wei zhao

If you want more details, please refer to https://github.com/pytorch/fairseq.

The raw data are available online.

--copy-ext-dict is a parameter that you can use and this should not happen. Maybe you can provide more details about how you use generate.py.

You can download CoNLL-2014 dataset for the test.m2 file. The generate.sh file don't rely on the test.label.src.txt and the test.label.tgt.txt. The name is incorrect and please check the script that...

> I want to use the pretrained model checkpoint9.pt directly to process the new data, how do I organize the code.(The methods I find on baidu cannot be used, so...

> I have exactly the same question, my main issues is that I can not find the script that preprocesses the oriinal CONLL dataset. We didn't release the preprocess scripts...

Thanks @kevbp5. We did use 4 different random initializations for the models without DA. For the models with DA, we also used pre-trained checkpoints from different pre-training stages.

Try the "interactive.sh" script.