Incoherent Babble

I found a post over at Download Squad today, entitled Google wins at machine translation, which links to this story from CNet’s News.com.

From that article:

Google scored the highest in Arabic-to-English and Chinese-to-English translation tests conducted by the National Institute of Science and Technology.

Well, even though their tests were in Arabic-to-English and Chinese-to-English, I thought I’d give it my own test in German-to-English mode, just because that’s the only recent real-world example I’ve actually got. Just for fun, we’ll also compare the results of a paragraph with Yahoo!’s translation (which is powered by Systran, which always makes me think of the line from “Pirates of Sillicon Valley” where Balmer goes “Fortran, ohh, Fortran!” in a mockingly sexual voice… Anyway…) . Here we go:

Original Content

Wenn ihr mal eine Geschichte wie aus einem Meg Ryan Film lesen wollt, dann seid ihr hier genau richtig. Alle einmal zusammen: *awwww*

Source: Cigarettes & Alcohol

Google

If times a history as from a Meg Ryan film want to read it, then are you here exactly correctly. Everything once together: * awwww *

Source: Google Translation

Yahoo!

If times a history as from a Meg Ryan film want to read it, then are you here exactly correctly. Everything once together: * awwww *

Source: Yahoo! Translation

Actual Author’s Human Translation

While the Yahoo translation is quite close to what I wrote, it’s mixed things up a bit. A more suited translation would be “If you want to read a story just like a Meg Ryan movie then this is the place to go�?

Source: Post Comments

Uhh… yyyeeaaahhhh… I can’t say either machine-translations are even close to what is claimed to be the actual translation / intent. Nor is Google particularly more accurate than Yahoo! (uhh, they’re exactly the same?). Anyone that speaks Chinese want to translate that same sentence from English to Chinese for us so we can perform the same test again, with the same specs as the cited test?

As with anything, we have to remember that any such test is purely subjective. Since human speach can be taken / intended in so many different ways, based on intonation and attitude, we can’t expect a machine translation to be perfect, can we? Boy I wish we could… I remember trying to cheat on my French homework in high school using Altavista’s service at the time… Yeah, that didn’t go over terribly well with the teacher…