amir-zeldes / gum

Repository for the Georgetown University Multilayer Corpus (GUM)
https://gucorpling.org/gum/
Other
89 stars 50 forks source link

passive VERBs with copulas: should actually be ADJ? #182

Closed nschneid closed 7 months ago

nschneid commented 7 months ago

https://universal.grew.fr/?custom=653dccf9ed623

Looks like these are mostly in new genres.

Here's the neaten.py code for validating passives: https://github.com/UniversalDependencies/UD_English-EWT/blob/7d3e96c3331620edcf78e332e892686fc9530afa/not-to-release/tools/neaten.py#L336-L401

amir-zeldes commented 7 months ago

huh.. Thanks for reporting - I thought we had code to catch these, but I notice the script makes them all into nsubj:outer, so I'm guessing the script checks only for plain nsubj or something. I'll investigate when I get a chance!

amir-zeldes commented 7 months ago

OK, true issues are fixed, and as discussed offline, compound cases like "self inflicted" or "tailor made" are allowed VBN + cop. I've also added ExtPos to fixed as discussed, we could consider whether these should get ExtPos as well. The GUM validator is updated and allows these exceptions.