Format: include :l if identical to :correct?

[nschneid]

Lemmas are specified if distinct from the word form, but what if there is a correction and the lemma is not distinct from the correction? It seems the trees are not consistent on this point. align_tokens.py is adding lemmas from UD despite identical corrections.

Consider insertions and deletions. Is it odd to specify an empty-lemma for a deleted word?

Perhaps the most intuitive policy is that if there is a :correct field, that value takes precedence when deciding whether the lemma needs to be explicit (and any deletion, i.e. :correct "", is assumed to have no lemma).