Semantic Label Drift in Cross-Cultural Translation

TL;DR

This paper investigates how cultural differences cause semantic label drift in machine translation, revealing that modern LLMs and cultural divergence significantly impact label preservation in sensitive domains.

Contribution

It uncovers the influence of cultural factors on label drift in MT, especially with LLMs, and emphasizes the importance of cultural alignment for accurate translation.

Findings

LLMs induce label drift in culturally sensitive domains

Cultural knowledge in LLMs can amplify label drift

Cultural similarity affects label preservation

Abstract

Machine Translation (MT) is widely employed to address resource scarcity in low-resource languages by generating synthetic data from high-resource counterparts. While sentiment preservation in translation has long been studied, a critical but underexplored factor is the role of cultural alignment between source and target languages. In this paper, we hypothesize that semantic labels are drifted or altered during MT due to cultural divergence. Through a series of experiments across culturally sensitive and neutral domains, we establish three key findings: (1) MT systems, including modern Large Language Models (LLMs), induce label drift during translation, particularly in culturally sensitive domains; (2) unlike earlier statistical MT tools, LLMs encode cultural knowledge, and leveraging this knowledge can amplify label drift; and (3) cultural similarity or dissimilarity between source and target languages is a crucial determinant of label preservation. Our findings highlight that neglecting cultural factors in MT not only undermines label fidelity but also risks misinterpretation and cultural conflict in downstream applications.