In response to the problem of insufficient translation accuracy of traditional statistical and rule-based machine translation methods, this paper introduces the Transformer algorithm, which aims to improve the accuracy and naturalness of English translation through a deep learning model. Firstly, this paper designs a translation model suitable for English to Chinese based on the basic principles of Transformer architecture, using self attention mechanism to capture long-distance dependencies. Then, a large-scale bilingual parallel corpus is used for training, and the order information of words in sentences is processed through positional encoding to ensure that the model can better understand grammar and semantics. Finally, to address the issue of overfitting that may occur during the training process, this paper introduces Dropout mechanism and learning rate decay strategy to optimize the model’s generalization ability. Through comparative experiments, this paper also evaluates the performance of the model and compares it with existing translation models based on RNN (Recurrent Neural Network) and CNN (Convolutional Neural Networks). The experimental results show that the Transformer based translation model has significantly higher BLEU (Bilateral Evaluation Understudy) scores than traditional translation models, especially in the translation of long sentences and complex structures, demonstrating stronger contextual understanding and language generation abilities. Firstly, Transformer generally outperforms RNN in translating sentences per minute, resulting in higher translation efficiency. For example, in test sentence 1, the Transformer translates 350 sentences per minute, while the RNN translates 250 sentences. Therefore, the Transformer architecture provides a broader prospect for future machine translation research and applications.