Gist Summaries for Visually Impaired Surfers
Simon Harper, Neha Patel · 2005 · Proceedings of the 7th International ACM SIGACCESS Conference on Computers and Accessibility (Assets '05) · doi:10.1145/1090785.1090804
Summary
This paper from the University of Manchester addresses a fundamental problem in web accessibility for visually impaired users: the inability to quickly assess a web page's relevance before committing to having the entire page read aloud by a screen reader. The authors draw parallels between sighted and visually impaired browsing, noting that sighted users also cannot see beyond a hyperlink's anchor text — both groups face uncertainty about what lies on the other side of a link. The paper builds on the authors' prior formative evaluation that compared four automatic summarisation algorithms (query-based, link-based, text and media extraction, and gist creation) against manually created summaries. That evaluation found users preferred small, automatically generated gist summaries that capture the essential focus of a page. Based on these findings, the authors developed Summate, a Firefox browser extension that generates on-the-fly gist summaries of web pages. The tool uses a DOM-traversal algorithm that extracts text nodes longer than 70 characters, identifies sentence boundaries, and then ranks sentences into quartiles (HIGH, MEDIUM, LOW) based on iterative goodness measures, outputting a compact summary displayed as a JavaScript alert. The paper provides the full algorithm in pseudocode and discusses the browsing context through five analytical views: context, behavioural, motivational, cognitive, and resource.
Key findings
The formative evaluation with 42 participants found that users preferred simple automatically generated gist summaries over more complex or manually created ones, with the gist approach reducing cognitive overload and increasing awareness of a page's focus. Statistical analysis suggested summaries should be between 4 and 6 sentences, with 4 sentences found to be optimal. An informal re-evaluation with six sighted volunteers comparing Summate-generated summaries against a conventional summariser across random BBC News pages showed that users on the whole preferred Summate summaries — respondents understood the gist faster and could infer content and switch contexts more easily. The algorithm performed well generally but had limitations: it could produce incorrect summaries for pages like CNN that had extraneous advertising content prefixed to the summary, and it struggled with Weblog-style pages containing multiple disjoint full texts where the summary became too fragmented. The authors identified two paths for future work: extending the system to handle incorrect summaries, information overload, and docuverse clutter; and conducting a formal summative evaluation with visually impaired users.
Relevance
This paper tackles a problem that remains highly relevant today: the serial, time-consuming nature of screen reader interaction with web content. While sighted users can visually scan a page in seconds to assess relevance, screen reader users must often listen to substantial portions before making that judgement. The concept of providing quick gist summaries as decision points anticipates modern developments in AI-powered content summarisation and browser assistants. For accessibility practitioners, the paper's analytical framework of five browsing views (context, behavioural, motivational, cognitive, resource) provides a useful lens for understanding how visually impaired users interact with web content. The finding that simple, short summaries outperform complex ones aligns with plain language principles in accessible content design. The work also highlights that reducing cognitive load is as important as technical screen reader compatibility — a page can be technically accessible yet cognitively overwhelming if users cannot efficiently triage its relevance.
Tags: visual impairment · web accessibility · text summarization · screen readers · cognitive overload · web browsing · document engineering · browser extension · information architecture