UniversalDependencies / tools

Various utilities for processing the data.
GNU General Public License v2.0
205 stars 44 forks source link

ref in enhanced dependencies #24

Closed gossebouma closed 6 years ago

gossebouma commented 6 years ago

I am working on adding enhanced dependencies. For relative pronouns, the guidelines propose to add a 'ref' relation to the antecedent noun. However, validate.py stumbles over this, as it is not a known dependency label. Adding it to the language specific set does not help either, as it is not an extension of a known relation. Should it be added to deprel.ud?

jnivre commented 6 years ago

We are aware of this problem and are working on it. For the next release, the validator will apply different rules to the DEPREL and DEPS field, and there will be two sets of relations: deprel.ud and edeprel.ud. In addition to the "ref" case, "orphan" is a valid relation in DEPREL but not in DEPS.

dan-zeman commented 6 years ago

"orphan" is a valid relation in DEPREL but not in DEPS.

Is it? If we maintain that all types of enhanced annotation are optional, then you could skip enhanced annotation of ellipsis (and thus keep orphan) but use DEPS for something else, e.g., propagation of dependencies across conjuncts.

jnivre commented 6 years ago

Okay. My bad. The rest of my reply was hopefully correct. :)

dan-zeman commented 6 years ago

I think it was :) Anyways, the way how validator finds out about ref should be transparent for most users.

dan-zeman commented 6 years ago

The current version of validate.py should not complain about ref.

https://github.com/UniversalDependencies/tools/commit/5906a1c2629536bc98c3a3aee54757d20df61d73