ACCURACY ANALYSIS OF LEXICAL RESOURCES FOUND IN ECONOMIC NEWS ARTICLES TRANSLATED USING DEEPL

Abstract

This study aimed to explore how accurately DeepL translates lexical resources on specific topics written in news articles. This research employs Mona Baker's equivalence theory to analyze and identify the equivalence of translation produced by DeepL. The data were taken from 7 randomly selected news articles with the economy or business genre published under a digital news platform in Indonesia established in 2016 and was located in Central Jakarta. The selection of these articles was intended to reflect authentic language use that was recently used in economic reporting. The lexical resources for this study were specific terms used in economic domains identified using Cambridge Dictionary's "Business English" section as well as an Indonesian Glosarium Ekonomi. Based on the result of the analysis, the findings showcased a 100% accuracy for 70 lexical resources, indicating that DeepL is a reliable neural machine translation in handling economic terminology. The high accuracy that DeepL produced was linked to the nature of economic terms, where economic vocabulary was globally recognized and was also consistently used across languages, hence allowing universally used terms across nations. The second reason linked to its high accuracy was due to how advanced DeepL’s advanced neural networks are. DeepL employs a Transformer architecture and attention mechanism, therefore allowing DeepL to be able to analyse sequences of texts by capturing contextual information efficiently. Additionally, with its system, DeepL was able to analyze more accurately in comparison to past models like Recurrent Neural Network, where it was showcased that past models struggled in translating a text, leading to loss of context and information. The final reason linked to its high accuracy is due to the large numbers of datasets. The growing number of users allows the developer to stay up-to-date and strengthen the model’s ability in translating different languages. It further allows the developer to identify opportunities for improvement and innovation.