Augmentation

[KennethEnevoldsen]

• Entity augmentation
  • Gender augmentation (awareness of gender)
  • Second order person augmentation (Lastname, Firstname)
  • Usernames (autogenerates e.g. WhiteTruffle101 or Kenneth Enevoldsen -> KennethEnevoldsen)
• Mispellings Augmentations, se e.g. this repo
  • Keystroke error based on keyboard distance
• Historic augmentations
  •   æ->ae, å -> aa (and a), ø->oe
  • uppercasing of nouns
• Social media
  • Adding hashtags augmentation
• Others, potentially see this tweet or this kaggle summary

[martincjespersen]

• Add entity augmentation for abbreviated first names e.g., Martin Jespersen -> M. Jespersen or Jespersen, M.

[KennethEnevoldsen]

Hei @martincjespersen entity augmentation has just been added in the 1.0.0 update. I recommend you check out the documentation for robustness and biases in Danish NLP models. There will also be a paper coming out soon (probably Monday) I will make sure to send it your way. Parts of this work have very much been inspired by your fairness benchmark 🙏 - so I very much hope you find the results interesting.

[KennethEnevoldsen]

As seen in issue #59 augmenters will be replaced with using the augmenty package.

@martincjespersen you might be especially interested in this project. It is still under development, but already looking very promising.