In-Domain African Languages Translation Using LLMs and Multi-armed Bandits
Abstract
Neural Machine Translation (NMT) systems face significant challenges when working with low-resource languages, particularly in domain adaptation tasks. These difficulties arise due to limited training data and suboptimal model generalization, As a result, selecting an opti- mal model for translation is crucial for achiev- ing strong performance on in-domain data, par- ticularly in scenarios where fine-tuning is not feasible or practical. In this paper, we investi- gate strategies for selecting the most suitable NMT model for a given domain using bandit- based algorithms, including Upper Confidence Bound, Linear UCB, Neural Linear Bandit, and Thompson Sampling. Our method effectively addresses the resource constraints by facilitat- ing optimal model selection with high confi- dence. We evaluate the approach across three African languages and domains, demonstrating its robustness and effectiveness in both scenar- ios where target data is available and where it is absent.