Applying the mBART Multilingual Model to Optimize Translation Software’s Intelligent Error Correction Capabilities in English Multi-domain Terminology Conversion

Abstract

Existing translation systems generally lack dynamic disambiguation capabilities across domains, leading to frequent mistranslations in semantically ambiguous or interdisciplinary scenarios, while their terminology retranslation and error correction mechanisms remain limited. To address these issues, this study proposes an intelligent terminology correction framework based on the mBART multilingual model. By integrating multilingual context modeling, multihead attention mechanisms, and a terminology correction candidate list, the proposed approach dynamically identifies domain-specific meanings and generates accurate correction suggestions. Fine-tuning on parallel corpora from medicine, law, engineering, and electrical engineering enables effective intelligent error correction across multiple professional domains. Such capability is particularly valuable for technical communication involving electromagnetic engineering and related disciplines, where precise terminology translation supports reliable knowledge dissemination and interdisciplinary collaboration. Experimental results demonstrate that the proposed method significantly improves translation quality. Across 15 domains, the average BLEU score reaches 57.9, compared with 38.4 for the baseline model, while the error rate for ambiguous terms decreases to 24.3%. In the electrical engineering domain, terminology recognition accuracy reaches 0.94, and semantic fidelity in engineering applications reaches 96.1%, effectively enhancing context awareness and translation consistency for multi-domain terminology.