Evri, a startup spun out of Microsoft co-founder Paul Allen’s investment firm Vulcan Capital and headed by a long-time software executive from Amazon.com, has an idea for how natural language processing and semantic technology can help people navigate the Internet. And their idea has nothing to do with search.
Like all semantic startups, Evri is all about helping machines connect concepts, using the structure and meaning of human language to create some order. The first wave of similar companies, like Powerset, Hakia and the still-stealthy Cuill, are all about using that concept to out-do Google and return better results. But where traditional search requires a definite aim and returns answers — good or bad — a better metaphor for Evri is Internet tourism, with an eye toward serendipitous discovery.
When surfing content on the Internet, you’ll be able to click a link for a person, place or thing you’re interested in and move into a sort of parallel web built by Evri. The company automatically constructs “nodes” for concepts like Paris or Megan Fox based on their relationships to other nodes in the database. Each node contains some information, like a Wikipedia summary or video, about the thing you’re interested in, and links to associated content.
That may sound something like an automated Mahalo, but CEO Neil Roseman says there’s a key difference. On sites like Mahalo and Wikipedia, “You go to a page on the Holy Grail there, you read it and you’re done. With us, we want to help people look into the content deeper,” he says.
For example, if you were reading a news article about Paris and clicked through, you might find links out to more content about Marais, the neighborhood the article deals with, or a local musician, Julien Ribot. The most closely associated content comes up first, so you can read more about the subjects the article deals directly with if you want. But because the semantic linking recognizes any connection, you could also find yourself following a link to another person, place or thing that happened to have been mentioned alongside Marais or Ribot elsewhere.
The result should be an entertaining journey for the casual surfer, reminiscent of a myth my dad told me about how the local roads were planned in the rural Virginia county where I grew up. First, he said, engineers dipped a cow’s tail in paint. Then they let it free to roam, and cleared a road anywhere they found paint. The result were winding, meandering roads that did a better job of showing you the local scenery than getting to a definite destination. That’s basically what Evri promises to do for the Internet.
Of course, any Wikipedia user who has spent hours clicking through related links already knows the concept is good. The difference is the delivery. Evri plans to launch first with publishers like newspapers, offering either links in the content or a separate widget. And because its nodes are permanent pages, they should show up in search results, just like Wikipedia pages.
Given Google’s continuing dominance of search, this angle seems somewhat more likely to have a broad appeal. The only question is the technology. It looked reasonably good when Roseman gave me a tour of the site, but also contained flaws and confusing or seemingly unrelated links. Steve Hall, a partner at Vulcan, says the technology is still coming together.
“You’ve got the equivalent of a bunch of 7th grade grammar students going out on the web and collecting information bases on verbs and nouns,” he says.
The aim is to make the software almost as smart as the web surfer using it. Roseman thinks that the basic concept is going to quickly find traction, and that there’s a big win in store for the company that does best. He says he expects to have only about 18 months to perfect the current iteration before needing to roll out even bigger and better implementations.
Evri is based in Seattle, and has been funded entirely by Vulcan to date. It’s planning on opening up to other investors soon.