Extreme Adaptation for Personalized Neural Machine Translation

This repository contains the code for the paper Extreme Adaptation for Personalized Neural Machine Translation.

Data

The data used in the paper is the SATED dataset, available at this url: https://pmichel31415.github.io/sated/.

Additional experiments were performed on the gender annotated europarl corpus from [1], available at this url https://www.kaggle.com/ellarabi/europarl-annotated-for-speaker-gender-and-age.

You can download all the data by running:

# SATED
wget http://www.cs.cmu.edu/~pmichel1/hosting/sated-release-0.9.0.tar.gz
tar xvzf sated-release-0.9.0.tar.gz
# Europarl
wget https://www.kaggle.com/ellarabi/europarl-annotated-for-speaker-gender-and-age/downloads/europarl-annotated-for-speaker-gender-and-age.zip
unzip europarl-annotated-for-speaker-gender-and-age.zip

Requirements

This project was coded in Dynet. It should be working with the 2.0.3 release which you can install by running:

pip install dynet==2.0.3

References

If you use this code or the SATED dataset in you research, consider citing the original paper:

TBD

Other references:

[1]: https://aclanthology.coli.uni-saarland.de/pdf/E/E17/E17-1101.pdf