Do LLMs Understand Wine Descriptors Across Cultures? A Benchmark for Cultural Adaptions of Wine Reviews

Chenye Zou, Xingyue Wen, Qian Janice Wang, Daniel Hershcovich

Keywords: Machine Translation, Cross-Cultural Adaptation, Cultural Evaluation Metrics, Wine Reviews

Abstract: Leveraging remarkable advancements in Large Language Models (LLMs), we are now poised to tackle increasingly complex challenges requiring deep comprehension of multifaceted domains and contexts. A specific application scenario is wine reviews adaptation. Wine reviews usually describe a wine's appearance, aroma, and flavor to help consumers appreciate its characteristics. However, the adaptation of wine reviews transcends mere translation; it requires consideration of regional preferences, flavor descriptors, and cultural nuances that shape wine perception. We introduce the first-ever task involving the translation and cultural adaptation of wine reviews between Chinese and English. In a case study on cross-cultural wine review adaptation, we compile the \textbf{first} dataset of 8k Chinese and 16k Western professional wine reviews. We evaluated various methods, including LLMs and traditional machine translation techniques, using both automatic and human metrics. For human assessments, we introduce three novel culture-related metrics—Cultural Proximity, Cultural Neutrality, and Cultural Genuineness—to gauge the success of different approaches in achieving authentic cross-cultural adaptation. Our analysis shows that current models struggle to capture cultural nuances, especially in translating wine descriptions across different cultures. This highlights the challenges and limitations of translation models in handling cultural content.

Archival Status: Non‑archival

Paper Length: Long Paper (up to 8 pages of content)

Submission Number: 333