We discussed on our calls: * extract tsv files ready for BERT-like training * extract label files (one label per line) for the evaluation script * combine train and test so that it is relatively clear that they are not intended to train
We discussed on our calls: