The World’s Leading Microsoft .NET Magazine
   
 
The .NET Addict's Blog

My Top Tags

                                                           

My RSS Feeds








I heart FeedBurner

Latest Diggs - Programming

Computers Blogs - Blog Top Sites

Site Hits

Total: 4,895,612
since: 19 Jan 2005

Does Google Translate Suck Just as Bad as all the other Machine Translators?

posted Thu 17 Sep 09

Before I get into the main part of the blog post, I want to preface this whole thing by saying that I am well aware of the problems of doing language translation by a machine. It's an inherently difficult problem.

My issue here isn't so much with Google Translate as it is with the public perception of Google. I've read a few reviews of Google Translate and seen some blog posts. The general consensus is that because Google built it, its awesome and flawless. People think that Google is building this massive, soon-to-become-self-aware giant brain in the cloud that is sucking intelligence out of search patterns to be used for some world dominating purpose in the future. This is rubbish because we know Google doesn't start Judgement Day, Cyberdyne does. Sheesh, anybody knows that.

So, what's different about Google Translate than other previous attempts at machine translation? The biggest is that Google is using collective intelligence like it uses on its search engine to cull human-supplied translation bits and add those to its existing translation rules. The theory is that it makes for a better engine and should make the translations more in line with what a human would expect.

So how does it fare? You tell me:

My Input Google's Output  Actual Meaning
 内閣は倒れるだろうということだCabinet and they will fallThey say that the cabinet (govt) will fall
 その先生は病人だということが分かったThe teacher found that it's sickThe teacher turned out to be sick
 私達は彼の死を悲しんだWe mourned the death of hisWe mourned his death
 彼の殺人の動機は何だ
What's his motive for murder
What's his motive for murder?

So this is actually pretty enlightening. I have to admit that I was really skeptical when I tried it out. Just like every other machine sentence translator on the planet, if you attempt to use a Google translation to glean meaning from a web page or from a paragraph or even a single sentence, you're probably screwed. The first three sentences I have in the chart show that while verbs and nouns are translated properly, you actually lose a lot of meaning and context in translation. For example, the second sentence - relying on the Google translation I don't know if the teacher is sick or something else - but it's pretty damn close to accurate. The second to last sentence is almost spot on and finally, the last sentence is a perfect translation.

So the verdict is this: the current state of machine language translation sucks and will continue to suck for some time. However, I now know that if I want to get a rough stab at a translation of something I am going to try Google Translate first. Among all the other engines I could find, Google's was far and away the best.

For example, when I plug in the Japanese for "what's his motive for murder?" Babelfish spits out the following: Motive of at that homicide is what.

Someday this won't matter and we'll all have universal translators in our ears. It also doesn't matter much because, as Hollywood has shown us time and time again, all aliens from far and distant worlds speak English.

tags:        

links: digg this    del.icio.us    technorati    reddit

AddThis Social Bookmark Button




1. Keith left...
Thu 17 Sep 09 1:07 pm

The thing is they market it as a way to remove the barriers between different cultures etc... I've just finished reading a book called China etc. and the author gave there an example of a Chinese e-Bay seller that truly believes that using machine translation he can sell his products abroad. Tell me: are you going to buy when the product descriptions are written in bad English? Are you going to use these tools to translate your business documents to other language and spread them to your customers ??? Call me traditional, but I'm a fan of old-fashioned human translators, who actually studied language and translation at academic institutions. Yes it's not free and it takes time but you get good results to work with. In this matter, I'd like to recommend a new breed of translation service that most people are not familiar with. It's called OneHourTranslation and it's online and working 24/7 (am not affiliated with them - I really recommend this). They have hundreds of translators available at any time around the world for any language. This way you get a faster translation than in your local translation agency, and the translations are always done by native speakers in the target language, which are professional translators.


2. Kevin Hoffman left...
Thu 17 Sep 09 6:47 pm

Totally agree. I don't think there ever will be a time when we truly get all the nuances and true intentions and most importantly, context, from a machine translation. Look at all the horrible subtitle work that happens even when actual human beings do the translation...


3. jamie left...
Fri 18 Sep 09 1:02 am

I seem to recall a major reason for the Fat Man bomb being dropped on Nagasaki was on account of an ambiguous translation of a statement from the Japanese general staff. They had responded to the allied demands of *unconditional surrender with threat of future atomic attack* by stating they accepted the statement by "killing with silence" (I believe the phrase was "moku setsu") but I'm pretty rusty.

Truman interpreted this to mean that the Imperial Staff was ignoring the demand, but the concept the Japanese were trying to convey was that they received the message with mollifying shock. They didn't really accept the statement on the merits though, so it's questionable how the outcome might have been any different.

On the topic, it seems that the best way to have a somewhat reasonable mechanical translation is by having a huge corpus of language that's been pre-translated, like the Canadian hansard.

I once had the idea that you could start a translation service that would accept typing from the user, and every few words it would pop up an image (or three) of the word of phrase, and require the user to select exactly what concept they meant. Since all human language tends to parse into the same grammar, translating the picto-stream from one language into another should be easy, given sufficient crowdsource.


Tag Related Posts

JLPT3の試験があります。

Tue 11 Nov 08 1:44 P GMT-05
tags:      

JLPT3のしけんが申し込みました

Sat 02 Aug 08 3:40 P GMT-05

Will Google's Android sink or swim?

Fri 16 Nov 07 7:08 P GMT-05
tags:        

Yet another half-app from Google

Tue 18 Sep 07 7:08 P GMT-05
tags:      

Expliosion of Exciting Super Shoot!!

Mon 30 Jul 07 6:35 P GMT-05

I'm not drinking Kool-Aid anymore

Sun 03 Jun 07 3:07 P GMT-05