The CISNA model of accessible adaptive hypermedia
Robert Dodd, Steve Green, Elaine Pearson · 2008 · Proceedings of the 2008 International Cross-Disciplinary Conference on Web Accessibility (W4A) · doi:10.1145/1368044.1368052
Summary
This paper presents CISNA (Content, Inventory, Semantics, Navigation, Adaptation), a five-layer document model designed to make web content adaptable to individual user needs and capabilities. The research addresses a fundamental problem: as web pages become increasingly script-driven — particularly with the rise of Ajax — assistive technologies like screen readers lose the ability to understand and interact with the content. The Dexter Model of Hypertext, which underpins HTML, describes navigation between and within documents but treats content structure as opaque. The Amsterdam Model extended Dexter with multimedia synchronization but still lacks mechanisms for adapting content across different modalities and user needs. CISNA decomposes a document into five independent layers of increasing abstraction. The Content Layer holds raw media assets externally. The Inventory Layer indexes all content and separates raw from formatted media elements, enabling the same content to be presented differently (as text, speech, or Braille). The Semantics Layer uses a rule-based meta-model inspired by Prolog, with nouns, verbs, ontologies, and statements that describe what content means and how it relates to other content. The Navigation Layer describes how users move between navigable nodes and views. The Adaptation Layer applies named instances — ordered transactions of add/modify/delete operations — that transform the document to match a user profile. Bridges between adjacent layers express the relationships without creating dependencies, allowing each layer to be modelled and modified independently.
Key findings
The paper demonstrates CISNA through a working Java prototype that adapts a fragment of Google Maps between two user profiles: a default visual user and a low vision user requiring text-to-speech augmentation. The prototype renders both an abstract model view and a concrete output, with the low vision profile producing spoken output such as "Application Google Maps. Menu Start. Item Web. Item Images. Item More. End Menu." The Semantics Layer's rule-based approach, using conditional multiple inheritance, allows content relationships to be formally validated — analogous to XML schema validation but for semantic structure. A key architectural insight is that adaptations are expressed as simple transactions (instances) rather than as entirely separate document versions, meaning the difference between a default interface and an adapted one can be precisely described and potentially measured. The paper suggests Levenshtein Distance as a possible metric for quantifying accessibility adaptations by comparing XML representations of different user profile instances.
Relevance
CISNA tackles a problem that remains largely unsolved in mainstream web development: how to formally describe and automate the adaptation of content for users with different abilities. While modern approaches like ARIA and responsive design address parts of this problem, they operate within the constraints of HTML rather than offering a complete separation of content, meaning, navigation, and adaptation. The model's concept of named adaptation instances — "Default to Low Vision User", "Blind User to Blind User with Restricted Mobility" — offers a way to think about accessibility as composable transformations rather than fixed alternative versions. For practitioners, the key takeaway is that truly adaptive interfaces require semantic knowledge that goes well beyond what HTML and ARIA currently provide. The prototype, while modest in scope, demonstrates that a formal model can produce working adaptations across modalities.
Tags: adaptive hypermedia · abstract user interface · hypertext models · assistive technology · content adaptation
Standards referenced: HTML · SMIL · XHTML · XForms · SVG