Are predicative adjectives with non-copulas `xcomp`?

[chiarcos]

Question: In constructions in which a lexical verb (not a copula) takes a predicative adjective (like the copula does) as an argument, what is the dependency label that links the predicative adjective with its head? Or can we use the predicative adjective as head and annotate the verb as cop?

Wondering how to annotate the following example from Old Saxon

   sô mi    mina sundia nu  suîđaron thunkiat
   so to.me my   sins   now great    seem
   "So, my sins now seem great to me"

Note that thunkian is not as semantically bleached as English seem, but it designates cognitive processing, as it is transparently linked to Old Saxon thenkian "to think", cf. German mich dünkt and English methinks. It seems to me that this pattern also applies to other verbs of perception and cognition.

The semantics of the adjective (suîđaron/great) is not so different from those in So, my sins now are great, but whereas the latter is clearly a copular construction, I am not sure about sure about thunkian "to seem, appear". But if we do not annotate thunkiat as cop, but root, what would the dependency label for suîđaron? Because of the copula rule, UD doesn't seem to have a dependency for predicative adjectives, does it?

However, I have seen xcomp for such cases, and https://universaldependencies.org/u/dep/xcomp.html may actually have an example for that ("You look great"), but the current definition under https://universaldependencies.org/u/dep/xcomp.html is almost exclusively about clauses (and secondary predicates).

If my interpretation is correct (please confirm), I suggest the following modifications to https://universaldependencies.org/u/dep/xcomp.html:

• Replace the first sentence by a merger of the third and first line "An xcomp can also be described as a predicative complement." as follows:

The dependency xcomp designates a predicative complement of a verb or adjective. This can be a predicative adjective (as in example 4) or an open clausal complement (as in example 1).

• Put examples 4 and 1 right after this.

• Continue with the text as currently is. Do not repeat the third sentence or examples 4 and 1.

• In addition to example 4, maybe add another example with English seem to.

Note that (if I am correct in my interpretation), these changes do not affect the actual content of the text, but only its presentation. My problem with the current reading of the text is that I have difficulties to see that adjectives are inherently clausal. In languages without grammaticalized adjectives (such as Inuktitut or Sumerian), adjectives may be verbs, morphologically, and there, this is certainly the case, but for many (incl. -- AFAIK -- all IE) languages, morphological and syntactic ties between nouns and adjectives are strong enough to ask for a a clarification.

[Stormur]

I can confirm that what seem to be the conventional tag for these occurrences is xcomp, while not annotating the copular verb as root. This is at least what we stick to in Latin, as it was understood from the guidelines and from annotating experience.

I agree anyway that this is far from ideal, especially with respect to what xcomp should be and even more in relation to ccomp... it's pretty contradictory and undefined by now.

I would not restrict the definition to adjectives, because in that position you can really find anything. Then, the current interpretation (as for secondary predications with advcl[:pred]) is that these are all some sorts of clauses, nominal or not. Adjectives do not need verbal morphology to be seen as clausal, they are seen as such in that they are inherently predicative. But in this case I do not think it is that important, as anything can be the head of a predicate and as such be the "predicative complement" of something.

[ellepannitto]

I can also confirm that this is how it's done in Italian.
I personally like the interpretation of these cases as secondary predication so I don't see issues with xcomp, but I agree with @chiarcos that it could be phrased more clearly on the guidelines and with @Stormur that adjectives are not the only case.

[chiarcos]

I would not restrict the definition to adjectives, because in that position you can really find anything.

True, I would change my suggested text from "This can be a ..." to "For example, this can be a ..." and maybe add more examples.

Adjectives do not need verbal morphology to be seen as clausal, they are seen as such in that they are inherently predicative.

True, but this also depends on the generally accepted definition of clause (which we don't really have ... nor need for UD), but one way of doing is by "relationship between verb and subject", and then, they are predicative, but not necessarily clausal. I'm not arguing for one way or the other, just for a less confusing formulation.

[chiarcos]

Thank you, for the moment, and unless anyone views it drastically different, I would continue to use xcomp then. But I also suggest to keep the issue open for some time to see what people think about the suggested changes to the text.

[Stormur]

Adjectives do not need verbal morphology to be seen as clausal, they are seen as such in that they are inherently predicative.

True, but this also depends on the generally accepted definition of clause (which we don't really have ... nor need for UD), but one way of doing is by "relationship between verb and subject", and then, they are predicative, but not necessarily clausal. I'm not arguing for one way or the other, just for a less confusing formulation.

Uhm, I do not know if I follow you completely here, but I don't see a definition working that requires some particular argument like a subject. A clause corresponds to a predicate together with any possible arguments (let's not talk about ellispes now, OK 😬), so we have to define a predicate. But once we agree that adjectives can head a predicate, it follows that when they do they are "clausal" by definition. This is actually valid for any among NOUN/PROPN/PRON/ADJ/DET/NUM/VERB/AUX... but ADJs surely tend to head predicates "more easily" because of their relational nature. Are you maybe thinking of a more restricted definition of clause? Something on the spectra finite-nonfinite, transitive-intransitive...?

[chiarcos]

It harks back to the old GB discussion on whether small clauses exist or not [1,2]. But trying not to be dogmatic. My personal view on syntax was massively influenced by Role and Reference Grammar -- after a brief and discouraging encounter with generative grammar --, and there, the clause is seen as a structure on top of a predicate, so, on a different ontological level than the predicate itself. I would say that adjectives are universally accepted as being predicative, but whether they are clausal is theory-specific. And IMHO, the traditional view still being taught to teachers of, say, English is that a clause requires the presence of a verb. (I wouldn't go so far, but that's what my students seem to believe after their introduction to syntax -- not by me, though.)

[1] Williams, E. S. (1983). Against small clauses. Linguistic inquiry, 287-308.
[2] Citko, B. (2008). Small clauses reconsidered: Not so small and not all alike. Lingua, 118(3), 261-295.

[jnivre]

I can confirm that xcomp is the recommended option here (and also what we use in the Swedish treebanks). It is perhaps not an ideal solution in that it makes these constructions biclausal, but I think the argument for not treating these verbs as copulas is that it is unlikely that any language would use a zero strategy for this kind of predication when there is a lexical verb involved, and the existence of the zero strategy in many languages (that is, something like "John nice" with no copula) is the main motivation for not treating the basic copula as a predicate in languages where a copula is required.

[nschneid]

There is probably room for improvement in the current guidelines, but I would be careful not to suggest that anything traditionally considered a predicate complement of a verb or adjective is xcomp, because copular predications are treated differently (as @jnivre notes).

[Stormur]

[...] but I think the argument for not treating these verbs as copulas is that it is unlikely that any language would use a zero strategy for this kind of predication when there is a lexical verb involved, and the existence of the zero strategy in many languages (that is, something like "John nice" with no copula) is the main motivation for not treating the basic copula as a predicate in languages where a copula is required.

I understand the logic, even if I do not see it workin g "in the long term". There is indeed some grammaticalisation/semantic gradient from "absolute nothing", the copula as is defined in UD now, up to elements bearing some other modal or (lexical) aspectual categories.

But anyway the major source of confusion, the major offender, as it were, in the definition and use of xcomp is that it should be the clausal equivalent of an object in alternation with ccomp, something which decidedly is not when used for cases like John seems nice etc..

[dan-zeman]

it should be the clausal equivalent of an object in alternation with ccomp

I don't think this is true (although I also used to think of xcomp this way in the past). If it were true, then xcomp could not co-occur with obj in the same clause. But that's exactly what it does in object control construction (Mary made John(obj) wash(xcomp) the dishes).

[Stormur]

it should be the clausal equivalent of an object in alternation with ccomp

I don't think this is true (although I also used to think of xcomp this way in the past). If it were true, then xcomp could not co-occur with obj in the same clause. But that's exactly what it does in object control construction (Mary made John(obj) wash(xcomp) the dishes).

I mean, I wanted to highlight that this is a source of confusion because you have that xcomp is put in opposition to ccomp, and ccomp is defined as a clausal core argument, but more precisely as an object (or adjective?!) of the verb since there is already csubj. So it seems straightforward to conclude that xcomp is an object, else it is not clear which "core argument" would be meant. So one is induced to think this, and I am pretty sure it was defined with (also) this in mind, while its use is partly wildly different.

I just see annotators having problems wrapping their heads around it, and this is because I think there are terminological and conceptual problems behind this tag.