NMT-Research-Reproductions-PyTorch

This is a repository for reproducing research results in neural machine translation. All details are implemented in PyTorch.

xnmt

Define general interfaces and modules for NMT. Heavily borrowed from OpenNMT-py.

text

Contains the preprocess script for texts.

RNNsearch

Reference paper:

Neural Machine Translation by Jointly Learning to Align and Translate

Comment: I can not get expected performance in a Chinese-English translation task.

Base-Model

Reference paper:

Effective Approaches to Attention-based Neural Machine Translation

Comment: Self-defined multi-layer encoder-decoder with attention and input feeding. The model can get good performance in a Chinese-English translation task. And it is adopted as baseline for following experiments.

BPE

Reference paper:

Neural Machine Translation of Rare Words with Subword Units

Comment: It just has the preprocess script since all others are just the same as normal procedure.