Unknown labels in decoding

[olzama]

With large tag sets (e.g. lexical types, of which there can be thousands), unknown labels are possible at decoding time. In production, it is not possible to include the test set into the alphabet during training.

Is there any reason to explicitly disallow unknown labels then?

NCRFpp/utils/alphabet.py

Line 29 in 105a53a

if not self.label: