Global PIQA is a participatory commonsense reasoning benchmark for over 100 languages, constructed by hand by over 350 researchers from over 65 countries around the world.
The parallel split is a multi-parallel dataset for 131 language varieties, covering five continents, 16 language families, and 23 writing systems.
In this parallel split, each example was machine-translated from English, then manually corrected by a native speaker of the target language.… See the full description on the dataset page:
https://huggingface.co/datasets/mrlbenchmarks/global-piqa-parallel.
