Hi, I’m Nick Doiron (monsoon-nlp here, mapmeld on Twitter).
A year ago I combined Arabic dialect datasets to make a classifier; I also have a GPT-2 model finetuned on dialect markers but I can’t test the accuracy.
In 2019 I visited Re-Coded’s code schools in Turkey and Iraq. I’d be interested in projects which would support and engage the Arabic speakers in their programs.
I’m also attempting to test model bias with a seq2seq model which flips male and female words. Still a work in progress - need to find additional sources for training data
3 Likes