A Domain Specific Language Framework for Non-Visual Browsing of Complex HTML Structures
E. Pontelli, W. Xiong, G. Gupta, A. I. Karshmer · 2000 · Proceedings of the Fourth International ACM Conference on Assistive Technologies (Assets '00) · doi:10.1145/354324.354373
Summary
This paper from New Mexico State University, University of Texas at Dallas, and University of South Florida presents a framework for navigating complex HTML structures — specifically tables, frames, and forms — that are particularly challenging for screen reader users. The core problem addressed is that screen readers at the time (and largely still today) linearise tables into a sequential reading order, destroying the two-dimensional relationships between cells that give tabular data its meaning. The framework has two novel components: a semantic representation of the table as an annotated conceptual graph with multiple hierarchical viewpoints, and a Domain Specific Language (DSL) for expressing navigation strategies over that representation. The semantic representation is built through a multi-phase analysis: first, syntactic analysis extracts structural information from the HTML (identifying header cells, scope and axis attributes from HTML 4.0); second, semantic analysis adds additional knowledge about the table's content structure, either from human annotators, automated scripts, or content analysis techniques like background colour grouping and latent semantic indexing; third, a human annotator can refine the representation using a GUI tool. The DSL itself is built on Horn Logic Denotations, using Definite Clause Grammars for syntax specification and logic programming for semantic interpretation. DSL programs consist of creation commands (grouping cells into meaningful abstractions), inner navigation (moving within an abstraction layer), outer navigation (moving between abstraction levels), extension commands (creating new viewpoints on the fly), and inspector commands (querying current position).
Key findings
The paper demonstrates that viewing table navigation as execution of a DSL program provides significant advantages over existing approaches. The semantic representation as a conceptual graph with multiple viewpoints allows the same table to be navigated in fundamentally different ways — by rows, by columns, by semantic groupings, or by custom abstractions defined by an annotator. For a travel expenses table with data for San Jose and Seattle across multiple dates, the framework enables navigation that preserves the three-dimensional conceptual structure (city x date x expense category) that is lost when HTML forces it into a 2D grid. The DSL approach means that the semantic representation can itself be expressed as a program, allowing automatic generation of navigation scripts from table structure analysis. The authors note that none of the existing proposals in the literature had considered viewing table navigation in terms of DSL program execution. The framework supports both browser-level and server-level integration, meaning annotations can be provided either by the content creator or by third parties. The implementation uses logic programming (Horn Logic), which provides automatic derivation of an executable interpreter and supports partial evaluation for efficient compiled code.
Relevance
This paper addresses what remains one of the most persistent challenges in web accessibility: making complex tabular data meaningful to screen reader users. While HTML 4.0 introduced structural attributes like scope, headers, and axis for table accessibility — and these were carried forward into modern HTML — adoption by web developers has been consistently poor, and screen reader support varies. The paper's insight that table navigation requires semantic understanding beyond syntactic structure anticipates the ongoing gap between what HTML markup can express and what assistive technologies actually convey. The multi-viewpoint approach — allowing the same data to be navigated through different semantic lenses — is more sophisticated than current screen reader table navigation, which typically offers only row-by-row or column-by-column traversal. For accessibility practitioners, the framework highlights that truly accessible tables require more than correct markup: they need semantic annotation that captures the conceptual structure of the data. The concept of third-party annotation to retrofit accessibility onto existing content remains relevant, as most web tables still lack proper structural markup. The DSL approach also demonstrates that accessibility tooling benefits from formal, composable abstractions rather than ad hoc solutions.
Tags: table accessibility · non-visual browsing · screen readers · domain specific language · HTML tables · semantic structure · web accessibility · blind users · conceptual graphs
Standards referenced: HTML 4.0 · WAI · Aural CSS