therefore, the present study attempts to describe this impact and show how the characteristics of economic texts are of great importance in BLEU and METEOR evaluations. Two machine translation tools were used in this study: DeepL and Google Translate. They operate using deep learning and neural machine translation and rely on significant advancements in deep learning and neural machine translation. The choice of these tools is not accidental. It is rooted in the contemporary context, characterized by growing interest in this technology. This paper targets translation professionals and researchers in translation and translation studies, aiming to provide a critical analysis of the BLEU and METEOR scores, widely used in the machine translation field but also subject to much debate. Hence, the study addresses some ambiguities by offering a more precise and nuanced view of these evaluation tools.