Better Literary Translation: A Multi-Aspect Data Generation and LLM Training Approach

2026-06-04

AI Summary

This study addresses the challenge of literary translation, where high-quality annotated data are scarce and it is difficult to simultaneously achieve fluency and literary expressiveness. The authors propose a multidimensional iterative optimization framework that leverages large language models specialized in distinct quality dimensions to generate high-quality translations and preference data. The framework employs a two-stage training strategy combining supervised fine-tuning (SFT) with GRPO, a reinforcement learning algorithm based on an explicit reward model. Experimental results demonstrate that GRPO outperforms DPO, and the proposed models LitMT-8B and LitMT-14B achieve CEA100 scores of 67.25 and 69.07, respectively, on the MetaphorTrans English-to-Chinese benchmark—surpassing Claude Sonnet 4.5—and exhibit strong out-of-domain generalization on O. Henry’s literary works.

Abstract

Literary translation poses unique challenges due to the scarcity of high-quality annotated data and the need to balance expression fluency with literary effect. We present a multi-aspect iterative refinement framework that generates high-quality translation references and preference data through specialized LLM translators, each targeting a distinct quality dimension. We leverage the generated data for supervised fine-tuning and reinforcement learning. Experiments show that our generated references outperform the original ground truth for SFT by 8.65 CEA100 points. For reinforcement learning, we find that DPO leads to performance degradation in this setting, while leveraging an explicit reward model for GRPO yields an additional 1.51 point improvement. We attribute this to the stability of two-stage training and GRPO's online exploration capability. Our resulting models, LitMT-8B and LitMT-14B, achieve 67.25 and 69.07 CEA100 respectively on the MetaphorTrans English-to-Chinese literary translation benchmark, competitive with Claude Sonnet 4.5 at 68.43, and demonstrate strong generalization to out-of-domain literary work (i.e., O. Henry).

Problem

• literary translation
• data scarcity
• fluency
• literary effect
• quality annotation

Innovation

• multi-aspect data generation
• LLM-based literary translation
• GRPO reinforcement learning
• supervised fine-tuning
• translation quality dimensions