In response to the errors that traditional translation models may make when processing complex language structures and identifying subtle semantic differences, this paper proposes a translation error correction method based on the Bidirectional Encoder Representations from Transformers (BERT) model. This paper uses the BERT model combined with specific error correction strategies and implementation steps to improve the error correction capability during the translation process. This paper first performs preliminary error detection on the translation output, uses the BERT model to encode the translated text and capture its semantic and grammatical features; designs a BERT-based error correction strategy to identify potential translation errors by comparing the distance between the translated text and the source text in the BERT encoding space. For the identified errors, this paper uses the Masked Language Model (MLM) function of the BERT model to correct the errors and generate more accurate translation texts; finally, iterative training and optimization of model parameters are used to further improve the error correction capability. The BERT model has a significantly higher error detection rate than the Errant rule system, with the average BLEU score increasing from 0.7055 to 0.917, and the TER score also decreasing. This paper successfully improves the error correction capability in the translation process, reduces translation errors, and improves translation quality.