Acrolinx: A Controlled-Language Checker Turned into an Accessibility Evaluation Tool for Image Text Alternatives
Silvia Rodríguez Vázquez, Sabine Lehmann · 2015 · Proceedings of the 12th International Web for All Conference (W4A) · doi:10.1145/2745555.2746676
Summary
This paper presents a novel application of controlled-language checking technology to a persistent accessibility problem: evaluating the quality of image text alternatives, not just their presence. The authors observe that while most web accessibility evaluation tools can detect whether an alt attribute exists on an image element, they rarely perform any linguistic analysis of its content — leading to high false positive rates where technically present but inappropriate alt text (such as filenames, placeholder text, or overly generic descriptions) passes automated checks. The researchers adapted Acrolinx, a commercial controlled-language software designed for content quality assessment, by developing a set of 40 accessibility-oriented style rules specifically for French-language alt text. These rules were derived from two sources: a comprehensive review of existing guidelines for creating high-quality text alternatives (including W3C techniques and ISO/IEC TS 20071-11:2012) and a detailed corpus analysis of 52 Swiss websites comprising approximately 2,000 pages with over 10,000 images, of which nearly 8,000 had unique text alternatives. The rules were implemented using the Acrolinx Linguistic IDE built on Eclipse, following a formal error description framework.
Key findings
The 40 style rules are classified according to image content type — functional, descriptive, or uninformative — and check for common alt text quality problems such as missing context (e.g., "Home" instead of "Homepage of [site owner]"), inadequate descriptions of social media actions (e.g., "RSS" instead of "Subscribe to RSS feeds"), and other linguistic patterns associated with poor alt text. When a rule is violated, the tool provides the specific error, the associated rule, and where possible an improvement suggestion. Rule documentation includes introductory context on image accessibility, a detailed explanation of the problem, and illustrative examples of appropriate and non-appropriate alternatives. The Acrolinx Batch Checker can process both live websites and local HTML/XML files, producing detailed per-page reports. Preliminary evaluation comparing Acrolinx against IBM's aDesigner tool in a study with web localisation professionals found that Acrolinx facilitated alt text quality improvement and provided easier-to-understand suggestions. The authors note the tool's potential extends beyond alt attributes to checking other attribute values or full page body content against accessibility-tailored rule sets.
Relevance
This paper addresses a genuine gap in accessibility tooling that persists today: automated tools are good at detecting missing alt text but poor at evaluating whether the alt text that exists is actually useful. The controlled-language approach — using linguistically-informed rules derived from corpus analysis — represents an intermediate step between simple presence/absence checking and full natural language understanding. For content authors and localisation professionals who are not accessibility specialists, this kind of guided feedback with concrete suggestions is more actionable than generic WCAG failure notices. The focus on French also highlights that alt text quality is a multilingual problem, not just an English one. The corpus analysis of nearly 8,000 unique alt texts on Swiss websites provides useful empirical grounding. However, this is a short extended abstract that reports only preliminary evaluation results, and 40 rules for one language represent early-stage work. The broader principle — that accessibility evaluation should assess content quality, not just structural compliance — is increasingly important as automated testing matures.
Tags: alt text · image accessibility · automated testing · controlled language · natural language processing · accessibility evaluation · web accessibility · content quality
Standards referenced: WCAG 2.0 · ISO/IEC TS 20071-11:2012 · HTML5