Accessibility of the Web with a Computer Game

Images on the Web present a major accessibilityassign proper descriptions to arbitrary images. A
issue for the visually impaired, mainly because the"proper" description is correct if it makes sense
majority of them do not have proper captions.with respect to the image, and sufficient if it gives
This paper addresses the problem of attachingenough information about its contents. Rather
proper explanatory text descriptions to arbitrarythan designing a computer vision algorithm that
images on the Web. To this end, we introducegenerates natural languagedescriptions for
Phetch, an enjoyable computer game thatarbitrary images (a feat still far from attainable),
collectsexplanatory descriptions of images. Peoplewe opt for harnessing humans. It is common
play the game because it is fun, and as a sideknowledge that humans have little difficulty in
effect of game play wecollect valuabledescribing the contents of images, although
information. Given any image from the Worldtypically they do not find this task particularly
Wide Web, Phetch can output a correctengaging. On the other hand, many people would
annotationfor it. The collected data can be appliedspend a considerable amount of time involved in
towards significantly improving Web accessibility.an activity they consider "fun." Thus, like the ESP
In addition toimproving accessibility, Phetch is anGame, we achieve our goal by working around
example of a new class of games that providethe problem, and creating a fun game that
entertainment in exchange forhuman processingproduces the data we aim to collect.
power. In essence, we solve a typical computerWe therefore introduce Phetch, a game which, as
vision problem with HCI tools alone.a side effect, generates explanatory sentences
The Web is not built for the blind. Only a smallfor randomly chosen images. As with the ESP
fraction of major corporate websites are fullyGame, we show that if our game is played as
accessible to the disabled, let alone those ofmuch as other popular online games, we can
smaller organizations or individuals . However,assign captions to all images on the Web in a
millions of blind people surf the Web every day,matter of months. Using the output of the game,
and Internet use by those with disabilities growswe mention how to build a system to improve
at twice the rate of the non-disabled .the accessibility of the Web.
One of the major accessibility problems is the lackDesign of a Useful Game
of descriptive captions for images. VisuallyA traditional algorithm is a series of steps that
impaired individuals commonly surf the Web usingmay be taken to solve a problem. We consider
"screen readers," programs that convert the textPhetch as a kind of algorithm. Analogous to one,
of a webpage into synthesized speech. AlthoughPhetch has well-defined input and output: an
screen readers are helpful, they cannot determinearbitrary image from the Web and its proper
the contents of images on the Web that do notdescription, respectively. Because it is designed as
have descriptive captions. Unfortunately the vasta game, Phetch needs to be proven enjoyable.
majority of images are not accompanied byWe do so by showing usage statistics of a
proper captions and therefore are inaccessible tooneweek trial period. Because it is designed to
the blind (as we show below, less than 25% ofcollect a specific kind of data, Phetch's output
the images on the Web have an HTML ALTneeds to be proven both correct and sufficient.
caption). Today, it is the responsibility of WebWe prove this through a specifically designed
designers to caption images. We want to take thisexperiment.
responsibility off their hands. We set our goal to