What do we do? Recreating SuperGlue benchmark, but for low resource language - Belarusian SuperGlue, with 3 tasks: BoolQ , COPA, CB.
1 - BoolQ (Yes/No Question Generation over a Paragraph)
Who is running? Haeshitha
Size: 800 (Yes 400 / No 400)
Details here:
2 - COPA (Causal Reasoning Generation)
Who is running? Amrit
Size: 600 (300 “cause”, 300 “effect”; within each, ensure 300 A-correct / 300 B-correct overall)
Details here:
COPA- Causal Reasoning Generation
3 - CB (Entailment / Contradiction / Neutral Generation)
Who is running? Poli
Size: 600 (200 / 200 / 200 (E/N/C))
Details here: