This repository is research for language specific machine translation.
The pivot model is transformer encoder-decoder model which is trained by denoising autoencoding. To train pivot model, run exe_pivot.sh file. We provide pretrained pivot model trained with 0.2M english data (download small_pivot). To test with trained pivot model, run predict_small_pivot.sh.