[Coda]

English, French, German, and Spanish are their best languages, since they have the largest databanks. Formal text in these languages will generally be translated very well and informal text does reasonably well. It can handle idioms and figures of speech because its databanks are built from statistical correlations of not just individual words but phrases, sentences, and contexts.

The big problem comes when you bring in languages that have different structure than these four languages. BECAUSE Google Translate uses statistical correlations instead of structural rules, it doesn't have any protection to make sure that the structure of the resulting translation matches the structure of the original sentence.

Japanese is particularly bad because it uses the negative inflection of a verb in many places where European languages would use a positive one. For example, in English you might say "I must go to school." In Japanese, the equivalent statement ("gakkou e ikanakute wa ikemasen") is written as a double negative ("It is unacceptable to not go to school"). Google Translate doesn't seem to be aware of this particular idiom -- it understands that "~te wa ikemasen" is an imperative structure but it doesn't know that it's a NEGATIVE, then it translates the other part of the phrase literally, so it comes out as "Do not go to school."

Ironically, one of the "casual" forms of that phrase ("gakkou e ikanakucha") DOES translate correctly ("Gotta go to school"), showing how Google Translate can handle idioms and casual speech.

(Edit: Side note: I work for a translation company as one of my side jobs, so it's my job to know these things. :P )

Edit 2: In Google Translate's defense, it does offer "Must" as an alternative to "Do not" if you click on it. It just, for whatever reason, considers that a less-likely translation.