Machine Learning applications in accessibility
summary
VMware builds vA11y (code-named vli) to automate currently manual accessibility tests, pushing the 30% automated / 70% manual split toward 50/50 by end of 2021. Prototype automates five checks: captions/transcripts, keyboard focus presence and contrast, and heading-content match; four run at 100% reliability and heading matching runs at ~88% accuracy. Models train with supervised learning on 1 → 30 → 300 sites, discover real bugs (two thirds of focus indicators broken on one homepage), and ship as an open-source tool with API and data models.
AI-written summary · about
chapters (13)
- 0:00Introduction to talk
- 0:48Current accessibility testing state
- 7:30Disability and ML intersection
- 9:15WCAG and automated testing limits
- 11:00Project origin and VMware pitch
- 13:31Open source and expansion plans
- 16:42Five automated accessibility checks
- 20:32Testing methodology for ally
- 21:58Demo of keyboard focus detection
- 24:22Demo of contrast measurement
- 26:14Using and contributing to VLI
- 27:39How to get involved
- 28:01Closing and Q&A
related talks
38mAutomated vs. Manual Accessibility Testing: Finding the Right Balance
37mAccessibility Testing Techniques, Issues and Tricks
56mAccessibility Testing 101
33mAutomated Accessibility Testing
25mYou can’t read this sentence - A11y automation
49mProgramming for Accessibility
34mAutomation meets UX: Perfect Partners for Accessibility
44mTesting Web Accessibility