ngohoanhkhoa / GAN-NMT

Generative Adversarial Networks in Neural Machine Translation
57 stars 11 forks source link

GAN-NMT

Generative Adversarial Networks in Neural Machine Translation

■ 03.2017 - 09.2017 ■ Neural Machine Translation ○ NLP

○ Internship: 6 months in Laboratoire d'Informatique pour la Mécanique et les Sciences de l'Ingénieur (LIMSI, France)

○ Supervisor: Alexandre ALLAUZEN {allauzen@limsi.fr}

○ Technical Keywords: Generative Adversarial Network, Reinforcement Learning, Convolutional Neural Network, Recurrent Neural Network, Python

○ Description: The objective of this project is improving neural machine translation with generative adversarial nets. It is based on the paper "Improving Neural Machine Translation with Conditional Sequence Generative Adversarial Nets" [Yang et al] (https://arxiv.org/abs/1703.04887) and the framework of NMTPY (https://github.com/lium-lst/nmtpy).

How to run

1: Setup NMTPY

2: Copy "GAN" and "models" folders into "nmtpy" folder

3: In "GAN_NMT_model/model_conf", check config files (file.conf)

4: In "GAN_NMT_model", run bash file (file.sh)

Note: Remember to check the path