Traditional Chinese Medicine (TCM) terminology embodies not only a rich conceptual system but also complex cultural, philosophical, and clinical attributes. Because many terms are highly condensed, semantically abstract, metaphorically structured, and culturally embedded, translating them into English has long posed substantial challenges for semantic fidelity, scientific accuracy, terminological consistency, and communicative effectiveness. Guided by Eco- translatology, which emphasizes multidimensional adaptation across linguistic, cultural, and communicative dimensions, this study proposes a data-driven and knowledge-enhanced neural approach to TCM terminology translation. Specifically, we construct a syntax-aware convolutional model to identify internal semantic structures, including predicate–object relations and central words, and integrate TCM knowledge-graph embeddings, including term–herb–compound–target–pathway relations, into a TCM-BERT-based translation architecture through a knowledge-attention mechanism. We compile a bilingual corpus containing 3,825 standardized TCM terms and approximately 48,000 parallel sentence pairs, supplemented by pharmacological annotations from established resources such as TCMID, HERB, and TCMSP. Experimental results show that the proposed model outperforms the baseline system, increasing BLEU-4 from 35.4 to 48.1, an improvement of 12.7 points, and increasing the terminology-consistency rating from 2.8/5 to 4.2/5. Qualitative evaluation of representative terms such as “气血两虚证” and “肝阳上亢” further demonstrates the model’s ability to preserve syntactic structure while incorporating biomedical semantic information into English interpretations.
