The accuracy of cultural words translation by DeepL in the novel Tarian Bumi

Abstract

These days, the usage of machine translation, such as DeepL, has increased in number. Furthermore, this study aims to assess the accuracy of DeepL in handling culturally nuanced words, in conjunction with Yildiz’s (2021) theory for evaluating translation accuracy. This study employs a descriptive-qualitative method to analyze the source data. Three independent professional translators assessed DeepL's accuracy. Of the 303 rich points identified, 95 of them are raters' consensus, followed by raters' dissent for 76 rich points. According to the assessment, DeepL can identify terms that have been administered globally. However, some shortcomings are apparent regarding words with culturally dense nuances. DeepL's inconsistency was also highlighted in translating a word that appeared 44 times but produced six different translations. For the accuracy of the translation, DeepL is given a final score of 0.54, within the range of -8 (Critical Error) to +4 (Acceptable). This score is considered partially acceptable and sufficient for initial understanding. Then, it concludes that DeepL's translation still requires human involvement in preserving cultural nuances. This research recommends enhancing MT systems with more diverse cultural words and phrases datasets. For future studies, it is suggested to use multiple MTs and quantitative metrics like BLEU scores for broader, comparative evaluations.