He surely didn’t know it at the time. But Adrian Aoun’s godson, who lives in London and just turned 5, has been supplying the Seattle entrepreneur with inspiration for years. The setting is their regular video chats, which have allowed Aoun to watch the kid’s mastery of language take hold—and even do a little casual research.
“The other day I was talking to him and he said, ‘I want a sandwich.’ And I’m thinking, how did he know that?” Aoun says. “So I asked him, how did you know that? And he just looks at me like, huh?”
The kid, of course, wasn’t deconstructing the way he learned language. He heard it all around him and, as youngsters do, put the pieces together. He probably started with a few simpler phrases and worked his way up, maybe garbling the first few attempts and listening as adults corrected him. Before long, he was chatting away.
So if a small human can do this—it happens every minute of every day, all over the world—why is online search still so bad at answering simple questions? In an age of ever-expanding computing power and growing engineering prowess, why isn’t the Web easier to sift?
That’s the question that motivates Wavii, a roughly two-year-old company founded by Aoun. Based in downtown Seattle, Wavii is applying machine-learning software to the floods of information online, seeking to turn all of those words into a more orderly, Facebook-like news feed.
The service is still in its early stages—after a long period of secrecy, Wavii is just now moving beyond a private beta test to sign up more users, available on the Web and iPhone. But the results so far offer a tantalizing look at how a social media-like user interface could be applied to pretty hard-core information-extraction technology, the kind of stuff that will undoubtedly be the future of search.
“It makes you wonder why something like what we’re building doesn’t live inside Google Plus,” says Aoun, an entrepreneur who previously worked for Microsoft and MySpace parent Fox Interactive.
Others have tried to unlock this puzzle, an area known broadly as “semantic search.” But as University of Washington professor and Decide co-founder Oren Etzioni has pointed out, there still isn’t enough progress being made. Ask Google a question, and you’re going to get back a list of Web pages that have part of the answer—unstructured stabs at the real solution that Etzioni likens to the index at the back of a book. (Etzioni, incidentally, is a fan of Wavii’s work and says its technology is a leap ahead.)
Other startups and innovators are likewise trying to create more personal, intelligent way to sift news and information online, from the mobile apps of Vulcan-backed Evri to social-signals aggregators like Summify, which was recently acquired by Twitter.
At this point, it seems highly unlikely that there will just be one winner to claim this prize—the pace at which Facebook, Google, Microsoft, Apple, and Amazon will battle each other for the data we consume and share doesn’t appear to be slowing down anytime soon. Wavii, with a long list of prominent early stage investors, isn’t focusing on business model yet—it’s in the users first, revenue later realm. But you can see a path for advertising based around users’ interests, if some big technology company doesn’t decide to bring the company onboard—Wavii has reportedly been courted for acquisition already.
“A lot of people have been interested in what we do,” Aoun says.
Wavii’s team of 25 engineers, which includes people with experience from major tech companies and academia, focuses on using modern machine learning technologies to train its software. That language-learning machine is pointed at the Web’s open-ended source of material, where it pores through countless sources of text and assembles related items into nice little packages for the user.
In practice, that means Wavii’s software can see four articles online discussing a possible shortage of MacBook Pro laptops, and put together its own summary sentence, image, and package of stories drawing those threads of information together.
It’s no mistake that Wavii’s shorthand for its service is “Making Facebook out of Google.” On the design side, Aoun says Facebook has trained connected consumers to expect their information in discrete chunks stacked on top of one another, with room for personal interaction and ways of sharing that information tucked underneath.
Not surprisingly, Wavii doesn’t just echo Facebook’s concept—it’s tightly integrated with the service, both for signups and sharing information (Twitter will be more fully integrated in the future, Aoun says).
Aoun thinks Wavii can stand apart on both design and technology. Too many of the people attacking the problem of decoding text, he says, have relied on older-generation versions of natural language processing that incorporates grammar rules from the study of linguistics, rather than a machine-learning approach that attempts to turn the translation of writing into a mathematical problem.
“We’re treating this with a fresh set of eyes, as opposed to the Microsofts and the Googles that have all looked at this with the same view for 20 years and are all going down the same path,” he says.
It’s no coincidence that Aoun, a software guy, is trying to decode language. His father, Joseph Aoun, is a career linguist who studied under Noam Chomsky at MIT and now serves as president of Northeastern University. That meant Adrian’s childhood was spent in the company of language researchers who often looked to him to settle debates about their theories.
“From an early age I knew that the rules they were using never made sense. Because whenever they got into a debate about which rule was right, they’d just turn to me—literally—and say, ‘Which one sounds right?'” Aoun says.
They were asking the kid because, as a young native speaker, his ear was perfectly tuned. It’s the same process that helped Aoun’s young godson piece his first sentences together, and—if the technology works as hoped—the kind of intuitive information sifting we’ll be using in a few years.
“The little kid who learned ‘I want a sandwich,’ they still don’t know all of language. They don’t know what being arrested is—you don’t understand that until the age of 7. They don’t understand white-collar crime until 14,” Aoun says. “We taught our system to come to us, much like the little kid, to say ‘I found a pattern—tell us what this pattern means.'”
