A Semantic Transcoding System to Adapt Web Services for Users with Disabilities
Anita W. Huang, Neel Sundaresan · 2000 · Proceedings of the Fourth International ACM Conference on Assistive Technologies (Assets '00) · doi:10.1145/354324.354363
Summary
This paper from IBM Almaden Research Center presents Aurora, a semantic transcoding system that adapts web-based services for users with disabilities by extracting the meaning of web content rather than simply reformatting its syntax. The key problem Aurora addresses is that most web pages are designed for visual display in graphical browsers, with their semantics implicit in the visual layout rather than explicitly encoded. When users with disabilities access these pages through screen readers, voice browsers, or simplified interfaces, the meaning is often lost. Unlike previous transcoding approaches that operated on individual pages by rearranging HTML elements, Aurora works at the service level — it understands the multi-page workflow of web services like auction sites or search engines. The system uses a schema-driven framework where "service domain schemas" define the abstract steps a user takes to complete a task (e.g., find an item, place a bid, confirm the bid), and XML-transformation rules map specific web pages to these abstract steps. Aurora acts as a web intermediary, intercepting HTTP requests, downloading pages, extracting relevant data into structured XML fragments using PatML pattern-matching rules, and then passing these fragments to interface adaptors that generate customised output for different user needs.
Key findings
Aurora demonstrated three main accessibility benefits through its semantic approach. First, consistency: by defining a common schema for each service domain (e.g., auctions), Aurora produces a similar functional interface across all websites in that domain (eBay, Yahoo Auctions, etc.), so users learn one interaction model rather than navigating different layouts. Second, simplicity: Aurora strips away irrelevant content (advertisements, site maps, banners) and presents only data relevant to the user's current goal, dramatically reducing the information users must process. Third, adaptability: because content is extracted into standardised XML, multiple interface adaptors can generate different presentations — including a text-only HTML adaptor for screen reader users and an icon-enhanced adaptor for users with motor impairments or reading disabilities that annotates each step with graphic icons showing the function of page objects. The prototype was implemented in Java as a plug-in for IBM's Web Intermediaries (WBI) infrastructure and defined schemas for auction and search engine domains, with transformation rules for eBay, Yahoo Auctions, AltaVista, Yahoo Search, and Google.
Relevance
Aurora's approach anticipated several ideas that became central to web accessibility and web development in the following decades. The separation of content semantics from presentation foreshadowed the semantic web movement and the push for structured data on the web. The concept of understanding web pages not as isolated documents but as steps in a service workflow relates to modern task-oriented web accessibility, where the goal is not just making individual pages readable but enabling users to complete real tasks. The system's extensible adaptor architecture — allowing disability-specific experts to create custom interface transformations — was a forward-thinking model for personalised accessibility. However, Aurora's main limitation was scalability: creating schemas and transformation rules for each web service required significant manual effort, and rules broke whenever sites changed their HTML structure. This tension between semantic understanding and maintenance cost remains relevant today, though modern approaches using AI and machine learning to understand web page structure are beginning to address it.
Tags: web accessibility · web transcoding · content adaptation · XML · semantic web · assistive technology · web navigation · adaptive interface
Standards referenced: WCAG 1.0