You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I am asking, as a matter of the CoNLL-U format, whether there can be enhanced edges sharing the same source and target: X-[E:rel1]->Y and X-[E:rel2]->Y.
I don't see it spelled out here one way or another (whether the contents of a DEPS field must have unique head indices). I presume the exact same edge cannot be duplicated.
I consider it legal when writing code that processes CoNLL-U and I believe the validator will not complain about it.* Strictly speaking it could be problematic if we cannot find an example where this results from one of the six officially defined enhancements or their combination (as the guidelines also say that released treebanks should not contain other enhancements). But technically it is not a problem.
*) I agree with you that repeating the same relation with the same label should be forbidden; there is no point in having multiple copies of the same edge. I don't remember whether the current validator flags it as an error. But I'm pretty sure it has some other restrictions on the enhanced graph, e.g., self-loops are not allowed.
I am asking, as a matter of the CoNLL-U format, whether there can be enhanced edges sharing the same source and target: X-[E:rel1]->Y and X-[E:rel2]->Y.
I don't see it spelled out here one way or another (whether the contents of a DEPS field must have unique head indices). I presume the exact same edge cannot be duplicated.
At least some tools for reading .conllu files require unique head indices: Arborator/arborator-frontend#433 (comment)
The text was updated successfully, but these errors were encountered: