Simple and Effective Input Reformulations for Translation
This addresses data efficiency in finetuning for translation tasks, enabling more effective training to improve state-of-the-art performance.
The paper tackles the problem of improving translation performance by reformulating inputs during finetuning of foundation language models, achieving up to 3.5 chrF++ improvement on the Flores200 benchmark.
Foundation language models learn from their finetuning input context in different ways. In this paper, we reformulate inputs during finetuning for challenging translation tasks, leveraging model strengths from pretraining in novel ways to improve downstream performance. These reformulations are simple data level modifications, require no additional collection of training data or modification of data at inference time. They can be applied either on single language pair translation tasks or massively multilingual translation tasks. Experiments with these techniques demonstrate significant performance improvements up to $\textbf{3.5 chrF++ on the Flores200 translation benchmark}$. We hope our research accessibly improves finetuning data efficiency, enabling more effective training to scalably improve state-of-the-art performance. Our code is released $\href{https://github.com/bri25yu/LanguageModelExperimentation}{here}.$