Two-Way Neural Machine Translation: A Proof of Concept for Bidirectional Translation Modeling using a Two-Dimensional Grid

Neural translation models have proven to be effective in capturing sufficient\ninformation from a source sentence and generating a high-quality target\nsentence. However, it is not easy to get the best effect for bidirectional\ntranslation, i.e., both source-to-target and target-to-source translation using\na single model. If we exclude some pioneering attempts, such as multilingual\nsystems, all other bidirectional translation approaches are required to train\ntwo individual models. This paper proposes to build a single end-to-end\nbidirectional translation model using a two-dimensional grid, where the\nleft-to-right decoding generates source-to-target, and the bottom-to-up\ndecoding creates target-to-source output. Instead of training two models\nindependently, our approach encourages a single network to jointly learn to\ntranslate in both directions. Experiments on the WMT 2018\nGerman$\\leftrightarrow$English and Turkish$\\leftrightarrow$English translation\ntasks show that the proposed model is capable of generating a good translation\nquality and has sufficient potential to direct the research.\n