Posted by: patenttranslator | June 25, 2014

Machine Translation Is Not Translation


In spite of what various snake oil salesmen and saleswomen have been trying to make us believe for many years now, machine translation (MT) does not make it possible for translators to increase their productivity by incorporating MT into their normal translating workflow. The general public generally does not yet understand it, but most translators at this point do understand that the main problem with machine translation is that machine translation is not translation.

MT is a translation tool, just like a dictionary or a bilingual database on Internet, but it is not translation. A bilingual, context based-database on Internet, with whole sentences, sometime accurate, but often not, and even with whole paragraphs that look like a real translation, would be a better description of MT than adding the word “translation” after the first accurate descriptor “machine”.

Sentences processed by MT sometime make sense, and sometime they make no sense. What is even worse is that sometime something that looks like it makes perfect sense is total nonsense. This is why MT cannot be used by translators to “increase their productivity by integrating MT into their work process” as those who need to sell this “linguistic technology” to gullible clients would want us to believe. It takes such a long time to separate the wheat from the chaff during so-called editing of MT, especially when chaff can at first glance look like wheat because “editing” of MT always involves a lot of retranslating, a conditio sine qua non if we want to avoid mistranslation.

But many companies have invested so heavily in machine translation in the hope that eventually they would be able to save the money that they used to have to pay to humans called translators that at this point they may be trapped in a labyrinth of illusions of their own making. (My heart goes out to them, of course).

MT has been 5 years away from becoming “almost as good as human translation” for so long now, at least for the last 5 decades. At this pace, it will still be 5 years away from this resplendent goal for at least the next 5 centuries.

As somebody who has been translating patents from Japanese, German, French and other languages into English for 27 years, I appreciate MT for what it is – an extremely useful tool for translators and non-translators alike. But from my viewpoint, MT is a most wonderful tool especially for translators.

I remember the days when there was no MT in the eighties and nineties. Not only that, the legibility of documents was often horrible when after a second or even third generation fax, Japanese characters that were originally beautifully clear and crisp became but poor shadows of what they used to be, like shadows of ideas on the wall of Plato’s cave.

Digital copy has then finally put an end to this particular torturing of translator’s eyes and brain, and MT has later made it still easier for translators to do their work, for example by quickly locating possible translations of technical terms, or even the correct pronunciations of personal and geographical names, which used to be very difficult to ascertain in Japanese before easy access to MT for the most part solved this and other problems.

At the beginning of 21st century, I consider myself very lucky because like everybody else, I now generally have access to clearly legible copies of documents in foreign languages, especially patents. And before I start translating a patent, I generally print out an MT version of the text in Japanese or another language because it helps me with my work. MT helps me in the same way that a specialized dictionary or a specialized online database helps, but it does not do my work for me.

I look at the printed text quite a bit when I start translating because at that point, I am still formulating in my mind the terms that I will be using, and suggestions from the MT tool are often useful (although they can be also misleading).

If it made sense for me to “edit” the MT-processed text and pretend that it is my own work – hey, I would do it myself! Why not if I could work faster and make more money in this manner?

But it so happens that things don’t work like this in translation, because, as I said at the beginning, while machine translation is a very useful tool, it is not translation. If I tried to edit the MT text, no matter how much and for how long I tried to edit the damn thing, the editing process would necessarily introduce abominations conceived in the brain of a machine into a human translation to such an extent that I would lose all of my clients very quickly.

The only real solution for all of the problems inherent in machine translation is a retranslation.

This is something that is difficult to explain to people who don’t know much about translation, especially since expecting that technology will keep delivering new fantastic miracles to us on a daily basis is deeply rooted now in the very essence of our so-called civilization.

But as more and more people now have direct experience with MT as it is available to anyone for free with access to a computer, tablet or smart phone, I am hoping that fewer people will be giving me the uncomprehending look that I sort of became used to over the years when I try to explain to non-translators why MT is a gift to human translators rather than a curse that will eventually do away with my profession.

As most people have already used MT on their own, or will be using it soon, more and more of them will hopefully start understanding what MT is and what it isn’t.



  1. ‘TV will see the demise of cinemas (1950s)’, ‘cars will be able to fly (1960s)’, ‘accounting software will do away with the need for accountants (1980s)’, etc. ad infinitum.

    There has never been a shortage of those who believe they can foretell the future, particularly those who have a vested interest in making people believe their prognostications.

    I cannot think of a profession based on knowledge, training and experience, that has been replaced by technology. Indeed, in most cases, technology has enhanced the performance, status and success of professionals.

    Of course, if your understanding of translating is ‘simply replacing the words of one language with the words in another language’, or if you have a vested interest in making the public think so, prognosticating about the future of MT will either be logical, or, in the case of the latter, profitable.

    Thanks Steve, spot on as always 🙂


  2. @ Louis

    And thanks for commenting.

    I read somewhere that accounting software did reduce the number of accountants by something like 16 percent over the years.

    Presumably, accountants who were only able to prepare a simple tax return were replaced by software, while their colleagues who are able to take on more complicated tasks are doing better than ever.

    The same thing may eventually happen also to the translating profession.


  3. Hi, Steve.

    Just one question: when you’re talking about MT, are you considering all kind of MTs (RBMT, EBMT and other hibrid types)? It seems to me that you’re considering only MTs such as Google Translate or Bing Translator, am I right?

    I use RBMT (Promt) myself as a tool to increase my productivity and it works very well. Of course, you have some points: sometimes I get some segments that are really not accurate, but in this kind of software there are some rules I can change to make it produce a better translation next time, so, when you translate similar subjects, there are lots of improvements in your productivity.

    I believe in this post you were more concerned about companies which wrongly use MT in an attemptive to improve their productivity, not about translators using it to improve their productivity, right?

    About translators using this online tools hopping to improve their productivity, I would like to call their attention to the NDA they probably have with their customers. If they use Google Translate they are allowing Google to make any use Google wants from that material, which include distribute such information to third parts. This is really clear in Google’s Agreement.

    Back to your post: could you tell me your opinion about MTs such as Promt and Systran (which are RBMT, not online tools)?


    William Cassemiro


    • I only look at Google Translate and sometime, when Google Translate does not seem to work, I look at Microsoft (Bing) Translator.

      Would you like to explain how these other MT tools like RBMT or Promt work?

      Are these proprietary MT tools?

      Also, what does “using MT” means? How can somebody prevent a translator from looking at a dictionary, at an online database, or at a particular MT tool?

      Is that within the scope of the definition of what “using” means?


  4. Steve, the proper term isn’t even machine translation these days. As you pointed out, this isn’t real translation, which is why the term “machine pseudo-translation” or MpT is slowly replacing it for better understanding. And it is true that many empty heads will promote the use of MpT as a “productivity tool” and an alternative to filling your head with a bit of knowledge and doing the job right.


    • 1. I have not seen the MpT abbreviation yet. But it would be a better name than MT.

      2. These empty heads may not be so empty. The MT propagandists are selling the world the latest version of snake oil in order to make much dinero.


  5. Great post! Here goes something I wrote last week, if you allow me to share:

    The types of translation that Roman Jakobson proposes are:

    1) intersemiotic: an interpretation of verbal signs by means of signs of nonverbal sign systems (example: adaptation of a literary work into a movie)
    2) intralingual: translation within a same language (example: paraphrasing)
    3) interlingual: also called “translation proper” (example: the translation of a document from English into French) This is what we, translators and interpreters, do for a living.

    With the arrival of machines, some gurus now talk about ‘machine translation’ and ‘human translation’.

    I am sorry, but I refuse to use the latter. This is about machine pseudo-translation (MpT) and… translation (REAL translation, professional translation, translation proper, Mr. Jakobson? ) If the differential is the machine, then I propose:

    1) Translation (i.e. translation proper, real professional translation)

    2) Machine pseudo-tranlation (MpT), a tool with a very limited use and full of very well known disadvantages. Best scenario for clients who resort to MpT: wrong translations that are funny and can negatively affect the image of their business. Worst scenario: wrong translations that can have serious legal consequences.


  6. […] In spite of what various snake oil salesmen and saleswomen have been trying to make us believe for many years now, machine translation (MT) does not make it possible for translators to incre…  […]


  7. Thanks for tweeting and commenting on my silly posts, Au.

    But there is also a third type of translation: human translation by people who can’t translate, although they may not know it. Possibly more than 50% of human translations would belong to this category. The reason: it is CHEAP! This type of translation can be sometime almost as ridiculous as MT (or MpT if you will).

    And there is probably also a fourth type of translation: machine translation “edited” by human (humanoid?) editors, who may or may not be able to translate, to make it look like a real translation. At this point it is not clear whether this type of translation will be viable in the long run.

    But it probably is viable as long as it is CHEAP!


  8. Machine Translation, or Computer Generated Transaltion as some like to call it now, is indeed not a translation. It is a Language Transformation Algorithm (LTA) and language transformation and translation are completely different things.

    Very simply put, language transformation is like that short glossary in a tourist guide book (or app, these days). The expressions there can help one find their way in a foreign country, but they don’t mean that one understands the language, that one can converse (i.e. think) in that language, and certainly not translate into it.
    If learning the basic grammar of a language and knowing which words in the target language correspond to the ones in the source was enough to translate, almost anyone could have translated using a dictionary and after a short crush course to familiarize oneself with the basic grammar of that language.

    So yes, a mathematical algorithm can transform the language to a certain degree based on previous translations. RBMT (Rule Based Machine Translation) could be even more grammatically accurate in very certain scenarios, but it s a lot more resource intensive to mainatain (diminishing ROI), and this is probably why it was not “chosen’ as the standard for commercial application. Both can be shown to mimic translation up to a certain point, but they are not translation.

    Let’s stop calling them that and start calling them what they are: MpT as Kevin has suggested, or LTA (Language Transformation Algorithm).


  9. “Let’s stop calling them that and start calling them what they are: MpT as Kevin has suggested, or LTA (Language Transformation Algorithm).”

    I wonder whether this is going to happen as the perception of what MT evolves among the general public.

    Although so many people now use MT, I personally doubt it. So I just use the abbreviation with the word “tool” to avoid having to use the word translation.


    • I like the Language Transformation Algorithm (LTA) term because it accurately describes what it is! Thank you Steve for posting yet another great article!


  10. And let’s not forget clownslation! (you guessed! when clownsourcing is involved).


  11. So this is the strategy to stop the onslaught of translators using computers to work for lower rates? Start changing the names from “lower end translators” to “nanolators” and “machine translation” to “machine “language transformation algorithm” or “machine pseudo translation”?

    Sounds like a plan.


    • You see Jeb, it is one thing to disagree with the content of the article and/or with some or all the comments, and to do that by explaining your opinion. It is completely another thing to repeat your “translators are doomed, machines are better, it is just a matter of time before you will become near obsolete as the computing power will grow to no end” mantra again and again without any real substance behind it.

      For example, what is your experience with translation in general and MpT in particular? What makes you an expert in the field of MpT application that support the claims that you make? Are you a translator, an agency owner, owner/employee of a translation supporting technology company, other (I asked you before, you didn’t bother to answer)?

      Being the contrarian in the discussion is nice, but gets tired real fast. You also comment under a pseudonym, which makes it that much more suspicious and makes me wonder what kind of motive is behind you commenting here.

      You are entitled to your opinion, of course, but if you want to engage in a conversation, you should at least ready to converse instead of repeating the same old predictions over and over again.


  12. I’ve been reading Steve’s blog on and off since he started. At first he just mocked MT and rarely discussed it. Then around 2011 he began to write about “zombie translators” and MT more frequently. In 2013, Steve began to write about the “subprime translators” and of course the “zombie translators” even more often and by late last year these almost identical posts became a staple, yet he then began to say how great a tool MT was and why all translators should embrace it.

    Just when you thought Steve’s ranting might have peaked, he ratcheted it up another notch this year with the “nanolator” talk. And while he mocks the name change to “Language Service Providers” he says he uses the phrase “MT tool” so that he doesn’t have to use the word “translation.” Now translators are playing word games using such catchy phrases as “MpT” and “Language Transformation Algorithm” as if this matters one iota.

    Here’s a philosophical question: Over at the Honyaku list, someone said that he is on his first year “as a professional translator” and gets 4 yen per character at one place and 5 yen per character at another. So, 8 yen and 10 yen per English word and that a third place’s rate is 2.5 yen per character, or about 5 yen per word,
    ” and I need more work, but . . . this (2.5/character) is way below average, correct? I heard that the average pay is lowering. . .but. . .”

    He calls himself a “professional translator” but is Steve correct that he is a lowly “nanolator”? Should this self-deluded “nanolator” take a long hard look in the mirror as Steve implores?

    What you call predictions are instead well known facts by anyone who has an interest in computers: 1) A cheap computer will have the power of IBM’s Watson by around 2018. 2) A cheap computer will be 1000 times as powerful in 2024 than a new computer today.

    I’m not happy with the trend translation-wise, but if any of you happen to be aging, then you definitely want the computer trend to continue at its blistering pace..


    • Oh well, whatever.


  13. […] projects? Business skills for translators: online courses 7 ways to get help on SDL Trados Studio Machine Translation Is Not Translation Game QA and Localisation Conference Customer Experience: 5 Practical Tips Trados shortcuts periodic […]


  14. […] In spite of what various snake oil salesmen and saleswomen have been trying to make us believe for many years now, machine translation (MT) does not make it possible for translators to incre…  […]


  15. […] people do not understand the simple fact that a major limitation of machine translation is …. that it is not a translation. Most people are led to believe that machine translation is similar to human translation, that it […]


  16. […] makes no sense, but they still don’t understand that it will never really make sense because the reason why machine translation often makes no sense is that it is not translation. It is basically just a bunch of words generated by a machine based on an algorithm and it is up to […]


  17. Hi Steve,
    First of all, thank you for your blog posts that I very much enjoy reading!
    I am a freelance translator trying to survive and make a living for me and my family, and I publish a newsletter (in Swedish) in which I write about languages and translation in various aspects.
    I had an idea to write about machine translation and to prove a point by actually using an MT tool to translate the post itself. I wouldn’t feel comfortable writing this post myself in English, since it’s not my first language I believe that my Swenglish constructions would shine through in the machine translation and create an ok Swedish text.

    This is obviously not what I’m trying to achieve here, I would like to show the results of MT.

    To my question: would you agree to me using just a part of your blog post on MT (from June 25, 2014) “Machine Translation is Not Translation” to do a machine translation of it and publish in my newsletter?
    I would of course refer to you as the creator of the original text and if you’d like me to, I could also post your original text in English.

    What do you think, does this sound like something you’d be willing to do?

    Thank you!

    Sasha Branner


  18. Hi Sasha:

    Sure, that’s fine, you have my permission.

    I hope you send me a link to it.


  19. Hey this is somewhat of off topic but I was wondering if blogs use WYSIWYG editors or if
    you have to manually code with HTML. I’m starting a blog
    soon but have no coding know-how so I wanted to get advice from someone with experience.
    Any help would be greatly appreciated!


  20. […] the most important thing to understand is that machine translation is not really a translation; just a suggestion made by a dumb machine. Google Translate sometime makes very good suggestions because it picks a similar translation, […]


  21. […] and value of the patent substantially. In short: as long as drafting patents is skilled work, so too will the translation of those patents be skilled […]


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


%d bloggers like this: