Tagging of partitive `ne` in Italian (and maybe French?)

[ellepannitto]

Hello, @EleonoraZucchini and I are dealing with some instances of partitive ne in Italian and we came up with some doubts.

Our sentence is the following: "Io ne scelgo uno di questi", which translates to "I choose one of these". The "ne" functions as a partitive, reinforcing the expression "di questi". In fact it would be also acceptable to only have "Io ne scelgo uno", with "di questi" not expressed, or "Io scelgo uno di questi" (although we find it slightly less common maybe).

The same construction should also exist in French (e.g., "Prends-en un parmi ceux-ci"), but we don't recall anything like this in other Romance languages.

Our issue is that currently we only find instances of "ne" tagged as iobj in the treebanks, which is fine for sentences like:
"Ne abbiamo parlato a lungo" (we talked a lot about it, where "ne" replaces "about it")

In our sentence, however, it is an internal modifier of the object ("uno"). Se we would rather treat it as dislocated for instance, or else attach it to "uno".
At the same time it is also an issue of grammaticalization, as "sceglierne" is becoming more acceptable than the bare "scegliere".

What do you suggest?

[EleonoraZucchini]

This is another example, with a proposal of annotation https://arborator.grew.fr/#/KIParla-Forest/PBB004?sent=PBB004_67
The translation of this unit would be: here not many (buses) run

[amir-zeldes]

These seem to be a double realization of the same valency slot filler, so I think dislocated is appropriate. Standard practice is to give the 'real' deprel to the item occupying the canonical position, so you'd need to decide which is more 'normal'. In case of ties (for example in free word order cases) I've seen preferences to 1. take a pronoun as the 'real' representative; 2. prefer the earlier realization as 'real'. In this case, both of these heuristics would favor "ne" as the real deprel and "di questi" as dislocated, but if you strongly feel that "ne" is the optional doubled element, you could go the other way (and ideally document it as a rule in UD Italian for consistency).

[sylvainkahane]

dislocated supposes that the constituent is prosodically detached.
If It. ne is similar to Fr. en, I think there are several possible interpretations, and that the solution depends on the position of the object and its prosody.
Fr. en is a genitive pronoun that can be obl, obj or nmod: https://universal.grew.fr/?custom=6a5fdb3eafc80. I think you're interest by these last cases, where en depends on a pronoun or a noun which is obj. But it is also possible that the object be dislocated: https://universal.grew.fr/?custom=6a5fdb3eafc80.

[amir-zeldes]

dislocated supposes that the constituent is prosodically detached.

Actually the guidelines don't mention prosody - they refer to "fronted or postposed elements that do not fulfill the usual core grammatical relations of a sentence", and specifically mention the case of double realization (e.g. clitic doubling). I completely agree that these constructions often have interesting prosodic properties in spoken language, but since UD is also used for written language, prosody cannot be a necessary criterion.

I think you're interest by these last cases, where en depends on a pronoun or a noun which is obj. But it is also possible that the object be dislocated: https://universal.grew.fr/?custom=6a5fdb3eafc80.

I agree, dislocation can apply to any underlying function, but we should then use the dislocated deprel and not obj etc. (though it's possible to use something like dislocated:obj). I had a look at the results of the last query, and it looks like Rhapsodie does not annotate clitic doubling with "en" as dislocated - it assigns the same deprel to both duplicates, which make it look like there are multiple dependent slots with the same function, when in fact both are saturating the same valency slot. For example:

• qu' est -ce que vous en pensez de la boule magique? "What do you think of the magic ball?"

Here "en" is the same argument as "de la boule magique", and I think one of them should be dislocated. My choice would be "de la boule magique", as a postponed lexical realization, and this would be consistent with the English and Modern Greek examples in the guidelines.

[ellepannitto]

I also agree that dislocated can be a good annotation, otherwise expl in case it is fully grammaticalized (I'm not sure about this case).
Our annotation would thus look like this: https://arborator.grew.fr/#/KIParla-Forest/BOA1003?sent=BOA1003_37

And I also agree that prosody is a cue but not the only one. In informal Italian sometimes there is doubling with no particular prosodic marking or in writing. The prosody comes more from the position of the dislocation maybe (e.g., "Glielo hai detto a loro?" vs. "A loro glielo hai detto?" -- both translate to "did you tell them?", the "glie-" has the same function of "a loro" but, as far as I'm thinking now, only in the second case "a loro" would be prosodically marked, the first case is just informal Italian).

In case of ties (for example in free word order cases) I've seen preferences to 1. take a pronoun as the 'real' representative; 2. prefer the earlier realization as 'real'.

Generally we're following 1, our doubt here was that:

• it created a non projective structure
• the dislocation appears inside a noun phrase, but clitic pronouns (such as "ne" in Italian) attach to the VERB
• (this is a more general remark that I have): dislocation feels a very odd name to use when the actual lexicalized element is not dislocated at all from its "unmarked" position. But maybe this is a discussion about the use of dislocated in general in languages such as Italian with very free word order

[amir-zeldes]

Yes, I think the tree is right, assuming that's also how you would annotate "ne scelgo uno" without the "di questi". Non-projectivity is fine if that's how the argument structure works, and I can see the case that "ne" is really a dependent of uno. The only other decision is "which one is dislocated?", and I agree the pronoun being primary seems to be the default behavior in other UD languages as well (incl. Balkan clitic doubling, e.g. in Greek).

dislocation feels a very odd name

I agree 100%, but that's how it's been used in UD for a long time now. I sort of get it in some examples, where the doubling is the result of a hanging topic, which is often also called "right dislocation" in European languages. But at this point it is the standard UD analysis of clitic doubling as well, which is a pretty good match for what I think is happening in these examples too.

[Stormur]

ne is a "brutta bestia" (i.e. nasty)... II do not think that making it depend on the object purportedly taking the partitive is the right solution.

Like all the other clitic pronouns, it is indexing something on the predicate, so it should depend on it. Surely not as iobj. The problem is that it seems difficult to find an appropriate relation, as it is not really working as an argument, but rather indexing something about some arguments: it is really close to a movable affix, even more so than the other clitics. Barring aux because as of now it is reserved for verbal elements, I would venture to suggest expl. The present description comes close to what I meant: "These are nominals that appear in an argument position of a predicate but which do not themselves satisfy any of the semantic roles of the predicate.", and also it seems to fit the descriptio nas "nominal" and "non core".

By the way, it seems to be a trend of contemporary Italian to lose the use of ne, which goes quite well together with the more grammatical nature of expl.

Annotating ne as expl removes the problem of what to do with di questi and other similar phrases, which would depend as usual on their heads: here, as nmod of uno. Anyway, in a similar sentence I would not speak of dislocation as it is really clear that ne is acting as some pleonastic marker, so something has to take expl, but there is no dislocated.

What would be interesting, then, is finding a nice subtype for ne, maybe expl:part for partitive or along this line.

[sylvainkahane]

since UD is also used for written language, prosody cannot be a necessary criterion

almost all written languages are a spoken language before all. Written utterances can be read aloud. So I think that for most languages, prosody can be convoked also for the annotation of written corpora. Moreover punctuation is generally used to mark prosodic breaks; for instance, commas are used for dislocated elements.

About qu' est -ce que vous en pensez de la boule magique? "What do you think of the magic ball?", I agree that the expected annotation would be to have de la boule magique as a dislocated element. But the phrase is prosodically integrated, so it is possible that it is an example of double marking, which would be a non-standard construction in French. It would be nice to have a way to indicate that the construction is in a transitional space between two constructions.

[amir-zeldes]

it is really clear that ne is acting as some pleonastic marker

That is pretty much what clitic doubling is, which the guidelines analyze as dislocated

it is indexing something on the predicate, so it should depend on it.

I don't think so - if we didn't have "di questi", it would clearly substitute that meaning exactly, and I think everyone agrees "di questi" is a dependent of "uno". So I think if we had just "ne scelgo uno", we should still have "ne" as a dependent of "uno", because "ne" is pronominalizing that dependent of the nominal argument.

Written utterances can be read aloud. So I think that for most languages, prosody can be convoked also for the annotation of written corpora.

That's true, but there is ambiguity in writing, and we can't always be sure how something should be pronounced (think about contrastive fpcus emphasis, which can be expressed by all caps in English, but is rarely spelled in practice). I think it could be invoked as a hint or "tie breaker", but I wouldn't want to see it posed as a necessary criterion for any deprel.

the phrase is prosodically integrated, so it is possible that it is an example of double marking, which would be a non-standard construction in French. It would be nice to have a way to indicate that the construction is in a transitional space between two constructions.

Yes, and I think that may be true for Balkan clitic doubling as well. One option would be to use subtypes, but TBH I think that could be disruptive to easy comparisons between dialects and would also clash with cases where people might want to annotate dislocated subtypes with the major deprel being duplicated (like dislocated:obj). A more flexible option is to use some kind of MISC annotation, or, if it's viewed as a distinct construction, specifically to use the notation proposed for constructions in UCxn.

[Stormur]

it is really clear that ne is acting as some pleonastic marker

That is pretty much what clitic doubling is, which the guidelines analyze as dislocated

In the guidelines clitic doubling is pretty clearly treated through expl.

[amir-zeldes]

I hadn't realized Greek was cited there as well, thanks for pointing it out! But for Romance, the page at that link explicitly says:

For languages in which clitics and lexical nominals are usually in complementary distribution – languages, such as French, which obey “Kayne’s generalization” – then whichever of a clitic or a lexical nominal occurs will get the appropriate role, such as obj or iobj. In such languages, when doubling does occur, such as in spoken French, the right analysis is to regard the lexical nominal as dislocated

So it looks like both the dislocated and expl guidelines confirm we should be using dislocated here. As for Greek, unless I'm missing something, the guidelines appear to be internally contradictory. Under dislocated we have this example:

• Τον/dislocated ξέρουμε όλοι μας εδώ πολύ καλά, τον Γιάνη "We all know him very well, Yanni"

And under expl:

• Της/expl τον/expl έδωσε της Καίτης τον αναπτήρα "He gave it to her, the lighter to Kaiti"

I don't really see how they're different, notwithstanding the fact that two arguments are doubled in the second one. I had a look at UD_Greek-GDT and they are indeed using expl for it there, not dislocated.