Automated translation? What could possibly go wrong?

It was 1968. Fully automated translation was just around the corner.

Or so I learned in a computer science class.

Of course there was officially no such discipline at the time, so it was actually an applied math class, but never mind that technicality. Basically we were studying artificial intelligence with an admixture of linguistics and logic. Keep in mind that this was more than half a century ago, before Google and email and even before the internet! But people were optimistic. Computer scientists are notoriously (over)optimistic.

We learned about this new movie called 2001: A Space OdysseyWe discussed the plausibility of the computer HAL, which spoke and understood English. If you’ve never seen this great film (IMHO), go watch it right now! You’ll get one answer to the “What could possibly go wrong?” question. And do note that we’re now two decades past the time frame of the movie, and the capabilities of Siri and Alexa and so forth don’t even approach HAL’s. Computer scientists are notoriously (over)optimistic.

Back in 1968 people often looked for humorous errors in what could be considered the precursors of Google Translate. They weren’t hard to find. Professors and other researchers considered them to be cautionary tales that suggested promising lines of research. The errors would surely be overcome before 2001. Computer scientists are notoriously (over)optimistic.

Two examples still stick in my mind today. They both came from an experimental protocol that grew out of the Cold War: automated translation between English and Russian. Since the American scientists didn’t speak Russian, the easiest way to test the quality of the translations was to have the computer translate from English to Russian and then back to English.

  • One example that I remember was the statistical term “standard error.” The bidirectional translation of “What was your standard error?” was “What was your usual mistake?”
  • One scientist got the idea of translating proverbs and metaphors as a particularly tough challenge. The computer translated “The spirit is willing but the flesh is weak” as “There is enough vodka but not enough meat.”

Maybe those stories are apocryphal; just because my professor quoted them doesn’t make them true. But still, “si non è vero, è ben trovato,” as they say in Italy.

A recent example that may or may not involve computers comes from translating from Chinese, a notoriously difficult endeavor. Check out the link! Warning: NSFW.


Categories: Linguistics