J-Shuwa

J-Shuwa is a parallel corpus for Japanese Sign Language (JSL) and Japanese, collected from YouTube videos accessible as of approximately June 2023. It is designed to support research on Japanese Sign Language translation and related multimodal language tasks.

Because the original videos and associated textual content cannot be redistributed, this Hugging Face release provides only the redistributable metadata layer: YouTube video IDs, segment timestamps, and a source label indicating whether the segment was aligned from YouTube closed captions (cc) or from hard-coded subtitles (hardsub).

Dataset Summary

The current release contains a single split with:

• 198,866 segment references
• 6,331 unique YouTube videos
• Approximately 348.78 hours of referenced segment duration (with overlaps)
• 161,228 segments from hard subtitles (hardsub)
• 37,638 segments from closed captions (cc)

Data Fields

Each sample has the following fields:

Field	Type	Description
vid	string	Unique segment identifier, formatted from the YouTube video ID and segment timestamps.
yid	string	YouTube video ID.
start	float	Segment start time in seconds.
end	float	Segment end time in seconds.
source	string	Source of the subtitle cue used to define the segment: cc indicates a YouTube closed-caption file, and hardsub indicates subtitles extracted from the video via OCR.

Not Included in This Hugging Face Release

This Hugging Face release contains only segment-level metadata and annotations. It does not include or redistribute any YouTube videos, audio, frames, thumbnails, subtitles, transcripts, OCR text, translations, comments, or other copied media or textual content.

All referenced materials remain hosted on YouTube and are controlled by their respective creators, channels, or rights holders.

Limitations

Some referenced YouTube videos may become unavailable, private, deleted, region-restricted, age-restricted, or otherwise inaccessible over time. As a result, not every row is guaranteed to resolve to a playable video in the future.

The video selection were NOT performed by sign-language experts, so the dataset may reflect subjective visual judgments and may not fully represent language use within the Deaf community. In particular, YouTube videos may contain a mixture of Japanese Sign Language (JSL, 日本手話) and Manually Coded Japanese / Signed Japanese (MCJ/SJ, 日本語対応手話), which can look similar at the lexical level but differ in grammar and sentence-level structure. Users should consider this potential bias when training models, analyzing results, or drawing conclusions from the dataset.

Citation

If you use this corpus, please cite:

@inproceedings{mo-etal-2026-j,
    title = "{J}-Shuwa: A Large-Scale Web-Collected {J}apanese {S}ign {L}anguage-{J}apanese Parallel Corpus",
    author = "Mo, Junwen  and
      Vo, MinhDuc  and
      Nishida, Noriki  and
      Satoh, Shin{'}ichi  and
      Nakayama, Hideki",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
    month = jul,
    year = "2026",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.findings-acl.1821/",
    doi = "10.18653/v1/2026.findings-acl.1821"
}

License

The metadata and annotation layer created by the dataset maintainer, including segment identifiers, timestamps, and source labels, is released under the Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).

This license applies only to the metadata contained in this repository. It does not grant any rights to the referenced YouTube videos, audio, subtitles, transcripts, thumbnails, comments, OCR outputs, translations, or other third-party content.

All referenced YouTube videos remain owned by their respective creators, channels, or rights holders. Users who access or use the referenced videos or any related third-party content are solely responsible for doing so through YouTube or other authorized channels, and for complying with YouTube’s Terms of Service, applicable licenses, copyright law, privacy requirements, and local laws.