Toward supporting quality alt text in computing publications
Candace Williams, Lilian de Greef, Ed Harris, Leah Findlater, Amy Pavel, Cynthia Bennett · 2022 · Proceedings of the 19th International Web for All Conference (W4A) · doi:10.1145/3493612.3520449
Summary
This paper investigates the state of alternative text for figures in computing research publications, an understudied context where figures often convey dense, complex visual information quite different from social media or news imagery. The researchers conducted two complementary studies. First, they analyzed 300 figures randomly sampled from ACM CHI papers published between 2019 and 2021, categorizing each figure by type (data visualizations, diagrams, images, tables/text blocks), assessing whether it was a composite of multiple elements, and grading the quality of its alt text, caption, and combined description on a 0-4 scale (from Blank to Very Descriptive). Second, they interviewed 10 researchers in HCI and related fields with varying levels of alt text experience, observing them write alt text for their own figures using published guidelines and discussing their workflows, challenges, and support needs. The research team included a blind researcher as one of the lead mentors, grounding the work in lived experience with screen reader access to academic papers. The study was motivated by the fact that while ACM CHI had been increasing its accessibility resources for authors — including alt text guidelines and submission form fields — little was known about whether these efforts were actually producing quality descriptions.
Key findings
The figure analysis revealed that while 81% of figures had alt text in the short or long fields, 20% of that alt text was completely uninformative (e.g., just "Figure 1"), meaning only 65% of figures had any substantive alt text. Considering the full description (caption and alt text combined), only 23% of figures scored 4 (Very Descriptive — describing most visual content needed to understand the figure), while 38% scored just 2 (a summary or single piece of information). Quality varied significantly by figure type: data visualizations were most likely to receive high-quality alt text (26% scored 4), while tables and text blocks captured as inaccessible images mostly had blank alt text. A substantial 38% of figures were composites of multiple elements — a format largely absent from current alt text guidelines. Longer descriptions generally scored higher, contrasting with common guideline advice to keep alt text brief. Quality improved year over year from 2019 to 2021, correlating with increased author resources. The interviews revealed that authors spent far more time creating figures than writing alt text, with seven of eight experienced authors waiting until the day the paper was due. Authors struggled with where to put information (caption vs. alt text vs. body text), when interpretive descriptions were appropriate, and how to structure alt text for composite figures. Workflow barriers were significant: alt text was invisible during drafting in LaTeX, the ACM publishing pipeline created inconsistent behavior between LaTeX and PDF alt text, and using Acrobat Pro to add accessibility tags was described as tedious and error-prone. All 10 participants found the guidelines helpful but felt under-prepared, even those with years of experience.
Relevance
This research is directly relevant to anyone producing or consuming academic publications, and its findings generalize to any context involving complex figure accessibility — technical documentation, reports, educational materials. The finding that brevity guidelines may actually work against quality is important: for dense scientific figures, more words generally meant better descriptions, suggesting that alt text guidance needs to be figure-type-specific rather than one-size-fits-all. The 38% composite figure finding exposes a major gap in existing guidelines, which typically show examples of single-element images. For accessibility practitioners, the paper highlights that tooling and workflow integration matter as much as guidelines — authors knew alt text was important but their tools made it invisible and their deadlines made it an afterthought. The research agenda the authors propose — better process integration, figure-type-specific education with good and bad examples, and authoring tools that surface alt text alongside figures — provides a practical roadmap for improving document accessibility in any publishing context.
Tags: alternative text · scientific figures · data visualization · screen readers · document accessibility · academic publishing · blind and low vision
Standards referenced: WCAG 2.1