Skip to content
Assistivox

Aggressively Accessible

Most digital content was authored by sighted people for sighted audiences. The structure is visual. The meaning is often carried by layout, images, and formatting that was never described in words. Federal accessibility mandates now require public entities and universities to provide that description, and content still arrives with images that have no useful description and structure that only makes sense on a screen. Some of this is a failure of authoring practice. Some of it is a property of the format itself. PDF and PowerPoint are visual representations. They encode position and appearance, not structure. Even a carefully authored PDF may carry no information about reading order, heading hierarchy, or the meaning of a diagram. There is no tagging practice that fully compensates for a format that was not designed to express structure. A system that depends on that tagging work being done correctly is conditionally accessible, and the condition is one the unsighted user cannot control.

Aggressively accessible is a philosophy about what an accessible system should do instead. It has two parts.

  1. A spoken language interface must exist for all input and output. Every action the system can perform must be reachable by voice. Every point of feedback from the application to the user must be spoken. Screen readers describe the visual screen for a nonvisual user. They do not give the user a voice command for each action the user could select.
  2. The system must do the work that authors did not do. Documents arrive as they are. An aggressively accessible system uses all available computation, OCR, vision models, language models, structural analysis, to extract meaning, reconstruct structure, and generate descriptions from content that was never prepared for a spoken interface.

These two parts depend on each other. A native spoken language interface is what makes the second commitment meaningful. Without a stable, first-class spoken path, there is nowhere to deliver the results of that computational work.