Triangular Architecture for Rare Language Translation
This addresses the challenge of translating rare languages, which is crucial for accessibility and communication in underserved linguistic communities, representing an incremental advancement in low-resource NMT methods.
The paper tackles the problem of poor neural machine translation performance for low-resource language pairs, especially when one language is rare, by proposing a triangular training architecture that leverages bilingual data involving a rich language to improve translation quality, achieving significant improvements on MultiUN and IWSLT2012 datasets.
Neural Machine Translation (NMT) performs poor on the low-resource language pair $(X,Z)$, especially when $Z$ is a rare language. By introducing another rich language $Y$, we propose a novel triangular training architecture (TA-NMT) to leverage bilingual data $(Y,Z)$ (may be small) and $(X,Y)$ (can be rich) to improve the translation performance of low-resource pairs. In this triangular architecture, $Z$ is taken as the intermediate latent variable, and translation models of $Z$ are jointly optimized with a unified bidirectional EM algorithm under the goal of maximizing the translation likelihood of $(X,Y)$. Empirical results demonstrate that our method significantly improves the translation quality of rare languages on MultiUN and IWSLT2012 datasets, and achieves even better performance combining back-translation methods.