UniversalDependencies / UD_English-EWT

English data
Creative Commons Attribution Share Alike 4.0 International
199 stars 42 forks source link

Edeprels from reduced relative clauses #392

Open nschneid opened 1 year ago

nschneid commented 1 year ago

RRCs lack a relativizer to receive the relative clause dependency in Basic Dependencies, so we have to use heuristics to infer it.

http://universal.grew.fr/?custom=643c5c7377d9f

Need to check for whether the RC is passive, whether it has a stranded preposition, whether it has a subject, etc.

If the RC has a subordinate clause on its own, manual checking may be necessary.

nschneid commented 9 months ago

^ this commit supports the script in #474 by avoiding ambiguity between multiple edeps, one of which is for the RC

amir-zeldes commented 9 months ago

This sounds interesting - would you be able to make a depedit version of this addition for the GU corpora enhancer?

nschneid commented 9 months ago

You mean for amir-zeldes/gum#114? I don't know the extent to which adding the missing edeps can be automated. @xiulinyang is looking at this.

amir-zeldes commented 9 months ago

Oh, this is manual? nvm I somehow understood that there was a script from the above.

nschneid commented 9 months ago

The script uses the edeprels that are already there to add construction information. But a lot are still missing, and with relative clauses there are many corner cases.

nschneid commented 9 months ago

See also #388