Due to licensing restrictions, we cannot share the CONLL 2003 dataset.

It can be built from the Reuters corpus following the directions here:

https://www.clips.uantwerpen.be/conll2003/ner/

In addition, several github repositories also provide the CONLL 2003 dataset.


Place files eng.train, eng.testa (dev data), eng.testb (test data) here,
and then delete this file.


