Note Manually translated and culturally adapted IFEval.

Note Manually translated and culturally adapted WinoGrande test set.

Note Manually annotated extractive question answering.

Note Manually annotated named entity recognition.

Note Machine-translated and post-edited choice of plausible alternatives.

Note Grammatical error correction. Text with errors → corrected text.

Note ERR news story transcripts generated by an ASR pipeline paired with the human written summary.

Note Newspaper article sentiment.

Note Exam questions on several subjects in multiple choice format.

Note Machine-translated MMLU.

Note Matching words with their dictionary definitions.

Note Noun phrase inflection.

Note Trivia dataset extracted from the "Eesti Mäng" board game.

Note Machine-translated MMLU-Pro.

Note Extension of the MT-Bench dataset including Estonian, Livonian, Komi, and Võro.

Note Extension of the original Flores dataset adding additional translation directions, such as Estonian to Võro, Estonian to Komi-Zyrian, and some others.

Note Multiple choice reading comprehension. Includes Estonian as one of the languages.

Note Machine-translated multilingual COPA. Includes validation and test for Estonian.

Note Linguistic acceptability dataset derived from Universal Dependencies with heuristics. Includes Estonian as one of the languages.

Note Includes Estonian as one of the languages. English sources are translated to 200 languages. Any translation direction is possible potentially.

Note Includes English to Estonian translation direction.

Note Includes Estonian as one of the languages, classification labels remain in English.