DEPRELs for "går att VERB" in Swedish

[harisont]

@AleksandrsBerdicevskis and I encountered the sentence fragment

huruvida detta är optimalt går att debattera

and annotated it as

går -[csubj]-> debattera -[ccomp]-> optimalt

considering that we would analyze

det går att debattera huruvida detta är optimalt

in the same way (+ det<-[expl]- går).

Cross-checking with Talbanken and LinES, we found that the pattern

går -[csubj]-> VERB -[ccomp/obj]-> (det som går att VERBa)

does occur both in Talbanken and LinES, but we also found:

• går -[csubj]-> VERB -[xcomp]-> (det som går att VERBa)
• (det som går att VERBa) <-[nsubj]- går -[advcl]-> VERB
• (det som går att VERBa) <-[nsubj]- går -[xcomp]-> VERB

(query results here)

So, my questions are:

1. are these annotation inconsistencies, or am I missing some nuances?
2. if these are inconsistencies, do you agree on går -[csubj]-> VERB -[ccomp/obj]-> (det som går att VERBa)?

[LarsAhrenberg]

To take something simpler like "Detta går att debattera", I would treat "Detta" as an nsubj and "debattera" as xcomp taking "Detta" as a surface subject. A motivation for this is that clefting the sentence into "Det är detta som går att debattera", you cannot drop "som" indicating that it is a subject.

In analogy, in "Huruvida detta är optimalt går att diskutera" I would treat "optimalt" as csubj, and "debattera" still being xcomp.

This is the annotation logic I've tried to follow In LinES though I see now that the annotation for the example doc4-1473 is incorrect.

What does @jnivre think?

[harisont]

I'm not convinced: even in the simpler example, isn't "detta" the object of "debattera", as in

att debattera detta går

and

det går att debattera detta

?

[LarsAhrenberg]

Well, it is both, I would say. How would you analyse "Ja, det går (bra)" as an answer to the question "Går detta att diskutera?"

[jnivre]

I share the intuition of @LarsAhrenberg that "går att V" is an xcomp-construction. It is true that in the near-synonymous sentence "man kan debattera huruvida detta är optimalt", there is clearly a ccomp relation from "debattera" to "optimalt", but I am not sure that it follows from this that the same relation holds in the "går att V" construction.

This seems to be related to tough-movement examples like "John is easy to please". Even though "John" is clearly the direct object of "please" in "To please John is easy", it does not follow that "John" is the object of "please" in "John is easy to please". At least in UD, this would be analyzed as:

nsubj(easy, John)
xcomp(easy, please)

[nschneid]

At least in UD, this would be analyzed as:

nsubj(easy, John)
xcomp(easy, please)

Tough constructions are an open issue for the guidelines: #923

I think when we last discussed it we were debating between ccomp and advcl. Not xcomp because the subject of "please" is not controlled by the matrix clause.

[harisont]

I think I see your point now @LarsAhrenberg and @jnivre. After all, if the first element was a personal pronoun, it would be a nominative, e.g.

hon går att ringa

and neither

henne går att ringa

nor

hon går att ringas

(right?), even though when the word order changes the accusative is used instead

det går att ringa henne

and as soon as "går att" is replaced with "kan" you get

hon kan ringas

So, I guess this is just a case where logical and syntactic subject don't coincide.

On the other hand, I also agree with @nschneid that xcomp doesn't really work. ccomp seems more convincing than advcl because "att VERBa" cannot be omitted without the meaning of "går" changing.

[jnivre]

Thanks, @nschneid and @harisont. I agree that xcomp is not appropriate because it is the object, rather than the subject, of the lower verb that is co-referential with the higher subject. So ccomp seems plausible.

[LarsAhrenberg]

I have put it on my list for v2.17 to change xcomp to ccomp for the second argument.

[jnivre]

And I will make the necessary changes to Talbanken.