2M-Flores-ASL

# 2M-Flores

As part of the [2M-Belebele](https://github.com/facebookresearch/belebele) project, we have produced video recodings of ASL signing for all the `dev` and `devtest`
sentences in the original [flores200](https://github.com/facebookresearch/flores/tree/main/flores200) dataset.

To obtain ASL sign recordings, we provide translators of ASL and native signers with the English text version of the sentences to be recorded.
The interpreters are then asked to translate these sentences into ASL, create glosses for all sentences, and record their interpretations into ASL one sentence at a time.
The glosses are subjected to an additional quality check by expert annotators to harmonize the glossing format. To harmonize the recording conditions and eliminate visual bias,
the videos are recorded against plain monochrome backgrounds (e.g., white or green), and signers are requested to wear monochrome upper body clothing (e.g., black).
All videos are captured in 1920x1080p resolution with all of the signing space covered in FOV. The recordings are done in 60 frames per second to address most of the motion
blur that happens during signing.

### Columns

- `id`: copied from flores
- `URL`: copied from flores
- `domain`: copied from flores
- `topic`: copied from flores
- `has_image`: copied from flores
- `has_hyperlink`: copied from flores
- `sentence`: copied from flores
- `gloss`: the gloss for the signed video
- `signer`: some sentences have multiple recordings, this is not a global id.

## Citation

If you use this data in your work, please cite the 2M-Belebele paper:

```bibtex
@article{2mbelebele,
author = {Marta R. Costa-jussà and Bokai Yu and Pierre Andrews and Belen Alastruey and Necati Cihan Camgoz and Joe Chuang and Jean Maillard and Christophe Ropers and Arina Turkantenko and Carleigh Wood},
journal = {Arxiv},
= {https://arxiv.org/abs/2412.08274},
title = {{2M-BELEBELE}: Highly-Multilingual Speech and American Sign Language
Comprehension Dataset},
year = {2024},
}
```

## License

2M-Flores is released under CC-BY-SA4.0, it is composed based on Flores200 (CC-BY-SA 4.0).

# 2M-Flores

作为[2M-Belebele](https://github.com/facebookresearch/belebele)项目的组成部分，我们为原始[flores200](https://github.com/facebookresearch/flores/tree/main/flores200)数据集中所有`dev`与`devtest`语句制作了美国手语（American Sign Language, ASL）手语视频录制。

为获取手语录制素材，我们向美国手语译员及母语手语使用者提供待录制语句的英文文本。随后要求译员将这些语句译为美国手语，为所有语句编写手势标注（gloss），并逐句录制为美国手语视频。专业标注人员将对这些手势标注开展额外质量校验，以统一标注格式。为统一录制条件并消除视觉偏差，所有视频均以纯色单色背景（如白色或绿色）录制，同时要求手语者身着纯色上身衣物（如黑色）。所有视频均采用1920×1080p分辨率录制，覆盖手语动作的全部视觉空间，并以每秒60帧的帧率录制，以缓解手语动作过程中产生的大部分运动模糊。

### 字段说明
- `id`：源自flores数据集
- `URL`：源自flores数据集
- `domain`：源自flores数据集
- `topic`：源自flores数据集
- `has_image`：源自flores数据集
- `has_hyperlink`：源自flores数据集
- `sentence`：源自flores数据集
- `gloss`：对应手语视频的手势标注
- `signer`：部分语句存在多份录制素材，该字段并非全局唯一标识符。

## 引用
若您在研究工作中使用本数据集，请引用2M-Belebele相关论文：
bibtex
@article{2mbelebele,
author = {Marta R. Costa-jussà and Bokai Yu and Pierre Andrews and Belen Alastruey and Necati Cihan Camgoz and Joe Chuang and Jean Maillard and Christophe Ropers and Arina Turkantenko and Carleigh Wood},
journal = {Arxiv},
url = {https://arxiv.org/abs/2412.08274},
title = {{2M-BELEBELE}: Highly-Multilingual Speech and American Sign Language
Comprehension Dataset},
year = {2024},
}

## 许可协议
2M-Flores采用CC-BY-SA 4.0协议发布，其基于Flores200（CC-BY-SA 4.0）构建。