Computer Translation #2

Google TranslateOver a year ago, in my post Computer Translation, I wrote about how some of my students use translation software when writing letters or reports. I wasn’t particularly impressed with the results, and urged my students not to trust such software.

So, one year on, I was playing with Google Translate and tried a little experiment. First, I copied some text from a Japan Today article, and used the site to translate it into Japanese:

The 20th Tokyo International Film Festival began Saturday in Tokyo with some 300 actors, directors and others making the red carpet walk at Roppongi Hills in front of hordes of fans. Among the actors and actresses were Takao Osawa and Yuko Takeuchi, who appeared in the Japanese movie “Midnight Eagle” directed by Izuru Narushima. Some 300 films will be shown during the festival through Oct 28, according to organizers of the event.

I can’t show you the Japanese translation because it will look like gobbledegook if your computer doesn’t support Japanese fonts. However, when I copied and pasted the Japanese text back into the box and translated it back into English, I got this:

The 20th Tokyo International Film Festival started on Saturday in Tokyo, about 300 actors, directors and other red carpet walk at Roppongi Hills in front of thousands of fans. Among the actors and actresses are Takeuti Takao Yuko Osawa, who appeared in the Japanese film “Midnight Eagle” Shima Shigeru izuru director. Some 300 films will be shown during the festival through October 28, the event’s organizers said.

That’s not too bad, is it? Now, what makes Google Translate so great, even though it’s still in Beta, is you can suggest a better translation. What this really means is that if you copy a block of Japanese text and the resulting English looks a bit funny, you can tidy up the English yourself so Google remembers it next time. That’s right, you can teach Google Translate to make better translations. Imagine how smart it would become if thousands of people did this.

Now before you get too excited. I wanted to run the exact same experiment I did a year ago in my first post on the subject of computer translation. Back then, I used Babel Fish to translate a paragraph from the About page into Japanese and then back into English.

Here’s the original text:

I first came to Japan after finishing university in 1997. My first three months was a homestay-type arrangement with the family of a Japanese friend I had back in the U.K. The following year, after getting my teaching certification, I came back to Japan and have been here ever since.

Here’s the Sept. 2006 Babel Fish translation:

I finished in 1997 and first came to Japan after the university. My first 3 months were the homestay type rearrangement to which series of the Japanese friend who in me has the back section in England has been attached. After obtaining the proof of my professor, the following year, I return to Japan, after that it was here.

Now, here’s the Oct. 2007 Babel Fish translation:

I finished in 1997 and first came to Japan after the university. My first 3 months were the homestay type rearrangement to which series of the Japanese friend who in me has the back section in England has been attached. After obtaining the proof of my professor, the following year, I return to Japan, after that it was here.

Clearly, Babel Fish has made absolutely no changes to their English-Japanese, Japanese-English translation “engine”. I find it quite sad that a whole year has passed and there has been no change whatsoever. Do the people behind Babel Fish believe they already have the perfect translator?

Let’s see what Google Translate makes of the paragraph:

The first time I came to Japan in 1997, the university said. My first three months Homestay type of agreement with the families of the Japanese friends to return to the United States k. I am of the following year, my instructor’s certification later, I come back to Japan where it has been since.

Hmm, it’s not nearly as good as the Film Festival translation, but it’s no better than the Babel Fish attempt. Fortunately, with Google’s “suggest a better translation” feature, I hope it will be even better if I run this teast again next year.

Of their new system, Google says:

We have a system that can learn to translate better if we know where the problems are. In the past, there was no way to tell us about problem translations. Now there is. Next time you see a sentence that makes you go “hmmm,” click the “Suggest a better translation” link. Tell us what it should have said, and we’ll use your suggestion to improve translation quality in future updates to our service. So the next time you can think of a better translation, be sure to use this new interface and share your wisdom! That way, everyone will benefit from it. (Source) 

It may not be as fun as Google Image Labeler, an online game which Google has developed to get thousands of people labeling all the images on the internet, but mass participation in teaching our languages to computers could really kickstart the automatic translation era.

If you like, you can find me on Twitter at @nick_ramsay. I'd love to hear from you!

4 thoughts on “Computer Translation #2

  1. I’m curious to know how Google might actually make use of these alternate translations. Are there checks in place to make sure that bogus translations aren’t entered? Some people spend quite a bit of time entering false information into Wikipedia … could Google’s engines fall prey to this tactic as well?

    I can already see it … someone is trying to translate “I hate Coca-cola” into Portuguese or Spanish, and it turns into “I can’t get enough of Coca-cola” (in Portuguese or Spanish, of course).

    I’m all for making use of the wealth of personal knowledge and skill-sets from the people using the internet, but it’s always that 0.005% of the population that can mess something up for the rest of us 😕

    1. With Google Image Labeler, they won’t accept a description unless two random people give the same one. Whether or not there’s some behind the scenes validation going on, I can only hope so, but I guess the number of people actually suggesting better translations is very small. I didn’t bother and I’m sure I’m not alone!

  2. I’ve used the translate function a lot for Korean texts.
    I could certainly use some help. The best mis translation was
    about something having “large genitals” because it was not using the whole phrase.

    I’ve not seen the suggest new translation option yet.

    I’ve seen the google image labeller before and it’s quite addictive. Must stop now, I’m at work 🙂

Comments are closed.