Linguistic challenges to algorithms: an error analysis of German-Chinese machine translation from a second language acquisition perspective

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Machine translation (MT) is crucial in today’s globalized world for enhancing efficient international communication, particularly for non-English-speaking communities. Despite advances in natural language processing, the complexity and diversity of human languages remain major challenges for MT. Meanwhile, the predominance of English in MT research and training resources constrains a more comprehensive understanding of MT performance. To bridge this gap, the current study conducted an exploratory study on MT errors in German-Chinese translation using Google Translate as the primary MT tool, complemented by a comparative analysis with Youdao Translate. Grounded in the error analysis approach of second language acquisition (SLA) research, a structured error dataset was collected through error annotation, covering parameters of translation errors, linguistic levels, and language features. Quantitative analysis was conducted on the error dataset to explore the vital error types, their distribution patterns, and associated linguistic factors in German-Chinese MT. The results showed that mistranslation was the dominant error type and occurred primarily at the lexical level. Besides, syntactic misalignments and contextual additions/hallucinations were also commonly observed. These error patterns were frequently associated with lexis with high semantic density (e.g., compound words) and complex sentence architectures (e.g., embedded clauses). Notably, while MT appears fluent and natural at a syntactic level, it often compromises semantic accuracy and logical coherence. The comparative analysis further revealed similar findings across both MT tools, suggesting that diverse word forms and complex sentence structures may pose broader challenges for German-Chinese MT. This exploratory study contributes to a better understanding of current MT in German-Chinese translation tasks and provides a methodological foundation and preliminary findings for future research based on larger datasets and comparative analysis.