slovak-nlp/scio-testy

SCIO Tests (Slovak)

Dataset Summary

A multiple-choice question-answering dataset extracted from Slovak-language SCIO National Comparative Exams (Národné porovnávacie skúšky, NPS) used for university admissions. It contains 2,714 questions from test booklets spanning roughly 2019–2026, covering four subject areas:

Subject (source prefix)	Approx. questions
General Study Prerequisites (VSP)	~1,250
Biology (BIO)	~516
Mathematics (MAT)	~562
Chemistry (CHE)	~369
English/other (Aj, …)	~17

• Language: Slovak (sk)
• Total questions: 2,714 (dev: 929, test: 1,785)
• Source documents: 73 official SCIO NPS test booklets (PDF)
• Answer format: single correct answer, typically 4 options (also 3, 5, or 8 for some question types)
• All questions in this release are text-only (requires_image = False).

Supported Tasks

• Multiple-choice question answering — benchmarking Slovak LLMs on university-admission-level reasoning and subject knowledge.
• Subject-specific evaluation — filter by source_file prefix (VSP, BIO, MAT, CHE) to test analytical reasoning, biology, mathematics, or chemistry separately.

Dataset Structure

Data Instances

{
    "question_number": "14.",
    "question": "Pri výskume infekčných chorôb v laboratórnych podmienkach ...",
    "answer": "(B) kultivácia vírusov vo vodnom roztoku s obsahom cukrov a anorganických látok",
    "year": "",
    "source_file": "BIO_NPS_2020_2021_T5.pdf",
    "reference": "None",
    "options": [
        "(A) kultivácia baktérií vo vodnom roztoku s obsahom cukrov",
        "(B) kultivácia vírusov vo vodnom roztoku s obsahom cukrov a anorganických látok",
        "(C) ...",
        "(D) ..."
    ],
    "page": 5,
    "requires_image": False
}

Data Fields

Field	Type	Description
question_number	string	Question number within the source test booklet.
question	string	Question text in Slovak (may contain line breaks from PDF extraction).
options	list[string]	Answer options, usually labelled (A)–(D) (3–8 options depending on the question type).
answer	string	The correct option (verbatim copy of one of options, including its label).
year	string	School year of the test (empty when encoded in source_file instead).
source_file	string	Name of the source PDF booklet; its prefix encodes the subject (VSP, BIO, MAT, CHE) and the school year.
reference	string	Reference passage/context, if any (almost always "None" in this release).
page	int64	Page number of the question in the source PDF.
requires_image	bool	Whether the question needs a figure to be answerable (False for all included questions — image-dependent questions were excluded).

Data Splits

Split	Examples
dev	929
test	1,785

How to Use

from datasets import load_dataset

ds = load_dataset("slovak-nlp/scio-testy")

print(ds["test"][0])

# Filter mathematics questions only
mat = ds["test"].filter(lambda x: x["source_file"].upper().startswith("MAT"))

Dataset Creation

Questions were automatically extracted from publicly released Slovak-language SCIO NPS test booklets (PDF) together with their official answer keys, then parsed into a structured format. Questions requiring an accompanying figure were filtered out (requires_image = False throughout). Minor text artifacts from PDF extraction (line breaks, hyphenation) may remain.

Considerations for Using the Data

• The correct answer is a verbatim copy of one of the options (including the (A)–(D) label), so accuracy can be computed by exact string match.
• Some questions may retain PDF-extraction artifacts (line breaks, diacritics irregularities).
• The source test materials are authored and published by Scio (www.scio.cz / www.scio.sk). This dataset is distributed for research and evaluation purposes; please verify the terms of use of the original materials before any other use.

Citation

If you use this dataset, please cite it as:

@misc{scio_testy_slovak,
  title        = {SCIO Tests (Slovak)},
  author       = {{Slovak NLP}},
  year         = {2025},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/slovak-nlp/scio-testy}}
}