Chainer-based implementation of Attention-based seq2seq model.

See "Neural Machine Translation by Jointly Learning to Align and Translate", Dzmitry Bahdanau, Kyunghyun Cho, Yoshua Bengio, arxiv 2014.

This repository is partly derived from this repository and Chainer's official seq2seq example.

• Ubuntu 16.04
• Python 3.5.2
• Chainer 3.1.0
• numpy 1.13.3
• cupy 2.1.0
• nltk
• progressbar
• and their dependencies

First you need to prepare parallel corpus. Download 10^9 French-English corpus from WMT15 website.

http://www.statmt.org/wmt15/translation-task.html

$ sh download_wmt.sh

Now you can get six files:

• Source sentence file: giga-fren.preprocess.en
• Source vocabulary file: vocab.en
• Target sentence file: giga-fren.preprocess.fr
• Source vocabulary file: vocab.fr
• Source sentence file (validation): newstest2013.preprocess.en
• Target sentence file (validation): newstest2013.preprocess.fr

Then, let's start training.

$ python train.py giga-fren.preprocess.en giga-fren.preprocess.fr vocab.en vocab.fr --validation-source newstest2013.preprocess.en --validation-target newstest2013.preprocess.fr

See command line help for other options.