Construction of English Language and Literature Translation Quality Assessment Model Based on Deep Neural Networks

Abstract

English literature translation is a systematic and complex project, the study is based on the assessment of English language and literature translation quality, mBERT, XLM-R, mBART are selected for cross-language pre-training, and a translation quality estimation model based on the pre-training model of Transformer encoder-decoder structure (mBART) is constructed. The mainstream pre-trained language models are selected for performance comparison, and the mBART-based translation quality estimation model proposed in this paper achieves the best results, surpassing the baseline model under different translation tasks in multiple datasets. The average absolute error and root mean square error of the model training in this paper are less than 0.15, and the correlation coefficients in English and German translations and English and Chinese translations are higher than those of the baseline model by 0~4% and 0~13.3%, and are less affected by the number of translated sentence types and utterances, which indicates the accuracy and stability of the translation quality estimation model of English language and literature in this paper. In conclusion, the proposed model is suitable for use in the assessment of English language and literature translation quality.