Google translate – where a dot means a lot

Posted by & filed under Oddities, Technology.

Google Translate is a fun service — both as something that helps you to get a vague idea about what some news article in foreign language says, as well as an endless source of amusement and OMG WTF moments (without BBQ, unfortunately).

Lesson number one: a measly dot, or an absence of one can spoil the whole translation. The wildest example: a colloquial phrase in Russian “Отмывание пола просто супер!” (“[For] cleaning of the floor, it’s just super!”) gets translated into something quite shocking:

Trust me, nobody with even rudimentary knowledge of Russian would translate that into that. I can theoretically understand how shredding phrase into individual words and picking the worst possible versions of translation can lead to such “artificially non-intelligent” result — отмывание is frequently used in отмывание денег — money laundering. Except Translate sometimes arbitrarily swaps the words around, so in this case washing/cleaning/laundering got turned into “money” (fro the money laundering). And noun “пол” can mean both a gender and a floor, and sometimes a half (of something). Every robot knows that gender means sex and voila, we suddenly turned a phrase about floor washing into rambling of a rich sex-obsessed maniac. The most curious aspect is that if you add a dot at the end of the sentence, then sex suddenly does turn into floor. Perhaps AI does know something about humans that don’t finish their sentences with a dot…

Other oddness is, for example, complete dislike of the city of Amsterdam when it’s spelled in all capitals. In many cases translating from Dutch into English results in Amsterdam being translated as London. Yes, both are capitals, but totally not the same!

The same oddity of city substitution persists if you’re translating into Russian.

German used to cause duplicity of “positive/negative” translations, where English version would show the opposite meaning fro time to time (from “was not opened” to “was opened” etc), but at least that issue seems to be rare these days.

The moral of the story? Do not rely on automatic translation. Otherwise you risk of getting something quite odd and, from time to time, completely different meaning.

Update: Jul 31 2012 The London/Amsterdam problem has been fixed for a while, though the translation system continues to be sensitive to case and punctuation where it shouldn’t be. For example:
“Нет выхода.” = “There is no way out.”
“нет выхода.” = “there is no escape.”
“нет выхода” = “no exit”
While some might argue that certain languages do need case sensitivity (German, for example), I would expect basic rules to be coded from the beginning, and Russian is not really case sensitive.
Submitting corrected translation is possible, but, obviously, it needs many corrections until AI changes its mind. Additionally, in some cases system simply doesn’t allow you to replace the phrase it cooked up with a shorter translation:
Aside from that, do try to submit the corrected translation when you see one (even if it does seem to be counterintuitive, as people that know the language enough to be able to detect wrong translations probably won’t use the auto-translation)

Leave a Reply