Addressing Sociolinguistic Challenges in Machine Translation: An LLM-Based Approach for Politeness and Formality

Machine translation (MT) has advanced significantly, yet nuances such as politeness, formality, and humbleness are often lost due to language structure and cultural differences. This challenge is notable in translations between English and Japanese, where sociolinguistic norms are central to communication. This paper presents a novel approach integrating Large Language Models (LLMs), specifically GPT-4, into the translation pipeline to preserve these communication nuances. We detail our engineered GPT prompts, which control politeness and formality levels, address-ing limitations of traditional Neural Machine Translation (NMT) systems. Initial evaluations indicate GPT-4 outperforms NMT in maintaining politeness and formality, though challenges remain with humbleness, especially in Japanese-to-English translations.