Which roles can become orphan?

[lauma]

Is there something like a conclusive list which roles must become orphan in case when the predicate they are dependant of is elided? https://universaldependencies.org/u/overview/specific-syntax.html#ellipsis says following:

"If the elided element is a predicate and the promoted element is one of its arguments or adjuncts, we use the orphan relation when attaching other non-functional dependents to the promoted head."

Would that mean that all roles from blocks Nominals, Clauses, Modifier words listed in https://universaldependencies.org/u/dep/index.html are eligible to become orphan, i.e., full list would look like this?

• Nominals: core arguments - nsubj obj, iobj
• Clauses: core arguments - csubj, ccomp, xcomp
• Nominals: non-core dependents - obl, vocative, expl, dislocated
• Clauses: non-core dependents - advcl
• Modifier words: non-core dependents - advmod, discourse
• Nominals: nominal dependents - nmod, appos, nummod
• Clauses: nominal dependents - acl
• Modifier words: nominal dependents - amod

The inclusion of discourse in this list seems to be excessive, but what about others?

Having such complete list would be very helpfull, when building validation or convertion tools for UD data.

[dan-zeman]

No, I do not think such a list exists, at least not with the intention to strictly exclude everything else. You could say that the obliqueness ranking that is mentioned in the guidelines lists dependent types that were foreseen to become orphans: nsubj > obj > iobj > obl > advmod > csubj > xcomp > ccomp > advcl.

I also do not think that discourse should be included, and I am not sure about expl. The relations vocative and dislocated probably should be included.

On the other hand I am pretty sure that dependents of nominals are excluded: nmod, appos, nummod, amod, acl. If the modified noun is elided, one of the “orphans” should be promoted as the new head of the noun phrase, and the other orphans are attached to the promoted one via the same relation as if they were attached to the elided noun. They are not attached via orphan. The promotion occurs regardless whether the noun phrase is a predicate or not.

[lauma]

Thanks! I will tune the convertor code used for Latvian to follow these prelaminary guidelines.

[dan-zeman]

Unless there are objections I propose to add dislocated and vocative to the end of the obliqueness hierarchy in the guidelines.

[jnivre]

+1