README.md

metadata

configs:
  - config_name: kyrgyz_rc_all
    data_files:
      - split: validation
        path: kyrgyz_rc_all/validation-00000-of-00001.parquet
      - split: test
        path: kyrgyz_rc_all/test-00000-of-00001.parquet
  - config_name: kyrgyz_rc_literature
    data_files:
      - split: validation
        path: kyrgyz_rc_literature/validation-00000-of-00001.parquet
      - split: test
        path: kyrgyz_rc_literature/test-00000-of-00001.parquet
  - config_name: kyrgyz_rc_math
    data_files:
      - split: validation
        path: kyrgyz_rc_math/validation-00000-of-00001.parquet
      - split: test
        path: kyrgyz_rc_math/test-00000-of-00001.parquet
  - config_name: kyrgyz_rc_news
    data_files:
      - split: validation
        path: kyrgyz_rc_news/validation-00000-of-00001.parquet
      - split: test
        path: kyrgyz_rc_news/test-00000-of-00001.parquet
  - config_name: kyrgyz_rc_wiki
    data_files:
      - split: validation
        path: kyrgyz_rc_wiki/validation-00000-of-00001.parquet
      - split: test
        path: kyrgyz_rc_wiki/test-00000-of-00001.parquet
tags:
  - kyrgyz
  - llm-benchmark
  - evaluation
  - reading-comprehension
  - kyrgyzllm-bench
language:
  - ky

KyrgyzRC (Kyrgyz Reading Comprehension)

KyrgyzRC is a native, multiple-choice reading comprehension benchmark for the Kyrgyz language. It contains 400 questions designed to test factual understanding, inference, and reasoning.

🏔️ Part of the KyrgyzLLM-Bench

This dataset is a core component of the KyrgyzLLM-Bench, a comprehensive suite for evaluating LLMs in Kyrgyz.

• Main Paper: Bridging the Gap in Less-Resourced Languages: Building a Benchmark for Kyrgyz Language Models
• Hugging Face Hub: https://huggingface.co/TTimur
• GitHub Project: https://github.com/golden-ratio/kyrgyzLLM_bench

📋 Dataset Description

This benchmark evaluates a model's ability to understand and reason about text passages. Each question has 4 options, with 1 correct answer.

• Sources: Passages are sourced from Kyrgyz Wikipedia, national news articles, literature, and school-style math word problems.
• Skills Evaluated: Factual understanding, inference, vocabulary-in-context, and multi-sentence reasoning.

📜 Citation

If you find this dataset useful in your research, please cite the main project paper:

@article{KyrgyzLLM-Bench,
  title={Bridging the Gap in Less-Resourced Languages: Building a Benchmark for Kyrgyz Language Models},
  author={Timur Turatali, Aida Turdubaeva, Islam Zhenishbekov, Zhoomart Suranbaev, Anton Alekseev, Rustem Izmailov},
  year={2025},
  url={[https://huggingface.co/datasets/TTimur/kyrgyzRC]}
}