Document amod:poss for Italian #1158
Conversation
|
Why is there only one 's' in the filename (and in the title inside)? |
|
fixed, sorry! My mistake, obviously |
nschneid
changed the title
|
OK, this looks better :) Otherwise I am not going to act upon this PR, that's up to the maintainers of the Italian guidelines. I personally do not see a problem in |
|
What part of speech are you going to use for such cases? If you’re interested in further discussing this proposal (or similar ones), I’d be happy to contribute to the discussion. I think that discussing this and similar issues collectively within the Italian treebank community would be beneficial, as such matters are best addressed jointly to prevent inconsistencies and noise in the Italian data. |
|
It's just about the possessives "mio", "tuo" etc., which are categorized as adjectives by most online dictionaries (see De Mauro). We're following the principle of leaving the functional annotation to syntax while sticking to the morphological category at POS level, so we'd like to tag them as ADJ. Functionally, even if the semantics is closer to reference than to qualitative modification, they occupy the slot of modifiers in Italian as the article/determiner in most cases is anyway required ("il mio libro" vs. "*mio libro" is the same alternation as "il libro rosso" vs. "*libro rosso"). In Italian treebanks we already treat elements such as "altro" or "stesso" as ADJ+amod, which are quite similar. This was also proposed in line with treatment of modifiers such as "tout" in French Rhapsodie |
|
So what is left as determiners in Italian? |
|
Everything in that paradigm: articles, indefinites, demonstratives... |
|
We already had a discussion about the subrelation poss in #1097. This subrelation has been introduced in English, because English has a special construction, |
|
I am extremely skeptical to the least in not treating possessives or other elements like altro, stesso, tutto, etc. as I would like to highlight the incoherence of the dictionary criterion: all these elements are usually labeled as adjectives (further: as "adjectives or pronouns"...) in dictionaries. So why treat questo as Why should position and/or presence of articles be relevant? They are more based on wider patterns of definiteness/specificty marking in Italian, not on the items themselves. So we have that Italian mio is an |
|
@sylvainkahane yes, thanks! We proposed the poss subrel just to identify the class more easily. Still the issue of uniformity among treebanks exists, as possessive adjectives have been considered determiners until now in Italian, no matter the rest of the context |
|
@Stormur I don't think that DET is a category used in any dictionary. As far as your argument is concerned, I'm not an expert on the topic and I feel it's mainly an issue of grammatical tradition. As I see it, this is one of the cases where the issue of parallelism has to be solved at the higher level of construction and strategies rather than trying to be super consistent with language-specific tagging. It is true that possessive adjectives fall is the description of determiners, but they can also fall in the category of adjectival modifiers, as they can coexist with different definiteness (il mio libro vs. un mio libro) and specificity (questi miei libri vs. dei miei libri vs. alcuni miei libri). Differently from articles, demonstratives etc the modification of possessive can also be relativized ("tra tutti i ibri, il mio libro è rosso" en. "among all books, my book is red" - "tra tutti i libri, il libro che è mio è rosso" en. "among all books, the book which is mine is red" is exactly the same, while "tra tutti i libri, questo libro è rosso" en. "among all books, this book is red" - "tra tutti i libri, il libro che è questo è rosso" en. "among all books, the book which is this one is red" is not) and can be modified ("il libro quasi mio è quello rosso" en "the book that is almost mine is the red one", e.g., if I'm approaching the counter to pay for the book I'm buying. "quasi" is modifiying possession, not the book). |
|
The guidelines specifically use possessives as typical examples of Many grammatical traditions do not recognise
These are all facts specific to Italian and are also partly semantically motivated: continuing on this note, then why do we not put articles in a different class than demonstratives, maybe It is not such criteria (co-occurrence of other determiners, of articles, position) which define the If I have to apply the proposed annotation coherently, than the mio in mia madre will be a * By the way, using made-up examples, I can also envision a sentence like il libro, che è questo (qua) [pointing to the book/picking it up and showing it to the audience], parla di... 'the book, which is this (one here), speaks about...' in a given context. So its presence in a copular relative clause is allowed, as long as that clause is not restrictive. |
|
If I'm being completely honest I don't think that And I don't think that the fact that meaning is contextually defined is a good argument either, otherwise we should be tagging personal pronouns with some special category instead of Of course the modifier-determiner distinction is a continuum (see here for instance) and somewhere we have to draw a line, my point is that I'd like to use the formalism to describe what happens in everyday language use and there possessives behave as adjectives, on the road to grammaticalization maybe. And of course it's a language-specific decision to make: I would use the Last note, also acceptability and grammaticality are very difficult to define and nothing is fully acceptable or grammatical without a context. I find the "quasi mio" example quite acceptable if you think of the less-grammaticalized contexts, but you can also think of coordination as an example, like "guarda la mia e bellissima macchina nuova" (also made up, but I swear I pronounced this sentence yesterday while bragging about my new car with a friend). The difference between the role of "mia" and the role of "bellissima" can be made somewhere else, like in the features where we say that one of the two expresses possession, but they both qualify the car as being mine and being very beautiful. |
|
I think that POS must be mainly defined by distributional properties. In other word the criteria to define a POS must be language-specific. It does not contradict the fact UD POS are comparative concepts: among your language-specific classes, you will choose to name some of them VERB, NOUN, etc. according to universal properties. The question of functional POS such as DET or AUX allows you to identify some particular subclasses of grammatical elements combining with nouns and verbs respectively. For DET, you will put in it articles (if the language has such elements) and the elements that behave in a similar way. If the language has no articles and all elements that depends on a noun have basically the same behavior, DET will be define on semantic ground. But for languages like Italian or English, which have articles and where the articles have distributional properties distinguishing them from adjectives, you can define DET based on some of these properties. It is the traditional definition of determiners in such languages, I think: in some syntactic positions (such as subject), the noun must be accompanied of a grammatical element; moreover these elements are generally not predicative (*we are the; *I'm each). Everything provided by @ellepannitto about "mio" shows that distributionally "mio" behaves as an adjective and not as an article. So I strongly recommend to analyze it as |
So you are OK with the same exact element being The arguments given above actually do not tell us that mio is particularly "more
It is actually pretty well defined: synsemantic modifiers. In other words (guidelines): "Determiners are words that modify nouns or noun phrases and express the reference of the noun phrase in context". The feature The fact that for some reason there is incredible resistance in accepting it, one reason for which being that many traditional grammars and dictionaries do not contemplate this word class (not directly, at least, because they do recreate similar classes when discussing hybrids like "adjectival pronouns" etc.) is another thing, and always baffles me. Even if we want to cherry-pick some distributional facts ("methodological opportunism" as W. Croft puts it) in order to claim that mio is an
The tag
It is true that this redundant annotation is not ideal (by the way, this was partly addressed in this article at UDW 2025), but you can also have non-
Notwithstanding some issue with the Time "adverbs" are a different issue still and anyway, as of now they have their own dependency
Yes, this is the point of the whole grammaticalisation status, autosemantic vs synsemantic, in UD. It has (also) semantic bases. There is lots of semantics in the examples given!!! The first is that Italian does not require the article with possessives with kinship terms (mio padre, tua madre, suo fratello), and some other "inherently specific" terms like casa, but not with most others. We can say molte persone 'many persons' but then define them as le molte persone in attesa 'the many waiting persons'. You do not use the article with questo 'this', because if you point to something in the context, it is specific and defined by default, so Italian does not require the article. And so on, these patterns are all purely semantically determined, and further language-specific.
There are gray zones, but mio is definitley not in one of them. I would like to point out that the "type of phrasal unit" of And then again, like which other adjectives do they behave the same? Not "prototypical" qualifying adjectives for sure, because I cannot say gentile madre the same way I say mia madre (no article). And there is a semantic difference between grande casa and casa grande, while there is not between la mia casa and casa mia. Examples are neverending. You cannot do this because you would end up with 100+ classes of modifiers, each and every one of them idiosyncratic to that specific language.
The distinction at the
Using made-up examples is always questionable. But in this specific case, no wonder they can be co-ordinated (even if it does not sound so good to me personally), as they are both modifiers. Exactly the same way as you can co-ordinate |
With regard to the guidelines, I think the best place to look is https://universaldependencies.org/u/overview/morphology.html#pronominal-words, which says:
The first item may give the impression that possessives can ONLY be |
In fact, I don't think that possessives have the same meaning in different languages. For instance, French mon is intrinsically definite. It doesn't correspond to mio, but to il mio. Moreover mio, can be used predicativelly (e mio 'it's mine') and it is not possible to say that in French. You have to decide if it is indefinite (c'est à moi ‘e un mio') or definite (c'est le mien 'e il mio') (Fr. mien can only be used in collocation with the definite article le.) So definitely It. mio and Fr. mon do not have the same meaning. English is different from both Italian and French, with my and mine. |
|
If we are using criteria from the syntax-semantics interface, another criterion to consider is referentiality, which has reflexes in syntactic behaviors such as pronominalization and interrogability. Possessive pronouns, both the definite "mon" kind and "mio" (though "mio" can be definite without an article, for example with family members as in "mio padre") correspond to entity terms - we can refer to them with another pronoun, ask who they are and extract them in interrogation ("whose father?"). This suggests that a treatment as PRON should also be considered (and in languages with genitive possessor pronouns, that is also what is done in UD). |
Then no words ever are comparable. We cannot have UD anymore. What you are describing is some "contour variation". For what concerns "referring possession to a person", they are the same. In French we distinguish two series two encode this, a determiner (mon) and a pronoun (mien). This has no bearings on Italian mio still being a
What was in the mind of the compiler here were probably Slavic possessive forms like Karlův 'of Charles', which inflects agreeing with the modified element, for Karel 'Charles'. Being such forms contentful words derived from other contentful words, we can concur on their But those parts of the guidelines is under
It is said:
For personal possessives, the variation considered is among synsemantic classes. No
Yes, it is a matter of |
I don't think I ever suggested that. For essere there is no ability to pronominalize or interrogate it in an isolated way that produces an entity term. This is not so for possessives: "Whose book? Father's" (the answer is a nominal). Notice also that this distinguishes it from regular determiners: "the book" has only one referring expression and we cannot separately ask about or pronominalize the article. In "Mary's/her book", there are two entities, and either can be prominalized or interrogated separately (and in some languages also extracted). |
Strange conclusion. Of course, we can have UD and universal criteria. But languages behave differently. It. mio, En. my, and Fr. mon behave differently. They express similar meanings and all have the common feature It is quite common that languages express similar meaning by words from different POS. |
Right. But The choice is between |
I was trying to extend the logic. Probably what I find confusing here:
is that two things are conflated: the semantic part of "corresponding to an entity" and some possible syntactic operations. This made me think again of the previously mentioned article, where it is said that "all possessives are pronouns" because "from a semantic and referential point of view they behave like pronouns". The layers of analysis are conflated here, creating a strange ouroboros (I am ignoring the fact that there no definitions whatsoever are given). For a verbal form, I could ask who? and get the answer with a verb which indexes this person.
How is that different from her book and whose? If the person is there, whatever its form it can be asked about, pronominalised... or am I not getting it?
Is this not just the same as saying that referents can be expressed as nominals, and this is independent from where we find them? Because as whose does not mark any person, it is not interrogating it in a more isolated way than for any kind of referent. |
That's why I said syntax-semantics interface :)
No, I don't think "she knows that" is a normal answer to the question "who?". The answer to the question "who" is just "she". Of course if I ask "who knows that" as a whole sentence you can say "Mary knows that", but it is clear that the part that corresponds to "who" is "Mary" - in fact in an in-situ WH (without 'movement') the paradigmatic principle also shows they are equivalent, even without the interrogration or pronominalization tests (which I take to be syntactic, not semantic - these are the classic tests for phrasehood next to movement and coordination).
It's definitely a different way - "whose" is the possessive WH pronoun, and the response can be a nominal (Mike's) or a pronoun (his). In the case of "his" there is also no adjectival etymology, it is literally the old genitive of the pronoun. |
I still do not understand the point and still see a short circuit like in the aforementioned paper, where a semantic observation forces a morphosyntactic analysis despite the data at hand. Yes, we see that the expression of a person entails a referent, which can then be referred to in other ways. But the considered words here, mio and its ilk, behave as modifiers. The feature she and other pronouns bear a What I meant is that you can always extract any referent and express it with a
The 3rd person singular is indexed by the -e ending on legge. So, applying the previous reasoning, do we have to split this e and treat it as a
We already discussed this in an other thread. It is highly questionable that "the two his" are the same element instead of showing a synchronic syncretism between two paradigms. |
|
This discussion is getting kind of long and I think we've made our points above, so I'll try to be brief and just add that being a modifier, or predicative etc. does not determine part of speech. All of the POS we are discussing can be either modifiers or predicates:
The guidelines say that "Pronouns are words that substitute for nouns or noun phrases", and interrogation is one way of testing that. Of course the guidelines also say that pronominal adjectives should not be tagged PRON, so the question is just where to draw the line. For English, and especially the etymologically genitive pronoun forms "his/her/its", I feel strongly that PRON is the right class (otherwise I wouldn't know why Greek αὐτοῦ should be a PRON either). For Romance it's indeed less compelling due to the agreement pattern with the modified noun, which is similar to adjectives, though there are also differences (lack of negation, comparison, word order, etc.). |
Yes, but in fact, this sentence was referring to adjectives that have nothing pronominal in them (they do not refer to the |
During work on spoken Italian, we introduced
amod:possfor cases of possessive modifiers, formerly only tagged asdet:poss(mio,tuo,suo, etc.), when they do not occupy the determiner slot.Especially (but not exclusively) in spoken Italian, the analysis as
det:possis not satisfactory as they can also appear after the noun, while determiners can only precede the noun in Italian.This PR adds it to the Italian docs.