Cross-Linguistic Transfer Learning for Low-Resource Machine Translation Across Language Families
An empirical demonstration that transfer learning improves low-resource machine translation even across typologically diverse language families, by fine-tuning a model pre-trained on a related high-resource language pair using limited target-language data. Students learn how linguistic typology and a shared source or target side shape transfer effectiveness, and how hyperparameters such as batch size and learning rate should be adapted to how similar the language pairs are.
2501.00045
An empirical study of transfer learning for low-resource machine translation, fine-tuning transformer models pre-trained on a typologically similar high-resource language pair with limited target-lan…