Accessible Skimming: Faster Screen Reading of Web Pages
Faisal Ahmed · 2012 · Proceedings of the 14th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS 2012) · doi:10.1145/2384916.2384998
Summary
This doctoral consortium paper presents an automated approach to enable non-visual skimming of web pages for screen reader users. Sighted people routinely skim web content through quick eye movements (saccades) that let them glance over headlines and text to extract the gist of information. Blind screen reader users lack this ability — they must either listen to all content sequentially or skip content entirely without knowing what they are missing. The existing skimming feature in JAWS screen reader only reads the first line or sentence of each paragraph, which the author argues does not replicate the actual skimming process used by sighted people. Drawing on prior research with 20 screen reader users showing that extractive summarization is the most suitable technique for non-visual skimming, the author developed a novel summarization algorithm. The algorithm parses sentences to extract grammatical relations, constructs lexical trees, extracts grammatical (POS tags) and structural features (in-degree/out-degree) for each word, uses a trained classifier to determine which words to include in the summary, and constructs a subtree of selected words as the skimming summary.
Key findings
A user study with 23 screen reader users demonstrated that the automated summarization algorithm can be successfully used for non-visual skimming. Earlier experiments with 20 participants using human-generated summaries showed that skimming enabled blind users to read faster while maintaining high comprehension and retention levels. Interviews with those participants confirmed that extractive summarization — which preserves the original content rather than generating new text — is the most suitable technique for non-visual skimming. The approach differs from simple paragraph-first-sentence extraction by using NLP techniques to identify the most informative words within sentences based on grammatical structure and tree features. The resulting summaries are presented through a non-visual interface that screen reader users interact with aurally. The work frames visual skimming as equivalent to extractive summarization, since both involve scanning content to extract key information while preserving the original context.
Relevance
This research addresses a fundamental asymmetry between sighted and blind web users: the ability to quickly assess and filter information. As information overload worsens, the gap between sighted users who can skim and screen reader users who must process content linearly becomes increasingly consequential — affecting productivity, learning, and participation. For accessibility practitioners, the work highlights that screen reader accessibility is not just about making content readable but about supporting the full range of reading strategies that sighted users take for granted. The NLP-based approach to generating skimmable summaries foreshadows current interest in AI-powered accessibility tools. The concept remains highly relevant as web pages continue to grow in length and complexity, and as large language models offer new possibilities for intelligent content summarization tailored to screen reader users.
Tags: screen readers · blind users · web accessibility · text summarization · natural language processing · skimming · information overload · aural browsing · cognitive load