Arabizi Dataset
Any plans on curating an arabizi hatespeech dataset? If yes I'd like to contribute. If there's any existing datasets you can point me to, refer them. thanks!
Hello Ismail,
At the moment, I don’t have plans to curate an Arabizi hate speech dataset. I’m also not aware of any well-established public datasets specifically for Arabizi hate speech.
That said, one promising approach you could explore is automatic transliteration. For example, the UBC NLP team created UBC-NLP/nilechat-arabizi-egy by using CohereLabs/c4ai-command-r-plus-08-2024 to convert Egyptian Arabic text (in Arabic script) into Arabizi. You could potentially apply a similar pipeline to IbrahimAmin/egyptian-arabic-hate-speech to generate an Arabizi version.
It might also be worth experimenting with newer multilingual models (e.g., https://huggingface.co/collections/google/gemma-4, https://huggingface.co/collections/google/translategemma) to see if they improve transliteration quality.
Additionally, you can check this curated list of Arabic datasets:
https://arbml.github.io/masader/
You might come across something relevant there.
If you decide to go down this route, I’d be very interested to hear how it goes!