Neural machine translation has been widely concerned in recent years. The traditional sequential neural network framework of English translation has obvious disadvantages because of its poor ability to capture long-distance information, and the current improved framework, such as the recurrent neural network, still cannot solve this problem very well. In this paper, we propose a hybrid neural network that combines the convolutional neural network (CNN) and long short-term memory (LSTM) and introduce the attention mechanism based on the encoder-decoder structure to improve the translation accuracy, especially for long sentences. In the experiment, this model is implemented based on TensorFlow, and the results show that the BLEU value of the proposed method is obviously improved compared with the traditional machine learning model, which proves the effectiveness of our method in English-Chinese translation.

the designed system can interact in real time. It improves the fun of students' English practice and teaching effect, and the performance of the system is stable.

English has become one of the most widely used languages in the world. If there is no good translation mechanism for such a widely used language, it will bring trouble to both study and life. At present, the world’s major platforms are committed to the study of English translation strategies. There are translation platforms from different regions and different translation mechanisms. These translation data from different translation platforms have the characteristics of large-scale, multisource, heterogeneity, high dimensions, and poor quality. However, such inconsistent translation data will increase the translation difficulty and translation time. Therefore, it is necessary to improve the quality of translation data to achieve a better translation effect. How to provide a large-scale and efficient translation strategy needs to integrate the translation strategies of various platforms to perform heterogeneous translation data cleaning and fusion based on machine learning. At first, this paper represents the multisource, heterogeneous translation data model as tree-augmented naive Bayes networks (TANs) and naturally captures the relationship between the datasets through the learning of TANs structure and the probability distribution of input attributes and tuples, using data probability value to complete the classification of translation data cleaning. Then, a multisource, heterogeneous translation data fusion model based on recurrent neural network (RNN) is constructed, and RNN is used to control the node data of hidden layer to enhance the fault-tolerant ability in the fusion process and complete the construction of fusion model. Finally, experimental results show that TANs-based translation data cleaning method can effectively improve the cleaning rate with an average improvement of approximately 10% and cleaning time with an average reduce about 5%. In addition, RNN-based multisource translation data fusion method improves the shortcomings of the traditional fusion model and improves the practicability of the fusion model in terms of root mean square error (RMSE), mean absolute percentage error (MAPE), fusion time, and integrity.