Copulas, again: Are there clear-cut criteria for using either `cop`+`root` or `root`+`xcomp`?

[chiarcos]

This is a general, cross-linguistic question, but my concrete problem is about the treatment of Old Saxon werđan "to become" as either a future-tense copula or as a lexical verb with clausal complement. It has a different lemma than the present tense copula, but that does not speak against its interpretation as future-tense copula, because present-tense copula also uses multiple lemmas (wesan and sin "to be"). Linguistically, this situation is not specific to Old Saxon, but similar to that of modern West Germanic languages.

Old Saxon werđan is effectively used in the same way as German werden and English will, but with UDpipe v2.5 models, we get different analyses for German and English:

Das Wetter   wird    schlecht.
the wheather will.be bad      
             root    xcomp

Das Wetter   wird schlecht sein.
the wheather will bad      be
             root xcomp   (cop)

The wheather will be  bad.
             aux  cop root

Overall, it seems that in the spectrum between a semantically empty copula and lexical verbs and that different languages put the border between both very differently. Is there any general recommendation? (I am also asking because Old Saxon is as far away from Modern English as from Modern German, so the analysis should actually be equivalent across all three languages.)

[chiarcos]

Some intermediate results of my own research: I imagine we operate on a spectrum ranging from a semantically empty copula to a full lexical verb, and on this spectrum I am looking for morphosyntactic criteria to disentangle both cases (not semantic criteria). The languages I am thinking of are primarily Slavic and Germanic, but the question is for a cross-linguistically applicable criterion (in theory, at least).

I assume there is some systemacity with respect to categories such as:

1. empty copula (predicate is root)
2. optional copula (predicate is root)
3. obligatory copula with default TAM (= present indicative imperfect), default polarity (= non-negated), default epistemicity (= factual), etc.
4. obligatory copula with non-default TAM (= non-present or non-indicative, etc.) and default polarity/epistemicity
5. obligatory copula with non-default polarity (= negated) and default epistemicity
6. obligatory copula with non-default epistemicity (e.g., hearsay)
7. raising verb with clausal complement
8. full lexical verb

I checked three Germanic and three Slavic languages for (variants of) these (see attached csv file for details), and ran UDpipe over them. I used the old v2.5 models, but I don't think the situation drastically changed.

	German	Swedish	English	Russian	Czech	Bulgarian
no copula				root
copula in default TAM	cop+root	cop+root	cop+root		cop+root	cop+root
copula in past	cop+root	cop+root	cop+root	cop+root	cop+root	cop+root
copula in future (single word)	root+xcomp			cop+root	cop+root
copula in future (multi-word)	root+xcomp	root+xcomp	cop+root			cop+root
copula in medium modality (can)	cop+root	root+xcomp	cop+root	root+obl	root+xcomp	root+xcomp
negated copula (single-word)				root+advmod	cop+root
negated copula (multi-word)	cop+root	cop+root	cop+root			cop+root
hearsay	cop+root	cop+root	root+xcomp	root+xcomp	root+xcomp	root+csubj:pass
raising verb	root+advmod	root+xcomp	root+xcomp	root+obl	root+obj	root+obj

(The table is oversimplifying and may be specific to the example sentence used for illustration. See attached csv file for details. Translations were created with Google translate from German. I am confident about the English and Russian translations; Czech, Swedish and Bulgarian look plausible to me, but please feel free to correct them, me and this table.)

Overall, it seems that the degree to which the cop analysis extends into constructions with obligatory copula varies considerably across languages and does not seem to follow a very consistent pattern. There seems to be some consensus on tensed copula annotated as cop, but this is with the exception of the analytic future in German and Swedish. There is no consistent pattern for modality (the historical link between modals and the analytic future may have led to the xcomp analysis of the future tense). Negation has no effect if the negative particle is a word formally distinct from the copula, but negated copulas (as in Russian and Czech) are analysed differently in different languages. In fact, this also occurs in Old Saxon.
Raising verbs generally follow the root+xcomp pattern, with variations (which might need to be corrected towards xcomp, depending on how its cross-linguistic definition (see issue #1200) is improved and enforced).

Any suggestions how to systematize that cross-linguistically?

[jnivre]

Verbs like English "become" and its synonyms in other languages (like Swedish "bli") are not regarded as copulas, not because they are future (they can obviously occur in any tense) but because they have an inchoative change-of-state meaning. This is different from a periphrastic form like "will be", where the nonfinite verb is the ordinary copula "be" and where "will" is a future auxiliary.

I am therefore curious to know what you refer to when you say that Swedish uses root+xcomp for "copula in future (multiword)" and "copula in medium modality (can)". If what you have in mind is "kommer att vara" and "kan vara", respectively, then this is incorrect, since these are both given a cop+root analysis in Swedish, with the respective auxiliary ("kommer", "kan") attached as aux to the root.

[amir-zeldes]

Yes, I think I'm with @jnivre regarding the inchoative change of state criterion - I looked at the CSV, and I wouldn't say that Russian "Погода будет плохая" is equivalent to "Das Wetter wird schlecht" - it's more like "Das Wetter wird schlecht sein". The German version without sein implies a change of state, while the Russian version is a continuous imperfective future. I think a better translation into Russian for "das Wetter wird schlecht" without sein is something like "погода становится плохой", which would indeed have a full verb, and I would analyze the complement as xcomp.

[chiarcos]

Thank you. I can agree on the inchoative change of state criterion (and apologies for my mistakes in this table -- this was just to get an idea for myself, hence the request for corrections). Although it probably applies to Old Saxon skulan "shall" (used in Heliand but apparently also used for indicating analytic future tense in some modern dialects), the problem is that this is not an ideal diagnostic for a corpus language, as the differentiation can be interpretative. It is clear that the later grammaticalized future tense developed out of earlier inchoatives, and there are examples that can be read like this, but often, either a reading as inchoative or TAM marking auxiliary are possible, e.g.,

       Thô  uuarð   thar  an thene gastseli   megincraft  mikil manno gesamnod
       then will.be there in the   guest.hall crowd/power great men   gathered
 (1.a) "Then a great power, men gathered, will be there in the guest hall."
 (1.b) "Then a great power[Sg], men[Pl], will be gathered there in the guest hall."

For the inchoative interpretation (1.a), you read manno gesamnod as an apposition to megincraft mikil, with uuarð roughly meaning "to appear in". In the reading as TAM auxiliary (1.b), you read only manno as an apposition to megincraft mikil, with uuarð being an auxiliary to gesamnod. The latter is the reading of Köne (1855: "Da ward dar in den Gastsaal eine große Machtkraft von Mannen versammelt"). This still is inchoative, but that's from the semantics of gesamnod, not necessarily uuarð. But there is another use of werđan which does not involve a change of state, and that is in the formation of passives. For this reason, I am not convinced that the inchoative reading of werđan can be taken for granted. This is why I was asking for morphosyntactic criteria. Of course, the simplest criterion would be to decide on grounds of the lemma, but this feels a bit ad hoc.

As for morphosyntactic criteria, there never was a synthetic future tense in Germanic, but all languages developed analytic tenses. I'm not sure the timing of this process has ever been studied for Low German (neither Modern nor Middle Low German), but Old English and Old Saxon were closely related dialects, and I think for English, it is indeed assumed that the grammaticalization of will occurred later (and contributed to the gradual disappearance of weorþan, the OE cognate of werđan). An analysis more in line with the OE data is thus to extend the xcomp analysis to werđan. But this is very indirect reasoning.

[chiarcos]

There is a relevant recent paper by Jäger (2024) on the origins of the German future tense with werden. She identifies the dominant usage of Old High German werdan (658/1701) and Old Saxon werđan (167/422) as "Kopula mit prädikativer NP, AP etc.", with the following illustrative examples (both Old High German):

    Sidh  uuarth her guot man
    since became he  good man
    "since then, he became a good man"

    thera bruoder Lazarus uuard  cumig
    their brother Lazarus became sick
    "their brother Lazarus got sick"

Apparently, her notion of "copula" includes cases with inchoative reading. Likewise, Eide (2020) refers to (Norwegian) bli "become" as "inchoative copula". If neither of these are within the UD definition of cop, it should be made more explicit in the documentation. At the moment, we read "pure copulas that add at most TAME categories", but isn't inchoative aspect covered by that?

Jäger, Agnes. "Die Entstehung des deutschen werden+Infinitiv-Futurs – ein alternatives Szenario" Beiträge zur Geschichte der deutschen Sprache und Literatur, vol. 146, no. 2, 2024, pp. 181-236. https://doi.org/10.1515/bgsl-2024-0015

Eide KM. Tense and Aspect in Germanic Languages. In: Putnam MT, Page BR, eds. The Cambridge Handbook of Germanic Linguistics. Cambridge Handbooks in Language and Linguistics. Cambridge University Press; 2020:591-614.

[jnivre]

The rule of thumb is that there should normally be at most one copula verb per language, a completely neutral verb that may correspond to zero or a nonverbal copula in other languages.

[chiarcos]

I think the solution to this is to follow the definitions under https://universaldependencies.org/v2/copula.html#guidelines-for-udv2, but not the definition under https://universaldependencies.org/u/dep/cop.html, i.e., update https://universaldependencies.org/u/dep/cop.html to match https://universaldependencies.org/v2/copula.html#guidelines-for-udv2 as follows:

Replace the following sentence

The cop relation should only be used for pure copulas that add at most TAME categories to the meaning of the predicate, which means that most languages have at most one copula, and only when the nonverbal predicate is treated as the head of the clause.

by

The cop relation is restricted to function words (verbal or nonverbal) whose sole function is to link a nonverbal predicate to its subject and which do not add any meaning other than grammaticalized TAME categories (only one word in most languages).

This solves the problem in the way you're suggesting, as inchoative is not a grammaticalized aspect here.

However, then, the negated copulas may need to be reconsidered, because they add meaning other than grammaticalized TAME categories.

[jnivre]

Thanks for your suggestion. I will take it to the core guidelines group.

As for negation, maybe "TAME" should be changed to "TAMP" (or "TAMPE") to cover polarity as well.

[amir-zeldes]

the negated copulas may need to be reconsidered, because they add meaning other than grammaticalized TAME categories

I think this is mainly a tokenization issue. In languages where there is a tradition of tokenizing negation-containing copula-like strings, such as English "isn't", the copula is generally taken to be the positive form and the negation is just added on. By contrast, languages that have a negative copula that is less obviously tokenizable into two parts are treated as cop since they stand in paradigmatic alternation with the positive copula, and it seems more important to indicated the copular relation than the negative adverbial one (and the negation can also be expressed using a Polarity feature). If we don't want to force an unconventional tokenization on such languages, I think we should continue to allow negative copulas as cop.

[chiarcos]

the negated copulas may need to be reconsidered, because they add meaning other than grammaticalized TAME categories

I think this is mainly a tokenization issue. ....

For Germanic and Slavic, probably. But in some languages, the negative copula consists of a stem (expressing negation) with verbal inflection. So, there is no distinct non-negative copula that this is cliticized with. Not 100% sure because of my lack of background in that language, but that seems to be the case for Turkish değil, annotated as cop in the IMST corpus. (I searched for [lemma = "değil" & deprel = "cop.*"] over UD 2.16.)

(I am not suggesting to change the practice, just trying to understand it ... but also to clarify the documentation along the way. Here, we might just need to say a word about polarity, as suggested by @jnivre. As for the handling of inchoatives, there indeed is the line "In contrast, the equivalents of to become are not copulas despite the fact that traditional grammar may label them as such." in the documentation, but it is buried in a chunk of otherwise unrelated text, and, even worse, the departure from traditional grammar is definitorial information stuck in a paragraph starting with As a concrete example. This could be put out of this paragraph and into a list, instead, to make it more findable.)

[amir-zeldes]

Yes, exactly - that's what I meant by "If we don't want to force an unconventional tokenization on such languages". I think Turkish treebank users would be very surprised if we did something like trying to decompose değil into a plain positive copula with zero surface realization and a separate negation:

# text = Ali öğrenci değil
# text_en = Ali is not a student
1	Ali	ali	PROPN	_	Case=Nom|Number=Sing	2	nsubj	_	_
2	öğrenci	öğrenci	NOUN	_	Case=Nom|Number=Sing|Person=3	0	root	_	_
3-4	değil	_	_	_	_	_	_	_	_
3	değil	değil	AUX	_	Polarity=Neg	2	advmod	_	_
4	dir	dir	AUX	_	_	2	cop	_	_

So if we want to keep inherently negative items like "değil" as copulas and not use a strange solution like splitting every negative copula into a copula and a negation, even in cases where they are not morphologically decomposable, then we have to allow cop with Polarity=Neg, as sanctioned by the current guidelines.

[chiarcos]

I don't think the alternative to negative copula in the case of değil is to split it. The alternative is to not to annotate that as copula at all, but to use the xcomp style of annotation as also applied to inchoative copula. In this case, this is also well sanctioned by the "only one word in most languages" clause, because the non-negated copula is (zero or) imek, and I don't think this has any direct relation with değil.

# text = Ali öğrenci değil
# text_en = Ali is not a student
1	Ali	ali	PROPN	_	Case=Nom|Number=Sing	3	nsubj	_	_
2	öğrenci	öğrenci	NOUN	_	Case=Nom|Number=Sing|Person=3	3	xcomp	_	_
3	değil	değil	AUX	_	Polarity=Neg	0	root	_	_

I'm not actually arguing for this kind of representation, I'm just noting that the different treatment of "semantically non-empty copulas" feels rather ad hoc. In principle, that's ok with me, too, but if that really is where UD should be going, the treatment of negated copula deserves an explicit mention under https://universaldependencies.org/u/dep/cop.html. At the moment, the treatment of negative copula varies wildly.

• This is even the case within Turkish, where we frequently find annotation of değil as aux, also for cases I would consider to be copular clauses (e.g., https://lindat.mff.cuni.cz/services/teitok/ud216/index.php?action=deptree&cid=tr/pud/tr-pud-test.xml&jmp=w-13294), but actually, in some corpora, a significant portion is root, hinting towards a xcomp-style analysis (numbers over https://lindat.mff.cuni.cz/services/teitok/ud216):

  • Kenet aux 95% (214/222), no cop
  • Penn aux 94% (237/251), no cop
  • TueCL aux 100% (13/13), no cop
  • Atis aux 100% (3/3), no cop
  • GB aux 89% (33/37), no cop
  • BOUN aux 50% (79/157), cop 29%, root 10%
  • PUD aux 56% (13/23), cop 17%
  • IMST cop 45% (45/100), root 25%

• Just for comparison: Old Church Slavonic (TOROT) and Russian (SynTagRus) seem to systematically annotate (verbal) нѣтъ and нет as lexical verbs, but never as copula.