Modern glosses improve machine translation of ancient Chinese records

Researchers introduced YRC-Bench, a human-curated benchmark designed to test how machine translation systems handle historical cultural heritage.

The tool addresses the difficulty of translating ancient texts, such as those from the Yellow River region, where classical syntax often hides specific cultural knowledge. The benchmark evaluates translations between Classical Chinese, Modern Chinese, and nine other foreign languages. The study found that using modern Chinese explanatory glosses as semantic bridges significantly improves the accuracy of translations.

When Large Language Models used these human-curated explanations, they produced results with higher lexical fidelity and better semantic adequacy. This method helps bridge the gap between ancient structures and modern understanding. Human experts also identified distinct preferences in how different language groups perceive translations. These preferences varied between Sinographic-tradition languages, which use Chinese characters, and non-Sinographic groups. The results suggest that human-curated bridges are essential for maintaining cultural nuance in automated systems.