대규모 번역 작업에서 GPT 도구의 활용 효과 : 한국어-러시아어 의료 교재 번역 사례를 중심으로

다국어 초록 (Multilingual Abstract)

The purpose of this study is to empirically examine how GPT’s Project, Memory, Canvas, and Branch features affect the operation and management of the translation workspace in a large-scale Korean–Russian medical textbook translation project. Whereas previous studies on LLM-based translation have mainly focused on translation quality, this study approaches GPT not simply as a translation generation tool, but as a workflow environment that organises guidelines, reference materials, terminology decisions, work history, and post-editing processes. The central problem addressed in this study is how context maintenance, norm application, terminology consistency, and workflow continuity can be supported or destabilised during a long-term specialised translation task.
The study is based on the Korean–Russian translation of the two volumes ofPractice! Medical Korean Translation and Interpreting for Professional Training, a textbook designed for Russian-speaking prospective Korean–Russian medical interpreters. The project included translation, revision, and final correction stages. The textbook is organised by medical department and contains terminology-learning sections, definitions, explanatory texts, dialogues, Q＆A texts, and medical interpreting practice dialogues. Methodologically, this study is a case-based comparative study with a mixed-methods design. It compares the two volumes as different GPT-use conditions within the same specialised domain and text structure. The analysis triangulates project work logs, chat records, draft and final translations, revision and proofreading records, glossaries, and lists of standardised expressions.
The findings show that the four GPT features supported different layers of the translation workflow. The Project feature reduced the operational cost of repeatedly restoring guidelines, reference materials, terminology criteria, genre-specific rules, and work history. The Memory feature supported the maintenance of approved Korean–Russian terminology decisions across chapters and helped stabilise the reuse of terminology and fixed expressions, although it did not function as a fully controlled termbase. The Canvas feature reorganised post-editing from a sequence of chat-based responses into the updating of a stable text state, making local revision, external-check integration, and subsequent review more manageable and observable. The Branch feature contributed to workflow continuity under long-chat conditions by mitigating response delays and allowing the work to continue while maintaining previously established rules and source-reference criteria.
Overall, the findings suggest that GPT workflow features should be understood not only in terms of individual translation outputs, but also in terms of changes in the operational structure of the translation process. In the medical textbook translation examined in this study, productivity was linked to the reduction of operational costs, including repeated explanation of rules, retrieval of previous decisions, terminology verification, reassembly of revised texts, and technical delays in long chats. At the same time, the study confirms that these features do not independently guarantee final translation quality. Human review, chapter-level glossaries, external verification, and MQM-based follow-up checks remained necessary. The contribution of this study lies in showing that GPT can be used in large-scale specialised translation as a workflow-support environment that helps organise the translator’s judgement, revision process, and long-term management of translation criteria.

목차 (Table of Contents)

• 1. 서론 13
• 1.1 연구 배경 13
• 1.2 선행 연구 15
• 1.3 연구 목적 및 필요성 17
• 1.4 연구 대상 및 방법 20
• 2. 이론적 배경 25
• 2.1 대규모 텍스트 번역과 번역 워크플로 25
• 2.2 의료 번역의 특성 36
• 2.2.1 의료 교재 39
• 2.2.2. 한국어‒러시아어 의료 번역 43
• 2.3 GPT 기반 번역 지원과 도구 개요 48
• 3. 교재 번역에서의 GPT 활용 절차 54
• 3.1 번역 작업 개요 54
• 3.2 초기 단계 GPT 활용 양상 60
• 3.3 후반 단계 GPT 활용 양상 64
• 4. GPT 프로젝트 활용 방식과 분석 72
• 4.1 비교 기준과 분석 자료 72
• 4.2 분석 결과 79
• 5장 GPT 메모리 활용 방식과 분석 103
• 5.1 비교 기준과 분석 자료 103
• 5.2 분석 결과 114
• 6장 GPT 캔버스 활용 방식과 분석 127
• 6.1 비교 기준과 분석 자료 127
• 6.2 분석 결과 137
• 7장 GPT 브랜치 활용 방식과 분석 229
• 7.1 비교 기준과 분석 자료 229
• 7.2 분석 결과 237
• 8장 결론 249
• 참고문헌 256
• 부록 1. 프로젝트 분석표 269
• 부록 2. 메모리 분석표 269
• 부록 3. 캔버스 분석표 260
• 부록 4. 브랜치 분석표 270