Improved Neural Machine Translation Model Based on Improved BLE-Transformer Architecture

With the continuous development of economic globalization and information globalization, the importance of translation has also increased. Neural machine translation technology uses deep learning methods to train and obtain translation models on parallel corpora, which has become the current mainstream machine translation method. This paper developed a Chinese-English translation system based on the Transformer translation model. A Seq2Seq translation framework based on the Transformer translation model is built, and the BPE embedding layer and prior knowledge encoder are added to solve the problem of excessive dependence on corpus in traditional translation, and the self-learning and training of corpus are realized. To solve the problem of mistranslating and missing translation, we improve the long short-term memory network, and propose a neural machine translation model with BLE-transformer architecture. The results show that compared with other translation models, the Chinese-English machine translation system proposed in this paper has improved the smoothness of sentences and the accuracy of semantics. The Test loss of the improved model reaches 3.95 and the Bleu Score reaches 26.57, which is significantly improved compared with the model before improvement, and helps to provide new ideas for the solution design of neural machine translation problems.