As an important carrier of scientific communication, the abstract part of medical literature carries the key function of conveying core information. However, the differences in academic writing traditions between China and the West have led to significant differences in the semantic expression, rhetorical structure and information organization of Chinese and English medical paper abstracts. This study combines semantic role annotation technology and CARS model to explore the semantic expression differences between Chinese and English medical abstracts, and proposes a corresponding translation optimization method. By constructing a dataset containing 13,791 medical papers with a total of 249,762 corpora, the improved LSTM-CRF model is used for semantic role annotation, and the modified CARS model is applied to analyze the rhetorical structure of abstracts. The experimental results show that the LSTM-CRF model proposed in this paper performs well in the semantic role annotation task, with a precision rate of 86.6%, a recall rate of 87.1%, and an F1 value of 86.8%, which is an improvement of more than 9% compared with the comparison model. The speech step analysis shows that Chinese dissertation abstracts are used 8124 times in speech step 2, which is more than twice as many as 4016 times in English dissertation abstracts. In the translation performance evaluation, the BLEU value of the translation model incorporating semantic role features is improved by 5.47% to 7.50% compared with the comparison model, and the TER metric is reduced by 0.257 to 0.452. In the semantic component recognition experiments, the recognition accuracies of the eight major types of medical semantic components are over 90%. The results demonstrate that the combination of semantic role annotation and CARS model can effectively identify the expression differences between Chinese and English medical abstracts and significantly improve the quality of machine translation.
