Search papers, labs, and topics across Lattice.
The paper introduces WALAR, a reinforcement learning method that uses only monolingual text to improve LLM translation capabilities for low-resource languages while maintaining performance on high-resource languages. It addresses the issue of reward hacking in multilingual quality estimation (QE) models used for RL training, which can lead to poorer LLMs. WALAR mitigates these issues by incorporating word alignment and language alignment techniques into the reward function, achieving significant performance gains over LLaMAx on the Flores-101 dataset across 1400 language directions.
Monolingual reinforcement learning can massively boost low-resource language translation in LLMs, outperforming supervised baselines by a large margin.
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.
CMU ML