Transcoding Proxy for Nonvisual Web Access
Hironobu Takagi, Chieko Asakawa · 2000 · Proceedings of the Fourth International ACM Conference on Assistive Technologies (Assets '00) · doi:10.1145/354324.354371
Summary
This paper presents a transcoding proxy system developed at IBM Japan that transforms existing web pages to be more accessible for blind users. The authors address the growing problem of web pages being designed with increasingly complex layouts that cram multiple functions and links into single pages to improve usability for sighted users, but which create significant barriers for nonvisual access. Rather than relying on web authors to make their pages accessible, the system works as an intermediary proxy between web servers and users, automatically transforming pages on the fly. The proxy architecture consists of five modules built on IBM WebSphere Transcoding Publisher: a portal module for user authentication, a simplification module that uses differential comparison to strip pages down to changed content, a full-text module that reorganizes content based on volunteer annotations, a user module that applies user-specified annotations, and an experience-based rules module that applies heuristics derived from nonvisual web access experience. The system uses three types of annotations: volunteer-specified annotations where sighted volunteers divide pages into visually fragmented groups and annotate their roles and importance; automatically-created annotations that insert alternative text for images and flag JavaScript issues; and user-specified annotations that allow blind users themselves to mark proper content starting points and select simplified forms. A key technical contribution is the simplification module, which uses Dynamic Programming matching to calculate differentials between two HTML documents, identifying only the changed content on frequently updated pages like news sites.
Key findings
The system demonstrates that proxy-based transcoding can make web pages more accessible without requiring changes from web authors, addressing the practical reality that most sites did not follow WAI guidelines despite their availability. The three-mode interface (simplification, full-text, and original) gives users control over how much transformation is applied, with simplification mode showing only changed content between page versions and full-text mode reorganizing the entire page based on annotated groupings. The differential method for finding changes between HTML documents uses a novel approach of converting DOM trees to serialized strings and applying longest common node string matching, with a sophisticated URL matching algorithm to find comparison pages even when URLs change over time. The experience-based rules module applies practical heuristics such as moving image maps and forms to the bottom of pages in full-text mode, and removing duplicate links that share the same URL. The volunteer annotation system was designed to be simple enough to encourage broad participation, while user-specified annotations empower blind users to customize their own experience without depending on sighted volunteers.
Relevance
This early research from IBM Japan represents a foundational approach to web accessibility that anticipated many challenges still relevant today: the tension between visual complexity and nonvisual access, the impracticality of relying solely on web authors for accessibility, and the value of intermediary tools that can retrofit accessibility onto existing content. The proxy-based approach prefigured modern accessibility overlays and browser extensions, though with more rigorous technical foundations. The annotation framework, which combines automatic, volunteer, and user-generated contributions, foreshadowed crowdsourced accessibility efforts. For practitioners, the paper demonstrates that accessibility can be improved through multiple complementary strategies rather than a single solution, and that empowering users with customization options is as important as automated fixes.
Tags: web accessibility · screen reader · transcoding · proxy · blind and low vision · non-visual access · annotations · content adaptation · web navigation
Standards referenced: WAI Guidelines