Underscore in CoNLL 2003 converter

[Ziinc]

Hi label studio team,

Was wondering what this underscore in the converter is supposed to represent.

https://github.com/heartexlabs/label-studio-converter/blob/83ff059a1257cf600559c09a3459483d54073f81/label_studio_converter/converter.py#L243

I am asking because this has led to issues when integrating label studio into other NLP toolchains.

From my understanding, this is not usual when building a CoNLL 2003 NER dataset.

Thanks!