20 years on: the Dexter Model of Hypertext and its impact on web accessibility
Robert Dodd · 2008 · SIGACCESS Access. Comput. · doi:10.1145/1340779.1340780
Summary
This paper examines the Dexter Reference Model of Hypertext, published in 1988, and asks whether the foundational assumptions it encodes still serve the needs of accessibility twenty years later. The Dexter Model was created as a superset description of what hypertext systems could do, capturing concepts like components, links, anchors, and presentation specifications. HTML, as one concrete hypertext notation, implements only a subset of the Dexter Model — notably lacking features like bidirectional links, runtime link resolution, and multiple simultaneous presentation specifications. The paper walks through the Dexter Model systematically, mapping each element to its HTML counterpart and identifying where HTML falls short from an assistive technology perspective. It also examines the Amsterdam Model of Hypermedia, which extended Dexter with synchronization arcs and channels to handle multimedia — concepts that become directly relevant when considering captioning, audio description, and multi-modal content delivery. A key concern is that scripting in HTML (particularly the rise of Ajax) obfuscates the underlying component structure that the Dexter Model describes. When navigation and content are constructed dynamically through JavaScript, assistive technologies lose visibility into the semantic relationships between components. The paper uses practical examples including Google Suggest, bus timetable websites, and template-driven content management systems to illustrate how real-world web development obscures the hypertext model from assistive technology.
Key findings
The paper identifies three fundamental limitations of the Dexter Model for accessibility. First, the model ties navigation to a fixed decomposition of content into components — once content is split across pages, there is no mechanism to reassemble it for alternative presentations (e.g., different screen sizes or cognitive needs). Second, the model lacks a formal mechanism for synchronizing contemporaneous content, which matters when text-to-speech must coexist with existing audio or when captions need precise timing. Third, identifying and selecting alternate content (such as replacing a video with a text description) requires content duplication rather than semantic equivalence. To address these limitations, the paper proposes an alternative five-layer model that positions the Dexter Model within a broader context: a Content Layer (physical assets), a Semantics Layer (document composition and relationships), a Navigation Layer (the Dexter Model itself), and an Exception Layer (handling broken links and adaptive system interventions). This extended model reveals that HTML 5, while adding semantic elements like section and nav, still concentrates improvements within individual pages rather than addressing cross-page or cross-component relationships.
Relevance
This paper provides a theoretical foundation for understanding why assistive technology often struggles with modern web applications. The core argument — that the web's foundational hypertext model was designed for content navigation, not content adaptation — remains relevant today. The rise of single-page applications, component-based frameworks, and dynamic rendering has only deepened the gap between what the underlying model describes and what assistive technology can perceive. For practitioners, the five-layer model offers a useful lens for analyzing where accessibility breaks down: is the issue at the content level (missing alt text), the semantic level (unclear relationships), the navigation level (inaccessible routing), or the exception level (broken states)? The paper's observations about scripting obfuscating the component model directly foreshadow the challenges that WAI-ARIA was designed to address.
Tags: hypertext models · HTML · assistive technology · adaptive content · web standards
Standards referenced: HTML 5 · WAI · SMIL