invariable gender ADJ in Spanish - include Gender feature?

[AngledLuffa]

Similar to the question of invariable days in Italian, what do we do with features for words that are invariable? For example, indígena does not change ending even for masc NOUNs, so it should presumably have the lemma indígena. Should it have Gender features based on context? GSD sometimes has those features, sometimes not, whereas AnCora never does.

[nschneid]

My personal opinion is that if the gender can be inferred from syntactic context, it should be recorded in a feature.

[AngledLuffa]

can do

[AngledLuffa]

On a similar topic - what to do with Fondo de Desarrollo Indígena / Indigenous Development Fund? Is Indígena a PROPN here or ADJ? It kinda feels like ADJ to me, even though it's in a proper name.

If PROPN, it should still have features, right? The vast majority of PROPN in AnCora do not have features

[sylvainkahane]

It's the discussion we had in #985. I don't see the point of adding the feature Gender on invariable adjectives. If you do implement it, add a feature to indicate that they are not inflected in Gender, such as Exponence[Gender]=Absent.

@AngledLuffa and @nschneid, can you explain what benefit there would be in adding the feature Gender to adjectives that do not vary in gender? What would be the purpose? What could you do that you couldn't do before?

On the other side, if you add the feature Gender on all adjectives, we will lose interesting information for linguists: What is the yield of the noun-adjective agreement in Spanish? In other word, if I am interested in how the syntactic relations are marked in the languages around the world, I would like to know whether Spanish really uses agreement or if it rather rely on other clues.

[nschneid]

@sylvainkahane posted my answer in #985.

[amir-zeldes]

What would be the purpose? What could you do that you couldn't do before?

Assuming we understand gender as something determined in the context of a specific sentence, then you can easily get the proportion of Masc vs. Fem adjectives in the corpus that way. Without this annotation, breaking down adjectives by gender for stats would give you Masc, Fem and a bunch of cases you can't easily put in either group. You seem to be more interested in the morphology (do they end in -a, -o, etc., or more precisely, is their Masc form different from their Fem form). But other people may be interested in their syntactic behavior - are they modifying a masculine or feminine noun? There are definitely questions in semantics, pragmatics and sociolinguistics in which identifying feminine and masculine modifiers (in context) is interesting.

if I am interested in how the syntactic relations are marked in the languages around the world, I would like to know whether Spanish really uses agreement or if it rather rely on other clues

That's totally fine - you can add another annotation to study that if you like, or you can break down words by their endings, which at least for Spanish would be a good indication. You can even use the context-sensitive Gender feature to identify adjectives that have the same form for Masc and Fem.

[Stormur]

Similar to the question of invariable days in Italian, what do we do with features for words that are invariable? For example, indígena does not change ending even for masc NOUNs, so it should presumably have the lemma indígena. Should it have Gender features based on context? GSD sometimes has those features, sometimes not, whereas AnCora never does.

I also answer by saying that if they do not mark Gender in the form, then that morphological feature receives no value, and so must not be tagged on that word. As simple as that. There is no point in annotating what is not there. Beside, there will surely be contexts where it is not possible to pinpoint one specific value beyond more or less based speculation.

[Stormur]

On a similar topic - what to do with Fondo de Desarrollo Indígena / Indigenous Development Fund? Is Indígena a PROPN here or ADJ? It kinda feels like ADJ to me, even though it's in a proper name.

If PROPN, it should still have features, right? The vast majority of PROPN in AnCora do not have features

It is always ADJ. Even beyond its faultiness, as discussed in #1208, PROPN is not a feature that "spreads morphosyntactically" over a whole phrase.

The fact that PROPNs do not receive fetures in AnCora looks like a common bug, since they should receive features just like normal NOUNs.

[Stormur]

What would be the purpose? What could you do that you couldn't do before?

Assuming we understand gender as something determined in the context of a specific sentence, then you can easily get the proportion of Masc vs. Fem adjectives in the corpus that way. Without this annotation, breaking down adjectives by gender for stats would give you Masc, Fem and a bunch of cases you can't easily put in either group.

Because it is exactly like that: there are adjectives which inflect and others which do not. Then again, you are thinking of some argument-level features, but for those you do not need a feature tag on the single word and have instead to gather the features from the whole phrase to get the picture.

You seem to be more interested in the morphology

Yes, we are discussing what are called morphological features, or better, a combination of lexical and inflectional features. See also the guidelines (bold is mine): "Analogically to part-of-speech tags, features describe the word form but not necessarily its exact function in the given sentence. Most of the features are for locating the form in a slot of a morphological paradigm, and are canonical labels for the slot. [...]" and also "Not mentioning a feature in the data implies the empty value, which means that the feature is either irrelevant for this part of speech, or its value cannot be determined for this word form due to language-specific reasons."

if I am interested in how the syntactic relations are marked in the languages around the world, I would like to know whether Spanish really uses agreement or if it rather rely on other clues

That's totally fine - you can add another annotation to study that if you like, or you can break down words by their endings, which at least for Spanish would be a good indication. You can even use the context-sensitive Gender feature to identify adjectives that have the same form for Masc and Fem.

We already have that annotation, one is FEAT and the other one are deprels.

One of the motivations for annotating is also that I could be not knowledgeable enough of the heuristics which can work for Spanish, but I lean on an informative annotation to perform investigations.

[amir-zeldes]

Most of the features are for locating the form in a slot of a morphological paradigm, and are canonical labels for the slot

@Stormur From the definition you cite, it seems pretty clear we should be filling out the properties of the relevant paradigm slot, which for a nominal in an IE language means its case, number and gender. Basic linguistic theories of IE languages generally assume that every nominal has a case, number and gender, even if the form happens to be syncretic. If the paradigms of the language generally distinguish these features, then FEATS should indeed provide the cell coordinates in the paradigm. And "Analogically to part-of-speech tags, features describe the word form but not necessarily its exact function in the given sentence" means that, just as we determine whether "shop" is a NOUN or a VERB in context (although in English there is no morphological form distinction), so we must also determine the gender of those Spanish adjectives for which Masc and Fem happen to look the same.

[AngledLuffa]

It is always ADJ. Even beyond its faultiness, as discussed in #1208, PROPN is not a feature that "spreads morphosyntactically" over a whole phrase.

Same applies to Nueva in Nueva York?

[Stormur]

It is always ADJ. Even beyond its faultiness, as discussed in #1208, PROPN is not a feature that "spreads morphosyntactically" over a whole phrase.

Same applies to Nueva in Nueva York?

Yes.

[Stormur]

Most of the features are for locating the form in a slot of a morphological paradigm, and are canonical labels for the slot

@Stormur From the definition you cite, it seems pretty clear we should be filling out the properties of the relevant paradigm slot, which for a nominal in an IE language means its case, number and gender. Basic linguistic theories of IE languages generally assume that every nominal has a case, number and gender, even if the form happens to be syncretic. If the paradigms of the language generally distinguish these features, then FEATS should indeed provide the cell coordinates in the paradigm. And "Analogically to part-of-speech tags, features describe the word form but not necessarily its exact function in the given sentence" means that, just as we determine whether "shop" is a NOUN or a VERB in context (although in English there is no morphological form distinction), so we must also determine the gender of those Spanish adjectives for which Masc and Fem happen to look the same.

No.

I stop discussing it here because it is obviously pointless. The basic misunderstandings are too big.