<![CDATA[The advancement of Artificial Intelligence (AI) technology has increased the use of automatic translation systems in cross-language communication, including Japanese-Indonesian business translation. The characteristics of Japanese business language, which involve honorific expressions (keigo), indirect communication, professional terminology, and cultural conventions, remain challenging for AI-based translation systems. This study aims to identify the types of AI translation errors, classify them using a Translation Quality Assessment (TQA) approach based on Multidimensional Quality Metrics (MQM), and analyze the factors contributing to these errors. This research employed a descriptive qualitative method using translation error analysis. The corpus consisted of 50 Japanese business texts translated into Indonesian using three AI translation systems-Google Translate, DeepL, and ChatGPT-resulting in 150 translation units. The data were analyzed using six MQM categories: Accuracy, Terminology, Fluency, Grammar, Style, and Locale Convention. The findings revealed 90 translation errors among the 150 units analyzed. The dominant error category was Style/Keigo, with 45 cases (50%), followed by Accuracy with 23 cases (25.5%), Fluency with 9 cases (10%), Locale Convention with 8 cases (8.9%), and Terminology with 5 cases (5.6%), while the Grammar category showed no significant errors. These findings indicate that although AI translation systems demonstrate strong linguistic and grammatical capability, they continue to face challenges in interpreting pragmatic meaning, politeness levels, and the cultural values embedded in Japanese business communication]]>
