Hugging Face's logo

Datasets:
andrian-kr
/
ethics_sc101_pva

Modalities:
Text
Formats:
parquet
Size:
< 1K
Libraries:
Datasets
pandas
Dataset card Data Studio Files Community
1

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

⚠️ Disclaimer
This dataset contains morally and socially sensitive scenarios, including content that may be offensive, harmful, or reference illegal behavior. It is intended solely for research purposes related to value alignment, cultural analysis, and AI safety. Use responsibly.

ETHICS_SC101_PVA

ETHICS_SC101_PVA is a bilingual (English–Ukrainian) dataset of 84 morally unacceptable scenarios curated for evaluating passive value alignment in language models. It draws from:

Each entry contains:

  • query_en: Scenario in English
  • query_ukr: Ukrainian translation
  • source: Dataset of origin (ethics or sc_101)

Scenarios include morally sensitive topics such as violence, abuse, self-harm, and antisocial behavior.

Intended Uses

  • Evaluate LLM refusal and safety mechanisms
  • Benchmark cross-lingual value alignment
  • Support alignment research in low-resource languages
Downloads last month
64

Collection including andrian-kr/ethics_sc101_pva