The SADIe Transcoding Platform
Darren Lunn, Sean Bechhofer, Simon Harper · 2008 · Proceedings of the 2008 International Cross-Disciplinary Conference on Web Accessibility (W4A) · doi:10.1145/1368044.1368073
Summary
This short paper presents the SADIe (Structural-Semantics for Accessibility and Device Independence) transcoding platform, developed at the University of Manchester. SADIe addresses a fundamental challenge for visually impaired screen reader users: web pages are designed primarily for visual presentation, but screen readers rely on underlying document structure rather than visual layout to generate audio output. The platform combines two existing approaches to transcoding — heuristic-based (applying predefined rules based on content patterns) and semantic (using manual annotations to identify element roles) — to create a system that is both accurate and scalable. The key insight is that CSS rendering information is closely associated with the intended role of each page element. Sighted users can infer that a styled list of links is a navigation menu from its visual presentation; SADIe captures this same semantic information by annotating CSS element roles within a two-part ontology. The upper ontology defines abstract concepts representing potential roles of web page elements, while website-specific lower ontologies map actual CSS classes to those abstract roles. This architecture means the upper ontology serves as a consistent interface — SADIe can query any number of site-specific ontologies through the same abstract role definitions, enabling scalable transcoding across diverse websites.
Key findings
SADIe performs three core transcoding operations: "Defluff" removes elements classified as providing little or no useful information (banners, advertisements, decorative elements); "Reorder" repositions important content areas to the top of the page for immediate screen reader access; and "Menu" moves the site navigation to the bottom of the page so it does not obstruct access to main content. These operations are driven by ontology queries — for example, the Defluff operation queries for all CSS classes classified as "Removable" and then traverses the DOM to remove matching elements. The paper demonstrates the approach using the ASSETS 2008 conference website, where the original page surrounded the main "Important Information" section with banners and navigation menus. After SADIe transcoding, the clutter was removed and the main content was promoted to the top. The CSS-based annotation approach provides high scalability because a single CSS annotation propagates to every page on a website that shares the same stylesheet, eliminating the need to annotate pages individually — a significant advantage over per-page semantic annotation approaches that prior research found to be "extremely tedious and time consuming."
Relevance
SADIe represents an important step in the evolution of web accessibility remediation, bridging the gap between scalable but imprecise heuristic approaches and accurate but labor-intensive manual annotation. The core principle — that visual presentation in CSS implicitly encodes semantic role information that can be made explicit for assistive technology — anticipated the later development of WAI-ARIA landmarks and HTML5 semantic elements, which formalized many of the same role concepts (navigation, main content, complementary) directly in markup. For accessibility practitioners today, SADIe offers a historical perspective on why semantic HTML and ARIA roles matter: they make explicit the same structural information that SADIe had to infer from CSS. The ontology-based architecture also demonstrates a pattern of separating accessibility rules from their application, allowing different transcoding strategies to use the same underlying semantic annotations — a flexibility that the same team later exploited by combining SADIe with Google AxsJAX framework for ARIA injection.
Tags: web accessibility · transcoding · screen readers · visual impairment · semantic web · CSS · ontology · assistive technology
Standards referenced: XHTML