Understanding and Improving Drilled-Down Information Extraction from Online Data Visualizations for Screen-Reader Users
Ather Sharif, Andrew M. Zhang, Katharina Reinecke, Jacob O. Wobbrock · 2023 · Proceedings of the 20th International Web for All Conference (W4A '23) · doi:10.1145/3587281.3587961
Summary
This extended abstract presents enhancements to VoxLens, an open-source JavaScript plug-in that improves the accessibility of online data visualizations for screen-reader users (SRUs) through a multimodal approach. Over 7.6 million people in the United States use screen readers, yet online data visualizations remain largely inaccessible — alternative text descriptions, when present, only provide high-level overviews and do not allow granular data exploration. Prior research found that SRUs spend 211% more time and are 61% less accurate in extracting information compared to non-SRUs. The researchers conducted role-based and longitudinal user studies with screen-reader users to understand the specific granular information they seek from both simple visualizations (single-series bar graphs) and complex ones (multi-series line graphs and geospatial maps). From these studies they developed taxonomies of information types sought during holistic and drilled-down explorations, which then guided the extension of VoxLens's Question-and-Answer mode to support complex visualizations.
Key findings
The enhanced VoxLens supports three interaction modes: Question-and-Answer (verbal queries), Summary (overview descriptions), and Sonification (data-to-audio mapping). Key enhancements include factor-level categorization for multi-series data (e.g., asking about a specific state's data in a multi-factor dataset), regional categorization for geospatial maps (grouping US data by regions like "east coast" or "New England" using National Geographic Society classifications), a two-step interaction for exploring factor levels and their counts, min/max range functionality, and a hidden data table appended to visualizations that is accessible to screen readers. After these enhancements, SRUs performed 5.6% more accurately than non-SRUs in task-based studies — a dramatic reversal from the original VoxLens where SRUs performed 15% less accurately. The tool requires only a single line of code for integration, making adoption straightforward for web developers.
Relevance
This work addresses a critical gap in web accessibility: while WCAG requires text alternatives for non-text content, simple alt text on data visualizations provides only surface-level information, effectively excluding screen-reader users from the granular data exploration that sighted users take for granted. The VoxLens approach — enabling natural language queries against visualization data — represents a shift from static descriptions to interactive, user-driven data access. The finding that enhanced VoxLens allowed SRUs to outperform non-SRUs in accuracy demonstrates that well-designed accessibility tools can fully close, and even reverse, performance gaps. For practitioners, this highlights that alt text alone is insufficient for complex visualizations and that interactive, multimodal approaches should be considered. The open-source nature and single-line integration of VoxLens make it a practical tool for improving data visualization accessibility on the web.
Tags: screen readers · data visualization accessibility · alternative text · sonification · blind and low vision · web accessibility · non-visual interaction · information retrieval · accessible maps
Standards referenced: WCAG