Fine tune a transformer model for POS tagging

Can the example scripts for token classification be used for POS tagging? The README for token classification currently shows benchmarks for NER datasets only. Are the steps(preprocessing and evaluation) for POS tagging similar? Also are there any results for POS tagging using any transformer architecture?
Thank you


Hi @kushalj001,

yes, it is possible! Just use the same data format (token, tag pair per line) like it is shown for the NER example.

However, what is currently missing is “accuracy” as metric, because for PoS tagging you normally report accuracy (instead of F1-score).

I’m working on a nice readme extension for PoS tagging, maybe it is ready until next week.


Hi @stefan-it, any update on the readme extension? :slight_smile: thanks

Functionality was for PoS tagging was added by @vblagoje in this PR :hugs:

Readme is currently missing, but you can follow the instructions in the shell scripts for training a PoS tagging model :slight_smile:

1 Like