Detoxify Roadmap

[laurahanu]

Issue to keep track of improvements we'd like to make (in no particular order). Feedback and suggestions welcome!

• better way to handle emojis (Detoxify doesn't work well on Emojis #27)
• train the unbiased model on the Wikipedia dataset from the first challenge as well
• add a multilingual light model (Multilingual light model #17)
• train the multilingual model with more languages
• add more datasets to training
• add new categories like personal attack (potentially using https://github.com/ewulczyn/wiki-detox)
• improve bias metrics & test on different benchmark like HateCheck

[erjonhub]

Hello,
What are the languages planned to be added in the roadmap? Is German one of them? Any target date?

Thank you