---
language:
  - bo
  - mn
  - ug
license: cc-by-4.0
dataset_info:
  features:
    - name: id
      dtype: string
    - name: title
      dtype: string
    - name: content
      dtype: string
    - name: title_match_1
      dtype: float64
    - name: title_match_2
      dtype: float64
    - name: tendency
      dtype: string
    - name: average_score
      dtype: float64
    - name: score_difference
      dtype: float64
  splits:
    - name: train
      num_bytes: 1415800000
      num_examples: 200000
  task_categories:
    - text-generation
  task_ids:
    - headline-generation
---
# CMHG Dataset

[Paper](https://arxiv.org/abs/2509.09990)

## Dataset Description

The CMHG (Chinese Minority Headline Generation) dataset contains headline generation data for three minority languages in China:
- Tibetan: 100,000 entries
- Mongolian: 50,000 entries
- Uyghur: 50,000 entries

This dataset is designed to support research and development in headline generation for these languages, providing a valuable resource for natural language processing tasks in low-resource languages.

## Annotation Process

For quality control, we annotated 3,000 entries for each language. Each entry was evaluated by two annotators who provided scores for the following attributes:
- `title_match_1`: First annotator's assessment of title-content relevance
- `title_match_2`: Second annotator's assessment of title-content relevance
- `tendency`: Sentiment or tendency classification
- `average_score`: Average score from both annotators
- `score_difference`: Difference between the two annotators' scores

## Data Quality Classification

Based on the annotation results, we classified the data into two quality categories:
- **High-quality data**: Entries with an average score of 4 or higher (`average_score_4_or_higher.csv`)
- **Lower-quality data**: Entries with an average score below 4 (`average_score_below_4.csv`)

This classification helps researchers and developers select appropriate data for their specific use cases, ensuring they work with data that meets their quality requirements.

## Directory Structure

The dataset is organized into language-specific directories:

- `bo/`: Tibetan language data
  - `average_score_4_or_higher.csv`: High-quality Tibetan data
  - `average_score_below_4.csv`: Lower-quality Tibetan data
  - `bo-all.csv`: Complete Tibetan dataset

- `mn/`: Mongolian language data
  - `average_score_4_or_higher.csv`: High-quality Mongolian data
  - `average_score_below_4.csv`: Lower-quality Mongolian data
  - `mn-all.csv`: Complete Mongolian dataset

- `ug/`: Uyghur language data
  - `average_score_4_or_higher.csv`: High-quality Uyghur data
  - `average_score_below_4.csv`: Lower-quality Uyghur data
  - `ug-3.csv`: Complete Uyghur dataset

## Data Format

All CSV files follow the same structure with these columns:
- `id`: Unique identifier for each entry
- `title`: Generated headline
- `content`: Original content/text
- `title_match_1`: First annotator's relevance score
- `title_match_2`: Second annotator's relevance score
- `tendency`: Sentiment/tendency label
- `average_score`: Average quality score
- `score_difference`: Difference between annotator scores



## License

This dataset is available under the CC BY 4.0 license for research and development purposes.

## Citation

If you use this dataset in your work, please cite the following paper:

```bibtex
@inproceedings{xu-etal-2025-cmhg,
    title = "{CMHG}: A Dataset and Benchmark for Headline Generation of Minority Languages in {C}hina",
    author = "Xu, Guixian  and
      Su, Zeli  and
      Zhang, Ziyin  and
      Liu, Jianing  and
      Han, Xu  and
      Zhang, Ting  and
      Dong, Yushuang",
    editor = "Christodoulopoulos, Christos  and
      Chakraborty, Tanmoy  and
      Rose, Carolyn  and
      Peng, Violet",
    booktitle = "Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing",
    month = nov,
    year = "2025",
    address = "Suzhou, China",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.emnlp-main.622/",
    doi = "10.18653/v1/2025.emnlp-main.622",
    pages = "12350--12357",
    ISBN = "979-8-89176-332-6"
}
```