New analysis for multiple/nested subjects due to clausal predicative complement

[nschneid]

The plan is still being finalized. Here are cases with multiple enhanced subjects that need looking at:

• http://match.grew.fr/?corpus=UD_English-EWT@dev&custom=6224e2b0b87cb (and csubj variants)

[nschneid]

• copula should no longer take ccomp (and UPOS should be AUX not VERB): http://universal.grew.fr/?custom=629a99f26acdd
  • (remaining ones are existentials, "be like" idiom)
• be with a subject but no ccomp, excluding existentials: http://universal.grew.fr/?custom=629a9c577a14a
  • subset ruling out be as promoted predicate in relative clause: http://universal.grew.fr/?custom=632fc19381f1d
• other non-existential uses of be tagged as a VERB: http://universal.grew.fr/?custom=629a9d23c2555

[nschneid]

Clarification: should it be considered an outer subject if the copular predicate is a gerund (VBG)? E.g. I encountered the sentence

• True narcissism is just telling everyone your idiolect is a language.

I think it should be nsubj:outer(telling, narcissism) because a local subject of "telling" could be inserted (e.g. "you") or expressed with a possessive ("your").

• EWT verbs with cop dependents: http://universal.grew.fr/?custom=62c6e5b56b8af
  • NEW: subset without an :outer subject

[amir-zeldes]

Hm, this would vastly increase the number of :outer relations, and may well make it impractical to automatically add :outer to datasets without manual inspection. As long as there are two subjects (and it was not an annotation error, in which case 'not our fault'), it is probably close to trivial to add :outer to any TB, but for cases like this I can imagine there are various constructions we would not consider to be truly nested but could look like this, and which probably vary quite a bit across languages.

But I'm probably the wrong person to be commenting on this, since I don't think of nesting as a type of grammatical function in general. Do you have thoughts on this @manning ?

[nschneid]

We already have nsubj:outer documented for "the important thing is to keep calm". I guess I'm wondering whether the VBG form is more gerund-like and whether this a cause for concern. But I guess if it has an object, that means we're treating it as projecting a clause.

[nschneid]

Just want to point out a beautiful example:

"we will have to look at Plan B which would be to put the metering on our ROW"

Currently, "which" is the basic subject of "be". It should become the outer subject of "put".

But it is a relativizer, so the edep that is changed is E:nsubj(be,B) -> E:nsubj(put,B) ("B" is considered the head of "Plan B").

[nschneid]

I notice that sometimes the clause-serving-as-predicate is sometimes a coordination. Should the outer subject be propagated as enhanced nsubj:outer for both verbs, or just the first one?

• The only neurological sign that he shows is that he barely drags his toes and sometimes has trouble with his right lead.
• The bottom line is that the food isn't great and is relatively high-priced.
• The first thing you notice when you arrive on location is that the waiting line literally goes out the door and spills into the parking lot.

Semantically I don't see a need to say that "sign" is separately a subject of both "drags" and "has", but perhaps we should propagate coordinations irrespective of meaning?

[nschneid]

• In EWT, 21 hits for 'be' with an xcomp dependent, mostly indicating conditionality/subjunctive mood? Should 'be' be an aux instead? This doesn't feel like a copula to me. GUM is inconsistent (a couple aux examples and a couple cop examples among other uses of cop).

  • if he were to actually read the letter
  • if Americans are to be removed from the picture entirely
  • he sought to take a crop dusting course that was to last up to 6 months

[amir-zeldes]

should the outer subject be propagated as enhanced nsubj:outer for both verbs

yes, I would expect that

regarding "X is to VERB" meaning "X must VERB", I think making 'be' the head would be odd, not just because be is rarely the head but also because it's modal - if "must" is not the head in such constructions, then neither should be be.

As for the deprel, I could see doing aux, but then we would have an infinitive governing an nsubj directly without cop, which doesn't explain the morphosyntax well. So maybe cop is actually the more elegant analysis overall. But if you think aux is much better I could go along with that (more faithful to the modality thing, less faithful to the agreement pattern).

[nschneid]

then we would have an infinitive governing an nsubj directly without cop

Right, though that also happens with for-to infinitivals.

aux seems more straightforward to me than cop.

[amir-zeldes]

I can live with that. Can implement in GUM.

[nschneid]

Fun example with subject-aux inversion: "it was very difficult to see what I would actually look like were I to purchase some of these dresses."

[amir-zeldes]

Implemented in GUM, it's really quite rare though

[nschneid]

@amir-zeldes also

• The first referendum is to be held from November 20 to December 11, selecting a single new flag design out of about four finalists.
• USI's exhibit, which is located in Evansville, Indiana, is to run from January 23 through March 9.
• But there has been no change in my concept of the goals which this country must strive for if it is to meet its responsibilities to its people and those who depend upon it.
• The most enduring of those was with John FitzGibbon, 2nd Earl of Clare — four years Byron's junior — whom he was to meet unexpectedly many years later in Italy (1821). [22]

[amir-zeldes]

Thanks, all fixed, plus some more stuff

[nschneid]

Update: the main work adding :outer is done. There are still open checkboxes above with queries that surface some issues with copulas in general. Would be good to fix these at some point.