Datasets:

yxdu
/

ccfqa

language:
  - zho
  - eng
  - fra
  - jpn
  - kor
  - rus
  - spa
  - yue
license: cc-by-nc-sa-4.0
task_categories:
  - question-answering
  - audio-text-to-text
pretty_name: CCFQA
library_name: datasets
tags:
  - factuality
  - evaluation

CCFQA

CCFQA is a speech and text factuality evaluation benchmark that measures language models’ ability to answer short, fact-seeking questions and assess their cross-lingual and cross-modal consistency. It consists of speech and text in 8 languages, containing 1,800 n-way parallel sentences and a total of 14,400 speech samples.

Language: Mandarin Chinese, English, French, Japanese, Korean, Russian, Spanish, Cantonese(HK)
ISO-3 Code: cmn, eng, fra, jpn, kor, rus, spa, yue
Data Size: 14,400 sample
Data Split: Test
Data Source: Native speakers (6 males and 6 females)
Domain: Factuality Evaluation
Task: Spoken Question Answering(SQA)
License: CC BY-NC-SA-4.0

📄Paper：https://arxiv.org/abs/2508.07295

How to use

from datasets import load_dataset

ccfqa = load_dataset("yxdu/ccfqa")
print(ccfqa)

⚖️ Evals

please visit github page.

🖊Citation

@misc{du2025ccfqabenchmarkcrosslingualcrossmodal,
      title={{CCFQA}: A Benchmark for Cross-Lingual and Cross-Modal Speech and Text Factuality Evaluation}, 
      author={Yexing Du and Kaiyuan Liu and Youcheng Pan and Zheng Chu and Bo Yang and Xiaocheng Feng and Ming Liu and Yang Xiang},
      year={2025},
      eprint={2508.07295},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2508.07295}, 
}