Closed khuangaf closed 4 years ago
Unfortunately no. Coref didn't help at all on ACE, so we didn't end up releasing the co-training code. I'll try to dig up the ontonotes data to see if I can train a model for this, but I'm not sure I've got it in the required format anymore.
I see. Thank you very much!
I have one quick question regarding ACE05 coreference resolution. In the paper you states that "the ACE
data set lacks coreference annotations". However, the original ACE05 annotation contains <entity>
and <entity_mention>
tags. I am wondering why it does not count as coreference for all <entity_mention>
under the same <entity>
tag?
Hmm that's a great point, I hadn't thought of that. I'd definitely welcome a PR to preprocess the data such that ACE mentions are linked via coreference clusters.
Just wanted to clarify -- the current ACE05 relations model doesn't include coreference propogation, so it's equivalent to the "lightweight" versions of the other datasets' models?
Thanks!
Yep, that's correct!
In the paper you mention as ACE does not have coreference annotation, you use OntoNotes for coreference propagation. I was wondering if it is possible to predict coreference on ACE using the current trained model?