Paul's scripts to get baselines: https://github.com/pmichel31415/jsalt-baseline. Feel free to modify in any way.
You can find trained model on the AWS cluster under /export/fs01/pmichel1/models
, and preprocessed data for en-{fr,ja,zh}
under /export/fs01/pmichel1/corpora
The following are various scripts you might use:
create-alignments
: A pipeline to create alignments for the WMT2019 robustness task data and outputstagging-scripts
: Scripts that can be used for tagging for further downstream analysiscompare-mt
: Scripts to run compare-mt to generate reports