Day 24: Privacy and Consent

Goal

Understand privacy risks in sign-language datasets.

Learn

• Sign-language video can reveal a person's face, body, identity, location, disability status, language community, and personal expression.
• Pose data can reduce some visual detail, but it is not automatically anonymous. Body motion, signer IDs, metadata, and source links can still be sensitive.
• Consent should explain what is collected, how it will be stored, who can access it, whether it can train models, and how removal requests work.

Example

• Consent checklist: purpose, dataset owner, recording type, storage location, access limits, model training permission, public demo permission, withdrawal process, contact person, license, retention period.
• A signer may consent to private research review but not public web demos. The metadata should preserve that distinction.

Practice

1. Write a consent checklist for collecting signer video.
2. Add one field that separates training permission from public display permission.

Checkpoint

Deaf-first note