One and the other construction in copula: singular subjects, plural copula, more than one subject?

[Stormur]

I don't know if the following problem has already surfaced, but I was not able to locate it here or in the guidelines, as it seems to me of a mixed nature, and I don't actually know how to aptly name it; I don't think it to be an instance of clefting.

The starting point is this Latin sentence (which I am slightly modifying and simplifying in its components to highlight its structure) which perplexed as to how it is best annotated:

libros habent, qui libri sunt alter
book.MASC-PL-ACC have that.MASC-PL-NOM book.MASC-PL-NOM are other.MASC-SING-NOM

bonus et alter malus
good.MASC-SING-NOM and other.MASC-SING-NOM bad.MASC-SING-NOM

i.e., quite literally, 'they have (some) books, which are books one good and the other bad', which I don't know how grammatical it is in English, but something very similar can happen at least in Italian, too.

Now, my first consideration is that there are too many elements. It seems we migh split it into two sentences:

qui libri sunt: alter (est) bonus et alter (est) malus
'that [they] are books: one is good and the other is bad'

Zero copula is allowed in Latin. One sure thing should be that qui is (part of) nsubj, and sunt cop, probably depending on libri, but how to join the other pieces? parataxis does not look as a good choice, as this isn't just a juxtaposition.

1. By the way, the problem would be solved if we had a relative pronoun as a connector, e.g. quorum 'of which [masculine plural genitive]':

libros habent, qui libri sunt quorum alter bonus et alter malus 'they have some books, which are books of which one is good and the other is bad'

Here, I see this analysis to be quite unproblematic:

1	Libros	_	NOUN	_	_	2	obj	_	_
2	habent	_	VERB	_	_	0	root	_	_
3	,	_	PUNCT	_	_	5	punct	_	_
4	qui	_	PRON	_	_	5	nsubj	_	_
5	libri	_	NOUN	_	_	1	acl:relcl	_	_
6	sunt	_	AUX	_	_	5	cop	_	_
7	quorum	_	PRON	_	_	8	nmod	_	_
8	alter	_	DET	_	_	9	nsubj	_	_
9	bonus	_	ADJ	_	_	5	acl:relcl	_	_
10	et	_	CCONJ	_	_	12	cc	_	_
11	alter	_	DET	_	_	12	nsubj	_	_
12	malus	_	ADJ	_	_	9	conj	_	_
13	.	_	PUNCT	_	_	2	punct	_	_

That is, we have the two copulae connected by means of an acl:relcl relation, each with its own subject: libri and alter/alter.

2. But without, I am not so sure about keeping acl, which does not seem justified by anything.

1	Libros	_	NOUN	_	_	2	obj	_	_
2	habent	_	VERB	_	_	0	root	_	_
3	,	_	PUNCT	_	_	5	punct	_	_
4	qui	_	PRON	_	_	5	nsubj	_	_
5	libri	_	NOUN	_	_	1	acl:relcl	_	_
6	sunt	_	AUX	_	_	5	cop	_	_
7	alter	_	DET	_	_	8	nsubj	_	_
8	bonus	_	ADJ	_	_	5	?	_	_
9	et	_	CCONJ	_	_	11	cc	_	_
10	alter	_	DET	_	_	11	nsubj	_	_
11	malus	_	ADJ	_	_	8	conj	_	_
12	.	_	PUNCT	_	_	2	punct	_	_

I was thinking rather of a very awkward ccomp, in vein with the treatment of copulae with a clause as their "nominal" part. One comparative consideration that contributed to this was that, at least in very colloquial Italian (not every speaker would agree on its grammaticality), it is possible to formulate this sentence with the universal complementiser che 'that' (not clear if it should be seen as a relative element): questi libri sono libri che uno è bello e l'altro brutto lit. 'these books are books that one is good and the other bad'. xcomp has to be excluded, since subjects can be different.

3. Another solution still I have thought of:

1	Libros	_	NOUN	_	_	2	obj	_	_
2	habent	_	VERB	_	_	0	root	_	_
3	,	_	PUNCT	_	_	5	punct	_	_
4	qui	_	DET	_	PronType=Rel	5	det	_	_
5	libri	_	NOUN	_	_	8	nsubj	_	_
6	sunt	_	AUX	_	_	8	cop	_	_
7	alter	_	DET	_	_	8	xcomp	_	_
8	bonus	_	ADJ	_	_	1	acl:relcl	_	_
9	et	_	CCONJ	_	_	11	cc	_	_
10	alter	_	DET	_	_	11	xcomp	_	_
11	malus	_	ADJ	_	_	8	conj	_	_
12	.	_	PUNCT	_	_	2	punct	_	_

That is, now the interpretation is: 'they have books, which books are one good and the other bad'. qui would act here as a relative determiner: the major problem is that this syntax of qui (the repetition of the antecedent) is very rare, even if attested. But then: would be the xcomp for the alter's sensible?

---

So, out of the three possibilities, my formal preference goes to the 3), as it is much cleaner, even if it presents a quite unusual syntax of qui (as it blurs the line between its use as an indefinite determiner and a relative pronoun), but I am not sure it represents the right interpretation.

Has anybody come across similar cases, or are there some indications on how to treat them buried in the guidelines? Thanks for any help and sorry for the lengthy post!

---

Still another variant, more similar to 3) :

1	Libros	_	NOUN	_	_	2	obj	_	_
2	habent	_	VERB	_	_	0	root	_	_
3	,	_	PUNCT	_	_	5	punct	_	_
4	qui	_	DET	_	PronType=Rel	5	det	_	_
5	libri	_	NOUN	_	_	7	nsubj	_	_
6	sunt	_	AUX	_	_	7	cop	_	_
7	alter	_	DET	_	_	1	acl:relcl	_	_
8	bonus	_	ADJ	_	_	7	amod	_	_
9	et	_	CCONJ	_	_	10	cc	_	_
10	alter	_	DET	_	_	7	conj	_	_
11	malus	_	ADJ	_	_	10	amod	_	_
12	.	_	PUNCT	_	_	2	punct	_	_

Here the alter's are seen as the predications, modified by different amods in the two conjuncts. The repeated element is the head, and this allows to preserve symmetry even if the adjective is replaced with a more complex element, like obls domi 'in the house' / ruri 'in the countryside' (i.e. 'two books, one at home and the other in the countryside')

[sylvainkahane]

French has similar constructions. I think that two unusual constructions are involved.

First, the possibility to have the noun with the relative pronoun (the pronoun lequel (plural lesquels) is very formal, in both examples):

_J'ai des livres, lesquels sont en bon état._
I have books, which are in good shape.

_J'ai des livres, lesquels livres sont en bon état._
I have books, which books are in good shape.

Second, the coordination of predicative adjectives with determiners:

_Certains livres sont abimés._
Some books are damaged.

_D'autres livres sont en bon état._
Other books are in good shape.

=> _Les livres sont certains abimés et d'autres en bon état._
The  books are some damaged et other in good shape.

Both constructions can be combined:

_J'ai des livres, lesquels livres sont certains abimés et d'autres en bon état._
I have book, which books are some damaged and other in good shape.

Both construction are complex and I'm not sure how to annotate them. Your solution 3 seems the best one, except for the determiner alter, which looks like a floating quantifier (#581). I'm not sure to know what solution has been adopted, but not xcomp. Probably advmod.

[Stormur]

Hm, about the solution with advmod, I have to say that I would find it undesirable, sic stantibus rebus (as things stand 😃) in UD, since it would mean to annotate alter as an ADV, with all consequences regarding its morphological traits and its concordance with liber, here. This actually ties in into a bigger problem wortthy of its own issue, namely to allow advmod deprel for non-ADV, but I am digressing.

[Stormur]

Regarding the discussion about floating quantifiers, as I have just commented there, I think that the right interpretation is xcomp; I feel that advmod would distort the syntax of the sentence on too many levels.

I have thought that these constructions might be probably simply seen as an instance of nominal ellipsis: what is left out is a repetition of the noun that already appears as a subject.

Taking your French example: we can see that we have both certains livres abimés 'some damaged books' and autres livres en bon état 'other books in a good state'. So, instead of saying:

J'ai des livres, lesquels livres sont certains livres abimés et d'autres (des?) livres en bon état
'I have some books, which books are some damaged books and other books in good shape'

, we just omit the livres following the first one.

If this is case, we then just promote according to the hierarchy amod > nummod > det > nmod > case. keeping the same dependencies. So (hoping to get everything right in French):

det(livres,lequels) , nsubj(abimés,livres)
cop(abimés,sont) , det(abimés,certains)
conj(abimés,autres) , nmod(autres,en bon état)

What is important is that both noun phrases stays co-ordinated and keep their internal relations.

By the way, might this be a case to move nmod up in the hierarchy?

[sylvainkahane]

Does alter work as a pronoun? Because for French, both certains 'some' and d'autres 'other(s)' work both as DET and PRON and in this construction they are pronouns. It is impossible to have what you propose (*J'ai des livres, lesquels livres sont certains livres abimés et d'autres livres en bon état.) with a noun livres after certains or d'autres. Exactly as in floating quatifiers :

_Mes livres sont tous abimés._
My  books are all damaged.
_*Mes livres sont tous les livres abimés._

Floating quantifiers are modifiers and they don't involve ellipsis. In SUD, I will just say that there are mod: examples. This mod relation is automatically converted in obl:mod in UD (examples), but we never really discussed this choice, because we don't know what is UD's solution.

[Stormur]

Does alter work as a pronoun? Because for French, both certains 'some' and d'autres 'other(s)' work both as DET and PRON and in this construction they are pronouns. It is impossible to have what you propose (*J'ai des livres, lesquels livres sont certains livres abimés et d'autres livres en bon état.) with a noun livres after certains or d'autres. Exactly as in floating quatifiers :

_Mes livres sont tous abimés._
My  books are all damaged.
_*Mes livres sont tous les livres abimés._

Am I wrong or can you generally say certains livres (abimés) 'some (damaged) books'? In any case, with alter 'other' in Latin you can, and it is a determiner. But then, every determiner DET, as a nominal, can stand alone and act as a "pronoun", however it always stays a DET (the same way that an ADJ used without a nominal head does not become a PRON or a NOUN). You might say:

Hic liber est alter liber bonus 'this book is another good book'

But an interpretation of alter as a "floating quantifier", or rather as a secondary predication, meaning something like 'this book, as the "other" of more than one book, is good', might still stand. alter might be replaced by totus 'every/all': 'the book, in its entirety, is good'. Maybe copulae have some restrictions regarding secondary predications, but one could also replace alter with apertus 'opened' and mean: 'this book, when it is open/considered as opened/..., is good', in contrast with 'the opened book is good', and so on. In Latin every secondary predication can have such ambiguity. The limit seems to be semantic/pragmatic (for example, absence/presence of contrast).

However, "italianizing" your example, this ambiguity vanishes, as we can have:

• I miei libri sono tutti (libri) danneggiati 'My books are all damaged (books)' - no article, "floating quantifier" - libri 'books' can be omitted.
• I miei libri sono tutti [i libri/quelli] danneggiati 'My books are all the damaged ones/books' - with article OR pronominal form of quei 'those (pl. masc.)' - tutti 'all (pl. masc.)' determines the phrase. The repetition of libri is pedantic, but still sounds well.

Now, with certi 'some (pl. masc.)', which should be the equivalent of fr. certains, it functions indeed differently. One difference is that certi never admits the article as tutti, so there is only one probable reading:

• I miei libri sono certi (libri) danneggiati... 'My books are some damaged ones/(books)...' - leaving out libri sounds awkward and very conditioned by the context ('some damaged things'...), at the limit of what is acceptable. But:
• i miei libri sono certi danneggiati e certi intatti 'My books are some damaged and some untouched' - Here the contrast makes the sentence good and at the same time clear that it is used as a "floating quantifier". And also:
• i miei libri sono certi libri danneggiati e certi libri intatti is OK, the repetition gives emphasis.There is again ambiguity: libri might be "assigned" to certi or to the adjective: 'some books damaged and some books untouched' vs. 'some (are) damaged books and some (are) untouched books'. Also certi as a determiner of the whole phrase might be possible, but very awkward, sounding somehow cryptic (which books are alluded to?).

---

Now, what I think this all says? That some interpretations are preferred over others and that a language might or might not present ways to disambiguate between them (in the Italian case with certi ambiguity is given by the absence of the article, as in Latin). Latin does not. So, in this case alter might work either ways, but in the end I think the preferred interpretation is that of a "floating quantifier", or secondary predications, as it is the only one creating the contrast (also etymologically, alter has an innate contrastive meaning). Contrast is symmetric, while nominal ellipsis is not (at least in UD).

Also, from a semantic point of view, libros habent, qui libri sunt alter bonus et alter malus and libros habent, qui libri sunt bonus et malus are nearly identical, the former highlighting contrast and the latter not.

[dan-zeman]

I think bonus/malus is something like a secondary predication about libri. Probably optional depictive, following the terminology laid out in the guidelines. That would make them acl of libri, and @Stormur's first option would apply, only I don't think they are necessarily acl:relcl:

1	Libros	_	NOUN	_	_	2	obj	_	_
2	habent	_	VERB	_	_	0	root	_	_
3	,	_	PUNCT	_	_	5	punct	_	_
4	qui	_	PRON	_	_	5	nsubj	_	_
5	libri	_	NOUN	_	_	1	acl:relcl	_	_
6	sunt	_	AUX	_	_	5	cop	_	_
7	quorum	_	PRON	_	_	8	nmod	_	_
8	alter	_	DET	_	_	9	nsubj	_	_
9	bonus	_	ADJ	_	_	5	acl	_	_
10	et	_	CCONJ	_	_	12	cc	_	_
11	alter	_	DET	_	_	12	nsubj	_	_
12	malus	_	ADJ	_	_	9	conj	_	_
13	.	_	PUNCT	_	_	2	punct	_	_

There are two points that make this example different from the usual cases of optional depictives though. First, the superordinate clause is copular and the modified nominal is its predicate rather than argument. And second, the "secondary" predicate gets its own subject, alter. Still I think I prefer this analysis. One could say that the three clauses with nonverbal predicates are at the same level as conjuncts or parataxis members (all would be relative clauses modifying libros) but that does not make much sense because one would not utter it without the bonus/malus part: ?Libros habent, qui libri sunt. I also don't like the xcomp solution (@Stormur's no. 3) because I think alter represents the subject of bonus rather than another predication about an invisible subject of bonus.

[Stormur]

So, if I understood well, you would prefer the analysis where the alter bonus/alter malus part depends on libri by means of a non-relative (because there would be no relative element quorum) acl, assuming a zero-copula there?

So:

1	Libros	_	NOUN	_	_	2	obj	_	_
2	habent	_	VERB	_	_	0	root	_	_
3	,	_	PUNCT	_	_	5	punct	_	_
4	qui	_	PRON	_	_	5	nsubj	_	_
5	libri	_	NOUN	_	_	1	acl:relcl	_	_
6	sunt	_	AUX	_	_	5	cop	_	_
7	alter	_	DET	_	_	9	nsubj	_	_
8	bonus	_	ADJ	_	_	5	acl	_	_
9	et	_	CCONJ	_	_	12	cc	_	_
10	alter	_	DET	_	_	12	nsubj	_	_
11	malus	_	ADJ	_	_	9	conj	_	_
12	.	_	PUNCT	_	_	2	punct	_	_

Anyway, in solution no. 3 the subject of alter would be visibile in libri, and the mismatch in Number given by the split of a plural subject over two different predications.

The problem now is if the treatment above should be different from another sentence with nearly identical meaning and very similar structure:

Libros habent, qui sunt alter bonus, alter malus, alter magnus, alter parvus, alter...

This is what led me to consider alter as a depictive, rather than bonus. Here it looks like cop(bonus,sunt) is the predicate with qui as subject, and all the others are linked by means of conj: conj(bonus,malus) conj(bonus,magnus) conj(bonus,parvus) conj(bonus,...) So I thought it should be possible to "fit" libri into this structure without radically altering it (i.e., libri replacing bonus as the predicate and shifting everything else to acl). Or, in such a case, should we assume that the predicate of sunt is itself a clause (alter bonus...)?

[dan-zeman]

I tried to find a similar case in Czech. I found one, and I don't completely agree with the analysis that my conversion script produced there (http://hdl.handle.net/11346/PMLTQ-RAAY). The equivalent of alter bonus is attached as an appos of the equivalent of libri. We shouldn't do that in UD because we restrict apposition to nominals (unlike in the Prague-style annotation, where it is occasionally used with clauses). But maybe it is not so bad to use conj instead in UD after all?

conj(libri, bonus); conj(bonus, malus); nsubj(bonus, alter) ...

[Stormur]

I have come again upon this construction, and I think that the solution might be quite simpler than thought. I also think that the copular construction is the most peculiar one of this kind, and this might have misled us.

So, the basic observation is that we have a block consisting of two or more, usually elliptical, co-ordinated clauses that has to be "hooked" to the main clause in some way. I would distinguish two main cases:

1. this block "expands" an element that is already present in the main clause, or
2. the first clause of the block coincides with the main clause.

So, based on the examples I used:

• lat. Habeo duos libros, alterum domi et alterum ruri / it. Ho due libri, uno a casa e l'altro in campagna 'I have two books, one at home and the other in the country', we have the first case;
• lat. Illi libri sunt alter domi et alter ruri / it. Quei libri sono uno a casa e l'altro in campagna 'Those books are one at home and the other in the country', we have the second case.

The first case is simple: it is not dissimilar from an id est clause of explanatory nature, which is nothing else than a particular type of co-ordination and for which we are using the subtyped relation conj:expl. So I would annotate:

conj:expl(habeo,alterum) orphan(alterum,domi) conj(domi,ruri)

as this is like saying I have two books; I have one at home and I have the other in the country. And the same goes for other sentences like They went in (two) different directions, Ann eastwards and Bob westwards, and also These are the books, one good and one bad: conj:expl(books,good) nsubj(good,one) conj(good,bad).
In the case of an attached copular clause, I think the best bet would be to simply keep conj:expl for symmetry (instead e.g. of acl) and to link it to the corrisponding noun (nominal clause <-> nominal element): I have two books, one good and one bad -> conj:expl(books,good) nsubj(good,one) conj(good,bad) (here I have to admit that I don't know how Latin would treat this with cases and thus underlying structures: alterum bonum et alterum malum or alter bonus et alter malus, or...? Is it allowed?).

The second case is trickier because of the role of the "contrasting" elements (one/other/...), but still follows the same logic.

(a) They went one eastwards and one westwards

has the same basic structure as (b) They went eastwards and westwards (ambiguous co-ordination) and/or (c) One went eastwards and the other westwards (no "common" subject expressed): it is a fusion of the two models, and so it happens that the subjects of version (a) become (are "demoted" to) "secondary subjects" in the resulting sentence, i.e. secondary predicates. One fact I think is interesting is that it seems to me that there are some preferential choices for such secondary elements which might depend from some notion of definiteness, compare for example They went Ann eastwards and Bob westwards to Ann and Bob went one eastwards and one westwards... but the discussion becomes too complicated here.

In any case, I would annotate (a) as

1	They	_	PRON	_	_	2	nsubj	_	_
2	went	_	VERB	_	_	0	root	_	_
3	one	_	PRON	_	_	2	advcl:pred?	_	_
4	eastwards	_	ADV	_	_	2	advmod	_	_
5	and	_	CCONJ	_	_	7	cc	_	_
6	one	_	PRON	_	_	7	(advcl:pred?)/orphan	_	_
7	westwards	_	ADV	_	_	2	conj	_	_

Now, the best relation for one is probably advcl, as discussed in #476, maybe with a subtype for this special use.
If we remove the advcl:preds from (a)'s annotation, we get the annotation for (b); if we remove the nsubj and "promote" the advcl:preds from secondary to "primary subjects", we get the annotation for (c).

And so, going back to the example (with copula) that prompted the whole discussion (Dante, De Monarchia):

[...] relationes, quae relationes sunt altera sub ambitu paternitatis et altera sub ambitu dominationis
"officially" translated as 'relationships [...] which are respectively relationships of 'paternity' and of 'lordship''

I would annotate:

12	relationes	relatio	NOUN	_	_7	obl	_	_
[...]
21	quae	qui	PRON	_	PronType=Rel	22	det	_	_
22	relationes	relatio	NOUN	_	_	26	nsubj	_	_
23	sunt	sum	AUX	_	_	26	cop	_	_
24	altera	alter	DET	_	_	26	advcl:pred	_	_
25	sub	sub	ADP	_	_	26	case	_	_
26	ambitu	ambitus	NOUN	_	_	12	acl:relcl	_	_
27	paternitatis	paternitas	NOUN	_	_	26	nmod	_	_
28	et	et	CCONJ	_	_	31	cc	_	_
29	altera	alter	DET	_	_	31	advcl:pred	_	_
30	sub	sub	ADP	_	_	31	case	_	_
31	ambitu	ambitus	NOUN	_	_	26	conj	_	_
32	dominationis	dominatio	NOUN	_	_	31	nmod	_	_

The subject quae relationes is meant to be common to both symmetrical co-ordinated clauses. As before, we have the related sentences quae relationes sunt sub ambitu paternitatis et sub ambitu dominationis and Altera (relatio) [est] sub ambitu paternitatis et altera (relatio) [est] sub ambitu dominationis.

---

So, to sum it up (sorry again for the lengthy post), I think that these kinds of constructions which confused us so in the end all boil down to co-ordinations and secondary predications, so conj(:expl) and advcl(:pred).