Navigation aids such as headers and internal links provide vital support for screen-reader users on web documents to grasp a document’s structure. However, when such navigation aids are unavailable or not appropriately marked up, this situation can cause serious difficulties. This paper presents the design and evaluation of a tool for automatically generating navigation aids with headers and internal links for screen readers with topicalisation and labelling algorithms. The proposed tool uses natural language processing techniques to divide a web document into topic segments and label each segment in two cycles based on its content. We conducted an initial user study in the first cycle with eight blind and partially-sighted screen reader users. The evaluation involved tasks with questions answered by participants with information from texts with and without automatically generated headers. The results in the first cycle provided preliminary indicators of performance improvement and cognitive load reduction. The second cycle involved co-designing an improved version with two blind experts in web accessibility, resulting in a browser extension which injects automatically generated headers and in-page navigation with internal links, along with improvements in the generation of labels using OpenAI’s ChatGPT. The browser extension was evaluated by seven blind participants using the same four texts used to evaluate the preliminary prototype developed in the first cycle. With the two development cycles, the study provided important insights into the design of navigation aids for screen-reader users using natural language processing techniques, including the potential use of generative artificial intelligence for assistive technologies and limitations that need to be explored in future research.
