..it parses kanji numbers backwards.

All the basic 0-10 numbers are fine, but:

Trying 十一 gives you 10 instead of 11 because I think it's being interpreted as "one ten"; in order to get 11 you have to do 一十.

Trying 二十 gives 12, while 十二 gives 20.

It goes on like this. Palindromic numbers (九十九, 八十八, etc.) are fine because they're read the same both forward and backwards.

Larger numbers don't fare much better. If I try 二万 I end up with "10000 2". To get 20000, I have to reverse it to 万二.

It also applies in sentences. Something simple like "私は十七歳です" turns into "I am 70 years old".

I can understand kind of where this is coming from, since older horizontal Japanese did used to be read right-to-left, but the translator doesn't do any other text that way, so why do the numbers?

I ran up on this problem the other day. I've sent a couple "better translation" reports but I don't think they'll help.

Online translators are (well, can be) bad, m'kay?

Google translate gives a probability-based translation using database analysis and user correction/retranslation. Thus, you'll probably find that if you translate the same document again, you'll get a slightly different translation. I think this is the way to better computer translations in the future, but it does produce its share of anomalies.
I hadn't noticed the right/left switcheroo, but one thing I *have* noticed is the mistranslation of numbers 1万 and higher. It seems there's like a 50/50 chance 万 will be converted into "million" and 億 into "billion." It seems like such a ridiculous error to come from a normal software translator; I think it must be the result of lots of incorrect "corrections" by users. God knows there are enough Japanese students of English, as well as English students of Japanese, who are under that mistaken impression.

Isn't Google Translate the best free online translator available though? Despite it's inability to translate longer sentences, it's still handy for non-colloquial short phrases

@snallygaster: Yeah I figured it might be due to the user-submitted "translations". The numbers are hard enough for me to understand as it is. :|

Though, I agree that this is the way to better automated translations. It's light-years better than it was even a year or two ago. I've been surprised more than once at how natural it's capable of translating.

@TheTrueBlue: I've not really found one to beat it. I used to use it a lot for things I didn't get and it usually gave me enough of a lead to understand.

There's trade offs for both approaches. Typically probability techniques work best for general solutions, while a rule-based give better translations when adapted to particular situations. You could find a number of Ph.D thesis on it if you try, I'm sure.

@mafried: It seems like I read something about that before, discussing that Yahoo's Babelfish uses rule-based translation and Google's is adaptable.

That'd be pretty interesting to research. If I can not be lazy and actually read the theses

I just checked how the translator I use fared with the same problem and it seems to give the correct results. It doesnt provide wonderful translations and sometimes a whole word is wrong but it usually gives me enough of a clue that the sentences means what I think it means when im only 90% sure and want to check.

http://www.reverso.net/text_translation....action=new

Just input your japanese and from the drop down menu select Japanese - English