Machine Translation: The Good, the Bad, and the Ugly


by Chris Durand

Machine translation is the use of software to translate text from one language to another, usually without assistance from a human translator.  It is a fascinating field that is changing rapidly, but here’s my take on where things stand today.

The Good

Machine translation is cheap, and it is getting better every day.  I was encouraged by the success of IBM’s Jeopardy-playing system Watson in drubbing its human challengers. Watson’s ability to “understand” idioms and natural language will contribute greatly to the future of machine translation.

Translation projects vary in requirements for accuracy as shown in the following diagram.  For projects jobs where accuracy is less important, machine translation is a workable alternative.  An example of this would be a company support forum, with huge amounts of user-supplied content.  It is not cost-effective to pay a human translator to translate every post by a user into numerous languages.  However, a machine translation engine that has been tuned to translate support issues for a particular product won’t create perfect results, but may still be a valuable resource to users.  And creating value is what translation is all about.

Of course there are many translation jobs where accuracy is critical, such as legal documents.  And translations of literature, poetry, and the like will remain difficult for machine translation software for years since there is much more to this sort of translation than accuracy, such as style and other artistic considerations.

But with continuing advances in computing and linguistics, the line shown in the above diagram will move steadily to the right over time.

The Bad

Given the high variability of machine translation results (read: really bad translation), human translators still must review the results, reducing the cost advantage of machine translation.  Some customers try to save money by running their material though a generic translation service like Google Translate and taking the results to human translators to “clean up”.  Don’t bother.  It’s usually cheaper to just let skilled translators do the job from scratch if you want a good translation.

Bad translations create a number of problems, such as damaging your brand and resulting in unhappy users.  Would you buy a blender from a company if the box had a bunch of obvious translation mistakes on it?  If they don’t care enough to get a professional translation done for their product, what does that say about the company’s commitment to quality?

And remember that paragraph or link text on your website that you forgot to get translated?  You know, the one you just ran through Google Translate and are about to post to your website hoping no one will notice?  Don’t. Can you identify an obvious error in Chinese?  Or Arabic?  Which brings us to…

The Ugly

It’s not hard to imagine catastrophic outcomes from bad machine translation, but here are two humorous examples:

  • From the Department of Not Recognizing Machine Translation Errors, we have the “Translation Server Error” café in China. This is legendary in translation circles.  Apparently the owners of the café were thoughtful enough to translate their restaurant name from Chinese to English.  Unfortunately, the online translation server they used was not working properly, and they printed up a bright, new sign with the English name “Translation Server Error”.  (Though they probably got far more publicity out of their failed translation than if a professional translator had done the job correctly, so perhaps this is not a gaffe at all.)

Years ago my father-in-law was taking a French class and had to write a short introduction of himself to the class (in French).  To check his work, he pasted his French about how his grandson Hans lives in Portland into an online translation site to check his work, and got back the following:

Our only grandson, Grunts, also lives has Portland.

We still call my nephew Grunts to this day.

Author: bridge360blog

Software Changes Everything.... Bridge360 improves and develops custom application software. We specialize in solving complex problems at every phase of the software development lifecycle, removing roadblocks to help our clients’ software and applications reach their full potential in any market. The Bridge360 customer base includes software companies and world technology leaders, leading system integrators, federal and state government agencies, and small to enterprise businesses across the globe. Clients spanning industries from legal to healthcare, automotive to energy, and high tech to high fashion count on us to clear a path for success. Bridge360 was founded in 2001 (as Austin Test) and is headquartered in Austin, Texas with offices in Beijing, China.

8 thoughts on “Machine Translation: The Good, the Bad, and the Ugly

  1. Hello! You’ve highlighted well the aspects of machine translation, useful article, thanks! I have also read another opinion, in comparison with the human translator: Thanks!

  2. I teach ESL, and my lazier students are constantly handing in work that has been machine translated. I can tell that a given essay is machine translated rather than written poorly, but organically. However, I have no objective means of proving it or even explaining it. Can you give any tips on how to verify or explain that something has been machine translated?

    • Michael,

      Thanks for your comment.

      That’s an interesting problem! I don’t know of any automatic method to identify machine translation. In limited instances it can be pretty good, and given that you are starting with a certain set of source material there is a limited number of reasonable translations that follow the source material. So if it is (mostly) correct, there’s not much you can do to tell if it was done by a person or a machine. Mistakes, however, would be quite revealing. It would be interesting to research the differences between human mistakes vs. machine mistakes in translation (perhaps you are planning on getting a doctorate someday and need a research topic?). I would guess the mistakes would be quite different.

      I would suggest running the assignment through a few translation engines and looking for one-to-one correspondence between those results and student submissions as a baseline. But I expect if a student doesn’t understand the assigned material well they would incapable of fixing the mistakes from machine translation. A small bonus from this might be some additional learning opportunities in discussing why particular mistakes are incorrect.

      Finally, I am reminded of services that test for plagiarism in student essays, but I’m not sure if there is a similar service available for translations.

      Good luck!


  3. Excellent items from you, man. I’ve understand your stuff previous to and you’re simply too wonderful. I actually like what you’ve acquired right here, certainly like what you are stating and the way wherein you say it. You are making it enjoyable and you still care for to stay it sensible. I can not wait to learn much more from you. That is really a great site.

  4. Steven, Thanks for your comment!

    Fortunately for human translators, the quality of translation is subjective
    which makes it much harder for computers to “solve”. Contrast this with
    chess where decades ago computer scientists thought it would take hundreds
    of years before computers were able to play chess well, yet today there are
    freely-available chess engines that run on your desktop PC that can beat the
    top human players. But in chess you know when you’ve won (checkmate),
    whereas in translation it is often difficult to get two professional
    translators to agree on the “best” translation for something. That makes
    the machine translation problem that much harder which means human
    translators will be needed for many years to come.

    – Chris

  5. This is a very interesting post. Nowadays, machine translation has become more and more ‘popular’ and more and more software programmers are trying to find new ways of improving it.

    What is important to clarify, is that machine translation will never become as professional as a human translator can be. It will be always necessary to rely on human translators who offer specialised and professional if we want the output quality to be high.

    The story of this ‘Translation Server Error Cafe’ is absolutely hilarious, but it has certainly given a lot of publicity to it! This is a clear example of the importance of translation providers these days, because a human translator will always be necessary to get a good job done.

  6. Pingback: Machine Translation: The Good, the Bad, and the Ugly | bridge360blog | Global-Ready Content |

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s