fandongmeng / BMI-NMT

BMI-NMT

Source code for the ACL 2021 short paper Bilingual Mutual Information Based Adaptive Training for Neural Machine Translation.

Contents

• Introduction
• Usage
• Requirements
• Citation

Introduction

• Implementation

Implemented based on THUMT-TensorFlow, an open-source toolkit for neural machine translation developed by the Natural Language Processing Group at Tsinghua University which was implemented strictly referring to Vaswani et al. (2017).

• Data

WMT14 English-German

WMT19 Chinese-English

Usage

Note: The usage is on the top of THUMT, for more details, please refer to the user manual of THUMT.

• Calculating BMI

python mi_calculate.py

• Training

sh ende_mi.sh

Requirements

• Python version >=3.6
• Tensorflow version >=1.12

Citation

Please cite the following paper if you use the code:

@InProceedings{Xu2021bmi,
  author    = {Yangyifan Xu, Yijin Liu, Fandong Meng, Jiajun Zhang, Jinan Xu, Jie Zhou},
  title     = {Bilingual Mutual Information Based Adaptive Training for Neural Machine Translation},
  booktitle = {Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics},
  year      = {2021}
}