In its relatively short life, Facebook has grown more and more visual; its users upload and share some 350 million photos every day. The social network says that evolution has made its platform a more “fun and expressive way” to communicate, but it has also marginalized a subset of its users: the blind.
On Tuesday, Facebook is introducing new technology to help bring that group back into its fold.
Facebook’s new automatic alternative text, or automatic alt text, feature will generate photo descriptions for users relying on a screen reader--a tool for the visually impaired that reads the web pages aloud. Before Tuesday, users of screen readers would hear the name of the person who shared the photo, followed by the term “photo” when they’d scroll past an image in their Facebook news feed. Now, those same users will hear a basic description of the photo such as, “image may contain three people, smiling, outdoors.”
The new feature relies on Facebook’s visual recognition engine, an artificial intelligence-powered technology that processes every single photo and video uploaded to the social network. So far, Facebook has trained the engine to recognize dozens of items--car, mountain, tree, sunset, basketball court, ice cream, pizza. Automatic alt text then mentions those objects in the special captions it generates for the blind.
Facebook users are, of course, able to post text along with their photos, but they often don’t describe what’s actually in a given picture. Someone may write “Sunday night splurge!” on a photo of a pizza. Facebook’s new tool will provide blind users with vital information--”image may contain: pizza, food”--that puts a vague description in better context.
Such descriptions will allow blind users to join conversations they were previously left out of, says Matt King, an accessibility specialist in user interface engineering and a member of Facebook’s accessibility team. The new tool promotes “equal access to information for people who are blind,” he says, which is especially important to young blind users whose peers live on social media.
At launch, the automatic alt text tool will feature about 100 concepts, which can be identified with accuracy of between 80% and 90%. Facebook is taking a conservative approach to adding more items, favoring precision over quantity for fear of misleading visually impaired users. The feature was introduced as a prototype in November, and on Tuesday it will be available to iOS users. The company plans to expand it to Android devices and desktop users in the future. The initial version of the technology will be in English; other languages will be added later on.