Construction of English Translation Model Based on Neural Network Fuzzy Semantic Optimal Control

This work addresses four aspects of the English translation model: consistency, model structure, semantic understanding, and knowledge fusion. To solve the problem of lack of personality consistency in the responses generated by neural networks in English translation models, an English translation model with fuzzy semantic optimal control of neural networks is proposed in this study. The model uses a fuzzy semantic optimal control retrieval mechanism to obtain appropriate information from an externally set English information table; to further improve the effectiveness of the model in retrieving correct information, this work adopts a two-stage training method, using ordinary English translation data for model pretraining and then fine-tuning the model using English translation data with optimal control containing fuzzy semantic information. The model consists of two parts, a sequence generation network that can output the probability distribution of words and an evaluation network that can predict future whole-sentence returns. In particular, the evaluation network can evaluate the impact of currently generated words on whole sentences using deep inheritance features so that the model can consider not only the optimal solution for the current words, as in other generative models, but also the optimal solution for future generated whole sentences. The experimental results show that the English translation model with fuzzy semantic optimal control of the neural network proposed in this study can obtain better semantic feature representation by using a novel bidirectional neural network and a masked language model to train sentence vectors; the combination of semantic features and fuzzy semantic similarity features can obtain higher scoring accuracy and better model generalization. In English translation applications, there are large improvements in scoring accuracy and generality.