UA-SQuAD

Dataset Summary

Ukrainian version of Stanford Question Answering Dataset that includes context, questions and corresponding answers. Current version of the datasets consists of 13 859 samples. Dataset development is still in progress.

Info

Number of samples: 13 859 Number of questions without answers: 2 927 File size: 17.1 MB

Data sample (HF-style)

{
    "id": "3d9f1c2e7a4b1f20",
    "title": "DONDA",
    "context": "5 січня 2012 року Вест оголосив про створення компанії ...",
    "question": "Якою була мета нової творчої компанії DONDA, створеної Каньє?",
    "answers": {"text": ["виготовлення продуктів та поширення досвіду, які люди хочуть отримати й можуть собі дозволити"], "answer_start": [123]},
    "is_impossible": false
}

Example of usage

Python API (HF-style examples)

from ua_datasets import UaSquadDataset

qa_dataset = UaSquadDataset("data/", split="train", download=True)

for ex in qa_dataset:  # each ex is a dict
    print("Question:", ex["question"])
    print("Answers:", ex["answers"]["text"])  # list (may be empty if is_impossible)
    if ex.get("is_impossible"):
        print("(No answer — impossible question)")
    break

Optional: DatasetDict helper (no external Hub required)

If you have the optional datasets library installed, you can build a local DatasetDict using the in-package helper (this does NOT call the Hugging Face Hub API if the JSON files are already cached locally):

from ua_datasets.question_answering.uasquad_question_answering import load_ua_squad_v2

dd = load_ua_squad_v2(root="data/ua_squad", download=True)  # returns a datasets.DatasetDict
row = dd["train"][0]
print(row["question"], row["answers"]["text"], row["is_impossible"])

If datasets is not installed this helper will raise a RuntimeError; install it with:

uv add datasets  # or: uv add datasets

If you don't need a Hugging Face Dataset, stick with the pure-Python iteration example above.

Migration Note

Legacy versions exposed (question, context, answer) tuples and keys Question/Context/Answer in raw JSON; these have been replaced by the standard SQuAD v2 schema shown above. Update loops to: for ex in ds: ex['question'], ex['answers']['text'].

We thank our contributors

Kyrpa Mykyta, Ivan Makarov, Tepla Sofiia, Chudnovska Daria, Fedenko Anna, Zaremba Anna, Krainia Daria, Budenkova Marharyta, Butunaieva Diana, Stanislavska Kateryna, Samorodova Sofiia, Martynyshyn Yuliia, Matviienko Iryna, Bezruka Anastasiia, Mostova Mariia, Stepanenko Liubomyr, Bondarenko Vitaliia, Fedorenko Polina, Sydorka Bohdana, Okhrimenko Mykhailo, Hryha Ruslana, Ustynova Olha, Kondratenko Dmytro, Chornomorets Yelyzaveta, Heresh Yuliia, Hynku Anna-Mariia, Tarasiuk Kateryna, Demian Biliavskyi, Piatushko Ruslana, Pakholchak Kateryna, Barabukha Mariia, Poltorak Yuliia, Yuliia Fedor, Usenko Viktoriia, Balanchuk Yana, Kramchenkov Dmytro, Yatsiuk Mariia, Melnyk Tetiana, Biloverbenko Illia, Boiko Khrystyna, Steshenko Kateryna, Korcheva Anna, Syzonenko Anastasiia, Malysheva Alina, Yaroslava Kushcheva, Valeriia Denysenko