Paper Reading #16: Mixture model based label association techniques for web accessibility

Reference Information
Title: Mixture Model based Label Association Techniques for Web Accessibility
Authors: Muhammad Asiful Islam, Yevgen Borodin, I. V. Ramakrishnan
Venue: UIST '10 Proceedings of the 23nd annual ACM symposium on User interface software and technology


Summary
With the continued growth of the internet, it becomes increasingly necessary for blind users to have access to it. The dominant method of accessibility to the web for people with difficulty seeing, is audio based browsing. Browsers designed to do this attempt to associate text labels with relevant elements on the page, and then read the labels to the user. This can range from very easy, to extremely difficult depending on how the labels are associated in the HTML code or if there are labels at all.


This article proposes a solution aimed to address this issue by using their Finite Mixture Model algorithm to associate text with nearby elements. Elements that don't have labels or candidate labels are assigned labels from a database. The most likely match is calculated by a similar algorithm. Once key elements on the page have labels, they can be read to the user to describe what is on the page.


Discussion
The article was interesting, and the approach is without a doubt effective at labeling elements, but I feel that the entire solution needs to be reworked. Reading small parts of pages designed to be visible to the user is a poor substitute for being able to actually see the page, and I have a feeling that getting anything done with audio based browsers, regardless of how well labels are associated, would take a very long time.


Instead perhaps one of the substitute sight technologies would be a better place to spend research money on this area. I know there is ongoing work to map visual elements to a person's tongue. While a bit awkward to use, perhaps perfecting a technique like or similar to that would be the best approach for blind people.