We are releasing the Belebele dataset, a first-of-its-kind multilingual reading comprehension dataset spanning 122 language variants, 27 language families, and 29 scripts. [Paper] [Github] [Tweet]