Nonvisual Support for Understanding and Reasoning about Data Structures
Brianna L. Wimer, Ritesh Kanchi, Kaija Frierson, Venkatesh Potluri, Ronald A. Metoyer, Jennifer Mankoff, Miya Natsuhara, Matt X. Wang · 2026 · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI '26) · doi:10.1145/3772318.3791656
Summary
Wimer, Kanchi, and colleagues present Arboretum, a web-based system that generates accessible representations of introductory data structure diagrams (arrays and binary trees) for blind and visually impaired (BVI) computer science students. The authors argue that current practice — converting diagrams to ALT text — preserves visual appearance but strips away the computational properties (ordering, hierarchy, parent–child relationships, left/right positioning) that students actually need to reason with data structures. Motivated by the proliferation of text-based diagram specification languages like Mermaid and Graphviz DOT, they observe that structural information is already captured at authoring time and can drive automatic generation of multiple accessible outputs. The paper reports two studies. Study one was a Wizard-of-Oz comparison of three nonvisual flowchart prototypes (ALT-text with LLM Q&A, keyboard-navigable digital graph, and swell-paper tactile) with eight BVI participants, yielding five design requirements: standardized screen reader navigation, tabular structural views, explicit encoding of relationships, separation of structural facts from narrative explanation, and integrated access across complementary modalities. Study two evaluated Arboretum with eight BVI participants across six tasks (element location, sorting, parent–child identification, leaf identification, BST property check, and binary search), using three synchronized outputs: a tabular HTML view, a WAI-ARIA-based navigable tree/list, and a Braille-labeled swell-paper tactile diagram produced from SVG. Analysis combined descriptive statistics with inductive thematic coding of transcripts.
Key findings
Participants achieved 100% accuracy on parent–child identification and binary search tasks, 91.67% on array element location, 87.5% on leaf identification, and 62.5% on BST property checking — the latter being the hardest task because it required simultaneous reasoning over multiple relationships. Tactile graphics were overwhelmingly preferred: seven of eight participants spent 89%+ of task time on the tactile format, and across 40 task-modality decisions tactile was selected 31 times. Participants rated tactile graphics at a mean of 5.0/5 on Likert scales for every data structure, while tabular and navigable ratings fell to 3.38 for binary trees. However, tactile was not sufficient alone: five participants supplemented it with at least one digital modality for cross-checking. The navigable view benefited from established WAI-ARIA tree/list patterns but the right-arrow-as-expand convention clashed with traversal semantics when children represented different decision paths. Tabular views suited arrays well but struggled to convey tree hierarchy. Tactile graphics remain constrained by sheet size (~\$1,475 Swell Form machine, ~\$1.38 per page), limiting scalability to large structures. The authors consolidate findings into four POUR-aligned design principles for structure-first accessible representations.
Relevance
For accessibility practitioners in education and tooling, the paper reframes diagram accessibility from translation (image → ALT text) to structural authoring (specification → synchronized multimodal outputs). This has direct implications for instructors preparing CS materials under WCAG's complex-image guidance and the 2024 DOJ Title II rule placing responsibility on educators for accessible content. Practitioners should evaluate diagram pipelines that start from Mermaid, Graphviz DOT, or similar structured sources rather than retrofitting accessibility onto exported images. The finding that multiple synchronized modalities outperform any single modality — and that WAI-ARIA tree patterns do not cleanly map to all structural relationships — should inform accessible data visualization work beyond CS education (process diagrams, graphs, scientific visualizations). Limitations include a small sample (n=8 in each study), mild/moderate complexity only (arrays and binary trees), screen-reader testing limited to JAWS/NVDA/VoiceOver macOS, Braille label length capped at three characters, and an introductory scope that leaves linked lists, graphs, and advanced algorithmic reasoning for future work.
Tags: blind and low vision · BVI · screen readers · tactile graphics · data structures · computer science education · accessible diagrams · diagramming · nonvisual interaction · multimodal access · WAI-ARIA · WCAG · Mermaid · Graphviz · wizard of oz · accessible education
Standards referenced: WCAG 2.1 AA · WAI-ARIA · ADA Title II