# Generating Adversarial Texts for Machine Translation via GRPO

arXiv: https://arxiv.org/abs/2609.06048 · HTML: https://mlsanity.com/paper/2609.06048/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Reinforcement learning for reasoning, one of 17 directions in LLM reasoning.
- Builds on 7 of its 27 references that are in the map.
- First posted on arXiv 5 Sep 2026.

## Placed in

- LLM reasoning / [Reinforcement learning for reasoning](https://mlsanity.com/lr/reinforcement-learning-for-reasoning/)

## Builds on (most cited first, 7 of 7)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [No Language Left Behind: Scaling Human-Centered Machine Translation](https://mlsanity.com/paper/2207.04672/) (2022)
- [Zero-shot Benchmarking: A Framework for Flexible and Scalable Automatic Evaluation of Language Models](https://mlsanity.com/paper/2504.01001/) (2025)
- [Pearmut: Human Evaluation of Translation Made Trivial](https://mlsanity.com/paper/2601.02933/) (2026)
- [Searching the Internet for Challenging Benchmarks at Scale](https://mlsanity.com/paper/2509.26619/) (2025)
- [Last Translation Benchmark](https://mlsanity.com/paper/2609.04173/) (2026)

## Abstract (arXiv)

As machine translation (MT) systems continue to improve, standard benchmarks become less informative for exposing remaining weaknesses. Traditional methods for creating challenging test sets rely on expensive manual creation or curation, while automated approaches struggle to produce sets with the necessary translation difficulty and linguistic diversity. We propose a scalable reinforcement-learning-based approach for rewriting existing source texts into instances that are more difficult to translate for MT systems. We fine-tune a large language model with Group Relative Policy Optimization (GRPO), using reward signals based on translation difficulty together with constraints for semantic similarity, grammaticality, and approximate length preservation. On WMT25, our approach substantially reduces average COMET translation quality from 0.63 to 0.48, while preserving grammaticality and readability, whereas the base model remains at 0.64. Evaluations on the unseen WMT19-WMT24 benchmarks confirm that this behavior generalizes beyond the training data, and human evaluation further shows that the rewrites substantially lower translation quality while incurring a moderate drop in naturalness and only a small change in grammaticality. We release our code to support reproducibility.