Forgotten your password?
background1
Science and Technology

A search engine for metaphors

Debbi Evans
By teaching computers to recognise metaphors, the YossarianLives! search engine hopes to replace an unthinking reliance on Google with something altogether more meaningful

Much like the Heinz of internet search, it’s said Google uses 57 different filters to personalise the information we see online. Yes, all of us. Even you over there. This may sound like a perfectly reasonable and relatively innocent way to parse a gigantic world of sources, but these filters carry a heavy bias and can stop you from seeing the bigger picture. For example, if Google knows which political party you support, based on your choice of online reading materials, it will continue to tailor your results to fit that world view. And lots of those filters continue to work when you’re not even online.

Furthermore, now that web searches are frequently conducted on mobiles, there is also an overlap between your activities on the web and your activities in the physical world. Next time you search for a café while you are walking down the road, ask yourself: How much of the city is Google stopping you from seeing?

For a few years now, analysts and academics (including John Hagel, Eli Pariser, Aleks Krotoski and Adam Greenfield to name a few) have fought to make people aware of the dangers of this over-reliance on Google. There’s another 
more philosophical argument surrounding all of this, too – 
on some level, missing out on random finds outside our sphere of reference detracts from the human experience. Conversations with random strangers often lead to exciting new ideas and discoveries – but what happens to those discoveries when there’s no randomness any more?

I’m feeling lucky

Enter a different kind of search engine. YossarianLives!! was built to generate random, serendipitous results as a counter to these kinds of hidden information filters. Here’s what happens when you search for something: a picture comes up. That’s it. No signposts, no key words, no explanations. Through some computational wizardry, the algorithm has decided that this picture is conceptually related to whatever it is you’re looking for. Can’t see the link? Well, you need to try harder. It’s there somewhere: find it.

This frequently painful, sometimes riddlesome process, 
the founders believe, is the only way we’ll ever get to new forms of knowledge. But how on earth do you teach a computer to recognise metaphor in the first place? We asked Yossarian’s Chief Technology Officer and computational linguist Katia Shutova: “By essentially running through the text, splitting out the literal from the metaphorical, and paraphrasing the metaphors with literal expressions,” she says. This sounds simple enough in theory, but the finer points of definition are quite complex.

Robot researchers

“You still need to think about the conceptual mechanisms of how metaphor works – the relation between two distinct concepts from different domains,” says Shutova. It was at a meeting with Yossarian’s co-founders Dan Foster-Smith and J. Paul Neeley, whose backgrounds are on the design side of technology, that she first understood how the visual elements of the service might work.

“Even before I met them, one of my long-term research goals was to make a machine that could think for you and make scientific discoveries.” Shutova’s oversight was assuming the machine would then use metaphor to generate a set of hypotheses and test them on its own – to go through the whole process of scientific discovery unassisted. Dan and J. Paul explained that they wanted to build an assistive tool that would help the user to discover ‘new knowledge’ rather than having the machine do all the work. “It was completely mind-blowing that you could create something useful without having the computer do the whole task.”

Visual logic

Yossarian offers different levels of search, depending on how abstract you want your results to be. Shutova uses the example of love to illustrate her point: “Love is a river, or a staircase to the sky, or a heavy stone around your neck. All of those images highlight a different aspect of love that you might not necessarily think of, but if you want to think about love creatively, you need to effectively understand them all.

“If you [selected] concrete concepts as an output for love, then you’ll come up with things like stone or a river. The more abstract metaphors will be something like ‘Love is a devil’ or ‘Love is a drug’. Concrete concepts might be better for an ad campaign, where you’re trying to really visualise the concept.”

And this is currently where Yossarian is pitching  it– the founders have worked out a deal with Getty Images, the photography library: use the free search, buy an image if you like the results. And there’s a premium version in the works with enhanced features, which the Yossarian team hopes will help creative agencies approach their work differently.

“There are definitely other sites it could be integrated with to give people a new search experience,” says Shutova. “In the future we also plan to work with text – but that’s more for proper research. Down the line we might even tailor it to scientists – input a concept from one scientific domain and it comes up with a structurally similar concept from another domain. This is how a lot of new information happens, right? When we put two domains together and learn something from both of them.  This could have a big impact on the world, which is the part which is really important to me.”

yossarianlives.com

Related

Advertisement

advert
To continue reading...
Join
It's free and only takes a moment. You'll get:
  • - Access to libertine content
  • - The weekly newsletter
  • - Alerts about events
  • - Competitions and offers (but we won't spam you, we promise)
Continue
Log in
Email address *
Password *
Remember me (see our Cookies Policy)
Forgotten your password? Privacy policy and Terms & Conditions Log in
Forgotten your password?
Enter the email address you joined up with and we'll send you an email with a password reset link.
Email address*