Navigation of HTML Tables, Frames, and XML Fragments
Enrico Pontelli, Douglas Gillan, Weiqiang Xiong, Elham Saad, Gopal Gupta, Arthur I. Karshmer · 2002 · Proceedings of the Fifth International ACM Conference on Assistive Technologies (Assets 02) · doi:10.1145/638249.638256
Summary
This paper provides a progress report on a multi-institution project developing technology to support non-visual navigation of complex HTML tables, frames, and XML structures. The authors identify tables, frames, and forms as the most difficult areas of web accessibility because they are inherently multi-dimensional and carry implicit semantic content that is lost when linearised by screen readers. The project's approach has two phases: first, creating semantic descriptions (SDs) that capture the navigational semantics of complex HTML structures using conceptual graphs; second, using these SDs to support interactive and intelligent non-visual navigation. For creating SDs, the system provides both manual annotation tools (a GUI where instructors can select cell groups and assign semantic descriptions) and a syntactic analyser that automatically extracts structural information from HTML tables by examining COLGROUP, COLSPAN, TH elements, header attributes, and visual formatting cues like colours and font sizes. The automatically detected groupings are presented to the human annotator for review and refinement. The SDs are then processed by a Pathfinder network algorithm — borrowed from cognitive psychology — that identifies the most important navigational relationships between table components and produces a weighted network representing meaningful navigation paths. A Domain Specific Language (DSL) allows the SD to be expressed as navigation commands, and speech synthesis provides audio output.
Key findings
User studies with blind participants demonstrated that the spatial and semantic organisation of table information significantly affects navigation performance. In experiments where subjects used tables to answer questions, providing spatial cues (row/column structure awareness) improved performance compared to purely temporal formats that presented information sequentially. Interestingly, non-spatial enhancements such as colour or auditory cues added to the learning phase did not improve performance over the basic temporal format when structural cues were absent — suggesting that understanding the table's organisation matters more than enriched presentation of individual cells. For frame-based pages, the Pathfinder algorithm was used to generate navigation networks based on how frequently users traversed between frames, enabling navigation ordered by importance rather than arbitrary source order. The system was validated in a coursework management context where instructors created SDs for course materials, and the DSL was extended to support frame and XML fragment navigation alongside tables. The complete system architecture uses transcoding proxy servers that intercept web requests, extract complex HTML structures, apply the syntactic analyser and Pathfinder algorithm, and deliver semantically enriched navigation to the screen reader user.
Relevance
This paper tackles a problem that persists today: complex data tables and multi-panel layouts remain among the most challenging web content for screen reader users. While HTML frames are now largely obsolete, the underlying challenge of navigating multi-dimensional visual layouts non-visually has simply shifted to CSS grid layouts, dashboards, and complex web applications. The research's emphasis on semantic description — capturing what table regions mean rather than just their position — anticipates modern efforts around ARIA table semantics and the ongoing difficulty of making complex data visualisations accessible. For practitioners, the key insight is that accessible table navigation requires understanding the semantic relationships between data regions, not just marking up headers. The finding that spatial organisation awareness outweighs enriched cell presentation reinforces the importance of conveying table structure to screen reader users through proper markup and navigation mechanisms.
Tags: table accessibility · web accessibility · screen readers · non-visual navigation · semantic navigation · HTML frames · XML accessibility · domain specific language
Standards referenced: WAI · ADA · Section 508