The evaluation of the translation quality of literary works has always been a challenging issue in translation studies. Both manual evaluation and traditional single index automatic evaluation have certain limitations. This paper uses the BERT deep learning model and a publicly available large scale corpus pre trained corpus to evaluate and analyze the English novel The Razor’s Edge and its Chinese translations from the perspectives of text similarity, sentiment scores, and the consistency of imagery and themes. Quantitative indicators are used to represent the translation quality. The research shows that the selected two Chinese translations faithfully reproduce the content of the original text and demonstrate a high level of translation quality. The study also confirms the feasibility of using BERT based word vector technology to quantitatively evaluate and analyze the quality of literary translation.