Spanish: remove `Case` feature from passive `se` clitics?

[AngledLuffa]

Is Case=Acc correct on se clitics when the subject is in the passive voice?

So for example:

# sent_id = 3LB-CAST-104_c-1-s3
# text = Los activos en divisas en poder del Banco Central y el Ministerio de Finanzas se calculan en dólares estadounidenses y su valor depende del cambio oficial rublo-dólar que establece el Banco Central.
1       Los     el      DET     da0mp0  Definite=Def|Gender=Masc|Number=Plur|PronType=Art       2       det     2:det   _
2       activos activo  NOUN    ncmp000 Gender=Masc|Number=Plur 17      nsubj   17:nsubj        ArgTem=arg1:pat
...
16      se      él      PRON    p0000000        Case=Acc|Person=3|PrepCase=Npr|PronType=Prs|Reflex=Yes  17      expl:pass       17:expl:pass    _
17      calculan        calcular        VERB    vmip3p0 Mood=Ind|Number=Plur|Person=3|Tense=Pres|VerbForm=Fin   0       root    0:root  _
18      en      en      ADP     sps00   _       19      case    19:case _
19      dólares dólar   NOUN    zm      Gender=Masc|Number=Plur 17      obl     17:obl  ArgTem=argM:adv|Currency=Yes

here los activos (the accounts) I think ought to be nsubj:pass, at the very least. On top of that though, do either of the Case markers typically for se really fit this usage? Typically the passive se markers are tagged Case=Acc, but this isn't really an object per se, but rather a passive subject.

full disclosure, I only came up with the doubts about the Case when using ChatGPT to break down the different usages of se to write up a proposal to retag the se in GSD with some annotation company (the ones in GSD are annotated with much less context than the ones in AnCora). The belief that nsubj:pass is more appropriate than nsubj is entirely my own.

[AngledLuffa]

... this construction is the pasiva refleja, btw

[Stormur]

Well, it is a reflexive construction. It might have a passive interpretation, but it is not the same. I don't know about using the tag pass, but Case=Acc is in a sense definitely alright. The action is expressed as if it were "internal", starting and ending with the subject. I mean, it is the same se everywhere, you only have the passive interpretation based on semantics.

[AngledLuffa]

If it starts and ends with the subject, though, is it really representing an object? Or is the analysis here that "se" is telling us the passive subject is actually the object of some action?

I would also be totally fine with saying that obviously the AnCora people thought this over, and we don't have linguistics fluent in Spanish on this list (do we?), so we shouldn't be making this kind of change.

OTOH I do feel pretty strongly about it properly being nsubj:pass. One of the examples on the nsubj:pass documentation even mentions reflexive passive:

https://universaldependencies.org/u/dep/nsubj-pass.html

[dan-zeman]

the passive subject is actually the object of some action?

The passive subject is the undergoer/patient of an action. But it is not object, because it is subject. Passive (both reflexive and periphrastic) is mostly about changing the mapping between semantic roles of the participants and the subject relation (plus, usually also demoting the actor to an oblique dependent or removing it completely).

[AngledLuffa]

Right, but doesn't Case=Acc represent that something is the direct object? So that is the reasoning that led to me asking if these se should even have the Case=Acc feature

[dan-zeman]

Crosslinguistically it doesn't.

For Spanish reflexive se it is somewhat debatable because unlike irreflexive 3rd person pronouns, it does not change form between accusative and dative. But the debatable point is whether to annotate it with Case at all: if not, then it would not be annotated even when se is obj.

In the reflexive passive construction, se has the form of a core argument (direct object) while actually not representing an argument (that's why it is a subtype of expl).

Right now the paradigm table encoded by features in AnCora looks roughly like this:

[AngledLuffa]

The main motivation for me to ask this is to have a plan for disambiguating the se in GSD... if we want to say Case=Acc includes the pasiva refleja that's fine, but the Case=Acc,Dat is really unsatisfactory in my opinion. Unless we want to tag all of AnCora with Case=Acc,Dat as well

What are your thoughts on the nsubj:pass suggestion? It almost exactly matches the reflexive passive example for Czech

[dan-zeman]

What are your thoughts on the nsubj:pass suggestion?

If you are asking whether I think that the subject of a verb that has an expl:pass dependent should be attached as nsubj:pass, then yes, I think so.

[AngledLuffa]

Yes, sounds good, thanks

The other half of my question is what to do with Case features on se, since IMO neither paradigm is really satisfactory but having two different paradigms in the two large Spanish treebanks is worse

[Stormur]

Once you use expl:pass for it I do not think it is an issue. It is actually needed to infer that Spanish uses that strategy ([expletive] reflexive object) to produce what is a passive in a broad sense. Think it this way: if you are interested in collecting passive constructions, you have the pass tag for that and will not really need to look at morphological feats. But if you are interested in the latter's distribution, you need the information that se has that Case, which you would otherwise obliterate.

[AngledLuffa]

Are you referring to updating Case, or updating nsubj -> nsubj:pass?