Construction of automatic error detection model for English-Chinese translation of cultural heritage terminology based on BERT-BiLSTM-CRF

Abstract

For a long time, there has been a lack of effective cultural semantic conversion and automatic detection methods for term inconsistency in the English Chinese translation of cultural heritage terminology. This paper constructs an automatic error detection model for cultural heritage terminology translation based on Bidirectional Encoder Representations from Transformers (BERT), Bidirectional Long Short Term Memory (BiLSTM), and Conditional Random Field (CRF). This paper first uses the English Chinese bilingual joint encoding input method based on special separators to achieve cross language semantic joint modeling. Then, this paper uses BERT to complete bilingual contextual semantic encoding, and then uses BiLSTM to enhance long-range time-dependent features and capture consistency constraints of the same term in different sentences. Finally, CRF ensures continuity and consistency across sentence boundaries through label transfer constraints, and combines transfer learning with domain fine-tuning to achieve adaptation in the cultural heritage field. The results showed that in the evaluation of error fragment level, the F1 score of the model was 92.8%; F1 achieved scores of 93.12%, 91.47%, and 90.86% in detecting mistranslations, omissions, and inconsistencies in terminology, respectively; while the illegal tag transfer rate had dropped to 0.03%. Research has shown that the proposed model has high structural stability and cross linguistic semantic discrimination ability in the task of detecting translation errors in cultural heritage terminology.