Performance Comparison of LLMs Using PHP and BERT-Based Similarity Models in Maritime English

• 발행기관: 한국항해항만학회
• 저자명: 설진기 박영수 신동수
• 간행물 정보: 『한국항해항만학회 학술대회논문집』2025 춘계학술대회논문집, 213~214쪽, 전체 2쪽
• 주제분류: 공학 > 해양공학
• 파일형태: PDF
• 발행일자: 2025.05.08

국문 초록

Effective communication is vital for safe navigation and to promote clarity and consistency in maritime interactions, the International Maritime Organization (IMO) developed the Standard Marine Communication Phrases (SMCP) for use in ship-to-ship and ship-to-shore exchanges. Unlike general English, Maritime English features distinct grammatical structures, vocabulary, and formatting designed for precision and brevity. This study investigates how well commercial Large Language Models (LLMs) can handle Maritime English communication and phrases using PHP Text Similarity algorithms and BERT-based models. Three LLMs (ChatGPT, Google Gemini, and Meta LLaMA 3) were tested using 60 SMCP-based tasks, including sentence formation, terminology interpretation, and fill-in-the-blank questions. Their performance was then compared with that of maritime high school students results. This research is expected to highlight the potential use of general LLMs in maritime contexts.

키워드

Maritime English Standard Marine Communication Phrases BERT based Transformer