<p>To address the issues such as semantic distortion and poor syntactic adaptability that often occur when neural machine translation processes long English sentences, this paper took semantic relations as the core and integrated a long-sentence segmentation module based on semantic vectors into the traditional gated recurrent unit (GRU) encoder-decoder machine translation algorithm, constructing an optimized model for long English machine translation model. The research first designed the process of semantic vector extraction and segmentation word recognition to achieve accurate segmentation and re-ordering of clauses in long English sentences. Subsequently, an experimental dataset was built based on the Tsinghua Chinese-English parallel corpus, and the performance of the model was verified through simulation experiments in two steps. First, the recognition effect of the long-sentence segmentation module on segmentation words in long English sentences of different lengths was tested. Second, the translation performance of the proposed model was compared with that of traditional algorithms based on long short-term memory (LSTM), GRU, and Transformer. The experimental results showed that the designed long-sentence segmentation module had stable recognition performance and high accuracy for segmentation words. The average precision, recall rate, and F-score reached 0.97, 0.96, and 0.96 respectively, and were not significantly affected by the increase in sentence length. The optimized machine translation model performed best in the long English sentence translation task. Its Bilingual Evaluation Understudy (BLEU) score remained stable between 0.483 and 0.486, which was significantly higher than that of the other algorithms. Moreover, the text translated by the optimized model had a higher degree of agreement with the manual standard translation, effectively improving the accuracy and syntactic adaptability of long English sentence translation.</p>