EnTable: Rewriting Web Data Sets as Accessible Tables
Steven Gardiner, Anthony Tomasic, John Zimmerman · 2015 · ASSETS '15: Proceedings of the 17th International ACM SIGACCESS Conference on Computers & Accessibility · doi:10.1145/2700648.2811344
Summary
This demonstration paper presents EnTable, a crowdsourcing system that transforms visually complex web data presentations into accessible HTML tables. The problem it addresses is fundamental: modern web designers communicate semantic information through sophisticated CSS and JavaScript layouts that are visually intuitive but structurally opaque to screen readers. A product comparison grid, for example, allows sighted users to quickly scan across items—but screen reader users cannot easily compare prices across products because the visual grid has no corresponding semantic structure. Screen readers provide sophisticated navigation for properly marked-up \<table\> elements, allowing users to move by row or column and understand relationships between data points. However, many data-rich pages avoid semantic tables in favor of visually flexible grid layouts. Since algorithmic extraction of table structure from arbitrary HTML remains beyond current capabilities, EnTable employs human intelligence through crowdsourcing. The system operates through a browser extension. When screen reader users encounter a confusing data presentation, they submit a request to the EnTable service. The request is forwarded to sighted crowdworkers who use the SmartWrap interface to mark up the page's semantic structure. SmartWrap makes this task accessible to non-programmers by leveraging spreadsheet familiarity: users drag data from the first two rows into a table template, and the system extrapolates the required XPath/CSS selectors automatically. The resulting table descriptions are stored centrally and reused for all subsequent visitors to that page—and the patterns can apply to similar pages across the same website.
Key findings
The paper demonstrates an end-to-end workflow connecting screen reader users who identify accessibility barriers with sighted volunteers who can resolve them. The SmartWrap interface successfully enables non-programmers to create what are essentially small programs (nested selector patterns), by translating the task into familiar spreadsheet operations. The architectural insight is that table description creation needs to happen only once per page template. E-commerce sites, search results, and other data-driven pages typically use consistent templates across thousands of individual pages. A single SmartWrap annotation can therefore benefit many users across many pages, creating efficiency that makes crowdsourcing viable. Prior research (cited in the paper) demonstrated that crowdworkers can effectively and inexpensively provide table descriptions for complex template layouts. The Social Accessibility project showed that substantial numbers of sighted users are willing to volunteer labor to benefit blind users. Whether the supply of willing annotators will meet the demand from screen reader users remains an open empirical question that EnTable's deployment enables studying.
Relevance
EnTable exemplifies a broader pattern in accessibility research: using human computation to bridge gaps that current algorithms cannot close. While machine learning has advanced significantly since 2015, the fundamental challenge persists—visual design communicates meaning through spatial relationships that require human interpretation to capture as structured data. For practitioners, the key insight is that proper HTML table markup transforms screen reader usability. The contrast between the grid layout and table layout in the paper's figures dramatically illustrates how the same underlying data can be navigable or incomprehensible depending on markup choices. This argues for semantic table markup during initial development rather than requiring retrofit. The crowdsourcing model raises sustainability questions common to human-powered accessibility solutions: Can volunteer supply meet user demand? How quickly can requests be fulfilled? What happens when templates change? These questions apply equally to image description services, captioning, and other crowd-powered accessibility tools. EnTable provides infrastructure for empirically studying these dynamics.
Tags: web accessibility · crowdsourcing · screen readers · data tables · transcoding · HTML tables · browser extension · remote sighted assistance