Language Support

[eschaffn]

Hi a few questions, does this model work only on English?

If so, what would it take to train it on another language or script type?

Would it need to be pretrained again using self-supervision and how expensive is the pre-training process computationally?

Thank you!

[CheungZeeCn]

+1

[ziyi-yang]

UDOP currently is English-only. To fully replicate this work on multilingual domain, you will need multilingual documents with OCR annotations, and labeled multilingual document for supervised pretraining.