Research Summary: Form without function: a reception-theoretic evaluation of LLM translations of classical Chinese poetry and its ethical implications

A study on Large Language Models (LLMs) used for translating classical Chinese poetry reveals that while LLMs can preserve the formal structures of poetic gaps, they frequently fail to retain the functional aspect of these gaps, which is to invite reader imagination. This indicates a limitation in LLMs' ability to translate nuanced artistic conception, despite their capacity for formal preservation.

Why it matters

This research highlights a fundamental limitation in current LLM capabilities, specifically their struggle with nuanced semantic and artistic interpretation despite formal accuracy. It signals a critical area for development for any organization leveraging AI for complex content generation, cultural preservation, or creative tasks where interpretative depth is paramount.

Key insights

• Classical Chinese poetry uses 'formal gaps' to create artistic conception and invite reader imagination, which are challenging to preserve in translation.
• Large Language Models (LLMs) can preserve the formal structures of these gaps in their translations.
• However, LLM translations often lose the essential function of these gaps, which is to invite imagination, leading to a 'form without function' outcome.
• The study evaluated three LLMs (GPT, GLM-5.1, DeepSeek) by comparing their translations against human benchmarks for two Tang poems.
• A reception-theoretic framework, using concepts of 'gaps' and 'indeterminacy', was employed with a four-point scale to assess five categories of gaps.

Source

Frontiers in Education — https://www.frontiersin.org/articles/10.3389/feduc.2026.1945132