Bi-Directional Differentiable Input Reconstruction for Low-Resource Neural Machine Translation
This is an incremental improvement for low-resource neural machine translation, addressing data scarcity issues in specific language pairs.
The paper tackles the problem of limited parallel text in low-resource neural machine translation by introducing a differentiable reconstruction loss that compares original inputs to reconstructed inputs from back-translation, achieving small but consistent BLEU improvements on four language pairs in both directions.
We aim to better exploit the limited amounts of parallel text available in low-resource settings by introducing a differentiable reconstruction loss for neural machine translation (NMT). This loss compares original inputs to reconstructed inputs, obtained by back-translating translation hypotheses into the input language. We leverage differentiable sampling and bi-directional NMT to train models end-to-end, without introducing additional parameters. This approach achieves small but consistent BLEU improvements on four language pairs in both translation directions, and outperforms an alternative differentiable reconstruction strategy based on hidden states.