Article: A method for improving English text translation quality by combining BERT and knowledge graphs Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.80 pp.1 - 25 Abstract: Facing global cross-language interaction scenarios, the demand for high-quality English machine translation is increasingly urgent. However, current research suffers from issues such as lack of structured knowledge and long-text semantic fragmentation. To tackle these issues, a method is proposed in this paper that leverages both bidirectional encoder representations from transformers (BERT) and a knowledge graph to improve English text translation quality. An improved TransR model is employed to extract entity embeddings from the knowledge graph, obtaining knowledge-enhanced representations. Multi-layer BERT outputs are fused and a masked knowledge matrix is constructed to suppress semantic noise. An adaptive fusion module is designed to dynamically integrate BERT knowledge with encoder context, jointly optimising translation and knowledge losses. Experimental results show that BTKG achieves BLEU scores of 34.6 and 33.2 on English-German and English-Chinese translation tasks, with translation accuracies of 92.7% and 90.1%, respectively, significantly outperforming existing baseline methods. Inderscience Publishers - linking academia, business and industry through research

Title: A method for improving English text translation quality by combining BERT and knowledge graphs

Authors: Hao Tang; Chang Li

Addresses: School of Foreign Studies, Xiangnan University, Chenzhou, 423000, China ' School of Foreign Studies, Xiangnan University, Chenzhou, 423000, China

Abstract: Facing global cross-language interaction scenarios, the demand for high-quality English machine translation is increasingly urgent. However, current research suffers from issues such as lack of structured knowledge and long-text semantic fragmentation. To tackle these issues, a method is proposed in this paper that leverages both bidirectional encoder representations from transformers (BERT) and a knowledge graph to improve English text translation quality. An improved TransR model is employed to extract entity embeddings from the knowledge graph, obtaining knowledge-enhanced representations. Multi-layer BERT outputs are fused and a masked knowledge matrix is constructed to suppress semantic noise. An adaptive fusion module is designed to dynamically integrate BERT knowledge with encoder context, jointly optimising translation and knowledge losses. Experimental results show that BTKG achieves BLEU scores of 34.6 and 33.2 on English-German and English-Chinese translation tasks, with translation accuracies of 92.7% and 90.1%, respectively, significantly outperforming existing baseline methods.

Keywords: English text translation; bidirectional encoder representations from transformers; BERT; knowledge graph; knowledge embedding; adaptive fusion.

DOI: 10.1504/IJICT.2026.155105

International Journal of Information and Communication Technology, 2026 Vol.27 No.80, pp.1 - 25

Received: 11 Apr 2026
Accepted: 14 May 2026
Published online: 27 Jul 2026 *

Title: A method for improving English text translation quality by combining BERT and knowledge graphs

Keep up-to-date