법률 도메인 특화 기계번역의 번역 품질 개선 분석– 계약서 중한 번역을 실례로 –

This study evaluates domain-adaptive machine translation for legal texts by developing a “contract-adaptive machine translation model”(contract-adaptive model). Using a 53,334-character corpus of Chinese contracts, we fine-tuned a neural translation system and assessed outputs using two semantic-oriented automatic metrics: BERTScore and BLEURT. The evaluation results demonstrate that the contract-adaptive model consistently achieved higher scores across both metrics. In histograms visualizing BERTScore and BLEURT, the contract-adaptive model was concentrated in the high-score regions, and its median values were higher in the corresponding box plots. Analysis of actual translation examples showed that the contract-adaptive model delivered improved translation quality tailored to contractual usage - in terms of syntactic and semantic features.