The CISNA Model of Accessible Adaptive Hypermedia

A five-layer reference model for adaptive hypermedia, published at W4A 2008 with Steve Green and Elaine Pearson (doi:10.1145/1368044.1368052), and implemented as Carnforth-Java during doctoral research at Teesside.

What it is

The CISNA Model rebuilds the Dexter Reference Model of Hypertext for an era of script-heavy, AJAX-driven content, distinguishing five layers: Adaptation, Navigation, Semantics, Inventory, and External Content. Each layer is a formal model with its own diagrams and notation; bridges between layers express how meaning composes upward from raw content through to adapted presentation.

The published version (W4A 2008) presents the layered architecture and the formal definitions; the doctoral Java implementation (Carnforth-Java) applies the model to real interfaces. The Google Maps interface is the worked example used in the published material.

The five layers

  1. Adaptation — selects from the alternatives below based on user, device, and operating context. The layer where the recursive-design choice actually happens.
  2. Navigation — links and traversal between content elements. The original Dexter Model lives at this layer.
  3. Semantics — content composition and meaning, with notion ontologies, predicates, and rules.
  4. Inventory — formatted media elements available for use by the layers above.
  5. External Content — raw content beneath everything, usually networked.

Bridges between layers — Inventory↔Semantics and Semantics↔Navigation — express how meaning composes from raw content to adapted presentation. The bridges are formal: not metadata annotations on the side of the system, but typed relationships that the model compiler operates on.

The lineage

The CISNA Model is the published version of a longer published track:

  • 2006— “The Effectiveness of Self-Adapting User Interfaces as Assistive Technology for Handheld Mobile Devices.” Argues for intrinsic accessibility over functional bolt-on assistive tech, and names the formal next step (game theory + autonomous agents) that the 2029 framework resumes.
  • 2008— “20 Years On: the Dexter Model of Hypertext and its impact on web accessibility,” ACM SIGACCESS Newsletter. Critiques the Dexter Model and its descendant HTML for being structured around a printer’s view of a book rather than an author’s or a reader’s. Proposes the five-layer alternative that became CISNA.
  • 2008— “The CISNA Model of Accessible Adaptive Hypermedia,” W4A 2008, with Steve Green and Elaine Pearson (doi:10.1145/1368044.1368052). Published version of the five-layer model with formal definitions. This is what the literature cites.
  • 2009— “User Capability in an Adaptive World,” MSIADU 2009, with Steve Green and Elaine Pearson (doi:10.1145/1631097.1631110). The user side of the same project: separate models of capability (what a person can perceive and do, scoped by design space), capacity (their settings across contexts, with functional dependencies expressed as actions), and preference (their arbitrary choices, held apart from capability on purpose). It re-uses CISNA’s adaptation model to version and merge profiles, which is what makes “Fred is like Jim except…” expressible. Where CISNA says how content adapts, this says what it adapts to.

What it became

The five-layer model is structurally a Shlaer-Mellor domain chart with peer relationships rather than the canonical SM client-server hierarchy. That deviation is itself a methodological contribution — see the Shlaer-Mellor lens for the recursive-design framing beneath it.

The Carnforth-Java XML notation and execution engine live on as Action Language, which now ships as a TypeScript port running in the browser with four worked examples.

The modern descendant in active use is the ActionLanguage intermediate representation (IR) inside Paradise — the same shape, the same threaded-interpreter execution model, applied to source-level accessibility analysis of JavaScript.

Reading on