Word Sense Disambiguation (WSD) with Hugging Face Models

Hi all, I would be interested in fine-tuning or training a model from Hugging Face to perform word sense disambiguation (WSD) with a dataset (dictionary) of my own. However, I am not sure how to do it. There is no much information available on how to implement WSD out there, at least other than using Lesk algorithm (nltk).

I have seen that this user (jpelhaw/t5-word-sense-disambiguation · Hugging Face) managed to fine-tune the T5 model on the SemCor dataset but there is no info on how he did it.

In particular, I have the following questions:

  • Is there a particular fine-tuning procedure for this type of task?
  • Could I use a model from Hugging Face that has not been trained to perform this task before?

Any help or reference documentation is welcome. Thank you in advance!