What's hot:
 

, , ,

Watson and natural language

Whether you know it or not, you have been trained to think like a web copywriter. Suppose you want to know what time the Museum of Natural History closes. What would you search for on Google?

You’d probably search for something very similar to “Museum of Natural History hours.” Obviously. But it’s not obvious. If you called the museum on the phone to learn the same thing, you’d ask the person who answered, “What time does the museum close?” You’d speak in natural language.

I was thinking about language tonight as I watched the IBM supercomputer Watson handily beat two human opponents on Jeopardy! Watson is very good at knowing the answers to trivia questions, but the most remarkable thing is that the computer is so adept at responding to natural language.

Watson understood the punny, creative questions that the Jeopardy! writers are so good at. When it buzzed in first, Watson answered in a friendly, synthesized voice. You could tell it was a computer, but it was also able to win using the human-biased interface devices that are a requirement of the game (a button and vocalized sentences in the form of questions).

Many futurists predicted that computer interfaces would take the form of natural language. By now we should be talking to our computers the way Picard talks to the Enterprise’s computer, or the way the Jetsons talk to Rosie. Heck, we were supposed to use natural language to search the web with Ask Jeeves over a decade ago.

Yet by any measure, natural language interfaces on the Internet have failed. Google is the most popular way people use language to tell a computer what to do, and it works much better if you just enter disjointed phrases. Do a Google search for “Museum of Natural History hours” and, bang, first result is exactly what you want—the official hours page for the Museum of Natural History in New York City. Enter “What time does the Museum of Natural History close?” and you get a muddle of low-quality results; the page you want isn’t even on the top 10.

Researchers have plenty of theories as to why natural language user interfaces don’t work.

  • Computers sometimes lack knowledge that humans take for granted that’s required to interpret questions correctly.
  • The user, speaking to the computer as an equal, thinks the computer is far more capable than it really is, which inevitably leads to disappointment.
  • Some people just find it off-putting to speak to a machine.

Whatever the case, most of us never ask a computer a question. Instead, we enter commands. When we perform a search, we unconsciously translate our question into the title of a document in which the answer is likely to appear. We are thinking in terms of keywords, titles and important phrases. Just like web copywriters do.

The danger here is that with the importance of Google and SEO, we’re caught in a vicious circle. Copywriters are trying to guess what web surfers guess copywriters will write. This will lead to ever-more-generic language and boring, vapid writing.

No wonder the computers are beating us on game shows.

* * * *
Bonus: On his Final Jeopardy! slate tonight, Ken Jennings wrote, “I for one welcome our new computer overlords.” The site Know Your Meme has a page about where that phrase originated.

— By Daryl Lang. Filed under Copywriting, Language

Facebook Conversations