Accurate translation of English medical terminology is crucial in healthcare, where errors can lead to significant semantic distortion. Traditional rule-based and statistical machine translation methods often fail to handle domain-specific language due to limited contextual understanding. This paper proposes BBNMTPET (BioBERT-Neural Machine Translation Professional English Term), a deep learning framework that integrates BioBidirectional Encoder Representations from Transformers with NMT to improve medical term translation accuracy. The BBNMTPET framework aims to utilize a BioBERT-enhanced NMT model to effectively capture the contextual semantic features of medical language. To ensure consistent bilingual term mapping, an attention-guided terminology alignment mechanism is incorporated. Training utilizes adaptive learning rate scheduling, combined with domain-sensitive loss functions, to prioritize accurate medical term translation while enhancing convergence and generalization. The model is trained and evaluated on the Kaggle Accurate Medical Translation Data, a large-scale bilingual medical corpus. Experiments demonstrate that BBNMTPET achieves a 7.6% improvement in BLEU scores and a 9.2% increase in medical term translation accuracy compared to baseline NMT models, significantly reducing the mistranslation of rare and ambiguous terms. The integration of domain-specific BERT, terminology alignment, and optimized training strategies significantly advances the quality of professional medical term translation.
Nan Ni (Mon,) studied this question.