Loading the case study

Patent · US 9,268,753 B2 · Apollo Education Group

Automated accessibilityfeatures in documents

A system that reads a document the way a screen reader would, finds what will be misread, and automatically produces an accessibility-enabled version alongside the original.

Role

Co-inventor, with Heidi Jungel

Filed / granted

Oct 24, 2011 · Feb 23, 2016

Assignee

Apollo Education Group (later The University of Phoenix)

01 — The problem

Screen readers were reading the layout, not the content

  • Tags read aloud. Screen readers could fail to separate formatting tags, such as layout tables, from the content they display.
  • Layout announced as data. A page laid out with a table was announced as "table, 2 columns, 3 rows" before a single word of content.
  • Illogical reading order. Cells were read in table order, not the order a sighted reader would follow.
  • Meaning carried by looks alone. Required fields shown only with an asterisk or red text, with nothing an assistive technology could detect.
Before
<table>
  <tr>
    <td><b>Name *</b></td>
    <td><input></td>
  </tr>
</table>

“Table with 2 columns and 1 row. Name star. Edit text.”

After
<div class="row">
  <label for="name">Name</label>
  <input id="name"
    aria-required="true">
</div>

“Name, required, edit text.”

02 — How it works

Analyze, model, produce, serve

The system first analyzes a document for tags and patterns known to trip up specific assistive technologies, from layout tables to formatting applied without CSS to controls that can’t be reached by keyboard. It builds an element-based model of the page, then produces a second, accessibility-enabled version, either by modifying the original or generating a new one. The server stores both and returns whichever a request calls for.

Analyzefind what assistive tech will misreadModelbuild an element tree (DOM)Producegenerate the accessibility-enabled versionServereturn either version on request

03 — What gets added

Eight kinds of automatic repair

Tag replacement

Formatting that assistive technology misreads is swapped for tags it understands; layout tables are rebuilt with <div> and <span>.

Classes and ids

Elements get classes, styled through generated CSS, and ids so they can be referenced, such as a heading id built from its text.

WAI-ARIA attributes

aria-required, aria-labelledby, aria-describedby, aria-live, aria-atomic and aria-relevant are added where the page needs them.

Link titles

Links without a title get one generated automatically.

Decorative images

Design-only images get alt="" so a screen reader skips them instead of reading the file name.

Structure and landmarks

Labels, fieldsets, legends and landmarks are added so the page can be navigated by keyboard.

Keyboard shortcuts

Shortcuts such as Esc to close a pop-up window.

Human review

The system can produce a draft for a person to review and correct before it ships.

04 — What’s claimed

45 claims, one core idea

The independent claims describe a server that keeps a first version of a document, generates an accessibility-enabled version with features the first lacked for a particular assistive technology, stores it, and answers each request with either version. A parallel claim covers the same steps as instructions on computer-readable media.

The dependent claims narrow it to tag replacement, parsing into a DOM-style tree, generated style sheets, modifying versus copying the original, adding labels and identifiers, keyboard navigation between elements, applying the process across a whole site, and running it in the browser or on request.

05 — Why it still matters

Accessibility as a system, not a checklist

The idea underneath the patent is the one I still design by: accessibility belongs in the system, applied automatically and consistently, with people reviewing the result, rather than retrofitted page by page. It is the same thinking behind WCAG 2.1 AA being built into every Radial component.

Source: US 9,268,753 B2 on Google Patents. Google lists the patent as active with an adjusted expiry of April 13, 2034; that status is Google’s assumption, not a legal finding.