Analysis of Translation Errors in Chinese-Russian Neural Machine Translation

With the continuous advancement of Sino-Russian scientific and technological cooperation, the demand for translating specialized texts such as technical literature and reports has surged, making it difficult for traditional human translation to meet the requirements for efficiency and scale. The current accuracy of Chinese-Russian Neural Machine Translation (NMT) systems in handling specialized scientific and technical texts remains suboptimal, exhibiting significant issues particularly in terminology, consistency, and structural conversion. The innovations of this paper are twofold: first, the construction of a Chinese-Russian parallel corpus covering domains such as mechanical engineering and computer science, addressing the limitations of general-domain corpora; second, the adoption of a multi-system comparative framework to conduct quantitative evaluation and qualitative analysis of the Chinese-to-Russian translation performance of three mainstream NMT systems. This study constructs a parallel corpus encompassing both scientific/technical and general domains, evaluates the Russian-to-Chinese translation performance of three mainstream NMT systems, and performs error type statistics and analysis based on BLEU scores and manual annotation. The results show that the average BLEU score for scientific and technical texts is 0.19 points lower than that for general texts, with terminology mistranslation, word order errors, and semantic confusion identified as the primary error types.