UD alignment: interaction of :correct and :subt for multiword token

[nschneid]

We have an instance of cant which is a misspelling of can't. In EWT, there are two words, the second of which is nt with CorrectForm=n't.

In CGEL, should this be

(V_aux :t "cant" :correct "can't" :subt "ca" :subt "n't")

or

(V_aux :t "cant" :correct "can't" :subt "ca" :subt "nt")

I.e. are subtoken corrections incorporated into the :subt? Or is :subt strictly for unnormalized surface strings? Note that the MWT string in UD is cant.

[nschneid]

align_tokens.py currently interprets :subt as part of the surface string, so it doesn't reflect any corrections.