Conllu data - handling multiple edge labels

[paolo-gajo]

Hello,

I was going through NERdataset while running train.py to see how the data is loaded into the model and I noticed that it seems there are no multiple head_indices for words that should have two, e.g. the following from line 916 of train.conllu.

7	yeast	_	NN1	B-F	_	1	t	[(18,	'f-eq')]

That is, the datapoint returned by process_datapoint only contains one value per word in item['head_indices'].

Maybe I am misunderstanding how this part of the code works, but I am missing how/where the multiple edge labels are handled.

Thanks in advance for the help!