README.md

---
license: cc-by-4.0
---
# Cross-Cultural Meme Transcreation Dataset

## 📄 **Read our paper: [Beyond Translation: Cross-Cultural Meme Transcreation with Vision-Language Models](https://arxiv.org/pdf/2602.02510)**

## Dataset Summary

This dataset contains **6,315 meme pairs** for cross-cultural meme transcreation between English/American and Chinese cultures. It includes both bidirectional translations, making it valuable for research in multimodal AI, cultural adaptation, and humor translation.

**Key Statistics:**
- **English → Chinese**: 3,150 meme pairs
- **Chinese → English**: 3,165 meme pairs
- **Total**: 6,315 image pairs 

## Dataset Description

This dataset was created to study how internet memes can be effectively adapted across cultures while preserving humor, intent, and cultural relevance. Each meme pair consists of:
- Original meme (source culture)
- Transcreated meme (target culture)
- Comprehensive evaluation metrics
- Cultural adaptation analysis

### Languages

- **English** (en): American internet culture and meme conventions
- **Chinese** (zh): Chinese internet culture and meme conventions

## Dataset Structure

### Data Fields

Each meme pair includes the following fields:

#### Core Information
- `index`: Unique identifier for each meme pair
- `subfolder_name`: Folder name containing the image pair
- `english_image`: Path to the English/original meme image
- `chinese_image`: Path to the Chinese/transcreated meme image
- `source_culture`: Original culture (English/American or Chinese)
- `target_culture`: Target culture (Chinese or English/American)
- `timestamp`: Evaluation timestamp

#### Evaluation Metrics (1-5 scale)
- `caption_quality`: Quality and naturalness of the transcreated text
- `image_quality`: Visual quality and appropriateness of the image
- `synergy`: How well the image and caption work together
- `cultural_fit`: Appropriateness and resonance with target culture
- `intent_preservation`: How well the original intent/humor is preserved
- `overall_score`: Composite quality score (1-5)

#### Qualitative Analysis
- `raw_evaluation`: Full detailed evaluation text
- `offensiveness`: Offensiveness assessment
- `usability`: Practical usability in target culture
- `strengths`: Key strengths of the transcreation
- `weaknesses`: Areas for improvement
- `recommendations`: Suggestions for enhancement

### Data Splits

The dataset is organized by translation direction:

| Split | Meme Pairs | Description |
|-------|-----------|-------------|
| EN→CH | 3,150 | English/American memes transcreated to Chinese |
| CH→EN | 3,165 | Chinese memes transcreated to English/American |


## Considerations for Using the Data

### Social Impact

This dataset captures cultural nuances and humor styles from both American and Chinese internet cultures. Users should be aware that:
- Meme content reflects internet culture at the time of collection
- Cultural interpretations may vary within each culture
- Some memes may contain cultural references that require context

### Limitations

- Memes represent snapshot of internet culture (2024-2025)
- Cultural fit scores reflect general patterns, not individual preferences


## Citation

---
If you find this work helpful, please cite:

**BibTeX:**
```bibtex
@article{zhao2026beyond,
  title={Beyond Translation: Cross-Cultural Meme Transcreation with Vision-Language Models},
  author={Zhao, Yuming and Zhang, Peiyi and Ignat, Oana},
  journal={arXiv preprint arXiv:2602.02510},
  year={2026}
}
```

---



## License

[CC BY 4.0]

## Contact

For questions or feedback about this dataset, please contact:
- [yzhao4@scu.edu]
- [https://github.com/AIM-SCU/MemeXGen]


## Acknowledgments

This dataset was created as part of research on cross-cultural AI and multimodal transcreation. We acknowledge the use of large language models and image generation systems in creating the transcreated memes and evaluation framework.