Adaptive Neural Architectures for Low-Resource Language Translation
A. Kumar, S. Mendes, L. Zhao
We propose a lightweight transformer variant that achieves competitive BLEU scores on low-resource language pairs while reducing training compute by 42%.