# Mending the Holes: Mitigating Reward Hacking in Reinforcement Learning for Multilingual Translation

arXiv: https://arxiv.org/abs/2603.13045 · HTML: https://mlsanity.com/paper/2603.13045/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Placed in LLM preference optimization methods, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 40 of its 50 references that are in the map.
- First posted on arXiv 13 Mar 2026.

## Placed in

- LLM alignment / RLHF / preference optimisation / [LLM preference optimization methods](https://mlsanity.com/alignment/llm-preference-optimization-methods/)
- Deep reinforcement learning methods / [RL training of language and agent models](https://mlsanity.com/deeprl/rl-training-of-language-and-agent-models/)

## Builds on (most cited first, 30 of 40)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [LLaMA: Open and Efficient Foundation Language Models](https://mlsanity.com/paper/2302.13971/) (2023)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [mT5: A massively multilingual pre-trained text-to-text transformer](https://mlsanity.com/paper/2010.11934/) (2020)
- [No Language Left Behind: Scaling Human-Centered Machine Translation](https://mlsanity.com/paper/2207.04672/) (2022)
- [M3-Embedding: Multi-Linguality, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation](https://mlsanity.com/paper/2402.03216/) (2024)
- [COMET: A Neural Framework for MT Evaluation](https://mlsanity.com/paper/2009.09025/) (2020)
- [The FLORES-101 Evaluation Benchmark for Low-Resource and Multilingual Machine Translation](https://mlsanity.com/paper/2106.03193/) (2021)
- [xCOMET: Transparent Machine Translation Evaluation through Fine-grained Error Detection](https://mlsanity.com/paper/2310.10482/) (2023)
- [CometKiwi: IST-Unbabel 2022 Submission for the Quality Estimation Shared Task](https://mlsanity.com/paper/2209.06243/) (2022)
- [Word Alignment by Fine-tuning Embeddings on Parallel Corpora](https://mlsanity.com/paper/2101.08231/) (2021)
- [Multilingual Machine Translation with Large Language Models: Empirical Results and Analysis](https://mlsanity.com/paper/2304.04675/) (2023)
- [Aya Expanse: Combining Research Breakthroughs for a New Multilingual Frontier](https://mlsanity.com/paper/2412.04261/) (2024)
- [The Interplay of Variant, Size, and Task Type in Arabic Pre-trained Language Models](https://mlsanity.com/paper/2103.06678/) (2021)
- [MetricX-24: The Google Submission to the WMT 2024 Metrics Shared Task](https://mlsanity.com/paper/2410.03983/) (2024)
- [Eliciting the Translation Ability of Large Language Models via Multilingual Finetuning with Translation Instructions](https://mlsanity.com/paper/2305.15083/) (2023)
- [Cross-lingual Retrieval for Iterative Self-Supervised Training](https://mlsanity.com/paper/2006.09526/) (2020)
- [LLaMAX: Scaling Linguistic Horizons of LLM by Enhancing Translation Capabilities Beyond 100 Languages](https://mlsanity.com/paper/2407.05975/) (2024)
- [MT-R1-Zero: Advancing LLM-based Machine Translation via R1-Zero-like Reinforcement Learning](https://mlsanity.com/paper/2504.10160/) (2025)
- [RLHF Can Speak Many Languages: Unlocking Multilingual Preference Optimization for LLMs](https://mlsanity.com/paper/2407.02552/) (2024)
- [Tower+: Bridging Generality and Translation Specialization in Multilingual LLMs](https://mlsanity.com/paper/2506.17080/) (2025)
- [GlotLID: Language Identification for Low-Resource Languages](https://mlsanity.com/paper/2310.16248/) (2023)
- [X-ALMA: Plug & Play Modules and Adaptive Rejection for Quality Translation at Scale](https://mlsanity.com/paper/2410.03115/) (2024)
- [Improving Machine Translation with Human Feedback: An Exploration of Quality Estimation as a Reward Model](https://mlsanity.com/paper/2401.12873/) (2024)
- [Reinforcement Learning based Curriculum Optimization for Neural Machine Translation](https://mlsanity.com/paper/1903.00041/) (2019)
- [Naamapadam: A Large-Scale Named Entity Annotated Data for Indic Languages](https://mlsanity.com/paper/2212.10168/) (2022)
- [BLEURT Has Universal Translations: An Analysis of Automatic Metrics by Minimum Risk Training](https://mlsanity.com/paper/2307.03131/) (2023)

## Built on by (most cited first, 4 of 4)

- [Reference-Free Post-Training of Open Large Language Models for Multilingual Machine Translation](https://mlsanity.com/paper/2608.10812/) (2026)
- [In the Blind: Building Pseudo-References for MT Evaluation](https://mlsanity.com/paper/2609.13611/) (2026)
- [Drift-Constrained Optimization: Only Direction Matters in Fine-Tuning Instruct Models](https://mlsanity.com/paper/2609.13680/) (2026)
- [MixLID: Token-Level Language Identification in Mixed-Language Text](https://mlsanity.com/paper/2610.16601/) (2026)

## Abstract (arXiv)

Large Language Models (LLMs) have demonstrated remarkable capability in machine translation on high-resource language pairs, yet their performance on low-resource translation still lags behind. Existing post-training methods rely heavily on high-quality parallel data, which are often scarce or unavailable for low-resource languages. In this paper, we introduce WALAR, a reinforcement training method using only monolingual text to elevate LLMs' translation capabilities on massive low-resource languages while retaining their performance on high-resource languages. Our key insight is based on the observation of failure modes (or "holes") in existing source-based multilingual quality estimation (QE) models. Reinforcement learning (RL) using these QE models tends to amplify such holes, resulting in poorer multilingual LLMs. We develop techniques including word alignment and language alignment to mitigate such holes in WALAR's reward for RL training. We continually trained an LLM supporting translation of 101 languages using WALAR. The experiments show that our new model outperforms LLaMAX, one of the strongest open-source multilingual LLMs by a large margin on 1400 language directions on Flores-101 dataset.