Why Read if You Can Skim: Towards Enabling Faster Screen Reading
Faisal Ahmed, Yevgen Borodin, Yury Puzis, I. V. Ramakrishnan · 2012 · Proceedings of the International Cross-Disciplinary Conference on Web Accessibility (W4A) · doi:10.1145/2207016.2207052
Summary
This paper investigates how to bring the speed-reading technique of skimming — widely used by sighted readers to quickly get the gist of content — to blind screen reader users who face chronic information overload when listening to web content sequentially. The authors observe that sighted people routinely scan text visually, picking out salient phrases and keywords without reading every word, but screen readers offer almost no equivalent capability. JAWS' existing "skim reading" feature (reading the first sentence or line of each paragraph) was found to be inadequate in interviews — 20 experienced JAWS users either did not know about it or considered it "not useful" and "useless." The research proceeded in multiple phases: first, 12 sighted participants created "gold-standard" extractive summaries of 6 news articles (constrained to one-third the original length, preserving original word order). Analysis of these summaries revealed that nouns dominated (54% of summary content), with verbs (12%), adjectives (11%), and adverbs (11%) also preserved, while punctuation was almost always retained as it preserves clause structure. From these gold standards, three summary types were derived: nouns only (Summary A), nouns plus prepositions (Summary B), and the full human-generated extractive summary with mixed parts of speech (Summary C). The skimming interface was built on the HearSay platform, featuring a single-shortcut toggle between the summary and full article that preserved reading position — when switching to skimming mode, the reading position snapped to the closest preceding word common to both texts, creating the impression of reading the same content with parts skipped.
Key findings
A user study with 20 blind participants (recruited from Arizona State University's Disability Resource Center, all experienced JAWS users) tested the interface in two scenarios: listening-and-comprehension (reading articles and answering questions) and ad-hoc searching (finding specific information within articles). For comprehension, there was no significant difference in question-answering accuracy between the gold-standard summary C and the full text — the combined summary "carried almost as much information as the full text." Including verbs, adjectives, and adverbs (VAAs) significantly improved comprehension: summaries with VAAs yielded 77.5-80.5% accuracy versus 56.5-62% for noun-only summaries (p < 0.0001). Perceived difficulty decreased linearly from summary A (3.98) through to full text D (1.80). For the searching scenario, skimming was 1.9 times faster than ad-hoc searching for reaching the answer location (76.35 vs 148.45 seconds, p < 0.0001), and 1.6 times faster for total task completion (103.70 vs 169.40 seconds, p < 0.0001). Skimming was also rated significantly easier (1.25 vs 2.30, p < 0.0001). Post-study questionnaire results showed strong enthusiasm: participants wanted to use skimming in the future (4.30/5), agreed it made reading faster (4.20/5), and wished they could browse articles faster than current screen readers allow (3.80/5). Ad-hoc skimming strategies observed during the study included paragraph navigation to move quickly through content, reading sentences halfway, and increasing speech rate — but these proved less effective than the structured skimming interface. The paper concludes that extractive summarization (preserving original words and order) is the most suitable automated technique for non-visual skimming, as it most closely matches how sighted people skim.
Relevance
This paper addresses a fundamental asymmetry in web accessibility that persists today: sighted users can glance at a page and decide in seconds whether content is relevant, while screen reader users must listen sequentially, creating massive time and cognitive costs. The finding that extractive summaries preserving nouns, verbs, adjectives, and adverbs carry "almost as much information as the full text" has direct implications for AI-powered accessibility tools — modern large language models could generate such summaries on the fly for screen reader users. The seamless toggle between summary and full text is a particularly elegant interface design that maintains user orientation and control. For practitioners, the paper reinforces why well-structured content with clear headings, informative first sentences, and scannable writing patterns matters so much for non-visual users — these are the very features that enable both human and automated skimming. The research also connects to the broader HearSay project from Stony Brook, extending the team's work on intelligent non-visual web browsing from page segmentation and voice browsing into content summarization.
Tags: screen readers · skimming · speed reading · blind users · text summarization · information overload · non-visual interaction · user study · cognitive load