hamiids/iranian-social-norms-dataset

Iranian Social Norms (ISN) Dataset

Overview

The Iranian Social Norms (ISN) dataset is a curated collection of social norms specific to Iranian culture. It provides a rich resource for studying cultural norms and their variations across different demographic groups. By incorporating demographic context, such as age, gender, ethnicity, and religion, this dataset enables a nuanced exploration of social expectations within Iranian society.

Dataset Structure

Each entry in the dataset includes:

• Norm Statement: A sentence describing a social norm in Farsi.
• English Translation: The corresponding norm translated into English.
• Context: A brief description of the situation in which the norm applies.
• Demographic Features: Information such as age, gender, ethnicity, and religion associated with the norm.
• Norm Label: One of three categories:
  • Expected: Socially encouraged behaviors.
  • Normal: Neutral behaviors that are commonly accepted.
  • Taboo: Actions that are widely disapproved of.
• Scope: Specifies whether the norm is specific to Iranian culture or more universally applicable.

Annotation Process

The dataset was annotated by three native Persian speakers, with different demographic backgrounds, ensuring diverse perspectives. The annotation process revealed interesting insights:

• Moderate overall agreement (Fleiss' Kappa = 0.558) highlights the subjective nature of norm interpretation.
• Gender differences influenced labeling, as the female annotator sometimes differed in her assessments from the two male annotators.
• Higher agreement on Taboo norms (Kappa = 0.7569) suggests that strong cultural and religious influences shape widely agreed-upon prohibitions.
• Lower agreement on Normal norms (Kappa = 0.3829) reflects flexibility and contextual variation in what is considered acceptable behavior.

Applications

This dataset can be used for:

• Cultural NLP research: Training and evaluating models to better understand Persian language and cultural context.
• Bias and fairness studies: Investigating how cultural norms affect AI model behavior.
• Cross-cultural comparisons: Studying differences between Iranian norms and those of other societies.
• LLM Benchmarking: Assessing how well large language models (LLMs) grasp Iranian cultural nuances.

Citation

If you use this dataset, please cite our work:

@inproceedings{your_citation,
  title={Can I introduce my boyfriend to my grandmother?
Evaluating Large Language Models Capabilities on Iranian Social Norm
Classification},
  author={Your Name and Co-Authors},
  booktitle={Conference Name},
  year={2024}
}

License

This dataset is released under [license type, e.g., CC BY 4.0].