Docs: Appos and Parataxis - guidelines vs annotated golden corpus (EWT)

[claudiafreitas]

1. Although guidelines on appos (http://universaldependencies.org/u/dep/appos.html) state that
   `
   > state senator Paul Mnuchin

should not be tagged as appos, the following cases are tagged as appos in the EWT corpus:

my _friend_ Nick Keck

Her _mother_ Ismat

The _hijacker_ Ahmed Alhaznawi

_bacteriologist_ Dr. Abdul Qadoos Khan

our youngest _daughter_ Elizabeth

We (in portuguese) are fine considering restrictive appositives as appos, but we thought we were diverging from the guidelines.. maybe not?

1. The following cases are tagged as appos as well - we were not sure if, according to guidelines, they should beapposor parataxis:

...and author of "Sacred Space and Holy War" (I.B. Tauris, 2002).

1. The example below, taken from the parataxis guidelines (http://universaldependencies.org/u/dep/parataxis.html) is tagged aslist, and not as parataxis

WASHINGTON (CNN)

I think we could benefit if both cases were treated as if they were the same...

WASHINGTON (CNN)
... and author of "Sacred Space and Holy War" (I.B. Tauris, 2002)

[sebschu]

Re 1: This distinction is a bit tricky since we only use appos if the descriptive noun phrase can stand on its own and be switched with the name but not in cases of "titles" such as senator which have to precede the name.

So I think all of the examples you provide should be tagged as appos, except perhaps _bacteriologist Dr. Abdul Qadoos Khan_ (unless there is a determiner in front of _bacteriologist_).

Note that for all the other ones, we can invert the order:

Nick Keck, my friend

Ismat, Her mother

Ahmed Alhaznawi, The hijacker

Elizabeth, our youngest daughter

[Stormur]

1. I think that the first group of examples should all be tagged flat, as per the guidelines. A construction like _Ismat, her mother_ seems to have another structure and nuance to me. In one case we have a title which "binds" with the noun, in the second case we have a specifying phrase.

However, the appos relation has indeed shown to be somewhat controversial (I'd say too restrictive and too English-oriented in its definition) and its use in the different treebanks varies wildly. A change should be discussed, and at least consider the use of apposition-introducing elements.

1. I don't understand what is tagged as appos here, since to me it all looks just like conjs and nmods:

conj(...,author)
cc(author,and)
nmod(author,Space)
case(Space,of)
...
conj(Space,War)
...

[amir-zeldes]

@Stormur I think the idea is that I.B. Tauris is the author, but this case actually doesn't obey the reversibility guidelines, due to the presence of the year:

* (I.B. Tauris, 2002) author of "Sacred Space and Holy War"

For this reason we use dep to attach author citations to names of works in UD_English-GUM.

As for "Ismat, Her mother", I think this is a perfectly normal apposition - we don't have separate labels for restrictive and non-restrictive amod or acl, so I don't think we need to make this distinction on appos (though it could certainly be a sublabel or MISC annotation, if some corpora have that information)

[Stormur]

@Stormur I think the idea is that I.B. Tauris is the author, but this case actually doesn't obey the reversibility guidelines, due to the presence of the year:

* (I.B. Tauris, 2002) author of "Sacred Space and Holy War"

My fault, I did not notice the name in parentheses! Well, it looks like a loose relation to me rather than an appos, with an internal flat structure, as typical for names. We need to see the entire sentence for clarity, though, to understand the reference and what should be the head of that element. If however _(I. B. Tauris, 2002)_ refers to the publishing house and not to the author, I see a parataxis, but not with _author_ as head (guidelines: "[...] placed side by side without any explicit coordination, subordination, or argument relation with the head word"). In any case I think that either loose-relation deprel or appos is preferable to an agnostic dep.

As for "Ismat, Her mother", I think this is a perfectly normal apposition - we don't have separate labels for restrictive and non-restrictive amod or acl, so I don't think we need to make this distinction on appos (though it could certainly be a sublabel or MISC annotation, if some corpora have that information)

_Ismat, her mother_ is an apposition, sure; but _her mother Ismat_ is a flat construction. The comma here is a sign that we have two different constructions (see also the guidelines: flat would not be appropriate most of the time if punctutation sets the two elements apart, exactly because punctuation is trying to convey a different relation). I would also treat _her mother, Ismat_ as a case of apposition, for the same reasons. Please correct me, but I don't think that the sentence

_Ismat her mother scolded me yesterday_

is grammatical, compared to

_Her mother Ismat scolded me yesterday_.

We have to rephrase it as

_Ismat, her mother, scolded me yesterday_.

So I argue that we can not really reverse the phrase without altering its structure (and the nuance of meaning, too), and that all the examples in point 1 are not appos.

[dan-zeman]

I agree that the examples _my friend Nick_ to _our youngest daughter Elizabeth_ are not appos. Also note that the current guidelines define appos as a left-to-right relation.

[amir-zeldes]

I think even if there is a slight nuance to not using the comma, it's risky to rely on this, as many languages do not have reliable comma usage rules (English is not very strict either), and for some languages (especially in historical treebanks) we don't have commas at all.

Incidentally I think "her mother Ismat" is actually grammatical, and maybe only slightly odd because a mother is usually unique, so it has to be non-restrictive. If it said "her sister Sarah" (as opposed to her sister Isabel), that would be 100% unmarked for me.

[Stormur]

The comma is of course a useful convention to detect such structures, but can not be the only criterion.

Incidentally I think "her mother Ismat" is actually grammatical, and maybe only slightly odd because a mother is usually unique, so it has to be non-restrictive. If it said "her sister Sarah" (as opposed to her sister Isabel), that would be 100% unmarked for me.

I considered only a form like _Ismat her mother scolded me_ as ungrammatical.

2. The following cases are tagged as `appos` as well - we were not sure if, according to guidelines, they should be`appos`or `parataxis`:

...and author of "Sacred Space and Holy War" (I.B. Tauris, 2002).

I searched for the whole sentence, which is:

_Juan Cole ( www.juancole.com ) is a professor of modern Middle East history at the University of Michigan and author of " Sacred Space and Holy War " ( I.B. Tauris , 2002 ) ._

I, too, am not convinced by both choices of appos, as they appear in the treebank:

appos(Juan,www.juancole.com)
...
appos(Space,I.B.)

Namely, I see both rather as loose relations, so either as list or parataxis; I would say, in the first case

list(Juan,www.juancole.com),

following the examples of lists constituted by names, e-mail addresses, and so on. And for the second one I agree with the interpretation as

parataxis(Space,I.B.).

I do not consider it an apposition, because it does not truly "define, modify, name, or describe" the title of the book. It's just extra information about the book itself "without any explicit [...] argument relation with the head word", in this case omitting something on the lines of _edited by I.B. Tauris in 2002_.

So, @claudiafreitas, answering point 3, yes, I totally agree with you: both cases should be treated as parataxis.

I admit though that the border between list and parataxis is quite slippery... Maybe we could think of a unique deprel for loose relations? Would it be too coarse?

[nschneid]

The examples at http://universaldependencies.org/u/dep/list.html seem to be outside of a (grammatical) sentence.

For citations and other extra metadata injected in sentences in writing, which would not normally be uttered as such in speech, parataxis seems better to me. Though there's probably room for clarifying or refining parataxis.

[amir-zeldes]

I like parataxis for these last two cases better than list - you could consider it to be a parenthetical, similar to:

Juan Cole -- you can find more info about him at www.xyz.com -- is author of ...
parataxis(Juan,find)

This is what PTB brackets would group under a PRN node, and I understand parataxis to cover these cases as well.

[sylvainkahane]

For UD_French_Spoken, we decided to split appos in two relations: appos:conj for _my sister, Mary_ or _Mary, my sister_ and appos:nmod for _my sister Mary_ or _le journal Le Monde_. See http://universaldependencies.org/fr/dep/index.html for details.
For written corpora the distinction can be recovered using the punctuation, but for a spoken corpus it needed to be explicitely encoded.
The distinction is based on the fact that in appos:conj we have two (prosodic) phrases, while in appos:nmod there is only one. In some sense, appos:conj is a particular case of paradigmatic list as the coordination. Two phrases X and Y forms a paradigmatic list if Y piles up on X and occupies the same syntactic position. On the other side, appos:nmod is just a case of nominal modification of X by Y, even if by chance Y can commute with X. These two relations could be better named conj:appos and nmod:appos, as we do in our SUD version of the corpus.

Note that we also consider a bunch of different parataxis relations, due to the variety of such relations in spoken French. Again, details on http://universaldependencies.org/fr/dep/index.html and in our paper Trois schémas d’annotation syntaxique en dépendance pour un même corpus de français oral : le cas de la macrosyntaxe.

[arademaker]

Why can we take a more semantic definition of appos? Like two NP can be linked by appos whenever they reference the same entity. In the case, my friend Alexandre... would be fine to have appos between 'friend' and 'Alexandre'.

[amir-zeldes]

Some other constructions for which appos is frequently used are non-coreferential, such as the city-state construction:

Boston, MA
appos(Boston,MA)

I'm not saying this is the best/correct solution for these phrases, but if coreference were a requirement for appos, we'd need to do something different with these. There are also cases involving set-member relations, such as:

They have many problems, droughts for example.
She knows everyone, Beth, Bobby, Jane...
(everyone is not necessarily strictly the same as these three people)