Posted by: patenttranslator | July 16, 2012

The War Between Robots and Humans Is Now Taking Place in Machine Translation (But Mostly Just in Human Imagination)

The war between robots and humans has been going on in human imagination for almost a century now, ever since the Czech writer Karel Čapek coined in 1920 the word “robot” in his science fiction play “R.U.R.” which stands for Rossum’s Universal Robots. The word “robot” itself is derived from the Czech word “robota” which means “forced labor” in Czech (the Russian word “rabota” means just “work”).

Čapek must have been inspired by an old Jewish legend about an artificial monster called Golem, whose creator’s grave can be still seen at the Old Jewish Cemetery in Prague. His name was Rabbi Löw and he died in 1609. I remember all this because I had a date with a pretty girl once at his grave a long time ago. I forgot her name but I still remember her face.

There was always the danger, in human imagination, that if the robots created by people were really as good as the people want them to be, the robots would eventually try to rule the world themselves by getting rid of humans. That is one reason why people were so fascinated by the role Arnold Schwarzenegger played in the Terminator movies.

The current human-made but artificial Terminator destined to erase certain types of humans, namely translators, from existence, is made of silicon and his name is MT as in Machine Translation. But this again is mostly just human imagination.

The people who imagine that one day soon MT will one day soon replace the person who types these words mostly know nothing about translation. They see that MT is available instantaneously and for free on Internet, has been for many years, and that despite incremental improvement, the words “translated” with MT still mostly don’t make sense.

And they don’t understand why.

Even with a different approach to MT, such as the statistical approach that rejects processing of the rules of grammar in both languages and instead relies on enormous amounts of data, the result is for the most part still a far cry from real translation.

I have been arguing in countless article written over the last two decades that MT that is as good as human translation is in fact an impossibility because the computers, no matter how powerful, and software, no matter how clever, can never get around the one category that is essential for real translation – the category of meaning. You can not translate something into another language without understanding what the text in the original language means, and machines by definition don’t “understand” anything. They just process numerical instructions.

Google Translate can be surprisingly good on occasion, sometime to an extent that frightens human translators. But if you take the text of this post and run it for example through Google Translator and Microsoft Translator to “translate” it into another language, the difference between these two MT programs will be probably quite small.

I use relatively simple sentences and relatively simple words in my posts, which makes them relatively easy to process with MT. But I am not necessarily saying things that have been said before, which is a problem with the statistical approach to machine translation.

An intentional rebellion of robots against humans will probably not take place in the next few centuries, although there will probably be a quite a few unintentional ones that humans call “failures”.

And perfectly cogent and intelligently written MT product will probably not be available in the next few centuries either. Machine translation will continue to be used by more and more humans, and the product will continue to be improved, but only incrementally. There may be a revolutionary improvement in MT technology at some point, but it is hard to tell when and what it would be based on. The statistical approach was a really good idea because it freed the software designers from being enslaved by rules of grammar, but it still resulted only in incremental improvement, not a revolutionary one.

The war between robots and humans thus exists only in human imagination. Robots are only machines, and as long as machines don’t understand the concept of “rebellion”, how could they rebel against us?

MT will and already has replaced some human translators, just like robots have replaced some human professions. Bank tellers have been replaced for the most parts by ATMs. There are still some human bank tellers in banks, but these are mostly people who are being trained for managerial positions. Most of the simple tasks, such as receiving deposits and giving cash to customers, have been fully automated decades ago.

Some human translators must have been already replaced by MT. In the field of patent translation, which happens to be my field, MT has been available for at least two decades. So what has been the impact of MT in my field so far?

Nobody has the data, of course, but as far as I can tell, the fluctuating amount of work that is available to this human translator is probably not influenced very much by MT.

On the one hand, MT has decreased the demand for translation of patents that did not really need to be translated because before peoples started using MT, how could they tell which document was relevant?

On the other hand, more documents that are relevant and that need to be translated are discovered now with MT, which increases the amount of work available to human translators.

As far as I can tell, instead of removing human translators from the translation process, the overall result of machine translation is probably a slight increase in the amount of work that is available to human translators, at least in my field.


  1. Finally a reasonable post on MT! I have been reading a lot and everyone is always so scared… As if MT could take your job, being so smart. Thank you for sharing your thoughts on this subject.


  2. Thank you very much.

    I really like the design of your website.


  3. Who is afraid of the Big Bad Wolf (of MT)?

    I read this article titled “Bye bye Babel: Breaking language barriers online” written by Anna Heim a few days ago: (with a nice picture of the Tower of Books during the Buenos Aires Book Fair 2011). It’s about MT tools that should help us breaking language barriers online. Well, my late mother-in-law used to say, “Wer glaubt, der wird selig. (Blessed are those who believe.)”

    For us, tools are tools. There are tools. There are people who use tools. And there are people who are tools. We translators shall know how to use tools which are constrained. We shall know that there can be quite a few shizos popping out of MT tools which can only be deciphered by human capabilities more competent than the tools themselves, so that they can be good enough to perform the needed jobs.

    Should we be as good as MT tools, we would be out of work. Not even post-editing jobs could be ours.

    As to the question of a probable decreasing translation rate through the application of MT tools, I guess that would happen to those who work for VLSLSPs or LSLSPs, as they call such and such translation agencies. I like small, specialized agencies who are likely to pay decent rates even when they provide MT versions of documents for reference, because they know what they are doing with translation colleagues and tools.


  4. I must admit I have no idea what VLSLSPs or LSLSPs means.

    But I have noticed that clients often send me machine translations with a request for a new translation. And I can usually find an existing MT version of a Japanese or German patent unless it is really old.

    I just finished translating a complicated Japanese article from a medical journal describing a medical study. The first part of the existing translation that the client included with the job was so good that I thought it was probably done by a human translator. But the second part was so bad that it must have been MT.

    I can’t figure out what happened in that case. Maybe the first part was translated by a human, probably a Japanese doctor. Or maybe the best MT programs get somehow thrown out of their regular groove in which they perform well and then start acting up and delivering the usual garbage.

    If that is the case, MT is so human!


    • Steve, I was making fun of the habit of creating abbreviations in our industry.

      A VLSLSP is a Very Large Scale Language Service Provider and a LSLSP is just a V less.


    • Sorry. Correction: I mean the habit of creating acronyms.


  5. Alles klar. MoULSP is Masters of Universe LSP. Any subject, any language, any direction, any planet!

    話しが違うだけど、 I was finally paid the 5 thousand that I was worried about.

    So my instincts are still performing well.

    On yet another topic, I tried to read at one point some of the things that Kurzweil is writing about, and I got the distinct impression that he has no idea what he’s talking about.

    But since he is probably monolingual or basically monolingual, it’s not really his fault.


  6. Happy for you that you’ll be taking your son to Oktoberfest!

    We have a MoULSP which is a machine translator carried by the crew members of Spaceship Enterprise (Raumschiff Endescheiß in German, as I used to call it when I was young). It translates any language of any creatures encountered. Fantastic, isn’t it? The best is that there is no human translators fighting against the MoULSP.

    As to Kurzweil’s expectation of machines’ reaching human levels of translation quality by 2029, I’d like to defend him. Human translation cannot be perfect, anyway. Dans la direction de l’acceptation générale de “low-cost, good-enough” human translation, it is no wonder that machines catch up human translators soon.

    Besides, MT is in some cases so human, as you said. The shizos are so cute as human translators. There are different demands on quality. Quality is in the eye of the reader. Since people can make money or even gain power with “low-cost and good-enough,” they are presumably well prepared to accomodate MT quality by 2029.

    This is how I’d like to understand Kurzweil. “Make love, make no war!” I make no war against machines, but definitely no love to machines, either. Vive, la traduction automatique!


  7. Vive, la traduction automatique ….. à bas la traduction humaine!

    (Rien ne nous rend si grands qu’une grande douleur)


  8. You see, Steve, we are still happy human translators who earn decent money to provide our families, while some “subprimes” lose their rates or even jobs to machines. While we translate/write to help machines processing statistically/stochastically what we translate/write and further outputting it to eliminate some “subprimes” until we ourselves are finally eliminated by machines or replaced by some other talents needed for the Age of Machines.

    The cheap rates of post-editing machine translated materials is no pain in comparison to the agony of being uncertain with what machines will do to us. That is why we yet happy human translators believe that the war against MT is mostly just in human imagination, because we are not yet personally concerned.

    In fact, I do profit from MT development somehow, because I use it as a tool, sometimes as a dictionary, sometimes as a help to analyse sentence structures of some unfamiliar foreign languages. I don’t make love to machines, but I don’t need to make war against machines, either.

    As I see it, MT and the consequential post-editing are not an actual problem in translation industry. The real problem lies in the everlasting unbalanced power structure of the economy of our industry. The worst is that the unbalanced power structure forces translators/agencies to buying or updating more tools for digging their own graves.

    That has been how it comes to “à bas la traduction humaine!” Definitely not that there is the invention and further development of MT.


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 )

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


%d bloggers like this: