Open bguil opened 6 years ago
This is indeed a bug. I fixed it by forcing deprels to be unique. It should be in the dev branch soon. (It is a bit of a brute force solution, but I am copying (modified) regular deps to the enhanced deps AND spread (controlled) subjects across conjuncts etc. Apparently, the rules sometimes produce enhanced deps that were already there in the regular deps. Instead of listing all of these as exceptions, filtering duplicates seems easier.)
There are several places where the same enhanced relation is declare twice. For instance:
All occurences (80) can be listed with:
egrep "([0-9]+):([a-z:]+)\|(\1):(\2)" *.conllu
NB: the same problem occurs also in the other Dutch corpus
UD_Dutch-LassySmall
(11 occurences)