search

UniversalDependencies / docs

Universal Dependencies online documentation
http://universaldependencies.org/
Apache License 2.0
273 stars 248 forks source link

Subjects of non-clausal deprels #1066

Open martinpopel opened 1 week ago

martinpopel commented 1 week ago

I would expect that nsubj and csubj can appear only as children of a predicate, i.e. a word with one of the "clausal deprels". According to the guidelines, clausal deprels are csubj, ccomp, xcomp, advcl and acl. Obviously root is also OK and in some cases also list, parataxis, conj and appos.

However, when I checked this hypothesis, I found many counterexamples. Many of those seem to be annotation errors, e.g. amod instead of acl. I could add this as another test into ud.MarkBugs (and maybe later also to the validator), but it would be nice to first reach an agreement that these are actually errors (against the current UD guidelines). Maybe there are some more deprels that should be allowed as parents of subjects (dislocated, orphan, discourse, reparandum).

I attach results.txt which I've generated using the following command:

cd universal-dependencies-2.14
for treebank in UD*; do
  echo ===== $treebank;
  cat $treebank/*.conllu | udapy util.Eval \
  node='if node.udeprel.endswith("subj") and node.parent.udeprel not in "csubj ccomp xcomp advcl acl root list parataxis conj appos".split(): print(node.parent.deprel)' \
  | sort | uniq -c| sort -rn;
done | tee results.txt
nschneid commented 1 week ago

Thanks! The English-EWT results below:

OK-looking

Fishy

Postponing changes until after the data freeze.

P.S. Also glanced at English-GUM, and 2 of the 5 obl ones look like errors to me, as well as the sole advmod one.

bguil commented 1 week ago

The following Grew request can be used to visualize the examples (with the clustering key e.label to cluster like in @martinpopel's command):

pattern { 
  e: X -[1<>csubj|ccomp|xcomp|advcl|acl|root|list|parataxis|conj|appos]-> Y;
  Y -[1=nsubj|csubj]-> Z 
}

For instance: on UD_French-GSD (in this treebank, most of the cases are related to ExtPos=PROPN as @nschneid explained above).

jnivre commented 1 week ago

Thanks, Bruno. But this pattern does not include “acl:relcl” as a valid parent, which presumably it should (since this is a subtype of “acl”). So I think one either needs to use a regexp or add all subtypes of the clausal relations as well.

Joakim

From: Bruno Guillaume @.> Reply to: UniversalDependencies/docs @.> Date: Sunday, 10 November 2024 at 10:59 To: UniversalDependencies/docs @.> Cc: Subscribed @.> Subject: Re: [UniversalDependencies/docs] Subjects of non-clausal deprels (Issue #1066)

The following Grew request can be used to visualize the examples (with the clustering key e.label to cluster like in @martinpopelhttps://github.com/martinpopel's command):

pattern {

e: X -[1<>csubj|ccomp|xcomp|advcl|acl|root|list|parataxis|conj|appos]-> Y;

Y -[1=nsubj|csubj]-> Z

}

For instance: on UD_French-GSDhttps://universal.grew.fr/?custom=67308f7e9416e (in this treebank, most of the caseshttps://universal.grew.fr/?custom=673091450c9d1 are related to ExtPos=PROPN as @nschneidhttps://github.com/nschneid explained above).

— Reply to this email directly, view it on GitHubhttps://github.com/UniversalDependencies/docs/issues/1066#issuecomment-2466687779, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ABZ7ZVXYUYLR4K6UZX7BKFDZ744B7AVCNFSM6AAAAABRPSMNWCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDINRWGY4DONZXHE. You are receiving this because you are subscribed to this thread.Message ID: @.***>

VARNING: Klicka inte på länkar och öppna inte bilagor om du inte känner igen avsändaren och vet att innehållet är säkert. CAUTION: Do not click on links or open attachments unless you recognise the sender and know the content is safe.

När du har kontakt med oss på Uppsala universitet med e-post så innebär det att vi behandlar dina personuppgifter. För att läsa mer om hur vi gör det kan du läsa här: http://www.uu.se/om-uu/dataskydd-personuppgifter/

E-mailing Uppsala University means that we will process your personal data. For more information on how this is performed, please read here: http://www.uu.se/en/about-uu/data-protection-policy

jnivre commented 1 week ago

I had a quick look at the examples in Swedish-PUD and the majority were indeed annotation errors, so thanks for finding those. The one valid annotation involved a song title, similar to some of Nathan’s examples. I will follow up with the larger Swedish-Talbanken when I have the time.

Joakim

From: Nathan Schneider @.> Reply to: UniversalDependencies/docs @.> Date: Sunday, 10 November 2024 at 01:03 To: UniversalDependencies/docs @.> Cc: Subscribed @.> Subject: Re: [UniversalDependencies/docs] Subjects of non-clausal deprels (Issue #1066)

Thanks! The English-EWT results below:

OK-looking

Fishy

— Reply to this email directly, view it on GitHubhttps://github.com/UniversalDependencies/docs/issues/1066#issuecomment-2466531594, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ABZ7ZVVSOHAPLXY64U5JPVDZ72WFVAVCNFSM6AAAAABRPSMNWCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDINRWGUZTCNJZGQ. You are receiving this because you are subscribed to this thread.Message ID: @.***>

VARNING: Klicka inte på länkar och öppna inte bilagor om du inte känner igen avsändaren och vet att innehållet är säkert. CAUTION: Do not click on links or open attachments unless you recognise the sender and know the content is safe.

När du har kontakt med oss på Uppsala universitet med e-post så innebär det att vi behandlar dina personuppgifter. För att läsa mer om hur vi gör det kan du läsa här: http://www.uu.se/om-uu/dataskydd-personuppgifter/

E-mailing Uppsala University means that we will process your personal data. For more information on how this is performed, please read here: http://www.uu.se/en/about-uu/data-protection-policy

bguil commented 1 week ago

But this pattern does not include “acl:relcl” as a valid parent, which presumably it should (since this is a subtype of “acl”). So I think one either needs to use a regexp or add all subtypes of the clausal relations as well.

The pattern I gave in the previsous message takes into account sub relations as well. I used the feature structure representation of the edge label (see Grew doc):

pattern { X -[1=acl]-> Y } means that the main relation is acl whether a subrelation is present of not.

But I see that in treebanks with enhanced dependencis like UD_Swedish-PUD, the pattern sould be updated like this to avoid reporting enhanced relations:

pattern { 
  e: X -[1<>csubj|ccomp|xcomp|advcl|acl|root|list|parataxis|conj|appos, !enhanced]-> Y;
  Y -[1=nsubj|csubj, !enhanced]-> Z 
}
jnivre commented 1 week ago

Thanks, Bruno. I realized I had misunderstood something. :) But I did note edeps, so thanks for fixing that.

Joakim

Skickat från Outlook för iOShttps://aka.ms/o0ukef

Från: Bruno Guillaume @.> Skickat: Sunday, November 10, 2024 1:42:36 PM Till: UniversalDependencies/docs @.> Kopia: Joakim Nivre @.>; Comment @.> Ämne: Re: [UniversalDependencies/docs] Subjects of non-clausal deprels (Issue #1066)

But this pattern does not include “acl:relcl” as a valid parent, which presumably it should (since this is a subtype of “acl”). So I think one either needs to use a regexp or add all subtypes of the clausal relations as well.

The pattern I gave in the previsous message takes into account sub relations as well. I used the feature structure representation of the edge label (see Grew dochttps://grew.fr/doc/graph/#edges):

pattern { X -[1=acl]-> Y } means that the main relation is acl whether a subrelation is present of not.

But I see that in treebanks with enhanced dependencis like UD_Swedish-PUD, the pattern sould be updated like thishttps://universal.grew.fr/?custom=6730b775dac0d to avoid reporting enhanced relations:

pattern { e: X -[1<>csubj|ccomp|xcomp|advcl|acl|root|list|parataxis|conj|appos, !enhanced]-> Y; Y -[1=nsubj|csubj]-> Z }

— Reply to this email directly, view it on GitHubhttps://github.com/UniversalDependencies/docs/issues/1066#issuecomment-2466741386, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ABZ7ZVUNX3ATUXXU7U73UEDZ75PEZAVCNFSM6AAAAABRPSMNWCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDINRWG42DCMZYGY. You are receiving this because you commented.Message ID: @.***>

VARNING: Klicka inte på länkar och öppna inte bilagor om du inte känner igen avsändaren och vet att innehållet är säkert. CAUTION: Do not click on links or open attachments unless you recognise the sender and know the content is safe.

När du har kontakt med oss på Uppsala universitet med e-post så innebär det att vi behandlar dina personuppgifter. För att läsa mer om hur vi gör det kan du läsa här: http://www.uu.se/om-uu/dataskydd-personuppgifter/

E-mailing Uppsala University means that we will process your personal data. For more information on how this is performed, please read here: http://www.uu.se/en/about-uu/data-protection-policy

mr-martian commented 1 week ago

Most of the cases flagged in Ancient Greek-PTNK are non-finite clauses with determiners. I'm not sure whether ExtPos=NOUN or changing nmod and obl to acl and advcl is the better solution here.

The rest are errors that I'll fix after the data freeze.

Stormur commented 1 week ago

I started by giving a look to these structures in Latin UDante.

Most fall under the label of dislocated, flat and orphan: I see that the subtype of the dislocated relations should be made more coherent, but we are in fact in presence of elements expanded with clauses or clauses juxtaposed to nouns, like "as the [proverb \<clause>] goes" (where we have flat). I would not count them as errors. As always, these relations are kinds of horizontal "non-relations"...

Then we have some genuine errors in two obls and one obj which should actually be advcl (two of them with non finite verb forms, specifically infinitives); and one nsubj which is erroneously repeated instead of having a conj.

Stormur commented 1 week ago

Updates for other Latin treebanks with which I have experience:

I also gave a look at

Stormur commented 1 week ago

Just a couple of comments and cents.

@mr-martian

Most of the cases flagged in Ancient Greek-PTNK are non-finite clauses with determiners. I'm not sure whether ExtPos=NOUN or changing nmod and obl to acl and advcl is the better solution here.

It is similar to what happens in Latin, but I guess you are referring to the article before infinitives, especially. Then I would suggest going for advcl. The infinitive is a verbal noun, but you are also acknowledging they are clauses. It is not so much different than having just a preposition without an article in Latin, no?

@nschneid

  • compound: quotative compounds, e.g. a "You've been yelped!" card. Maybe ExtPos would help here.

This is exactly the same case as the proverbs I mentioned above. I just notice that we use flat instead of compound. I do not see it much different than phrases like the king Pelayo (or Dwayne " The Rock " Johnson, from the universal guidelines).

nschneid commented 1 week ago
  • compound: quotative compounds, e.g. a "You've been yelped!" card. Maybe ExtPos would help here.

This is exactly the same case as the proverbs I mentioned above. I just notice that we use flat instead of compound. I do not see it much different than phrases like the king Pelayo (or Dwayne " The Rock " Johnson, from the universal guidelines).

In English a "You've been yelped!" card would be considered a compound, as the quoted phrase premodifies a noun representing the main meaning of the nominal (it is a kind of card).

the king Pelayo: not appos?

Stormur commented 1 week ago

In English a "You've been yelped!" card would be considered a compound, as the quoted phrase premodifies a noun representing the main meaning of the nominal (it is a kind of card).

Premodification cannot be the only factor here, though. Apparently English needs has constraints on it, but in Latin it comes mostly after, but could also come before. But it seems to me we are talking of the same exact phenomenon.

the king Pelayo: not appos?

Trying to reason about that: king cannot be appos because it precedes the name; and I do not think we want to have the name as appos (≅nmod). Admittedly, this stays a mined field.

Then why not appos for a "You've been yelped!" card ?

Anyway, I think that with respect to @martinpopel 's request, flat should probably be added to the allowed cases. While compound might be trickier: if it is indeed assimilated to the flat relation, then yes, but if one wants to speak of it as a (pre)modifier, it is the same old problem of underdefiniteness (why not nmod or acl then, since we distinguish them otherwise?) which I, right now, really do not want to start a discussion about again. Just pointing this fact out again.

nschneid commented 1 week ago

appos guidelines: "appos is intended to be used between two nominals. In general, modulo punctuation, the two halves of an apposition can be switched." This works (at least in English) for "the king Pelayo" ~ "Pelayo, the king". Either part of the apposition could stand on its own as a full nominal, which distinguishes it from nmod in general.

Stormur commented 1 week ago

appos guidelines: "appos is intended to be used between two nominals. In general, modulo punctuation, the two halves of an apposition can be switched." This works (at least in English) for "the king Pelayo" ~ "Pelayo, the king". Either part of the apposition could stand on its own as a full nominal, which distinguishes it from nmod in general.

Then here king should be the head and the name depend on it as appos?

There are some issues with the notion of swapping, because if, as it seems to imply, this means that either could equally be the head with no change in meaning (or better, reference?), why does this not lead us to the headless flat? Otherwise, "swapping" is too general and can be more or less always be applied between two nominals.

This is why I am reticent to use a hierarchical relation such as appos in this cases.

In any case, the problem for the original issue (*subj in a clause which does not take a clausal relation) stays whichever the hierarchical relation we choose to use (and is compound not hierarchical?). Whereas flat probably should not be a problem, as we would then refer to the true head of the whole combination, which counts now as a nominal phrase.

So in one of our Lati nexamples

iuxta illud (...) : «Omne regnum in se divisum desolabitur» 'beside that one [saying]: "Every divided kingdom will be forsaken"'

where illud 'that one' and the head of the saying desolabitur 'will be forsaken' (which has regnum 'kingdom' as subject) are joined by flat, we have actually to worry for the head of illud, which is attached with relation obl and thus "regular".

gossebouma commented 1 week ago

I started going over the Dutch data, and indeed it is a good heuristic for finding conversion/annotation errors! (grew ,atch query However, a major portion of the suspicious cases involve headless (free) relatives:

Wat vandaag gebeurde, was de ontlading van iets dat al dagen was opgekropt (What happened today, was the climax of something that was building up for days)

The underlying annotation marks these as wh-relatives (headless relatives) and the conversion treats them as regular nsubj subjects. One might think this is correct, as the phrase wat vandaag gebeurde can be paraphrased as the thing/event that happened today. Also, the distribution of headless relatives is identical to that of regular nominal phrases. Clausal subjects, on the other hand, are subordinate clauses like that this happened today or indirect questions like how this happened and not all verbs can take these as subjects.

However, while the documentation of csubj has a 'proper' subordinate clauses and an indirect question as examples, the documentation of complex clauses contains examples for csubj that are headless relatives: what she said makes sense. Also, English EWT has free relatives as csubj (approximate query) Finally, there is a discussion page on relative clauses that addresses free relatives. Unfortunately, no examples where they are subject, but the proposed analysis treats the relative pronoun as head, and assigns the external relation obj to them in cases such as you can eat what you want whih also suggests they are nominal and could be nsubj.

So should headless relatives be csubj (as the documentation implicitly suggests) or nsubj (as their semantics and distribution, as well as the discussion page, would suggest)?

_

Stormur commented 1 week ago

@gossebouma

However, a major portion of the suspicious cases involve headless (free) relatives:

I understand perfectly your concerns, since we also had quite some problems wrapping our head around these pesky "free relatives".

Since, at least from the current formal point of view, they should be csubj, ccomp and advcl, I can suggest you the compromise of the relcl subtype to distinguish them and at the same time making them groupable with classical acl:relcls.

Documentation here, here and here (mostly for latin, currently; please note that here advcl:relcl is unfortunately different from that of other treebanks, where it is meant to be a relative of a whole clause).

amir-zeldes commented 1 week ago

Also, English EWT has free relatives as csubj (approximate query) ... also suggests they are nominal and could be nsubj.

Yes, that is definitely the intended analysis in the English TBs. If there are exceptions to that in EWT then they must be conversion errors, and I see there are only 10 hits for your query there. You can also check in GUM which is manually annotated in UD, so it shouldn't have such cases - it looks like your query there produces non-free relative examples of various unusual constructions.

nschneid commented 1 week ago

This page mostly written in 2022–2023 and discussed by the Core Group has the most complete documentation of how we (in principle) handle relative constructions in English. (The typologically based documentation of relative clauses is still a work in progress.) For free relatives with a WH-pronoun, the head is the relative pronoun, and the predicate of the clause attaches to the relative word as acl:relcl. So the free relative would attach to the matrix clause as nsubj, not csubj. A clear example on the page is "What money we have left will go to charity." This is different from the Latin analysis shared by @Stormur.

Some caveats:

1) Sources say that WH-subjects are often ambiguous between a free relative interpretation and an interrogative clause interpretation, with negligible difference in meaning. So English annotators may disagree as to whether a particular phrase is a csubj headed by the predicate (interrogative interpretation), or a relative interpretation requiring nsubj.

2) @amir-zeldes above points to GUM, which has a strong preference for the free relative interpretation over the interrogative interpretation. In EWT the interrogative interpretation is preferred following a verb that licenses ccomps in other contexts. For subjects, I think the general preference is the free relative interpretation, but the new standard has not been consistently applied. A lot of relative clause analyses have been cleaned up but there are more on the TODO list (UniversalDependencies/UD_English-EWT#278). :)

nschneid commented 1 week ago

However, while the documentation of csubj has a 'proper' subordinate clauses and an indirect question as examples, the documentation of complex clauses contains examples for csubj that are headless relatives: what she said makes sense.

Thanks for pointing this out. I updated the examples to ones that cannot be construed as free relatives.

amir-zeldes commented 1 week ago

GUM, which has a strong preference for the free relative interpretation over the interrogative interpretation.

Yes, in the subj/obj position, our guidelines basically accept any case where the WH pronoun can be replaced by a split expression (e.g. "that which") as a free relative. We do accept the interrogative interpretation otherwise, for example with "wonder", compare:

For subject, "whether" clauses are csubj, while headless "what/who" follow the same guidelines:

And similarly for "what".

nschneid commented 1 week ago
  • who/nsubj comes/acl:relcl is my friend (the one who comes is my friend)

Noting that this is marginal/archaic with "who" (I don't see any valid matches in EWT or GUM). "Whoever" is preferable here.

For verbs like "know" with flexible complementation, I would treat the split expression test with some caution. "I know what horse you rode" (of all the horses, I can pinpoint the one you rode) probably doesn't correspond to "I know that horse which you rode" (meaning, I am personally acquainted with the horse). I'd go with ccomp for the former. ("Are you wondering what horse you rode, or do you know?")

amir-zeldes commented 1 week ago

Noting that this is marginal/archaic with "who"

Right, this is the "who steals my purse steals trash" type, but it works fine with "what", so I'd expect the same analysis:

"I know what horse you rode" (of all the horses, I can pinpoint the one you rode) probably doesn't correspond to "I know that horse which you rode"

That's a different construction, in which the WH pronoun is a determiner, rather than the head of the phrase, so it's not quite the same, and we can't directly apply the splitting test, though we can split it around "horse" with "I know that horse which you rode" as you pointed out, but also "I know the horse which you rode", which doesn't need to mean 'personally'. In any case, the "thing known" evaluates to an NP, not a clause, regardless of the sense of know, so I would go with a free relative analysis across the board. If there is some kind of difference here, I think it's subtle enough that I wouldn't expect annotators to make it reliably. This is different with "wonder", which has a simple test to rule out the relative analysis.

mr-martian commented 6 days ago

Looks like the PTNK data also has some dislocated that I'm not entirely sure what to do with:

ὁ    συγκρίνων   οὐκ  ἔστιν   αὐτό
the  interprets  not  is      he

"the one who can interpret, he is not (here)"

I currently have αὐτό as the predicate and συγκρίνων as dislocated:relcl.

amir-zeldes commented 6 days ago

@mr-martian I don't think there's dislocation here, and I'm not sure there's a relative clause (it's just a nominalized participle, no?). Either way I think the predicate should be ἔστιν, in the strong sense of 'exist', since we don't have an overt location expression or anything (if you want we can also think of it as a promoted copula with an unexpressed location). I'd do:

root(ἔστιν) nsubj(ἔστιν,συγκρίνων)

As for αὐτό, I don't think it means "he", since it's neuter - I think it's a back reference to Pharaoh's dream (this is Gen. 41, right?), which is neuter, so it's an object of the participle: "the one who can interpret it isn't (here)".

mr-martian commented 5 days ago

@amir-zeldes you're right - I completely misread that pronoun. Nominal participle covers one of the other two cases, leaving only Genesis 15:4

15  ἀλλ᾿    ἀλλά    CCONJ   _   _   21  cc  _   Gloss=but
16  ὃς  ὅς  PRON    _   Case=Nom|Gender=Masc|Number=Sing|PronType=Rel   17  nsubj   _   Gloss=who,which,that,what
17  ἐξελεύσεται ἐξέρχομαι   VERB    _   Aspect=Imp|Mood=Ind|Number=Sing|Person=3|Tense=Fut|VerbForm=Fin|Voice=Mid   21  dislocated:relcl    _   Gloss=to-come/go-out
18  ἐκ  ἐκ  ADP _   _   19  case    _   Gloss=out,out-of
19  σου σύ  PRON    _   Case=Gen|Number=Sing|Person=2|PronType=Prs  17  obl _   Gloss=you,your
20  οὗτος   οὗτος   PRON    _   Case=Nom|Gender=Masc|Number=Sing|PronType=Dem   21  nsubj   _   Gloss=this
21  κληρονομήσει    κληρονομέω  VERB    _   Aspect=Imp|Mood=Ind|Number=Sing|Person=3|Tense=Fut|VerbForm=Fin|Voice=Act   11  conj    _   Gloss=to-inherit
22  σε  σέ  PRON    _   Case=Acc|Number=Sing|Person=2|PronType=Prs  21  obj _   Gloss=you|SpaceAfter=No
23  .   .   PUNCT   _   _   22  punct   _   _
amir-zeldes commented 5 days ago

leaving only Genesis 15:4

Yeah, this example would be seen as a free relative in the English guidelines. If you want to use the free relative analysis, it would be:

dislocated:nsubj(κληρονομήσει,ὃς) acl:relcl(ὃς,ἐξελεύσεται)

Stormur commented 3 days ago

Yeah, this example would be seen as a free relative in the English guidelines. If you want to use the free relative analysis, it would be:

Let's remark that this is the way free relatives are done in English treebanks, not in general. And it is a rather questionable way.

That said, dislocated is indeed the choice for fronted, topicalised elements. Maybe even marking its role in the matrix with a subtype, i.e. dislocated:*subj (le aving aside the choice between nsubj and csubj).

Looks like the PTNK data also has some dislocated that I'm not entirely sure what to do with:

ὁ    συγκρίνων   οὐκ  ἔστιν   αὐτό
the  interprets  not  is      he

"the one who can interpret, he is not (here)"

I currently have αὐτό as the predicate and συγκρίνων as dislocated:relcl.

Independently on the reading of αὐτό, I would strongly suggest to keep ἔστιν as a functional dependent. But was this sentence part of the problems of the original post?

mr-martian commented 3 days ago 
ὁ    συγκρίνων   οὐκ  ἔστιν   αὐτό
the  interprets  not  is      he

"the one who can interpret, he is not (here)" I currently have αὐτό as the predicate and συγκρίνων as dislocated:relcl.

Independently on the reading of αὐτό, I would strongly suggest to keep ἔστιν as a functional dependent. But was this sentence part of the problems of the original post?

The original annotation of this sentence had nsubj(συγκρίνων, ὁ) and so was included in the search.

I'm not entirely sure what annotation you're suggesting as an alternative.

Stormur commented 3 days ago 
ὁ    συγκρίνων   οὐκ  ἔστιν   αὐτό
the  interprets  not  is      he

"the one who can interpret, he is not (here)" I currently have αὐτό as the predicate and συγκρίνων as dislocated:relcl.

Independently on the reading of αὐτό, I would strongly suggest to keep ἔστιν as a functional dependent. But was this sentence part of the problems of the original post?

The original annotation of this sentence had nsubj(συγκρίνων, ὁ) and so was included in the search.

I'm not entirely sure what annotation you're suggesting as an alternative.

If that συγκρίνων were fronted, to use dislocated. But if the sentence is "there is no interpret...", then have it head of copula ἔστιν. Not sure about the αὐτό.

amir-zeldes commented 3 days ago

I would strongly suggest to keep ἔστιν as a functional dependent

I don't really see another way of analyzing it except as the main verb - the relevant clause reads:

So word-for-word in English more or less:

Or a bit more freely to match English word order:

If there were a locative adverbial for ἔστιν (like "one interpreting it is not here"), I could see keeping the verb as a copula, with the adverbial as the head. But since there is no such adverbial, we are left with just the verb - whether you want to think of it as promotion, or simply acknowledge that "be" can have a strong sense meaning "exist", which is a VERB and not AUX, I don't see an alternative to making that word be the root of the clause.

If that συγκρίνων were fronted, to use dislocated. But if the sentence is "there is no interpret...", then have it head of copula ἔστιν.

Are you saying that ἔστιν should be a child of συγκρίνων? This doesn't make sense to me because συγκρίνων is the subject of ἔστιν, and not a predicate.

Stormur commented 3 days ago

I do not see why there should not be an "alternative". As a first thing, it is very strange to annotate, purely based on the contextual absence of a locative element, as a main verb a verb which is normally otherwise always a copula. The simplest explanation is that, possibly contrary to English, Italian, and others, Greek (and Latin, by the way) does not need such an element for an existential statement (by the way, I suppose that in a sentence like there is no interpret, there still is not the head despite being a locative element, right?).

The verb ἔστιν is really the same in all its occasion. Supposing it has a "stronger" meaning is just overinterpretaion, especially because there do exist words with a real existential meaning, e.g. ὑπάρχω. But ἔστιν no, it is simply the minimum, somewhat underdefined "tool" viable to the language in this case just to state that there is no interpret. The "negative interpret" is the centerpiece of this clause.

If that συγκρίνων were fronted, to use dislocated. But if the sentence is "there is no interpret...", then have it head of copula ἔστιν.

Are you saying that ἔστιν should be a child of συγκρίνων? This doesn't make sense to me because συγκρίνων is the subject of ἔστιν, and not a predicate.

Yes. This is a predication of the interpret being or not being there, not about him/her. Other languages would use slightly different constructions with some kind of expletive element, but it seems Greek is content with ἔστιν alone (word order might be a factor to be investigated, though). I don't think there is a real subject in such statements.

amir-zeldes commented 2 days ago

I don't think there is a real subject in such statements.

No, I'm sorry but that's simply not correct. συγκρίνων is in fact the subject of ἔστιν here, for several reasons:

  1. Agreement - the subject determines the number and person of the verb. If there were multiple interpreters, it would be συγκρίνοντες, etc. If there were a locative predicate like ἐνθάδε "here" and multiple interpreters, the verb would match the interpreters, and not the locative adverbial.
  2. The sense of the verse is to predicate about the subject, the interpreter, that he is not present. If "interpreter" is the predicate, that would mean the verse predicates of nothing that it is an interpreter, which makes no sense. In context, Pharaoh is looking for an interpreter and is dismayed that he is not there, so it makes sense that "interpreter" is the subject.
  3. In the original Hebrew, which the Septuagint translators used in making the translation, the interpreter is also the subject (פֹתֵ֖ר אֵ֣ין)
  4. In virtually any translation you look at the interpreter is the subject. For example, in Brenton's translation of the Septuagint, it reads: "And Pharao (sic) said to Joseph, I have seen a vision, and there is no one to interpret it". Again, "no one" is the subject of "is", and if you pluralize it, you would get subject agreement: "there are no people to interpret it"

it is very strange to annotate, purely based on the contextual absence of a locative element, as a main verb a verb which is normally otherwise always a copula

Those are the canonical UD guidelines for such situations, and in fact there are many situations in which no other choice exists. Consider an exchange like this:

In the second sentence, if we want to keep "I" as the subject (which again, it unambiguously is), then we have no choice but to promote what is otherwise definitely a copula to be the root - even if you don't believe in the distinction between a copular and existential sense of "be" in Greek.

Stormur commented 2 days ago

Admittedly there might still be something to investigate and define about these constructions, but I think it ultimately boils down to not forcing annotations on languages based on other languages requiring a slightly different set of elements.

First to sort out

  • Are you a teacher?

  • I am

Of course here I is subject and am is the promoted head of an ellipsis. We have a context and this is a clear case of ellipsis. By the way, English is a language which can do that, while another like Italian cannot, it has to put a pronominal element: Lo sono lit. 'it I-am' (subject is not explicit). Saying sono io lit. 'am-I I' would be different, and interestingly would be expressed as it's me in English, which brings us closer to the Greek sentence we have at hand... where no elements whatsoever speak of an ellipsis.

  1. Agreement - the subject determines the number and person of the verb. If there were multiple interpreters, it would be συγκρίνοντες, etc. If there were a locative predicate like ἐνθάδε "here" and multiple interpreters, the verb would match the interpreters, and not the locative adverbial.

This is an interesting fact. I see it as a case of "attraction": when there is just one element in the predication, the indexing on the verbal element tends to coincide. But on the other side, this is a third person, and as a possible "default" agreement it does not warrant too many conclusions (the number of the form of συγκρίνω is not a factor here).

Cross-lingusitically it is interesting to notice that existential constructions veer towards impersonality: in modern Greek we would always have a third-person singular έχει lit. '(it) has', in German es gibt lit. '(it) gives', in French il y a (also for plural elements), in Italian regionally and dialectally you observe a tendence to use c'è 'there is' even with plural referents, and so on. But Greek (and Latin) do not help themselves with expletive elements (it, es, y, ci, there) or a transitive construction, and make an agreement instead.

I again stress that word order might be relevant to distinguish between existential and other constructions. So the famous expression hic sunt leones in Latin means here, there are lions and not '(the) lions are here', which would probably be leones hic sunt / sunt hic. This to say that even if we have a locative adverbial, it might still not be the nonverbal predicate.

  • The sense of the verse is to predicate about the subject, the interpreter, that he is not present. If "interpreter" is the predicate, that would mean the verse predicates of nothing that it is an interpreter, which makes no sense. In context, Pharaoh is looking for an interpreter and is dismayed that he is not there, so it makes sense that "interpreter" is the subject.

  • In the original Hebrew, which the Septuagint translators used in making the translation, the interpreter is also the subject (פֹתֵ֖ר אֵ֣ין)

  • In virtually any translation you look at the interpreter is the subject. For example, in Brenton's translation of the Septuagint, it reads: "And Pharao (sic) said to Joseph, I have seen a vision, and there is no one to interpret it". Again, "no one" is the subject of "is", and if you pluralize it, you would get subject agreement: "there are no people to interpret it"

I do not think that any paraphrasable sense or translations can be used as an argument here. For example, one Italian version is ... e nessuno sa interpretarlo '... and no-one knows how to interpret it'. Hm. The predicate is totally different. So it just seems that that Greek version has chosen an existential construction, and this is what we have at hand. We are not interested in Biblical exegesis.

it is very strange to annotate, purely based on the contextual absence of a locative element, as a main verb a verb which is normally otherwise always a copula

Those are the canonical UD guidelines for such situations, and in fact there are many situations in which no other choice exists. Consider an exchange like this:

Not really. The guidelines do not give an ultimate choice for existential constructions. If we are specifically referring to this page, as I am assuming, nothing is really said about them. It actually seems to speak in favour of keeping a cop analysis. Some examples are shown, but no explanation is given. Just staying with English, a statement like "In languages with fixed SVO order (like English), the final nominal is the predicate and the first nominal is the subject" all speaks in favour of considering food as the head in there is food . The current opposite practice seems arbitrary, or at least not coherent with the rest. For Russian, it even seems the guidelines try to justify this non-coherent annotation: "Yes, it is indeed the same thing, but... we will just annotate it differently because it is existential".

amir-zeldes commented 20 hours ago

I do not think that any paraphrasable sense or translations can be used as an argument here

For other texts, sure, but in this case I disagree - the Septuagint translators tried pretty hard to stay faithful to the Hebrew text (at times creating very unnatural calques in Greek), and they certainly knew Biblical Hebrew and Greek better than any of us, so if something is ambiguous between a reading that mirrors the original and something that does not, I think assuming parity with the source material is a good tie breaker. Incidentally, your Italian example also has a phrase representing the interpreting person as the subject, though the distribution of the lexical material and negation is different of course ('interpretation' as a lexeme is realized in the verb, and negation is part of the subject NP, not a verbal negator).

So just to be clear, am I right to say my position is 1 and yours is 2?

  1. I think this clause is predicating about an interpreter (realized as a participle of 'interpret'), which is the subject, that that person is not present (no predicate other than "is", which is negated)
  2. You think this clause is predicating about an empty subject that it is (copula) not an interpreter (predicate noun)

Is that right? If so, then I don't think I have any other arguments to add, and maybe other people who know Ancient Greek can tell us which one they think it right.