Coordination of auxiliary clusters under a predicate

[nschneid]

E.g. in EWT:

• "actions have been and will be taken"
• whether something "is or is not an economic hardship" (copula + negated copula)

What is the principle here, since AUXes and "not" normally depend on the main verb/predicate rather than forming a constituent?

I can think of a few options:

Option A: Everything is shallow: everything modifies the main predicate ("taken", "hardship"), including the conjunction as cc. There is no conj relation.

Option B: Second conjunct + conjunction is a constituent under the last word of the first conjunct: aux(taken, have), aux(taken, been), conj(been, will), aux(will, be), cc(will, and). Here the first word of the second conjunct is promoted to be the head of the rest of the conjunct.

Option C: Everything in the second conjunct is attached directly to the last word of the first conjunct: aux(taken, have), aux(taken, been), conj(been, will), conj(been, be), cc(will, and)

...Or something else???

[dan-zeman]

The guidelines actually assume that function words can be coordinated, and this is one of the exceptions where function nodes can have children. So the copula example seems clear to me and I believe I have seen such examples in the data (not only English):

cop(hardship, is-1) conj(is-1, is-2) cc(is-2, or) advmod(is-2, not)

The first example is trickier because there are two auxiliaries in each conjunct, and they are not supposed to be connected with a relation. So I think it is easier to think of the full conjuncts have been [taken] and will be taken where the first occurrence of taken is elided. I would then promote been as the new head, then the rest is straightforward:

nsubj(been, actions) aux(been, have) conj(been, taken) cc(taken, and) aux(taken, will) aux(taken, be)

[nschneid]

The ellipsis analysis is interesting. It would mean that in the enhanced representation, the elided predicate node would be the head of the overt predicate, whereas it's usually the reverse.

[sebschu]

The ellipsis analysis is interesting. It would mean that in the enhanced representation, the elided predicate node would be the head of the overt predicate, whereas it's usually the reverse.

In English it's usually the reverse, but in head-final languages this is is more common. See, for example, ex. (5) in our paper on gapping.

[nschneid]

Updated the parse for the "have been and will be taken" example with the ellipsis analysis.

Can somebody please double-check it?

[dan-zeman]

Can somebody please double-check it?

I tried but this visualization which mixes basic with enhanced and duplicates edges is terrible. I had to switch to CoNLL-U source :-) It looks good to me (except that I'm not sure we use empty nodes to resolve constructions where an auxiliary is promoted in the basic tree; but it's useful, so maybe we should think of adding it to the enhanced guidelines).

[dan-zeman]

Changing milestone to 2.6 and leaving open until the possible addition to guidelines is decided.