11

Sometimes I use Google Language Tools or other translation sites to check sentences that I've formed on my own. Recently I've tried copying Japanese sentences out of textbooks into these sites and noticed the the translation tends to be way off.

Is this because Japanese can be hard to translate literally? Or indicative of the quality of these tools?

5
  • 3
    No. Not now, and not in the near future.. Programming is still a very young science, if you trust my estimate, perhaps in another 120 years you can trust the quality of non-human-aided translations.
    – Pacerier
    Jun 24, 2011 at 3:11
  • It is funny that you have an internal brain to finger translation problem when typing the title...
    – ogerard
    Jun 24, 2011 at 5:44
  • translate.google.com/about
    – Pacerier
    Oct 8, 2011 at 2:19
  • I have heard that Yahoo Jisho (dic.yahoo.co.jp) is better than google translate for Japanese <-> English from someone who is fluent in both.
    – Muhd
    Oct 29, 2011 at 21:04
  • This question should long have been closed/migrated. Moving it to Meta.
    – Dave
    Mar 15, 2012 at 1:04

9 Answers 9

15

MatthewD is correct--computer translations are unlikely to be very accurate, because computers aren't very good at understanding natural language yet. Think about it--when you use the word "fair" in a sentence, for example, a listener has to figure out if you mean "just," "mediocre," "reasonable," or even "festival." Sometimes there isn't enough information for even a fluent speaker to figure out what you mean, and computers are far from fluent speakers at the moment.

I would like to add that Japanese<->English translations are particularly prone to inaccuracy, for a couple of reasons:

  1. Japanese sentences are much more likely to be ambiguous (from an English speaker's perspective), just as English sentences are likely to be overly-specific (from a Japanese speaker's perspective).

  2. Japanese has a lot of words that don't easily translate into English, like まま, はず, こと, and the various set phrases like いただきます and よろしくおねがいします.

3
  • 3
    To add onto this, if you really have no other source, you can try using several sites to translate at the same time. I usually cross-check Google with WWWJDIC because both sites will often come up with different interpretations of a word or phrase. It's still not perfect, but it can make you aware of the parts of the translation that might be shaky.
    – Troyen
    Jun 24, 2011 at 2:02
  • 1
    @Troyen: do you know if there's websites that asks several websites for translation, like how dogpile allows you to google with multiple search engines?
    – Golden Cuy
    Sep 3, 2011 at 1:35
  • 1
    @AndrewGrimm: You can only google with Google. :P Unless google has become a real verb now, which I guess is possible, if confusing. Sep 28, 2011 at 22:45
9

Google Translate is horrible. Due to its statistical methods, it makes absurd mistakes that not even other machine translators would ever make. I've seen it do things like change the subject from "you" to "we", even in languages where the subject is entirely unambiguous in both languages. Even the other machine translators aren't that good, either, for reasons described in the other answers.

This is my favorite example of machine mistranslation. I wanted to know if there was a Spanish term equivalent to "hardball" -- i.e., tough and unforgiving. I put in "It's a hardball world." I got back, "Es un mundo del béisbol" -- "It's a world of baseball."

2
  • 1
    Well, "it's a softball world" really means "it's a world of baseball (for females)", so I consider the translation quite accurate and witty :)
    – Axioplase
    Jun 24, 2011 at 2:39
  • 3
    I agree there are a whole class of errors that are specifically due to the statistical method currently in favour in MT. Especially between languages which leave out a lot which is required in the destination language as in the case with ja->en. Jun 25, 2011 at 4:58
9

Not really sure this question falls into the scope of JLU. And 100% sure a thorough answer wouldn't.

Answer: No, you can't.

Machine translation (MT) is a very old problem, people have been at it for the past 60 years, and while algorithms have gone a long way over the past 10 years, they are still miles away from "trustworthy" in the language pair you are interested in.

Btw, contrary to what some people have written above, symbolic MT (which relies on 'understanding' the text in order to translate it) does not produce good results and is currently not the favoured approach. After being touted as the Holy Grail of MT (and one of the main goal of AI research) in the 1980s, it's been mostly discarded in practical applications nowadays. Most efficient modern systems (Google Translate etc) use a heavily statistical approach: they handle word relations and sentence structures in an abstract way, with little-to-no attempt at understanding the meaning behind them.

This may seem counter-intuitive, but it works extremely well for languages that are reasonably close together, linguistically speaking (even on very ambiguous phrasing): English<->French<->Spanish etc. is fast reaching commercial quality.

Unfortunately for you, Indo-European and Japonic languages could not be sitting any farther on the great language tree (wildly different syntax, handling of referential and all sorts of linguistic terms I am not familiar enough to give you). Which means it will probably take another decade or so before symbolic MT or (more realistically) advanced statistical models gives any usable result on Japanese <-> English.

3
  • I agree. Most model-driven MT attempts have failed miserably, while data-driven (in other words, statistical) MT has shown some rather useful results. Google Translate probably provides 90% useful (if not entirely accurate or well-phrased) output when you do translate between languages close enough as English, Spanish and French, but it would fail miserably if you try to apply it to English<->Japanese, English<->Chinese or in fact any language that does not satisfy all the following points
    – Boaz Yaniv
    Jun 24, 2011 at 14:25
  • 1
    1. Indo-European. 2. Belong to the European cultural sphere (so no Hindi or Armenian for instance). 3. Is written in the Latin script (Russian and Hebrew fail spectacularly well because of Google Translate misreading the quirks of the script, and languages with no spaces such as Chinese or Japanese are even harder). 4. Doesn't have a well-developed statistical data (some GT languages got a whole lot better since launch, since they're constantly improving their data).
    – Boaz Yaniv
    Jun 24, 2011 at 14:30
  • 2
    I wonder how Google Translate does for Japanese<->Korean? I'm far too poor at both languages to judge though I have successfully used it that way when the en->ko translation was crap but I knew what I wanted to say in Japanese. Jun 25, 2011 at 4:55
7

Even with a human translator, some things are impossible to translate without context.

  • How would you translate "Fruit flies like a banana." without context?

    That would depend on whether you meant "A fruit which flies like a banana" or "A banana which fruit flies like".

  • How would you translate "主人公の猫" ?

    That would depend on whether you meant "The protagonist's cat", or "The protagonist, the cat".

  • How would you translate "僕が知っている人は誰も来なかった。" ?

    That would depend on whether you meant the thematic は or the contrastive は.

Even with a human translator, some things are impossible to translate without losing much of the intended meaning (e.g. when there's a play on the words used). This is an inherent problem with translation, regardless of language.

4

Language translation is an extremely difficult problem for a computer (or rather, for a computer programmer). I find it quite amazing that the available tools are as good as the are.

That said though, their translations are sometimes quite bad and shouldn't be relied upon for accuracy.

4

No. But they can be fun! Try a round trip translation from E to J to E. Then you'll have an idea of what they are like to Japanese speakers.

1
4

I'd like to add my two cents to this, even though the answers everyone else have provided pretty much answer the question.

My first cent is that I completely agree with what everyone has said. If you want an accurate translation, do not trust a machine...everyone has explained why this is much better than I ever could.

My second cent is that I like on-line machine translators a lot. I think they do a great job when used as a learning tool. They can also speed things a long tremendously. These are the chores I use machine translators for:

  • When you have long lists of vocabulary I have to look up. It's so much faster using these tools than looking them up individually in a dictionary. (Yes, you have to be careful with words that have several meanings, but that just means you have to be careful and use your brain.) Sometimes I do this even with small lists, or when it's just eaiser to copy and paste from one browser tab to the next. The point is, if you take grammar out of the equation they're basically electronic dictionaries (and I know Google Translated even gives you a list of alternate translations to a word, which is fantastically useful in this regard).

  • Getting the gist of things. I would never write a letter in Englsih then run it through a machine translator and then post it, but the other way around works pretty well if you don't have the time to sit down and actually figure something out.

  • When you're stuck. I like to use machine translators when I just can't figure out how to say something. I might not (well, I hardly ever) use exactly what the machine spit out, but it helps figure out what I was trying to say nonetheless. On the flip side of this, if I come across a sentence in my non-native language that I just can't make out, I run it through a translator and it helps to see what it says.

  • Finding obvious mistakes. One thing I used to do in my Spanish classes when I had to write essays was copy what I had written in Spanish and paste it into a machine translator. It's a good way of finding spelling mistakes, or when you used the wrong word (spell check doesn't pick those up), or simple conjugation errors.

With all that said, online translators don't give a perfectly accurate translation, but I still do use them when I need to. I doubt they'll ever be a time when a machine will do a better job then a person when it comes to translating languages, but who knows?

2

To answer your question literally, yes you can trust Japanese-to-English machine translation sites to have very inconsistent and widely varying results that often but not always help you get the gist of the original.

You cannot trust these sites to do a good job every time. Not for any language. But the languages of East Asia seem to be especially difficult for these systems. Korean-to-English and Chinese-to-English are just as bad.

0

There are situations automatic translations are somewhat usable, such as technical documentations where the reader has intimate knowledge of the subject matter, for instance, computer programming in C#. But for general use, no. Translations or not, computers still do not do a good jobs of getting the context in which the text exists. Even I have hard time coming back with the right answer if someone asks for a translation of a Japanese line in a manga, if I don't know that particular story.

It also does terrible job dealing with regional dialects. Here, let me do this

おっとっととっとってっていっとったとになんでとっとってくれんかったとっていっとーと (DUDE What I am saying is, I had asked you to save those goldfish crackers for me, why didn't you save any for me?)

Google translate has

Whoops, whoops, whoops, why didn't you stay with me?

I rest my case.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .