acl for secondary predication

[msklvsk]

I find acl(She, sad) in She entered the room sad the most confusing part in UD.

This relation seems more semantic than syntactic: sad intuitively modifies entered (she entered being sad). And if it really is semantic, we should apply the same logic as for dislocated (see #439) and move it to the enhanced representation level.

Also, If the nominal head is missing, the secondary predicate must be attached as advcl of the verbal predicate, which will often happen for pro-drop languages, hurting parallelism even inside a language family, a very UD-unlike phenomenon.

Moreover, in Slavic, sad can have not only nominal case (uk: сумна), but also instrumental (сумною), thus making a state more temporary and acl even more unusual, disagreed with the nominal.

We also find examples, where sad-alikes can be treated as conj to an adverbial clause or oblique:
(translated from uk)

• Andrew sat in the middle, knowing what to expect from other convicts, and prepared for a fight.

• 

    She     is-standing without    a-shawl    ,         gray-haired ,       lush-haired

I suggest to always analyze optional depictives as advcl(entered, sad) for the following reasons.

1. It intuitively belongs to a verbal predicate.
2. Semantics does not belong in the basic dependency representation.
3. We already use advcl(entered, sad) if the nominal head is missing, which would often be the case for e.g. Polish.
4. In Slavic, sad can also be instrumental, a very “adverbial” case.
5. There are examples of coordination between optional depictives and adverbial clauses/obliques.
6. xcomp vs advcl for secondary predication would rhyme with obj vs obl.

[mcandito]

I fully agree with this proposal: syntactic handling of optional secondary predicates, and recovering the semantic link in enhanced dependencies
Marie

[jnivre]

I find this argumentation convincing as well. This should definitely be taken into account when working out the new and improved guidelines. I am not sure whether we should set up a working group specifically for secondary predication, or whether it can be subsumed under some other topic. I will add a note to the "brainstorming page" for now. (See my email on the UD list two days ago.)

[amir-zeldes]

+1
And I believe this is also the way the Stanford Dependencies handled adverbial vmod, so others have found such an analysis sensible before.

[jnivre]

I think it would be very useful to have a working group looking at this in the larger context of non-finite modification (and complementation). Examples like:

she entered the room sad
she entered the room crying
she entered the room holding a handkerchief
she entered the room (with) a handkerchief in her hand

[rafael75012]

Hello,
May I ask if we came to any decision?

[jnivre]

Unfortunately, nothing seems to have happened here. We need to follow up this and many other issues.

[dan-zeman]

Anyways, if there is a conclusion, it will be the next version of the guidelines, right? Shall I modify the milestone to "later"?

[jnivre]

Right.