Benthic: Perceptually Congruent Structures for Accessible Charts and Diagrams
Catherine Mei, Josh Pollock, Daniel Hajas, Jonathan Zong, Arvind Satyanarayan · 2025 · Proceedings of the 27th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS 2025) · doi:10.1145/3663547.3746342
Summary
This paper introduces Benthic, a system that creates perceptually congruent screen reader structures for graphical representations such as charts and diagrams. The core insight is that graphical representations communicate meaning through visual structure—Gestalt grouping principles like spatial proximity, connectedness, alignment, and containment convey relationships between elements that sighted readers perceive effortlessly. However, existing screen reader solutions rarely surface this structure for blind and low vision (BLV) readers. Current approaches use either hierarchical trees (like Olli) that capture containment but not adjacency, or node-link graphs (like Data Navigator) that capture connections but not hierarchy. Benthic addresses this limitation by modelling graphical representations as hypergraphs—a generalisation of both trees and graphs where a single hyperedge can connect a parent to a set of children nodes. This unified representation captures both hierarchical and adjacent visual relationships simultaneously, is domain-agnostic (working across charts, diagrams, and other graphical types), and enables fluid traversal that is concise and reversible. The screen reader interface was co-designed over six months with co-author Daniel Hajas, a blind researcher with experience in accessible diagrammatic communication. Users navigate using keyboard commands: arrow keys move laterally between sibling nodes, shift+arrow keys move up and down the hierarchy, and context switches allow users to regroup data around different organisational dimensions—for example, switching from viewing a stacked bar chart grouped by team to grouped by competition.
Key findings
In an evaluation with 15 blind participants, Benthic enabled flexible, goal-driven exploration of two types of graphical representations: physics pulley diagrams and stacked bar charts. Participants found the interface easy to learn (median scores of 4-5 on 5-point scales), reported that it supported building clear mental models of diagram structure, and required minimal cognitive effort. For the bar chart replication study (adapted from Boger & Franconeri), 12 of 15 participants identified an implausible data trend—compared to only 46% of sighted readers in the original study who missed the same trend. This striking result occurred because Benthic users could perform context switches to regroup data, enabling them to detect the anomaly regardless of which visual grouping was initially salient. Participants described navigation metaphors that reflected the structural relationships in the diagrams: shift+down was "digging deeper" while arrow keys meant "moving across." However, the study also revealed limitations: participants struggled to build understanding of overall spatial layout (where components were positioned relative to each other in 2D space), suggesting that structural navigation alone is insufficient for diagrams where spatial reasoning is essential. The parent context layer for handling multiple parents sometimes confused participants by breaking expectations of consistent bidirectional traversal.
Relevance
Benthic represents a significant advance in making graphical representations accessible by moving beyond the tree-versus-graph dichotomy that has limited prior screen reader systems. For web developers and visualisation tool creators, the hypergraph model offers a principled, domain-agnostic way to encode visual structure that works across charts, diagrams, and other graphical domains. The concept of perceptual congruence—aligning screen reader navigation with visual structure—provides a clear design principle for accessible visualisation tools. The finding that BLV users outperformed sighted users in detecting data anomalies demonstrates that well-designed accessible structures can provide analytical affordances comparable to (or exceeding) visual ones. Practical adoption challenges include the need for scalable methods to generate Benthic-compatible hypergraphs from existing diagrams, potential integration with diagramming libraries like Bluefish, and extending the approach to handle spatial relationships in continuous representations like maps and heatmaps. The discussion of congruent structures across modalities—sonification, magnification, and tactile graphics—opens promising directions for multimodal accessible representations.
Tags: data visualization accessibility · screen readers · blind and low vision · chart accessibility · diagram accessibility · Gestalt principles · hypergraph · perceptual congruence