Middleware to Expand Context and Preview in Hypertext
Simon Harper, Carole Goble, Robert Stevens, Yeliz Yesilada · 2004 · Proceedings of the 6th International ACM SIGACCESS Conference on Computers and Accessibility (Assets 04) · doi:10.1145/1028630.1028643
Summary
This paper presents a middleware tool that automatically annotates web pages with expanded context and preview information for hypertext links, addressing a core navigation problem for visually impaired web users. The authors draw an analogy between physical mobility and web mobility: just as a traveller in the physical world needs to know where they are, what surrounds them, and what lies ahead before moving, a screen reader user needs equivalent context and preview to navigate hypertext confidently. The problem is acute because many links use vague anchor text like "Read More..." or "Click Here" that provides no meaningful information about the destination, and screen reader users who tab through links hear these anchors stripped of the surrounding visual context that sighted users rely on. The tool is implemented as a Perl-based Apache content handler that processes web pages server-side, performing two key transformations. For context expansion, the algorithm searches outward from each anchor tag, collecting surrounding text until it reaches sentence boundaries, then replaces the original anchor text with this expanded context. For preview generation, the system fetches each link's destination page and extracts descriptive information from the first paragraph, the body text, meta tags, and Dublin Core metadata, then adds this as a title attribute on the anchor element. The system was designed to work transparently with existing websites without requiring content authors to modify their pages.
Key findings
Testing the tool on the University of Manchester's news site demonstrated that the context expansion algorithm successfully replaced ambiguous anchors like "Read More..." with the full surrounding sentence, making the link purpose clear when encountered in isolation via a screen reader. The preview generation correctly extracted destination page descriptions from multiple sources (first paragraph text, meta descriptions, Dublin Core abstracts) and presented them as title attributes accessible to assistive technologies. However, several issues emerged: some preview texts were longer than expected, which could cause information overload for users with cognitive impairments or those using screen readers where lengthy title attributes are read in full. The system also generated some "anomalies" — for example, when processing links on a W3C consortium page, some structural anchors like "Home page Mailing List Download Jigsaw" did not provide useful preview information. Initial testing with JAWS screen reader confirmed that the additional context and preview information was useful when tabbing through links, though the authors acknowledge this was not a full user evaluation. The approach preserved the visual presentation for sighted users while enhancing the non-visual experience.
Relevance
This research addresses a problem that persists today: poorly described links remain one of the most common web accessibility failures, and screen reader users routinely encounter lists of "Read More" or "Click Here" links with no context. For accessibility practitioners, the paper's framing of web navigation as a mobility problem provides a useful conceptual model — users need to know where they are (context), where they can go (link identification), and what they will find (preview) before committing to a navigation action. The middleware approach of transforming content server-side without requiring author cooperation is pragmatic, acknowledging that waiting for all web authors to write descriptive link text is unrealistic. The work anticipated modern approaches like browser extensions and AI-powered link description tools. Key limitations include the lack of formal user evaluation with visually impaired participants and the potential for information overload from lengthy preview text. The authors also note that for a truly universal solution, client-side transformation with user-configurable preferences would be more appropriate than server-side processing, a conclusion that aligns with current accessibility tool architecture.
Tags: web accessibility · visual impairment · web navigation · web transcoding · link context · screen reader · hypertext · web mobility
Standards referenced: Dublin Core