Hello,
One proposal I think is worth considering is re-focusing our inflection challenge (liked hosted at SiGMORPHON) on transfer learning. We did this in 2019, but I think it makes sense given our annotation desires.
Why transfer learning? We came to the consensus that cleaning up the UniMorph languages by their family is the best way to go. We will start by cleaning the Uralic, Turkic, Semitic, Slavic, and Austronesian languages such that tags are truly cross-lingually consistent. We will also annotate cognates among the lemmata in this process (this was Mans's suggestion.) A transfer-learning based task would fit nicely with this goal. Plus, transfer learning for low-resource morphology is a lot of fun.
Please let me know if you have any thoughts or you'd like to participate. Feel free to respond on the thread.
Cheers,
--Ryan