NormGenesis

This repo provides the dataset introduced by our EMNLP 2025 paper "NormGenesis: Multicultural Dialogue Generation via Exemplar-Guided Social Norm Modeling and Violation Recovery".

---

🗓️ Update History

• 🚀 November 2025: Code, Dataset, and Project Page have been released!
• 🏆 November 2025: We received the SAC Highlights Award at EMNLP 2025!
• 🏆 October 2025: Nominated for Outstanding Paper, SAC Highlight, and Resource Paper Award at EMNLP 2025.
• 📘 September 2025: Our paper was released on arXiv.
• 🎤 August 2025: Selected for Main Conference (Oral Presentation) at EMNLP 2025.

---

🎯 Project Overview

NormGenesis introduces a framework for multicultural dialogue generation that models social norms and recovers from violations using exemplar-guided methods.

Key Components

• Dataset: Multicultural social norm datasets (American, Chinese, Korean).
• Generation: Pipelines for generating dialogues that adhere to or violate social norms.
• Evaluation: Comprehensive evaluation of dialogue quality including consistency, naturalness, relevance, and norm appropriateness.

---

📁 Repository Structure

NormGenesis/
├── 📊 dataset/                 # Dataset files
│   ├── American/               # American culture dataset
│   ├── Chinese/                # Chinese culture dataset
│   └── Korean/                 # Korean culture dataset
├── 🔬 evaluation_code/         # Evaluation scripts
│   ├── evaluation_dialogue_quality.py    # Dialogue quality evaluation
│   └── evaluation_refinement_quality.py  # Refinement quality evaluation
├── 🏭 generation_code/         # Generation pipelines
│   ├── American/               # American generation scripts
│   ├── Chinese/                # Chinese generation scripts
│   ├── Korean/                 # Korean generation scripts
│   └── refine_situation.py     # Situation refinement script
├── 📄 labeling_dialogue.py     # Dialogue labeling script
└── 📄 README.md

---

🚀 Quick Start

Prerequisites

Ensure you have Python installed and the following dependencies:

pip install openai pandas tenacity tqdm

Setup

Set your OpenAI API key as an environment variable:

export OPENAI_API_KEY="your-api-key"

Usage

1. Generation: Navigate to generation_code and use the scripts for specific cultures or refine_situation.py to refine scenarios. Note: You may need to update input/output paths in the scripts.

2. Evaluation: Use evaluation_code/evaluation_dialogue_quality.py to assess generated dialogues. Note: Ensure you configure the evaluation parameters in the script.

---

🌍 Evaluation Scope

Cultures Covered

• 🇺🇸 American
• 🇨🇳 Chinese
• 🇰🇷 Korean

Evaluation Metrics

• Consistency: Logical and contextual consistency.
• Naturalness: Fluency and native-like expression.
• Relevance: Alignment with the scenario and situation.
• Coherence: Logical flow from scenario to dialogue.
• Emotion Appropriateness: Matching emotional tone.
• Social Norm Appropriateness: Adherence to social norms.

---

🚀 Release Plan

We plan to release both the code and dataset after the EMNLP 2025 conference.

• [✅] Code
• [✅] Dataset
• [✅] Project Page

---

📄 Paper

You can find our paper at the following link:
👉 https://arxiv.org/abs/2509.18395

---

📚 Citation

If you use this dataset or code in your research, please cite our paper:

@inproceedings{hong2025normgenesis,
  title={NormGenesis: Multicultural Dialogue Generation via Exemplar-Guided Social Norm Modeling and Violation Recovery},
  author={Hong, Minki and Choi, Jangho and Kim, Jihie},
  booktitle={Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing},
  pages={33781--33819},
  year={2025}
}