What is the appropriate annotation for bare wh-elements (in elliptical contexts?) ?

[ClaudiaCorbe]

Good morning,

I have a question regarding how to annotate instances of "how" (come in Old/Modern Italian) when it appears without further specification, for example in sentences like "And see how".
More precisely, the issue is whether such cases should be considered elliptical clauses in which only the wh-element is retained, and thus treated as instances of so-called sluicing (Vicente L., 2018 "Sluicing and its subtypes." in Van Craenenbroeck, Jeroen, and Tanja Temmerman (eds.) The Oxford handbook of ellipsis. Oxford University Press), e.g., "Phoebe ate something, but she doesn't know what. (=what she ate)."

My question arises from a specific instance of come "how" in my data (see Example 1 below), but similar issue also arise with other wh- element, such as "where", "why" etc.
Below, I describe the example that raised my question, along with some differing annotation approaches I came across. These made me unsure about how to handle such cases.

Example 1 (Paradiso XX, v. 89): How should I annotate come ("how")?

Io veggio che tu credi queste cose perch' io le dico, ma non vedi come. (OldItalian_Dante_Paradiso-608)
I see that you believe these things because I say them, but (you) don't see how.

In Modern Italian treebanks, come (“how”) is annotated as an advmod of vediamo (“see”):

Vediamo come. (ISDT- tut-3126)
(Let's) see how.

However, when I looked at similar constructions in English, I found:

Sent home with Amoxi, and Pred, so, we'll see how — (GUM_conversation_vet-43)

Here, "how" is annotated as a ccomp dependent on "see", which reflects a sluicing interpretation.
The same pattern appears with "why":

I don’t like the doctor, but I don’t know why. (GUM_fiction_oversite-23)

In this case, "why" also receives a ccomp dependency depending on "know".

What would you recommend in this kind of case?

Thank you,
Claudia

[dan-zeman]

The ccomp analysis seems correct to me.

I imagine that it may not be easy to detect automatically that such interpretation is available if one is converting annotation from a non-UD scheme which did not make the distinction – the obvious default when one sees an adverb depending on a verb would be advmod. But a human annotator should be able to distinguish these two:

• I don't know how (ccomp(know, how))
• How do I know (advmod(know, How))

[Stormur]

I also agree on the ccomp analysis.

It seems that the position before or after the verb is discriminatory enough also in Italian. I mean that unless you say something very prosodically marked such as "come, non lo so" (by the way, there is usually a pronoun to index the - clausal - object), there are no alternatives in the interpretation of a postposed come than sluicing.

[amir-zeldes]

I actually think those are errors in GUM, since those are reduced free relatives. In this context "why" and "how" work like "what" - in "I know what you did", there is an NP [what you did], and that functions as an object, based on Bresnan's (1978) free relative account. Similarly "I know why" is like "I know the reason why...". So the tree should be analogous to the case with the full clause, and the head should have the obj function externally. Compare these three sentences:

[sylvainkahane]

In French, the paradigms of interrogative and relative pronouns are different (even if some forms are common) and after savoir 'know' we have interrogative clauses:

il l'a fait mais je ne sais pas comment 'he did it but I don't know how'

comment is only an interrogative pronoun. So it will be ccomp in French.

[ClaudiaCorbe]

Thank you all for the clarification!

[francescomambrini]

Hi Claudia,
Instictively, I also agree with the idea of treating them as ccomp. To me both Italian cases (Dante's and the one from ISDT) are text-book cases of "sluicing", as you write. So, per the usual chain of elision and promotion, they become the head of the interrogative clause. Indeed, they are all that remains of the interrogative clause.

[amir-zeldes]

So it will be ccomp in French ... I also agree with the idea of treating them as ccomp

@francescomambrini @sylvainkahane if that's how the full version would be done, then I would also expect ccomp. But in English, WH-object clauses are analyzed as NPs based on Bresnan's free relative analysis, so "what you did" in "I know what you did" is headed by "what", not "did" (see trees above), so the normal promotion would not happen with ccomp. The nested verb "did" would be governed as acl:relcl by "what".

In French, the paradigms of interrogative and relative pronouns are different (even if some forms are common) and after savoir 'know' we have interrogative clauses

That's interesting and sounds relevant for French, so maybe French should not have the free relative analysis. But could you elaborate on what you mean? In French I would have expected "comment" in both the interrogative and object position:

• Comment as-tu fait ça?
• Je sais comment tu l’as fait.

To me this looks the same as the English situation.

[sylvainkahane]

For instance, in French, we have two interrogative pronouns, où 'where' and quand 'when', which correspond to only one relative pronoun où:

• l'endroit où il habite ‘the place where he left
• au moment où il est parti 'when he left', lit. at the time where he left

After savoir, I use quand and not où:

• Quand as-tu fait ça ?
• Je sais quand tu l'as fait.

Of course, a possible analysis is to say that pronouns for free relative are not the same as relative pronouns. But another argument is that free relatives with qui 'who' are almost impossible in Modern French, but qui is not a problem in the object position of savoir.

By the way, from a distributional point of view, it makes sense to consider that the wh-word is the head of the clause. But this is another question.

[amir-zeldes]

I see what you're saying about distinct forms, but that could also apply to English "that", which can be used as a relative pronoun but not as the head of a free relative argument:

• I know what you said
• I know the thing that you said
• * I know that you said

In English, as in French, the distribution suggests that the WH element in free relative arguments is the head, and indeed we can say "I know what!" without the clause, but not "*I know you said". So in sum, I think there is no strong difference between the languages on this count, even if in the free relative position "ou" is available for time as well as place. If I were starting UD annotation of French from scratch I would probably say that "quand" is obj in "je sais quand tu l'as fait" (by analogy to "je sais quand", where I would also not use ccomp).