Source-linked AI summary
Rich Screen Reader Experiences for Accessible Data Visualization
Jonathan Zong, Crystal Lee, Alan Lundgard, JiWoong Jang, Daniel Hajas, Arvind Satyanarayan
TL;DR
Alt text and data tables provide limited non-visual ways to explore interactive charts. This paper co-designs and prototypes screen-reader experiences using three design dimensions, then evaluates a subset with blind and low-vision readers. The study identifies design features and reading behaviors that support richer chart exploration.
Problem
Common chart alternatives such as alt text and data tables do not offer information-seeking modes comparable to interactive visualizations.
Source: §1 · p. 1Method
The authors use iterative co-design to develop prototypes organized around structure, navigation, and description, then evaluate a subset in a mixed-methods study with 13 blind and low-vision readers.
Source: §1 · pp. 1–2, §3 · p. 3Results
The exploratory study identifies useful features including hierarchical and segmented data, cursors, and navigation roadmaps, alongside reader behaviors such as hypothesis testing and validating mental models.
Source: §1 · p. 2Takeaways & Limitations
The prototypes explore richer screen-reader interactions for web visualizations than are possible through alt text, data tables, or the current ARIA specification.
Source: §6 · pp. 9–10Current authoring methods likely cannot scale these designs: authors must handcraft structures, navigation, and descriptions per visualization, with user preferences varying widely.
Source: §6.1 · p. 10
Abstract
from arXiv · showhide
Current web accessibility guidelines ask visualization designers to support screen readers via basic non-visual alternatives like textual descriptions and access to raw data tables. But charts do more than summarize data or reproduce tables; they afford interactive data exploration at varying levels of granularity -- from fine-grained datum-by-datum reading to skimming and surfacing high-level trends. In response to the lack of comparable non-visual affordances, we present a set of rich screen reader experiences for accessible data visualization and exploration. Through an iterative co-design process, we identify three key design dimensions for expressive screen reader accessibility: structure, or how chart entities should be organized for a screen reader to traverse; navigation, or the structural, spatial, and targeted operations a user might perform to step through the structure; and, description, or the semantic content, composition, and verbosity of the screen reader's narration. We operationalize these dimensions to prototype screen-reader-accessible visualizations that cover a diverse range of chart types and combinations of our design dimensions. We evaluate a subset of these prototypes in a mixed-methods study with 13 blind and low vision readers. Our findings demonstrate that these designs help users conceptualize data spatially, selectively attend to data of interest at different levels of granularity, and experience control and agency over their data analysis process. An accessible HTML version of this paper is available at: http://vis.csail.mit.edu/pubs/rich-screen-reader-vis-experiences.
1. Introduction
Basic alternatives such as alt text and data tables leave screen-reader users without information-seeking options comparable to interactive visualizations. The paper co-designs richer experiences and evaluates prototypes with blind and low-vision readers.
1. Introduction
- Alt text offers high-level takeaways without drill-down, while line-by-line tables make trends difficult to identify.Together, these approaches do not provide information-seeking modes comparable to interactive visualizations.Source: §1 · p. 1
- Screen readers linearize charts and limit users to author-provided steps, requiring them to remember or record earlier narration.Screen-reader users nevertheless seek an initial holistic overview followed by data-point comparisons.Source: §1 · p. 1
- The authors use iterative co-design to develop rich screen-reader experiences organized around structure, navigation, and description.The dimensions cover chart organization, movement through it, and the content, composition, and verbosity of narration.Source: §1 · p. 1
- An exploratory mixed-methods study with 13 blind or low-vision readers identifies useful features and reading behaviors.Reported features include hierarchical or segmented data presentations, cursors, and roadmaps; observed behaviors include hypothesis testing and mental-model validation.Source: §1 · p. 2
2. Background and Related Work
Screen-reader accessibility for charts has advanced through standards and tools, but common alternatives still constrain exploration, and existing richer features differ in scope. The paper focuses on rich interactions for web-based visualizations.
2. Background and Related Work
- Alt text and tables remain common recommendations, while ARIA adds semantic markup for structured graphics and data values.WAI-ARIA was adopted in 2014, and the WAI-ARIA Graphics Module followed in 2018.Source: §2 · p. 2
- Blind and low-vision users spend 211% more time and are 61% less accurate extracting chart information than non-screen-reader users.Many web charts are intentionally designed to be skipped by screen readers.Source: §2 · p. 2
- Alt text can impose an author’s interpretation, tables require sequential traversal, and ARIA labels offer limited information-seeking opportunities.These approaches respectively limit reader interpretation, trend discovery, and expressive interaction.Source: §2 · p. 2
- VoiceOver Data Comprehension provides selectable chart descriptions, statistics, data-series summaries, and audiographs, but is limited to single-line charts.SAS Graphics Accelerator supports a broader range of statistical charts with similar features.Source: §2 · p. 2
- Vega-Lite and Highcharts are beginning to provide ARIA support, while this work targets richer interactions for web visualizations than ARIA currently supports.The authors hope the work informs ongoing web-accessibility discussions, including an Accessibility Object Model.Source: §2 · p. 2
3. Design Dimensions for Rich Screen Reader Experiences
The paper organizes rich screen-reader experiences around structure, navigation, and description, developed through iterative co-design. These dimensions address how chart information is organized, traversed, and narrated.
3. Design Dimensions for Rich Screen Reader Experiences
- The authors identify structure, navigation, and description as dimensions for designing richer accessible reading experiences.They distinguish what is organized for traversal, how users move through it, and what semantic content is conveyed.Source: §3 · pp. 2–3
- Over six months, iterative co-design with a blind researcher produced 15 prototypes through weekly discussions, observations, and memos.The co-designer contributed lived experience and academic expertise, while the authors note that he does not represent all screen-reader users.Source: §3 · p. 3
3.1. Structure
- Accessible structure organizes chart data and visual elements into a traversable representation, with form and represented entities as design choices.Trees support branching and hierarchical browsing at varying detail levels; data, encodings, and annotations can appear as entities, including parallel branches.Source: §3.1 · pp. 3–4
- Hierarchical structures place overviews at branch nodes and individual data points at leaves, supporting movement from broad summaries to finer detail.A co-designer found the tree made a large scatterplot’s data points fairly readable despite sequential screen-reader reading.Source: §3.1 · p. 4
3.2. Navigation
- Navigation includes local and lateral structural moves, spatial moves by screen coordinates, and targeted jumps to a specified location.Bounded regions and known entry points can help users maintain their position; structural and spatial exploration support open-ended exploration, while targeted navigation aids quick searching.Source: §3.2 · pp. 4–5
3.3. Description
- Descriptions vary in semantic content, sentence order, and verbosity, with usefulness depending on the reader’s task and familiarity.Summary statistics and trends may be more useful than simply listing encodings; ordering can prioritize context or datum-level information, while added wayfinding detail can aid orientation but slow comprehension.Source: §3.3 · pp. 5–6
- Readers’ need for precise verification or orientation varies with task and familiarity, motivating context-sensitive descriptions and customization.Wayfinding descriptions may help build a mental map but become cumbersome as users grow accustomed to an interface.Source: §3.3 · p. 6
4. Example Gallery
The example gallery demonstrates combinations of structures and navigation schemes in interactive prototypes implemented over Vega-Lite. The examples span facets, maps, alternate drill-down paths, annotations, and binary search.
4. Example Gallery
- The prototypes use in-memory structures, keyboard event listeners, and an ARIA-Live region to update screen-reader descriptions, while rendering charts graphically.A tree selection also highlights the points receiving the screen-reader user’s attention.Source: §4 · p. 6
- Arrow keys support structural navigation, while WASD keys move spatially through chart grids and their data points.Left and right keys move between six subplots in the multiview example.Source: §4 · p. 6
- Examples combine lateral navigation across facets, state and county branches, alternative month-first or weather-first paths, annotation trees, and a year-based binary search tree.These examples show distinct compositions of structure and navigation schemes.Source: §4 · p. 6
5. Evaluation
The evaluation explores how three screen-reader visualization prototypes support data exploration, finding distinct strengths in ease of use, navigation, and building spatial understanding. Participants also identified the need to combine familiar tables with richer interactions.
5. Evaluation; 5.1. Evaluation Setup & Design
- The exploratory study used 90-minute sessions with 13 blind and low vision participants to compare three prototypes’ relative strengths, not identify a single best design.Participants explored TABLE, MULTI-VIEW, and TARGET, each with a different dataset; the set balanced time for learning and interaction against the session limit.Source: §5 · p. 6, §5.1 · p. 7
5.2. Quantitative Results
- MULTI-VIEW was significantly more difficult to use than TABLE (p < 0.01), while TARGET was not; no significant differences or large effects appeared on the other prompts.Participants rated all prototypes more than useful on average (median = 4, μ ≥ 4).Source: §5.2 · p. 8, Table 1 · p. 8
- TARGET had the highest median for browse or explore, while TABLE was easiest to learn and generally easiest for discovering and answering questions about data.Participants generally enjoyed MULTI-VIEW and TARGET more and found them easier for lookup or locating data.Source: §5.2 · p. 8
5.3. Qualitative Results
- Participants described tables as familiar and necessary, but poor for large datasets and cognitively demanding because values must be remembered across rows.Participants reported tabbing through at most 20–30 rows during the sessions, and said they would otherwise skim a few rows for summary statistics.Source: §5.3 · p. 8
- Prior experience with data analysis, tactile graphs, or hierarchical structures helped participants understand spatial representations; some participants without such experience struggled.Participants suggested that help menus and mini-tutorials could aid learning navigation shortcuts.Source: §5.3 · pp. 8–9
- Participants described hierarchical and segmented data presentation as a useful middle ground between overview and detail, supporting quick skimming and deeper exploration.Some also valued control over which data to inspect rather than relying only on an author-provided interpretation.Source: §5.3 · p. 9
- TARGET’s direct access to components supported hypothesis testing, while MULTI-VIEW and TARGET helped participants combine a data picture with mathematical patterns.Participants described structural, spatial, and targeted navigation as facilitating hypothesis-testing and pattern-making behaviors.Source: §5.3 · p. 9
- Cursors, roadmaps, and ways to return to an origin helped participants orient themselves; without navigation aids, some found it easy to get lost.Participants valued moving through MULTI-VIEW’s grid and jumping around TARGET, while one participant described difficulty backtracking without such tools.Source: §5.3 · p. 9
6. Discussion and Future Work
The discussion presents structure, navigation, and description as dimensions for richer screen-reader visualization experiences, while treating alt text and data tables as the accessibility baseline. It identifies authoring, standards, individual needs, modality, and interaction questions for future work.
6. Discussion and Future Work
- Structure, navigation, and description compose into screen-reader experiences richer than alt text, data tables, or current ARIA can provide.The discussion frames these dimensions as promising directions for accessible interaction and representation.Source: §6 · pp. 9–10
6.1. Enabling Richer Screen Reader Experiences Today
- Alt text and data tables remain a good accessibility baseline; authors should add richer information access patterns only after these are in place.Source: §6.1 · p. 10
- Current authoring methods leave authors to handcraft each visualization’s structure, navigation, and descriptions, risking friction and mismatches with user preferences.The authors propose toolkit defaults that authors can customize, including hierarchical structures derived from visualization specifications.Source: §6.1 · p. 10
- Current standards lack a standard way to identify the element selected by the screen-reader cursor, limiting support for the design dimensions.Source: §6.1 · p. 10
6.2. Studying and Refining the Design Dimensions
- Blind and low-vision users can have different preferences: magnifier users favored verbose written descriptions, terser speech, and in-situ tooltips.Screen readers combined with magnifiers also helped one participant detect a point hidden behind another point.Source: §6.2 · p. 10
- The study focused on screen-reader text-to-speech and excluded multisensory modalities, whose combinations may help some users but confuse others.Participants cited possible benefits from combining modalities and difficulty mapping between sound and text.Source: §6.2 · p. 10
6.3. What are Accessible Interactions for Data Visualizations?
- For screen-reader visualizations, the boundary between reading and interaction is unclear; some navigation resembles shifting attention, while a binary tree felt like zooming into data.Source: §6.3 · p. 10