Dataset Card for CSL-Clinic

CSL-Clinic is a Chinese Sign Language dataset for clinical and healthcare-related communication. Each example pairs a sign-language video with a gloss sequence and a Chinese text sentence, supporting research on sign language recognition, sign language translation, and accessible medical communication.

Dataset Details

Dataset Summary

The dataset contains short video clips of Chinese Sign Language expressions in clinical consultation scenarios. The annotations include:

• file_name: relative path to the video file
• gloss: segmented sign gloss sequence
• text: corresponding Chinese natural-language sentence

Uses

Direct Use

This dataset can be used for:

• Chinese Sign Language recognition
• Sign language translation from video to Chinese text
• Gloss-to-text modeling
• Multimodal learning with video and text
• Research on accessible communication in medical settings

Out-of-Scope Use

The dataset should not be used as a substitute for professional medical advice, clinical diagnosis, or emergency communication. Models trained on this dataset may make translation or recognition errors and should be carefully validated before being used in any real healthcare workflow.

Dataset Structure

The dataset is organized into three splits:

Split	Videos	Annotation file
train	5,000	train/train.csv
dev	472	dev/metadata.csv
test	500	test/metadata.csv

Each split stores videos under a video/ directory. Annotation rows reference videos through paths such as:

video/example.mp4

Data Fields

• file_name: path to the video file within the split directory
• gloss: Chinese Sign Language gloss annotation
• text: Chinese sentence aligned with the video

Data Splits

The current release contains 5,972 annotated video examples:

train: 5000
dev:    472
test:   500

Dataset Creation

Source Data

The videos and annotations are intended to represent clinical consultation questions and healthcare-related communication in Chinese Sign Language.

Annotation Process

Each video is paired with a gloss-level annotation and a Chinese text sentence. The gloss field provides a sign-oriented representation, while the text field provides the corresponding natural-language Chinese sentence.

Considerations

Social Impact

CSL-Clinic is designed to support research toward more accessible healthcare communication for Chinese Sign Language users. Work using this dataset should account for the high cost of mistakes in medical contexts and include human review when deployed beyond research.

Biases and Limitations

• The dataset may not cover all Chinese Sign Language variants, signer styles, medical specialties, or regional expressions.
• Performance measured on this dataset may not generalize to real clinical environments.
• Video quality, signer identity, background, and recording conditions may affect model behavior.
• Medical-domain language can be sensitive and high-stakes; model outputs should be treated as assistive rather than authoritative.

Privacy

Users of the dataset should avoid attempting to identify signers or infer private personal attributes. Any downstream release of models, predictions, or derived data should preserve signer privacy.

Citation

If you use this dataset, please cite the dataset repository:

@misc{csl_clinic,
  title = {CSL-Clinic},
  author = {Zhao, R.},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/rzhao/CSL-Clinic}}
}

Contact

For questions about the dataset, please use the Hugging Face repository discussion page.