Criteria for aux:pass: 'get'

[nschneid]

Most of the English treeabanks use aux:pass for passive auxiliaries. Canonically this is a form of "be", but "get" can also act as a passive aux:

• He got hit by a truck ≈ He was hit by a truck

However, "get" as a light verb is thorny and can also have causative and inchoative usages where no agent-by-phrase is possible:

• I got ready/old.
• I got everybody excited/on track.

In EWT (at least) there is perhaps an overuse of aux:pass for non-passive uses of "get". Some that seem dubious to me:

• get + ADJ: lucky, fussy, stuck
• get + participle:
  • get involved
  • get married (the agent of 'marry' can be an officiator, and "We got married by a/the priest" ~sounds odd in contemporary English~ is attested, but to me sounds a bit odd)
  • get fucked up ('drunk')
  • get rid of
  • get caught in traffic

etc.

There are others that could potentially have a by-phrase, but seem more inchoative than passive:

• get certified in Delaware ('become certified' > 'be certified by (authority)')
• get scared

[amir-zeldes]

Again offering GUM for comparison - we treat the ADJ class as xcomp, by analogy to "become", so "get mad" = xcomp(get,mad):

https://corpling.uis.georgetown.edu/annis/#_q=ImdldCIgLiAibWFkIg&_c=R1VN&cl=5&cr=5&s=0&l=10

get + VVN is interpreted as a passive auxiliary, though we probably also have some errors:

https://corpling.uis.georgetown.edu/annis/#_q=ImdldCIgXz1fIHRva19mdW5jPSJhdXhwYXNzIg&_c=R1VN&cl=5&cr=5&s=0&l=10

Examples like "involved" are probably considered JJ in most contexts based on the PTB tagging guidelines (Santorini 1990), so they should also be xcomp IMO. They usually pass the JJ tests:

• Get very involved (cf. *"get very caught")
• Get more involved (*"get more caught")
• ?Get uninvolved (*"get uncaught")

Not totally sure about the last one, but overall it looks like JJ to me.

[nschneid]

206 concerns which of these "gets" should be tagged AUX

[manning]

I agree, these should be fixed and changed to xcomp, as you describe. Maybe they will be for the next release, since we're getting a bit more updating done!

[dan-zeman]

Has this bug been fixed?

[nschneid]

Not entirely, as it's painful to go through all the tokens of "get" to decide which are actually passive auxes. But I will look into it for EWT.

[amir-zeldes]

Get-passive should be properly distinguished in GUM, so one could also try using a GUM trained parser to find likely errors.

[nschneid]

"I just got done looking at the underlying contract language": I'd say "get done" is an idiom meaning 'finish', but to analyze it compositionally would we say "looking at..." is an optional depictive advcl???

In general, looking at https://universal.grew.fr/?custom=655e761777b12, I think some are clear get-passives and others are borderline. @amir-zeldes maybe we should sit down and go through them sometime.

[amir-zeldes]

Sure, happy to talk about it. But as for this "done", I don't think it's a passive participle, since you can say "I'm done" using this kind of done, but you can't say "I'm done (*by them)". In GUM this reading of 'done' is tagged JJ and lemmatized 'done':

https://gucorpling.org/annis/#_q=bGVtbWE9ImRvbmUi&_c=R1VN&cl=5&cr=5&s=0&l=10

[nschneid]

Happy to change "done" to ADJ, but is advcl correct for the "looking" clause? Should it be a second xcomp?