orphan: orphaned dependent of a predicate attached to another orphan
The orphan relation is used to provide a satisfactory treatment of
ellipsis (in the case of gapping and stripping, where a predicational
or verbal head gets elided).
To develop motivation, consider first a sentence without ellipsis:
Maria foi para Paris e Miriam foi para Praga
nsubj(foi-2, Maria-1)
root(root-0, foi-2)
nmod(foi-2, Paris-4)
case(Paris-4, para-3)
cc(foi-7, e-5)
nsubj(foi-7, Miriam-6)
conj(foi-2, foi-7)
case(Praga-9, para-8)
nmod(foi-7, Praga-9)
The question is then how to treat the sentence “Maria foi para Paris e Miriam para Praga”
Maria foi para Paris e Miriam para Praga
nsubj(foi-2, Maria-1)
root(root-0, foi-2)
nmod(foi-2, Paris-4)
case(Paris-4, para-3)
case(Praga-8, para-7)
One option would be to pretend that there is an empty verb and to have the final elements be dependents of it: Maria foi para Paris e Miriam ∅ para Praga. This analysis has some appeal but also has some problems and at any rate stops the basic dependency graph from being simply a tree of dependencies over the words of a sentence. A second option is to simply promote the final elements and to have them as dependents of the main verb of the sentence (foi-2) or of root-0. But then (in general) one loses the ability to successfully reconstruct the correct predicate-argument structure of the sentence from the basic dependency representation.
In UD v2, the enhanced dependency representation uses empty nodes to
represent the elided predicates. In the basic representation, one
orphaned dependent is promoted as the head and it attaches to the
parent of the elided predicate, using its relation. The other orphans
are attached to the promoted orphan via the orphan relation. So, for this example, we have:
Maria foi para Paris e Miriam para Praga
nsubj(foi-2, Maria-1)
root(root-0, foi-2)
nmod(foi-2, Paris-4)
case(Paris-4, para-3)
conj(Maria-1, Miriam-6)
cc(Miriam-6, e-5)
orphan(Miriam-6, Praga-8)
case(Praga-8, para-7)
In VP-ellipsis, we keep the auxiliary as the head, as shown below:
João vai ganhar ouro e Maria vai também
nsubj(ganhar-3, João-1)
aux(ganhar-3, vai-2)
obj(ganhar-3, ouro-4)
conj(ganhar-3, vai-7)
cc(vai-7, e-5)
nsubj(vai-7, Maria-6)
advmod(vai-7, também-8)
orphan in other languages: [bm] [cop] [cs] [de] [el] [en] [fi] [fr] [fro] [ga] [gsw] [hy] [it] [kk] [no] [pcm] [pt] [qpm] [ro] [ru] [sl] [sv] [swl] [tr] [u] [vi] [xcl] [yue] [zh]