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

2026-06-04 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, quick

Summary

A new multi-aspect iterative refinement framework addresses challenges in literary translation. It tackles the scarcity of high-quality annotated data and the need to balance expression fluency with literary effect. This framework generates high-quality translation references and preference data. It uses specialized LLM translators, each targeting a distinct quality dimension. The generated data supports supervised fine-tuning (SFT) and reinforcement learning (RL). Experiments show generated references outperform original ground truth for SFT by 8.65 CEA100 points. For RL, DPO degraded performance, but GRPO yielded an additional 1.51 point improvement. This is attributed to GRPO's stability and online exploration. The resulting LitMT-8B and LitMT-14B models achieved 67.25 and 69.07 CEA100 respectively on MetaphorTrans English-to-Chinese. These scores are competitive with Claude Sonnet 4.5 at 68.43 CEA100 and generalize well to out-of-domain literary work.

Key takeaway

For NLP Engineers developing literary translation systems, this multi-aspect data generation and LLM training approach offers a robust method. It overcomes data scarcity and enhances model performance. You should consider implementing specialized LLM translators for iterative data refinement. Also, leverage GRPO for reinforcement learning. GRPO demonstrated superior stability and exploration compared to DPO, achieving competitive results against models like Claude Sonnet 4.5.

Key insights

Multi-aspect data generation and iterative refinement significantly enhance LLM literary translation.

Principles

• Specialized LLM translators can target distinct quality dimensions.
• Generated references can surpass original ground truth for SFT.
• GRPO offers stability and online exploration for RL in this context.

Method

A multi-aspect iterative refinement framework generates high-quality translation references and preference data via specialized LLM translators, then uses this data for supervised fine-tuning and GRPO-based reinforcement learning.

In practice

• Generate superior SFT data using specialized LLM translators.
• Employ GRPO for reinforcement learning in literary translation tasks.

Topics

• Literary Translation
• Large Language Models
• Data Generation
• Reinforcement Learning
• Supervised Fine-tuning
• GRPO
• MetaphorTrans Benchmark

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• Better Literary Translation: A Multi-Aspect Data Generation and LLM Training Approach — Multi-aspect iterative refinement generates superior literary translation data for LLM training.
• AFRILANGTUTOR: Advancing Language Tutoring and Culture Education in Low-Resource Languages with Large Language Models — Dictionary-based synthetic data generation and combined SFT+DPO significantly enhance LLM tutoring for low-resource African languages.
• AI-assisted cultural heritage dissemination: Comparing NMT and glossary-augmented LLM translation in rock art documents — Glossary-augmented LLMs significantly improve terminology accuracy in specialized translation without sacrificing overall quality.
• Translating Chinese with Data Science: StanfordNLP and Beyond — Integrating existing tools and broad IT understanding significantly accelerates complex translation tasks.
• Detector-Evasive LLM Paraphrasing via Constrained Policy Optimization — DEPO enables LLM paraphrasing to evade AI-text detectors while explicitly constraining semantic preservation using a Constrained Markov Decision Process.
• Improving Cross-Lingual Factual Recall via Consistency-Driven Reinforcement Learning — GRPO significantly improves cross-lingual factual recall and consistency in LLMs by fostering shared representations, outperforming SFT.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.