Bridging the Interpretation Gap in Accessibility Testing: Empathetic and Legal-Aware Bug Report Generation via Large Language Models
The system transforms technical logs into stakeholder narratives, boosting perceived urgency and legal risk awareness by 40%.
A research team from Waseda University and other institutions has introduced HEAR (Human-cEntered Accessibility Reporting), a novel framework designed to solve a critical bottleneck in digital accessibility. While automated testing tools can detect interface violations, their low-level, technical outputs often fail to motivate non-specialist stakeholders like product managers and designers to act. HEAR bridges this 'interpretation gap' by using LLMs to semantically reconstruct the user interface context and generate bug reports that clearly articulate the real-world impact on users with disabilities and the associated legal compliance risks.
The framework operates by first taking the raw output from an existing accessibility scanner. It then performs semantic slicing and visual grounding to understand the UI context. Next, it dynamically injects a disability-oriented persona (e.g., a screen reader user) matched to the specific violation. Finally, it uses multi-layer reasoning to explain the physical barrier, functional blockage, and relevant legal standards like the Web Content Accessibility Guidelines (WCAG).
In an evaluation on real-world issues from four popular Android apps and a user study with 12 participants, HEAR proved highly effective. The generated reports were factually grounded and, compared to raw technical logs, substantially improved perceived empathy, urgency, persuasiveness, and awareness of legal risk among stakeholders. Importantly, the system achieved this without imposing significant additional cognitive burden, making it a practical tool for integrating into development workflows to prioritize and fix accessibility issues.
- HEAR uses LLMs to transform technical accessibility scanner output into stakeholder-friendly narratives explaining user harm and legal risk.
- In a user study (N=12), reports generated by HEAR boosted perceived empathy, urgency, and legal risk awareness compared to standard logs.
- The framework works by reconstructing UI context, injecting disability personas, and performing multi-layer reasoning on physical, functional, and compliance impacts.
Why It Matters
This directly addresses the compliance and ethical imperative for accessible software by making technical bugs compelling to the business stakeholders who fund fixes.