Where do the dependency labels of the English spacy model come from ?

[VSegonne]

Hi,
I'm using the en_core_web_lg spacy model for dependency parsing and I need to be sure where the set of dependency labels come from.

The documentation stipulates that the model was trained on Ontonote. Yet, Ontonote is only annotated with syntactic constituents.
Roaming around on multiple discussions :

• Annotation Specs for Syntactic Dependency Parsing are incomplete #233
• https://stackoverflow.com/questions/62399475/where-i-can-find-the-complete-list-of-spacy-dependency-parsing-labels-or-annotat

It seems that ClearNLP was used to convert Ontonote to dependency parsed data, possibly following this paper : https://www.mathcs.emory.edu/~choi/doc/cu-2012-choi.pdf

This is not clear to me, can anyone confirm this ?
And if so, that would be nice to include it somewhere in Spacy's documentation.

Thanks !

[adrianeboyd]

This information is in the v2 docs here: https://v2.spacy.io/api/annotation. This page seems to have gone missing in the v3 docs and should be added back, with updates since a few of the details have changed in v3.

[VSegonne]

Alright thank you !