Automatic Translation of English News Text Based on Transformer Model

There are problems in translating English news content into the target language accurately and smoothly. This paper adopts the Transformer model with Encoder-Decoder structure, and uses the self-attention mechanism and multi-attention to capture the dependency in the text to achieve efficient translation. Firstly, the English news text and the corresponding text in the target language are preprocessed. After the data is loaded into memory, word segmentation is carried out, and the sentence is decomposed into individual lexical units. Based on these words, a glossary is constructed, and then the Transformer model is constructed. The encoder captures the dependency in the sentence, and the Decoder gradually generates the translation result of the target language according to the output. The final results show that the minimum translation accuracy of Transformer model is 90% and the maximum translation accuracy is 99.8% at most data points, while the accuracy of NMT model is between 81.1% and 94.9%. In terms of fluency, the TER value of the Transformer model is lower than that of the NMT model, which indicates that its translation results are more fluent. This paper not only promotes the development of translation technology, but also provides a powerful reference for the research and application of machine translation technology in the future, and provides a new solution to solve the challenges in cross-language communication.