The code is based on https://github.com/ZNLP/NCLS-Corpora.

word-level-supervised: supervised pre-training with word-level knowledge distillation.

sent-level-supervised: supervised pre-training with sentence-level knowledge distillation.

rl-sim: reinforcement learning fine-tuning with cross-lingual similarity and ROUGE scores as rewards.

Training:

python train.py -config train.json

Translating:

python translate.py -config translate.json