Automated Alt Text: Can a Computer Comprehend What Humans Can?

By Rachel Guise, PPS Editor  −   If you’re not sure what alternative (alt) text is, or simply need a refresher, take a look at our blog titled Alt-Textually Speaking: What Is Alt Text and Why Do We Need It? for a quick crash course on this publishing term. We also have a blog on the possibility of technology replacing an editor. But how do these two go hand-in-hand? Well, these go hand-in-hand mostly due to the fact publishing has never been easier than it is today, especially if you want to self-publish or be as hands-on as possible during the process. One of the most important parts of the publishing industry right now is making sure your content is fully accessible.

As noted in our most recent blog on computer-assisted editing vs. human editing, there is no doubt that software programs or artificial intelligence (AI) can greatly reduce the time spent editing content. This is also true when it comes to creating content, such as alt-text for photos or art used in your publication.

Let’s look at an example.

Does this computer-generated alt-text describe the photo accurately? Right off the bat, the word person is listed twice. That’s your first red flag. But the generated alt-text doesn’t convey the general emotion of the photograph. Acknowledging that, a lot could be implied here. Is he frustrated? Does he have a headache? Is he tired? Simply noting that the picture contains a person just isn’t enough. The word picture also doesn’t lend to the overall image. Is this a cave drawing? Is this a painting from the 1500s? Is it a sepia-toned photograph?

So many questions, so little alt-text.

This is just one example of how AI simply can’t interpret all aspects of a piece of art that is being used to enhance a publication. Sure, sometimes art is inserted just to take up space, but sometimes it’s crucial to the content being provided. And as we can see, there is a big difference in what a computer is going to describe to the reader!