Languages other than English?

[rainergo]

Hi,
do you plan to pretrain models for maverick in languages other than English?
Thanks.

[g185]

Hello @rainergo,
Maverick methodology is language agnostic therefore can be applied in multiple languages.
There is no imminent release of Maverick models trained in other languages, but this repository can be used to train a multilingual model (mDeberta for example) on different languages.
It is necessary to create a new model configuration under conf/models/ containing the settings for the multilingual encoder, and adding a new dataset configuration in conf/data/ containing a coreference resolution dataset in OntoNotes format.

Thanks,
I am available for further clarifications if needed.

[rainergo]

@g185 Thanks. I was looking for a pretrained Coreference-Resolution model in German as I do not want to train the model myself. Would be great if you could provide such a pretrained German model in the future as the few available models that are pip-installable (i.e. Coreferee, crosslingual-coreference) are pretty much outdated.

[tanurathee]

@g185 @rainergo hey! if you could share me code where you used mdeberta it would be really helpful.

[gnmarten]

@g185 @rainergo Thanks. I was looking for a pretrained Coreference-Resolution model in German as I do not want to train the model myself. Would be great if you could provide such a pretrained German model in the future as the few available models that are pip-installable (i.e. Coreferee, crosslingual-coreference) are pretty much outdated.

this team https://github.com/uhh-lt/maverick-coref-de trained a specific German model for cofererence based on maverick https://huggingface.co/fynnos/maverick-mes-de10 .