Extraposed objects in Swedish

[AleksandrsBerdicevskis]

What do we do with constructions with extraposed objects/complements like _X gör det Y att Z 'X makes it Y to Z' (göra 'make' is not the only verb which can form such constructions, but by far the most common one)?

All Swedish treebanks agree on:
expl (gör, det)
xcomp (gör, Y),
and that seems to be in accordance with the guidelines. With respect to Z, however, there seems to a different policy in every treebank.

PUD: ccomp (gör, Z)
Example: Till sist måste vi göra det lättare för folk att rapportera misstänkta säkerhetsluckor och skadliga e-postmeddelanden.

As far as I can see, English-EWT does the same, but the difference is that in PUD (and other Swedish treebanks), för folk is an obl on göra, while in English-EWT, it would be an nsubj of rapportera.

LinES: ccomp (gör, Z) and xcomp(gör, Z)
ccomp: Detta gör det svårt att ge andra användare tillträde till Access-projektet.
xcomp: Världen fanns utanför honom, omkring honom, framför honom, och den hastighet med vilken den hela tiden förändrades gjorde det omöjligt för honom att uppehålla sig någon längre stund vid varje enskildhet.

Talbanken: xcomp (gör, Z)
Prishöjningar gör det dyrare för oss att leva.

Swell: csubj (Y, Z)
Att behärska svenska gör det lättare att lära sig indoeuropiska språk.

I can see that the Swell solution is rather unconventional (to be honest, I do not remember whether we discussed it explicitly), so perhaps it is not to prefer (though there is certain simplicity in it, and it does not seem obviously wrong to me).

Between xcomp and ccomp, I'd say that xcomp is more suitable: if we view to Z as an argument of the verb, then it is obligatory, and it does inherit a subject from a higher clause (and from a fixed position, since as far as I can imagine, it will always be from an oblique).

It feels a bit weird, however, to have two xcomp with different meanings on the same verb (Y is a secondary predicate, while Z is not). But since it's not formally forbidden, it seems that xcomp should be preferred over ccomp. An alternative is to follow the English-EWT strategy, but I think that it is more natural to treat the för N construction as an oblique argument on the main verb rather than a subject of the subordinate verb.

[jnivre]

First of all, I don't understand what this has to do with extraposed objects, but that is maybe less important. :)

This reminds me of the discusison in #1128, and for similar reasons I think it needs to be ccomp rather than xcomp. It is true that in case like "... lättare för oss att leva", it is natural to understand the implicit subject of "leva" as co-referential with "oss", but the fact that we don't need a co-referential argument in the higher clause shows that it is not a genuine case of control: "... lättare att leva". Here the subject of "leva" is completely unspecified, isn't it?

[AleksandrsBerdicevskis]

That's what these constructions are called in (the original) Eukalyptus (I think the logic is that det is the formal/"usual" object, while the att-clause is the logical/extraposed object).

In your example (and in the first example from LinES, and in the example from Swell) it is indeed unspecified, but referentially/semantically it's still the same as the unspecified oblique argument of the main verb. As far as I understand the guidelines for xcomp, it is not required that the argument in the higher clause is overt, it is just required that it is not possible to think of an interpretation where the subject of the lower clause is different from it. I'd say this requirement is met (unless there are some counterexamples I didn't think about?)

Maybe it's a bit of a stretch to talk about coreferentiality between two non-overt entities, but that seems less of a stretch to me than using ccomp for cases like gör det dyrare för oss att leva where all the criteria for xcomp are met. (Another solution would be to use xcomp when the to-be-inherited subject is overt, and ccomp when it isn't, but I don't think that's optimal.)

As for går att in #1128, I am not sure. I think these constructions are a bit different, ccomp does seem more motivated there.

[jnivre]

I think xcomp is ruled out by the fact that the "controller", whether present or not, is not a core argument of the higher clause. According to the guidelines, the only cases where xcomp is licensed by an implicit controller are cases of pro-dropped subjects.

[LarsAhrenberg]

Maybe ccomp becomes more intuitive if we think of the infinitive as a subject of the adjective analogously to how it is treated with a copula:

Att leva nu är lättare: csubj(lättare, leva)
Det är lättare att leva nu: csubj(lättare, leva)
Det blir lättare att leva nu: xcomp(blir, lättare), csubj(blir, leva)
Detta gör att det är lättare att leva nu: ccomp(göra, lättare), csubj(lättare, leva)
Detta gör att det blir lättare att leva nu: ccomp(göra, blir), xcomp(blir, lättare) csubj(blir, leva)
Detta gör det lättare att leva nu: ccomp(göra, lättare), csubj(lättare, leva),

For some reason there must not be a finite verb in the _X gör det Y att Z ' construction but I think it makes semantic sense with csubj(lättare, leva). A problem may then be constructions with a non-expletive object:

Detta gör att han är lättare att leva med: ccomp(gör, lättare), nsubj(lättare, han), ?(lättare, leva)
Detta gör honom lättare att leva med: obj(gör, honom), ccomp(gör, lättare), ?(lättare, leva)

[AleksandrsBerdicevskis]

@jnivre With these requirements, ccomp does seem to be the only option. Might be good to spell it out more explicitly in the guidelines that the controller must be a core argument.

But:

@LarsAhrenberg So are you advocating for ccomp(göra, lättare), csubj(lättare, leva) in (1)?
(1) Detta gör det lättare att leva nu
I thought lättare was a clear xcomp, and I am still not entirely certain about leva, but I think @jnivre has a strong case with ccomp(gör, leva).

I agree that your examples with a non-expletive object are problematic. In Talbanken, contexts like
(2) sakerna är lätta att hitta
seem to have advcl(lätta, hitta), as opposed to
(3) Det är lätt att hitta sakerna
which have csubj(lätta, hitta).

Following this logic, I guess it should be advcl in (4-4)
(4) Detta gör att han är lättare att leva med
(5) Detta gör honom lättare att leva med

But then shouldn't it indeed be csubj(lättare, leva) in (1)?

[LarsAhrenberg]

@AleksandrsBerdicevskis I now agree with you, after consulting manual pages, that we should have xcomp(gör, lättare) in (1) detta gör det lättare att leva This would agree with how we treat a nominal such as livet in Detta gör livet lättare. I am a bit confused by your last question since it seems to contradict your earlier preference for @jnivre's ccomp(gör, leva).

I think the following sentence is grammatical in Swedish, although awkward:

(1') Detta gör att leva lättare: Here we would have ccomp(gör, leva). If extraposition does not change the role we would have the same dependency in (1). We can also have
(1") Detta gör livet lättare att leva -- same or different? Either ccomp(gör, leva) or ccomp(lättare, leva) or advcl(lättare, leva)

[AleksandrsBerdicevskis]

Yes, I know I've contradicted myself :) That's because I am not sure which analysis is best, and I'm also a bit confused now about who stands where.

(1) Detta gör det lättare att leva (where det is "empty" and not an anaphor referring to, say, "livet"): xcomp(gör, lättare) AND [ccomp(gör, leva) OR csubj(lättare, leva)]. If I understand correctly, @jnivre preferred the former structure and @LarsAhrenberg preferred the latter. Is that right?

I am not sure myself. The ccomp solution seems to be closer to the current practice, but it seems a bit weird that the construction "X att Y" gets a different analysis than in (3)

(3) Det är lätt att hitta sakerna: csubj(lätt, hitta)

As for the new examples:
(1') Detta gör att leva lättare: ccomp(gör, leva), xcomp(gör, lättare), right?
(1") Detta gör livet lättare att leva: obj(gör, livet), xcomp(gör,lättare), advcl(lättare,leva) (rather advcl than ccomp, because it's not a core argument?)

[jnivre]

First of all, I must apologize for not paying proper attention to the pleonastic nature of "det" in "göra det lättare att V". If we have a real object, as in "detta gör livet lättare att leva", then I think we have an xcomp-ccomp chain, because "lätt" is a tough predicate:

detta gör livet lättare att leva
nsubj(gör, detta)
obj(gör, livet)
xcomp(gör, lättare)
ccomp(lättare, leva)

With the pleonastic "det" I think I agree with @LarsAhrenberg that this should be analyzed as an extraposition variant of "detta gör att leva lättare", where "att leva" is ccomp (because it clausal and can't be obj) and "lättare" is xcomp, as in the example above. I am not sure what this implies for the other variants.