The Recurring Problem
Across Cisco and Everfox, accessibility and localization had to survive the entire publishing path. The source could be semantically correct and still produce an inaccessible PDF or HTML page if a transform discarded meaning, generated the wrong navigation, or handled a table poorly.
Global output introduced a related set of constraints. A layout that behaved well in English could fail with right-to-left text, CJK typography, complex scripts, text expansion, or translated navigation. Accessibility and localization therefore belonged in the source model and transformation pipeline, not in separate cleanup projects at the end.
Where Systems Fail
The system needed to preserve meaning while changing formats. Alternative text, table descriptions, headings, navigation, reading order, generated labels, and page structures all behaved differently across HTML and PDF. Authors also needed a practical way to supply the required information before publication; a checker could report a missing description, but it could not invent the correct one.
Formatter choice created another tradeoff. Lower-cost tools could handle simpler output, but multilingual typography placed harder requirements on shaping, direction, and layout. The publishing stack had to be evaluated against the languages and accessibility behavior it actually needed to support.
How I Approach It
I added accessibility information such as alternative text and table descriptions to the structured source models, then reinforced those requirements with authoring controls and prompts. That moved the decision to the point where the writer still understood the content instead of deferring it until a final output failed a check.
XSLT and XSL-FO carried the semantics into tagged PDF and HTML. The transformation logic handled format-appropriate alternative text and accessible navigation, while custom XSL-FO controlled page masters, headers and footers, keeps, tables, generated text, numbering, bookmarks, conditional layouts, branding, and global output. Antenna House Formatter was used where multilingual typography requirements exceeded what lower-cost alternatives could support reliably.
Automated checks were only one part of validation. I tested PDF and HTML with Section 508 tools and manually with JAWS, Windows Narrator, and Android TalkBack. Contrast, structural behavior, zoom and reflow, and publishing across more than 50 languages—including right-to-left, complex-script, and CJK requirements—were part of the test surface.
What This Approach Makes Possible
Accessibility became a property of the publishing system rather than a remediation step. Source models captured the necessary information, authoring guidance exposed it early, transforms preserved it, and test plans verified the final behavior.
The same architecture supported global publishing at scale. Reuse and conditional-content strategies also limited unnecessary localization changes, so accessibility and internationalization worked with the content architecture rather than competing with it.
The Principle
This case shows why accessible output cannot be separated from content modeling, transformation engineering, tool selection, and testing. Section 508 requirements, semantic source, tagged output, assistive technology, XSLT and XSL-FO, and RTL and CJK publishing all became connected design constraints in one system.