John Bauer

Under the hood it's building the internal coref format here: https://github.com/stanfordnlp/stanza/blob/8be9ce85fcbd02247f7fc21abfcdfbeadc4c0608/stanza/pipeline/coref_processor.py#L103 and that's not using any speaker information. So, if there's a connection from word (or sentence) speaker to the...

Seems like a short list of things to do: - [x] have speaker as an attribute on Sentence (do we want it on individual Words?) - [x] if there's a...

based on @Jemoka 's comments, perhaps also - [ ] rebuild a model with speakers

>> if there's a speaker on the sentences or words, pass that along to the coref model > > Yup, should already be done. Unless you mean in a branch,...

I handled speakers on a sentence level, not the formatting used in the more finegrained method described above. It's in the `speaker` branch or this PR here: https://github.com/stanfordnlp/stanza/pull/1525 @jemoka are...

I don't recall having made any changes to the speaker embedding / training. I also don't ever recall it being a particularly important part of the input. Is it possible...

Merged the PR... only other thing to do is post the model? Should I clobber the old GUM model or call this one something new, like gum_speakers? Is there anything...

There is now `gum-speakers`, the new model, and `gum-nospeakers`, the old model: https://huggingface.co/stanfordnlp/stanza-en/tree/main/models/coref [gum-nospeakers_roberta-large-lora](https://huggingface.co/stanfordnlp/stanza-en/blob/main/models/coref/gum-nospeakers_roberta-large-lora.pt) [gum-speakers_roberta-large-lora](https://huggingface.co/stanfordnlp/stanza-en/blob/main/models/coref/gum-speakers_roberta-large-lora.pt) The speaker information won't be passed to the model from the pipeline until a new...

That's very weird. Creating a basic pipeline is of course one of the things tested by our CI...

That would be surprising, given that `Hände` shows up many times in the training data. However, I do not get this result: ``` >>> import stanza >>> nlp = stanza.Pipeline("de")...