Fast Web by Using Updated Content Extraction and a Bookmark Facility
Tsuyoshi Ebina, Seiji Igi, Teruhisa Miyake · 2000 · Proceedings of the Fourth International ACM Conference on Assistive Technologies (Assets '00) · doi:10.1145/354324.354343
Summary
This paper from Japan's Communications Research Laboratory presents two techniques for improving speech-based web navigation for visually impaired users. The core problem is that accessing web pages through speech output is significantly slower than visual browsing — users must listen to content sequentially, cannot quickly scan for changes, and have no equivalent of a scrollbar for jumping to desired locations. The first technique is a sentence-level bookmark facility that allows users to mark specific positions within a web page and jump directly to them on subsequent visits. The system works by decomposing HTML pages into a "text skeleton" (the permanent textual content) and a "tag skeleton" (the HTML structural layout), then using these to match bookmarked positions across page updates even when content has changed. The second technique is an updated-sentence search facility that automatically identifies which sentences have changed since the user last visited a page, allowing them to skip directly between new content without re-listening to unchanged material. The prototype was built for Windows using Microsoft's speech API and operated via a numeric keypad with dedicated keys for navigation, bookmarking, and updated-sentence search modes.
Key findings
User testing with six visually impaired college students demonstrated statistically significant improvements (p < 0.05) for both techniques. The bookmark facility reduced text search times substantially — all subjects found target sentences more quickly with bookmarks than without, with particularly large improvements for users who were slower navigators. The updated-sentence search facility also showed significant time reductions for finding new content on previously visited pages. Critically, both features reduced not only access time but also the users' self-reported mental workload, with mean usability scores of 5.00 for bookmarks and 4.83 for updated-sentence search on a 1-7 scale (higher being better). The text skeleton and tag skeleton ratings for the test web pages (Yahoo Japan and Yomiuri Online) remained consistently high over the observation period, confirming that page layouts were stable enough for the bookmark matching algorithm to work reliably. The paper also demonstrated applications beyond news pages, including navigating search engine results where bookmarks could mark the "Next 5 results" link position across pages.
Relevance
This research addressed a fundamental asymmetry in web access that persists today: sighted users can visually scan a page in seconds to find what has changed or jump to a specific section, while screen reader users must navigate sequentially. The sentence-level bookmark concept anticipated modern features like screen reader landmarks and heading navigation, which similarly allow users to jump to known structural positions. The updated-content detection idea remains largely unimplemented in mainstream assistive technology, despite being potentially valuable for users who regularly monitor news sites, dashboards, or any frequently updated content. The paper's approach of separating page structure from content to track changes across visits was ahead of its time and relates to modern concepts like the DOM diffing used in web frameworks. The finding that reducing navigation effort also reduces mental workload — not just time — is important for understanding the cognitive burden of inefficient web access on screen reader users.
Tags: web accessibility · visual impairment · web navigation · screen reader · speech-based navigation · content extraction · assistive technology