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

• URL: http://arxiv.org/abs/2603.13045v1
• Date: Fri, 13 Mar 2026 14:52:51 GMT
• Title: Mending the Holes: Mitigating Reward Hacking in Reinforcement Learning for Multilingual Translation
• Abstract summary: Existing post-training methods rely heavily on high-quality parallel data.<n>We introduce WALAR, a reinforcement training method using only monolingual text.<n>We develop techniques including word alignment and language alignment to mitigate such holes in WALAR's reward for RL training.
• Score: 9.906839381314082
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: 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.

Related papers

• Is Small Language Model the Silver Bullet to Low-Resource Languages Machine Translation? [18.975160361440597]
  Low-resource languages (LRLs) lack sufficient linguistic resources and are underrepresented in benchmark datasets.<n>This study systematically evaluates state-of-the-art smaller Large Language Models in 200 languages.
  arXiv  Detail & Related papers  (2025-03-31T13:56:03Z)
• Mufu: Multilingual Fused Learning for Low-Resource Translation with LLM [26.768074909642678]
  We introduce Mufu, which includes a selection of automatically generated multilingual candidates and an instruction to correct inaccurate translations in the prompt.<n>Mufu prompts turn a translation task into a postediting one.<n>Our experiments on En-XX translations over the Flores-200 dataset show LLMs finetuned against Mufu-style prompts are robust to poor quality auxiliary translation candidates.
  arXiv  Detail & Related papers  (2024-09-20T23:48:47Z)
• Quality or Quantity? On Data Scale and Diversity in Adapting Large Language Models for Low-Resource Translation [62.202893186343935]
  We explore what it would take to adapt Large Language Models for low-resource languages. We show that parallel data is critical during both pre-training andSupervised Fine-Tuning (SFT) Our experiments with three LLMs across two low-resourced language groups reveal consistent trends, underscoring the generalizability of our findings.
  arXiv  Detail & Related papers  (2024-08-23T00:59:38Z)
• Getting More from Less: Large Language Models are Good Spontaneous Multilingual Learners [67.85635044939836]
  Large Language Models (LLMs) have shown impressive language capabilities. In this work, we investigate the spontaneous multilingual alignment improvement of LLMs. We find that LLMs instruction-tuned on the question translation data (i.e. without annotated answers) are able to encourage the alignment between English and a wide range of languages.
  arXiv  Detail & Related papers  (2024-05-22T16:46:19Z)
• Enhancing Multilingual Capabilities of Large Language Models through Self-Distillation from Resource-Rich Languages [60.162717568496355]
  Large language models (LLMs) have been pre-trained on multilingual corpora. Their performance still lags behind in most languages compared to a few resource-rich languages.
  arXiv  Detail & Related papers  (2024-02-19T15:07:32Z)
• Okapi: Instruction-tuned Large Language Models in Multiple Languages with Reinforcement Learning from Human Feedback [61.83548032416181]
  We present Okapi, the first system with instruction-tuned LLMs based on RLHF for multiple languages. Okapi introduces instruction and response-ranked data in 26 diverse languages to facilitate the experiments and development of future multilingual LLM research.
  arXiv  Detail & Related papers  (2023-07-29T18:01:46Z)
• Democratizing LLMs for Low-Resource Languages by Leveraging their English Dominant Abilities with Linguistically-Diverse Prompts [75.33019401706188]
  Large language models (LLMs) are known to effectively perform tasks by simply observing few exemplars. We propose to assemble synthetic exemplars from a diverse set of high-resource languages to prompt the LLMs to translate from any language into English. Our unsupervised prompting method performs on par with supervised few-shot learning in LLMs of different sizes for translations between English and 13 Indic and 21 African low-resource languages.
  arXiv  Detail & Related papers  (2023-06-20T08:27:47Z)
• Chain-of-Dictionary Prompting Elicits Translation in Large Language Models [100.47154959254937]
  Large language models (LLMs) have shown surprisingly good performance in multilingual neural machine translation (MNMT) We present a novel method, CoD, which augments LLMs with prior knowledge with the chains of multilingual dictionaries for a subset of input words to elicit translation abilities.
  arXiv  Detail & Related papers  (2023-05-11T05:19:47Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.