PronType=Rel vs. Int

[nschneid]

It appears there are instances of PronType=Int that should be Rel, assuming that PronType=Rel is correct for any relativizer in the presence of acl:relcl.

For example, here are WDT determiners within a relative clause: http://match.grew.fr/?corpus=UD_English-EWT@dev&custom=61bffccb7016f

See also #88

[nschneid]

Ack maybe these are not actually relativizers. They are just interrogative NPs extracted from the RC.

"offset whatever gains could be made"—"that" can be inserted I think: "offset whatever gains that could be made"

[amir-zeldes]

I think this one is a WH-determiner - whatever gains is like "what gains", or "those gains" in the non-interrogative context.

[nschneid]

Right. Even though the whole sentence isn't interrogative, I think PronType=Int is correct for non-relativizer WDTs:

[amir-zeldes]

OK, I think that's what we have in GUM ATM too

[nschneid]

OK, these queries should work better—they reveal a bunch of PronType=Int or Dem that look like they should be Rel:

• [x] q0 (E:ref)
• [x] q1 (1 hop)
• [x] q2 (2 hops)
• [x] q3 (3 hops)
• [x] q4 (4 hops) (all of these are interrogative clauses within relative clauses)
• [x] q5 (acl:relcl free relatives) (a tiny bit of overlap with above queries)
• [ ] q6 (advcl:relcl free relatives)

...and likewise for GUM.

Note that q1-q4 are needed due to errors in the E:ref annotations. Corrections need to be manually checked.

[nschneid]

Most of the missing E:ref edges that I'm having to add manually are due to cases where (1) multiple RCs share the same true head (coordinated RCs, successive RCs), or (2) the extraction is from a clause nested within the RC.

[nschneid]

• [x] from q2: I'm undecided about the structure of "it will give you some idea of what it might cost and what is fair"

[amir-zeldes]

I think it's a coordinate free relative to an nmod modifying "idea" ("idea of that which it might cost/of that which is fair" - both WH phrases stand for NPs):

nmod(idea,what1) acl:relcl(what1,cost) conj(what1,what2) acl:relcl(what2,fair)

[nschneid]

I discussed this with Brett Reynolds on Twitter and it convinced me it was interrogative. https://twitter.com/YanisLing/status/1485921746775683074

The difference between free relatives and interrogative content clauses (a.k.a. indirect questions) is extremely subtle and hard to tell just from the meaning/paraphrasing. But basically the idea is that if it's an environment that licenses "whether" clauses, it's probably interrogative.

[nschneid]

• [x] as to what the light provides it gives UV rays..which is important for health in lizards as it is in humans.

Do we think the "which"-RC refers to the NP headed by "rays" (despite number agreement mismatch), or just "UV"? The current annotation is inconsistent between these two interpretations.

[amir-zeldes]

But basically the idea is that if it's an environment that licenses "whether" clauses, it's probably interrogative.

No, sorry I don't get it: the sentence was "give you some idea of what it might cost", there is no whether here. Of course you can say "give you some idea of whether it would cost something", but the same is true for prototypical free relative environments:

• See what Kim brought
• See whether Kim brought something

I agree the second is a content clause, but the first is a regular free relative, right? It means "see the thing which Kim brought" (which is an NP, and the sense can be split into the two roles, i.e. the "thing which" paraphrase)

it gives UV rays..which is important

I think the agreement suggests that the relcl expands on the whole "gives" predication, i.e. the clause is a dependent of "gives". In GUM such cases can be found by looking for verbs dominating a relative clause directly:

https://corpling.uis.georgetown.edu/annis/#_q=dXBvcz0iVkVSQiIgLT5kZXAgdG9rX2Z1bmM9ImFjbDpyZWxjbCI&_c=R1VN&cl=5&cr=5&s=0&l=10&o=random

[nschneid]

Thanks, I'll adopt that analysis for UV rays.

Re: free relatives: This is tricky. I think "see" is ambiguous. Note that it can accept either a direct object ("I saw the book") or a complement clause ("I saw that Kim brought something").

If the sentence were

• I need to see what Kim brought and whether anyone else brought anything.

I would interpret that as coordination of two interrogative complement clauses. It is about trying to discover something, so it is like saying "I want to discover the answer to the questions, What did Kim bring? and Did anyone else bring anything?" (but when those questions are embedded their form changes).

A canonical example of a free relative is with a verb like eat that doesn't license complement clauses:

• I'll eat what you cook me.
  • *I'll eat whether you cook the food.

At this page I've documented some of the main syntactic tests, including else (which favors interrogatives) and -ever (which only works for free relatives):

• I'll eat whatever you cook me. *I'll eat what else you cook me. => Free relative
• *I need to see whatever Kim brought and whether anyone else brought anything. I need to see what else Kim brought and whether anyone else brought anything. => Interrogative

In isolation, "I need to see what Kim brought" is ambiguous. It can imply 'there was something that Kim brought, and I need to see it' => free relative, or it can indicate an investigation into what Kim brought ('there was something that Kim brought, and I need to see what') => interrogative. Which leads to yet another test:

• There was something that Kim cooked, and I need to see/explain/*eat what.

[nschneid]

"see" also #122

[nschneid]

Stumbled upon another piece of evidence regarding the free relative/interrogative distinction: with expressions like (not) give a damn (and maybe also have no idea though my intuitions there are less sharp), what follows has to be a clause, not an NP, because then it would have 2 direct objects. Thus:

• I don't give a damn [what you cooked]/ccomp
• I don't give a damn [whatever you cooked]. (whatever*-clauses are free relatives)
• *I don't give a damn [your choice of food].

Likewise, "what you cooked" can be parsed as ccomp in "I don't care/know what you cooked" (know licenses a complement clause or direct object; care maybe as well: ?I don't care your reasons!).

[nschneid]

• [x] "which" PronType=Int tokens need reviewing. A lot are missing an acl:relcl annotation.

[nschneid]

Looking at the non-matrix-subject "what" free relatives with PronType=Int, I interpret many of these as interrogative clauses instead. Changing the ones headed by: bad, certain, discuss, find, give, moment, remember, restaurant [elliptical for "WHAT MATTERS IS not the restaurant but what you order..."], reveal, say, see, tell.

(“explain”, “hear”, “understand” could license interrogatives but in these cases I think the free relative interpretation is better)

[nschneid]

Tricky one: "She...listens to what it is you would like to achieve". Sort of a mix of a pseudocleft and an it-cleft! We can treat it as such:

obj(listens, what) acl:relcl(what, is) expl(is, it) advcl:relcl(what, like) xcomp(like, achieve)

E:obj(achieve, what)

[nschneid]

matrix-subject "what" free relatives with PronType=Int

[nschneid]

matrix-copular "what" free relatives with PronType=Int

[nschneid]

• [ ] q6 but check for missing edeprel from embedded clause
  • remaining tricky ones:
    • one thing you can't hide is when you're crippled inside
    • the mindset from when it was only smokers sitting around and drinking (should it be nsubj(sitting, smokers)?)
    • Have a good trip where-ever you end up. (guideline for this construction TBD)

[nschneid]

• [ ] q7: WH-word is det or advmod
• [x] q8: WH-word is "whose": double-check that this is PronType=Rel