Combined use of artificial intelligence and crowdsourcing to provide alternative content for images on websites

Web accessibility can be efficiently evaluated in both automated and manual ways, but fixing accessibility issues on live websites is still a major challenge, as it requires attention, dedication and knowledge of Assistive Technology requirements. A website revision system is thus proposed, relying on Artificial Intelligence to produce alternative text, and on crowdsourcing to correct it, for images or other generic types of resources. This solution is integrated in a web application called Farfalla, which also offers easy integration of accessibility options in websites. A major advantage offered by this approach is the independence from a specific browser or operating system: the proposed architecture is highly portable and platform-agnostic. The basic idea that will be put forward here is that Artificial Intelligence can be used to generate basic information about images in websites, and crowdsourcing would serve as a way for refining that information. As the development of the project is still in progress, the most important implementation issues are analyzed: they include context-aware alternatives, localization, scalable deployment and input validation.