Accessibility Commons: A Metadata Repository for Web Accessibility
Shinya Kawanaka, Masatomo Kobayashi, Hironobu Takagi, Chieko Asakawa · 2009 · ACM SIGWEB Newsletter, Summer 2009 · doi:10.1145/1514984.1514985
Summary
This short SIGWEB Newsletter article proposes Accessibility Commons (AC), a shared federated repository for externally authored web-accessibility metadata. The authors argue that accessibility cannot scale as long as fixes depend on the original site owner inlining alt text, ARIA attributes, or landmark markers in server-side code — a bottleneck made worse by AJAX, heterogeneous runtimes, and governments or enterprises that cannot touch third-party sites. External metadata approaches (WebInSight, SADIe, aiBrowser, HearSay, Social Accessibility) circumvent this by describing fixes outside the page, but each research project has historically stored those descriptions in its own silo. AC is the result of a 2008 workshop between IBM, Washington, Stony Brook, and Manchester that agreed on a common metadata schema with four components — URI Addressing (which pages it applies to), Element Addressing (which DOM nodes), Condition (when it applies), and Semantics (what the fix says) — plus author and timestamp information. Clients post and retrieve metadata over HTTP in JSON; the server resolves wildcard URI patterns using an automaton-based index so a repository of ~100,000 records can be queried in microseconds. The paper walks through two consumers (Social Accessibility and HearSay), describes technical challenges of authoring quality and query performance, and sketches a federated future in which governments, mobile-transcoding services, and illiterate-user voice portals all draw on the same infrastructure.
Key findings
By May 2009, 400+ volunteer authors had contributed more than 19,000 metadata records through the Social Accessibility service alone, comprising roughly 12,500 alternative-text entries, 6,500 navigation-heading entries, and 400 landmark entries designating main content. Some contributors were blind users themselves, adding landmarks to pages they revisit or deducing alt text from link context. A site-wide authoring prototype allowed a single person to remediate 82 inaccessible pages — 3,500+ individual accessibility errors — in one hour, suggesting that pattern-based metadata authoring scales far beyond per-page annotation. The automaton-based URI index reduced lookup time to microseconds at 100,000 records, addressing the scaling problem that had blocked wildcard-URI storage in conventional databases. The Social Accessibility project is cited as proof that the proposed schema is expressive enough to support real production use, while HearSay's planned migration onto AC demonstrates cross-project reuse — one consumer's landmarks becoming another consumer's jump targets without re-annotation.
Relevance
Accessibility Commons anticipates the modern conversation about third-party remediation, overlay-style accessibility tools, and crowdsourced alt text (e.g., Be My Eyes, automated alt-text services, and the accessibility fixes that browser extensions and AI models now apply on the fly). For practitioners, the most transferable ideas are the four-part metadata schema (URI + element + condition + semantics) and the observation that declarative, client-agnostic fixes let one author serve many assistive technologies. The paper is also a useful historical anchor for understanding why later systems such as Sasayaki, BBeep's wayfinding work, and modern ARIA-injection tools took the shapes they did. Limitations include a lack of empirical quality-control data, no analysis of how federated repositories should resolve conflicting fixes, and only an aspirational treatment of governance, trust, and versioning — all of which have since become central concerns in third-party remediation debates.
Tags: web accessibility · metadata · semantic transcoding · social accessibility · crowdsourcing · screen readers · voice browser · assistive technology · blindness and low vision
Standards referenced: ARIA · Section 508