Swedish _Alla vi barn i Bullerbyn_ -- apposition?

[AleksandrsBerdicevskis]

How to annotate a structure like this?
The best I could come up with (with @harisont's help) is to take vi as head and barn as apposition, i.e.

det (vi, alla)
appos (vi, barn)
nmod (vi, Bullerbyn)

...but I'm not entirely sure. (Partly because the description of appos seems very confusing to me (and is always followed in practice, at least in the Swedish treebanks)).

[jnivre]

The structure looks fine to me, but I'm not sure of the appos relation. While it seems to fit the co-referentiality criterion, it is not a prototypical case of apposition, and I think nmod could be considered as well.

For what it's worth, I personally don't think appos should be a universal relation at all, but rather a subtype of nmod, but as long as we're working with v2, they have to be treated as distinct.

[nschneid]

It looks similar to English "us students", for which we treat the pronoun as head and use nmod:unmarked: https://universaldependencies.org/en/dep/nmod-unmarked.html (discussion: UniversalDependencies/UD_English-EWT#436)

[jnivre]

Yes, I think nmod is fine. (We don't use the :unmarked subtype in Swedish.)

[amir-zeldes]

Agreed on nmod for this, but RE appos should be a type of nmod - I would disagree with that, even in a future UD v3. Noun modification to me means a type of subordination, whereas apposition is more like a coordinating relation. For example, apposition retains case agreement across both parts in languages with case marking, whereas noun modification typically has a specific case indicating the kind of modification. I also think appos is a good universal label, since I don't know of any language that doesn't have appositions, and they can have distinct morphosyntactic and even phonological characteristics.

[LarsAhrenberg]

An argument against the appos analysis is that you cannot change the order: *alla barn vi i Bullerbyn is ungrammatical. So I prefer nmod(vi, barn) as well.

[sylvainkahane]

(Sorry for the incomplete previous message)

For what it's worth, I personally don't think appos should be a universal relation at all, but rather a subtype of nmod, but as long as we're working with v2, they have to be treated as distinct.

@jnivre @amir-zeldes There are two different constructions currently annotated with appos in UD. In SUD we decide to separate them in two relations. One is called conj:appos because it shares properties with coordination: You have two conjuncts that can be inverted, the prosody clearly separate them in two phrases and the two conjuncts are separate by a comma in written texts. Here are examples in Spoken French. The other is indeed nominal modification and has been annotated mod@appos in SUD. The first one becomes appos in UD, the second nmod:appos. (You can see the UD annotation if you click below an example.)

[sylvainkahane]

(Sorry for the incomplete previous message)

For what it's worth, I personally don't think appos should be a universal relation at all, but rather a subtype of nmod, but as long as we're working with v2, they have to be treated as distinct.

@jnivre @amir-zeldes There are two different constructions currently annotated with appos in UD. In SUD we decide to separate them in two relations. One is called conj:appos because it shares properties with coordination: You have two conjuncts that can be inverted, the prosody clearly separate them in two phrases and the two conjuncts are separate by a comma in written texts. Here are examples in Spoken French. The other is indeed nominal modification and has been annotated mod@appos in SUD. The first one becomes appos in UD, the second nmod:appos. (You can see the UD annotation if you click below an example.)

[amir-zeldes]

@sylvainkahane thanks for pointing out that construction in French - I would not call those appos in UD either. In English, the equivalents of "garage Fiat" and "place Victor Hugo" would be annotated as compound (Fiat garage, Victor Hugo place), and I think there is at least some tradition of considering such constructions to be left headed compounds in Romance languages as well.

[Stormur]

Personal pronouns can modify nouns "adding person". I use det in these cases (a solution contemplated by the validator and the guidelines), and the noun is the head as the contentful element. A determiner like alla is also a DET/det as usual.

I do not see an apposition here: but notice that if appos is meant as a variant of nmod, and so modification is meant, then you need the "functional modifier" det, and I think it is clear that the functional elements depend on the contentful ones as usual.

You might refer to this nice article.

[LarsAhrenberg]

@Stormur, thanks for the pointer. The main arguments against an NMOD analysis in that paper seems to be the following:

It again conflates APCs with different, unrelated constructions (PP modifiers, genitives), does not contribute to the goal
of crosslinguistic comparability and obfuscates the fact that even though English and Greek APCs have
clear structural differences, they are comparable at least on a descriptive or phenomenological level

As NMOD is used for other constructions in English and Swedish where there is no case-marking, (with or without the subrelation unmarked) I think the first argument loses some of its force. I don't know Greek but if it is true that English and Greek are comparable "at least on a descriptive or phenomenological level" the last objection may possibly be turned the other way.

Going back to the Swedish data we can note that both the pronoun and the noun can be dropped meaning that either one can function as head:

1. alla vi barn i Bullerbyn
2. alla vi i Bullerbyn
3. alla barn i Bullerbyn

A difference though is that (1) and (2) can only be said by an inhabitant of Bullerbyn, ie they are "first person", while (3) can be said by anyone. Also, (1) and (2) are subject phrases; when in an object position the pronoun has to be oss (us), as in

4. (De gillar) alla oss barn i Bullerbyn

The choice between the pronoun or the noun as head can then be phrased in terms of Person being, in GPSG parlance, a head-feature or a foot-feature, like wh- or negation. To me it seems more natural to let the semantic difference be carried by a head.