Adding the advcl:relcl subtype

UniversalDependencies / UD_English-EWT

English data

Creative Commons Attribution Share Alike 4.0 International

201 stars 43 forks source link

Adding the advcl:relcl subtype #346

Closed nschneid closed 2 years ago

nschneid commented 2 years ago

It has been decided that relative clauses modifying entire clauses, not nominals, should no longer be acl:relcl: UniversalDependencies/docs#886

This issue is to track English-specific implementation

nschneid commented 2 years ago

@amir-zeldes How are you handling edeprels given the type mismatch of "which" (PRON) and its clausal antecedent? I am thinking nsubj/obj for "which" at the basic level and csubj/ccomp as the enhanced dependency for its antecedent.

nschneid commented 2 years ago

some more GUM hits (as well as false positives) in: http://universal.grew.fr/?custom=62e9b21aa045b http://universal.grew.fr/?custom=62e9b7f6c21e5 http://universal.grew.fr/?custom=62e9b9167881c

nschneid commented 2 years ago

Was surprised how few adverbial acl:relcls were in EWT—turns out most of the adverbial relatives were parataxis. Need to go through:

nschneid commented 2 years ago

[x] http://universal.grew.fr/?custom=62eaac2fdaf97: "which" marked as PronType=Int (the default for WH-pronouns) but not appearing in a typical relative clause context. Some are errors (ccomp should be advcl:relcl); some are relative clauses as full sentences, so should be PronType=Rel; others are truly interrogative. I will go through them and mark the truly interrogative ones with a MISC feature ManuallyChecked=PronType. (all addressed)

amir-zeldes commented 2 years ago

some more GUM hits (as well as false positives) in: http://universal.grew.fr/?custom=62e9b21aa045b http://universal.grew.fr/?custom=62e9b7f6c21e5 http://universal.grew.fr/?custom=62e9b9167881c

Thanks for reporting, there were two instances I'd missed in there!

amir-zeldes commented 2 years ago

@amir-zeldes How are you handling edeprels given the type mismatch of "which" (PRON) and its clausal antecedent? I am thinking nsubj/obj for "which" at the basic level and csubj/ccomp as the enhanced dependency for its antecedent.

Yeah, that looks good, just implemented it. I also saw cases of obl ->advcl and even csubj:pass once...

(2nd part moved to other thread)