Bridging the Language Gap: Evaluating Machine Translation for Animal Health in Low-Resource Settings

Abstract

Machine translation (MT) has made significant progress in high-resource languages, but translating technical texts into low-resource languages remains an open challenge. This study investigates the ability of state-of-the-art multilingual models to translate animal health reports from English to Yoruba, a crucial task for enhancing veterinary communication in underserved regions. Although previous research has explored low-resource MT, domain-specific translation for animal health has been largely overlooked. Using a curated dataset of 1,468 parallel sentences, we evaluated several MT models in zero-shot and fine-tuned settings. Despite the promise of multilingual models, we find substantial limitations in their ability to generalize to this domain, raising concerns about their applicability in specialized, low-resource contexts. We analyze potential causes, including vocabulary mismatch, training data scarcity, and constraints of model architecture. Our findings highlight the need for more targeted approaches to low-resource domain-specific MT and emphasize the broader implications for AI deployment in real-world applications.