Research on English translation quality evaluation and automatic optimization algorithm based on deep learning

With the rapid development of machine translation technology, traditional translation quality evaluation methods have been unable to meet the needs of high efficiency and accuracy. In this study, an English translation quality evaluation and automatic optimization algorithm based on deep learning is proposed. Through the feature extraction ability of deep learning model and the dynamic adjustment of optimization algorithm, the accurate evaluation of translation quality and the significant improvement of translation model performance are realized. In this study, the Transformer model is used as the infrastructure, and the quality evaluation module with double encoder structure is combined to capture the semantic and grammatical information between the translated text and the reference text through self-attention mechanism and cross-attention mechanism, and a joint training strategy is designed to combine the loss of translation quality evaluation with the loss of cross-entropy of translation tasks, and dynamically adjust the optimization target weight. The experimental results show that the quality evaluation model has reached the highest value in automatic evaluation indexes such as BLEU-4, METEOR and BERTScore, and the Pearson correlation coefficient with manual evaluation is 0.85, showing superior evaluation consistency. The joint optimization algorithm is significantly superior to the baseline model and other mainstream methods in the scores of BLEU-4, TER and artificial fluency, which effectively balances the accuracy and naturalness of translation. In addition, ablation experiments verify the key roles of dynamic attention mechanism, KL divergence constraint and dynamic weight adjustment strategy on model performance.