Posted by: patenttranslator | May 1, 2017

The Consequences of the Coming Machine Translation Armageddon Are Difficult to Predict

The consequences of a nuclear war are difficult to predict.

Difficult, but not completely impossible to prepare for and anticipate.

In fact, the world has been preparing for nuclear Armageddon for about 70 years now. Regular people were being prepared for just such an eventuality by being instructed by their teachers from an early age to duck and hide under the desks in their classrooms, or by running in an orderly fashion to take shelter in the school’s basement. This was by far my favorite part of atomic bomb drills when I was a kid because the basement of the old school, which was built in 1886, was dark, spooky and totally awesome.

Not that running to the basement would help very much, of course, depending on how far the bomb would fall and the extent of the radioactive fallout.

Important people, such as presidents of various governments, have been preparing themselves by having huge anti-atomic bomb shelters dug, for themselves and their families, so that they can continue waging nuclear war from a safe underground location until the glorious victory.

In the long run, this would probably not help very much either, but building anti-nuclear shelters for important people has been a nice and constant source of income for the nuclear bomb shelter building and maintenance industry since the 1940s.

One good thing about a nuclear war is that since it can anticipated, it is possible to make preparations for something like that, although the preparations will probably be inadequate.

The consequences of the coming Armageddon and the fallout from machine translation are much more difficult to anticipate. As a result, nobody seems to be giving much thought to the bleak future that humankind is facing as a result of excessive reliance on machine translation, which the translation industry now prefers to call “language technology tools”, and I don’t see any individuals or organizations making any preparations for the havoc the machine translation Armageddon will most likely wreak on our civilization in the very near future.

One thing about a nuclear bomb is that is likely to fall on us only once, and although the consequences would then remain with us for a very long time, we would know that it had already happened, and we could then try to figure out what to do next.

But the bad thing about machine translation is that we simply can’t know anything for sure about it because it works like a slow-acting, invisible virus that may have no cure.

We kind of know that there is a lot of machine translation out there already, and once we know that what we are dealing with is machine translation, we are already pretty safe because most people at this point understand that machine translation is not really translation, even when it sometimes kind of looks like a real translation.

Sometimes, machine translation is immediately identifiable as such because it sounds as very poor translation, even worse than a really horrible human translation.

But sometimes it is very difficult to distinguish one from the other.

For example, is the text below the product of machine translation, or of human translation?

(I saw a photo of this translation of instructions on Facebook, so this is a genuine set of very helpful instructions in English that came with a phone case.)

This is the PCV Mobile phone Case of easy shlepping and more function. This case is made with import and defended radialization materials. And the appearance is so beautiful.

The main characteristic is easy schlepping, it can be hunged up at the waist, hunged up at the cervix, and free holding.

Hang up at the waist: This case has steel button and high strength PC clip. The mobile phone can be hunged up with the case at the waist. And it can be went round and round for 365° .

Hang up at the cervix: This case has high strength and defended snap soft fibre sling. The sling will defend hurt to the cervix and defend breaking.

Free holding: If you will put the mobile phone into your placket, or hold in your hand, you can twist off the button of the case. Because it is so easy.

I think this translation is most likely a sample of the carnage that can be inflicted on a relatively simple text by a human translator (for lack of a better word), who is probably using machine translation for inspiration. But the responsibility would in this case rest squarely with the human who thinks he is a translator, because the machine translation tool would in this case be completely innocent of any wrongdoing.

After all, it is clear that what the text means is that the phone case can be attached to something like a lanyard and worn around the neck. If you know some Latin, you would also know that cervix is Latin for neck. So a moderately educated and intelligent person would probably realize what the word cervix means neck in this case (although most men might have to look up the word cervix on Wikipedia to make sure what and where it is.)

So even a highly entertaining translation like the one above would be easy to figure out and would also be quite easy to edit, even without having access to the original text in the original language.

On the other hand, it is a very different story when we are dealing with machine translation.

Because machine translations have improved over the last decade, unlike in the recent past, it is now often very difficult, if not impossible, to identify mistranslations in them because those mistranslations sound logical and plausible.

After all, machine translation operating on the basis of the statistical model (as opposed to the grammatical model) is based on real translations of the same kinds of texts that were produced by human translators, usually very good ones.

I have been translating applications of German patents that are to be filed in English in the United States for several patent law firms for quite some time now. Some of the law firms send me a Microsoft Word file with the original German name, which is what I prefer because there is no room for error in this solution.

But some of them use a machine translation to translate the title of the patent because it is easier for a legal secretary to keep track of titles she can understand.

Since nobody has told me how the titles are translated into English, I used to think that I should try to incorporate the words in these titles into my translations, since those are the words my clients are using.

But eventually I understood that I should not try to do that. The titles are sometimes so clearly wrong that the only reason for such incomprehensible mistakes must be that the source is a machine translation.

At the same time, when I start translating a patent, the translated words helpfully supplied by a specialized machine translation program, although they are completely wrong, sound plausible even to me, an experienced patent translator who has access to the text in the original language.

I sometimes even use the wrong words for several pages before I realize where the problem is and correct the misleading English title.

Can you imagine what would happen if high ranking politicians, the ones who have had deep underground bunkers dug for themselves so that they can survive a nuclear attack, decided in their infinite wisdom that one way to reduce the government’s deficit would be to eliminate human translations and replace them instead with machine translations, perhaps only lightly edited by human post-processors, who would be paid much less than actual translators?

The actual meaning of this communication might no longer be identifiable.

And what about things like product instruction manuals, patents, laws and legal ordinances, or even novels, for example – what if big corporations and publishing houses decided to save money by switching to machine translation, lightly edited by poorly-paid humans called post-editors?

Based on what this patent translator knows about machine translation, namely how difficult it is to identify mistranslations in a text that was translated by a machine with a software package, I believe that the impact of such an unwise decision would be tantamount to nuclear war.

Better in some respects – we would not be dying in millions right away, for example, because the results of the machine translation Armageddon would take a long time to seep into the fabric of society.

But because it would take such a long time to identify what the problem is (politicians and CEOs of multinational corporations would of course not tell us they had switched to machine translation to save money), the consequences of such a switch for human civilization might be even more dire than an all-out nuclear war.

Personally, I think that this world could probably recover from a major nuclear conflict, although it would take a very long time.

But I don’t know whether recovery from an insipid shift in society that would take place very slowly, while most people might not be realizing it, if communication between humans were replaced by communication that is automatically carried out by machines operating much more efficiently on the basis of algorithms, would be even possible.


  1. Let’s be honest, there are certain types of machine translation that are totally acceptable, when they consist of stock phrases, such as weather forecasts. “Winds light to variable” or the Israeli favourite “the sea will be calm to choppy” can easily be strung together and repeated ad nauseam as stock phrases. Anything else, however, is an unmitigated disaster. I once had to edit a translation performed by human translators who had relied heavily on CAT tools. The word “vol” in French, in reference to the volume of a book, did not have a full stop after it in the original text that had been written in 1812. So the machine translated it as “flight” rather than “volume” and the human translators were too dumb to realise the mistake.

    Liked by 1 person

  2. Well, many things are acceptable for reasons I don’t always understand.

    For example, the Electoral College is still acceptable to more than 300 million people.


  3. “…, usually very good ones”.
    Unfortunately if it’s good it’s a TM-remnant of long passed times when large agencies hired competent highly experienced translators to build up their TMs and paid those translators decent rates. Most MT-projects nowadays border on con operations where the clients are made to believe they get MT while in reality it’s an army of slaving low paid PEers who turn the shit caused by incompetent project lay-out and blabla computational linguists into something only just acceptable. The whole thing is highly disruptive; not in terms of technology but in terms of the con-men and surely not to be overlooked the con-women getting a strangling grip on the community of independent language professionals. With a few exceptions It’s all about big investors’ money creating havoc to our livelihood by massaging clients into believing that MT works the same way as a sausage machine. We loose out, the clients loose out, and the investors loose out. The winners are the con people moving from CEO to CEO to yet another CEO position where they use investors’ money to grant themselves a salary and accompanying lifestyle which in no way represents any achievements.
    MT however has it’s merits for certain types of sports such as legal content, structured technical content, and yes … patents. There is only one solemn stand alone hard core condition: you the translator has to be and remain in full control of what you feed the MT-“engine”! Full stop.
    I therefore recommend trying out Slate Desktop ( )


    • * its merits


      • * lose out

        (sorry should have proofread before posting 🙂 )


      • Wouter, if one does feel inclined to look at Moses-style MpT, Slate is certainly an option which avoids most of the usual ethical concerns. And I suppose that catalogs and certain other text types might benefit, though I find regex-based insertion rules preferable. With regard to patents and legal text, however, which are a mainstay of my work and make up a large portion of my archived content, I saw no benefit whatsoever in my working pair. It was an interesting study, and I have nothing but praise for the company’s attitude, but the results made it clear that this was not a good investment of my time. Not surprising, really, since the best performance cited in various articles by Kirti Vashee in the past for HAMPsTr processes to produce “understandable” text cannot beat what I routinely see for good output by translators who know their subject matter well and use speech recognition. Competent, dictating translators beat MpT monkeys every time in both volume and quality. Where MpT is said to help, a careful analysis of the situation will likely reveal some serious limitations of the translator’s language mastery. Hm. Make that “translator”.

        Liked by 1 person

      • Thanks, Kevin, but clearly Slate is losing the battle. Kirti Vashee and company on one side and desired customers with closed minds on the other side. The Vashee and company crowd are clearly winning. It’s really not worth my time and energy any more. I’ll continue supporting customers but the handwriting is on the wall.


    • @ Kevin Lossner
      and anyone else who cares to read about an unwelcome truth.

      Sorry for the delay. I prefer to spend my Sundays away from the screens 🙂
      No I didn’t went gardening, though I’d loved to.

      This man facilitates MT operations that I find in one word despicable!
      2) I have seen the results of speech recognition far too often to be of substandard quality where one simply reads the syntax of the source text through the literal “translation” produced. I sincerely believe that there are exceptions to this acquired rule of thumb and I am sure you Kevin are one of these exceptions. People like you remain however a rarity in the shire of women’s lib dominated social media omnipresence where the effect of their pampered Dragon is hugely supporting the whirl of spiralling down rates that sucks in just everybody. Henzellian exceptions are delusional and may be attributed to tiny personalities standing in the shadow of their gigantic phalluses. I prefer to ignore and negate them.
      The suppliers of substandard translations use speech recognition for the very reason of maintaining their hourly rate while undercutting the word rates of perceived competitors.
      I can undercut much better and still deliver outstanding translations as well as transcreations of poorly written technical content. I can because my expenses are 3 to 4 times lower than those of my typical competitors with no technical background let alone experience whatsoever. I saw the current shambles coming after a private behind the scenes talk with Yves Champollion at the Proz Conference in Krakov. So when my then wife again pressed her desire to relocate to her hometown in the deep South of Thailand I started to concur. It wasn’t easy but it sure was worth it. Now I am here and my word rate and speed of delivery “competitors” are still there, with no means to establish themselves in this part of the world where feminism is a daily routine without losing out on femininity.


      • “Now I am here and my word rate and speed of delivery “competitors” are still there, with no means to establish themselves in this part of the world where feminism is a daily routine without losing out on femininity.”

        I don’t get this part, Wouter.

        Could you elaborate a little more?


      • Wouter – you can mention my name. I won’t be offended as you have a right to your opinion.

        I actually agree with Kevin Lossner that some high performing translators (especially those who have learned to use STT software to speed up input or typing) can easily outperform most MT systems. What good MT systems do sometimes is to raise the productivity of the less skilled translator, and thus is really a solution for the bulk market content or a way to get content that would simply NOT get translated otherwise. Kevin Hendzel offers a strategy that I think is pretty valid for those with higher skill levels.

        Liked by 1 person

      • Basically, I agree with Kirti – all those who still translate the conventional way (text in text out), risk to become obsolete, be it due to MT or someone who does it faster or cheaper or both (rather cutting cost than creating more value). The only exception is what I call “added value translation” – clearly the premium segment and far less exposed (even increasingly less exposed, the more MT and human translators who are qualitatively no better than MT are there around). Efficiency or, like Kirti says, “better productivity of the less skilled translator… in the bulk market” doesn’t promise any sustainable business. Only higher skills do.


  4. I’ll open by quoting Clayton Christensen, the tenured Harvard professor who coined the term ‘disruptive innovation’, “I made a mistake calling the phenomenon ‘disruptive’ because there are so many connotations of the word “disruptive” in the English language. So there are a lot of people who call anything that is a dramatic improvement ‘disruptive’ and that’s not true. The disruption is built in to the *business model of the enterprise*, not by developing the best technology.”

    “and that’s not true.”… interesting. Relative to Christensen’s thesis, dramatic improvements are not disruptive.

    Secondly, I want to apologize to Josephine Bacon for my rude and emotionally disruptive comments on Linkedin. I addressed that exchange here on my own blog: When we point a finger, there are three pointing back at us. I saw those three glaring at me! Josephine, I hope that this exchange can be that constructive and intelligent discussion now that I washed-off my own head.

    Continuing with Christensen’s thesis, *competitive response* — not a technology or its quality — plays the most significant role in determining whether a *business model* is disruptive. As I interpret his thesis, low-quality Google (et al) is only disruptive because high quality translators fail to engage in a competitive response. Professionals are only at risk when they continue to deny the phenomenon.

    If you search YouTube, you’ll find many Christensen examples of companies who stuck their heads in the sand, doubled-down on their top-end products and failed to offer a competitive response. I don’t want professional translation to follow that path and I’m working my butt off to back-up that claim.

    Christensen also gives one clear example of a successful competitive response. It actually saved Intel from bankruptcy. In the mid 1990’s, Intel’s was losing significant market share in the low-end computer market (i.e. the one that grew to today’s telephones). Intel could have continued its focus on top-end technology with higher capacity and greater speed at top-end prices and profits margins. To do so, under Christensen’s theory, would have been a failed competitive response and they surely would have gone out of business.

    Instead, they consulted with Christensen. He didn’t teach them how to make new chips. He taught them a new thought process. He taught them how to view their business model from a different perspective. Then, Andy Grove (Intel founder & CEO) figured out how to make the Celeron to compete at a low-end and simultaneously maintain their high-end business. In effect, Christensen taught them how to fight fire with fire.

    Slate Desktop is NOT advanced technology that surpasses human translators. It’s a tool that offers individuals the opportunity to fight fire with fire. It’s an opportunity for a new freelancer’s business model. It is the Celeron that offers you a competitive response to maintain your high-end business and fight Google.


    • There is a difference between disruptive innovation and disruptive misrepresentation.

      I know business is in love with the former, but sleazy merchants of machine translation are selling the latter while shamelessly pretending that it is the former.


      • Technological innovation very seldom is disruptive in itself, it is business models that tend to be disruptive.
        Disruptive is e.g. what RyanAir has done to the aviation industry, and also what Virgin usually does. This has very little to do with technology. Technology for them are just the tools. The disruptive character of their commercial actions is that they turn standards and concepts upside down. Disruptive is not necessarily a bad thing. Con-operations are however very disruptive while destructive by their very nature. In my experience most large scale MT-projects border con-operations in a multitude of ways. We as translators should adapt and come up with our own disruptivity — if I may say so — to prevent being turned into new-born 19th century journeyman by the market forces unleashed onto us by unscrupulous business people.


      • Thanks Steve. It appears that Christensen, you and I all agree that there are those who misrepresent his theorized phenomenon. I don’t see how those misrepresentations relate formulating a competitive response that’s designed to thwart the truthful or misrepresented phenomenon. Am I missing a subtext?


      • Wouter, I agree that the phenomenon is not necessarily a bad thing in terms of a greater good, but it is a bad thing when you’re on the losing side, like DEC computers and Bethlehem Steel.

        I spent the day pondering Steve’s post (here) and Kevin Hendzel’s latest “Creative Destruction.” They have totally flipped my perspective on Christensen’s theory of disruptive innovation phenomenon relating to Google Translate as potential disruption. So much so that most of my previous thought and writings are now out of date. I’m drafting a new article to post later this week.

        I’ll go a step further than Steve. It’s not necessary to anticipate a war because it’s already upon us. I’m just not sure if it will go nuclear. As such, translators need to formulate a competitive response because at this point, I believe Google (et. al.) is only “potentially disruptive” and the path to complete disruption is not certain. I disagree with Steve that the any response or “preparations will probably be inadequate.” Intel responded and they’re still the world’s leading CPU manufacturer. Failure to devise a competitive response is the only guarantee that potential disruption will actually become disruption.


  5. Good morning 🙂 Maybe because I am also a journalist by trade (although only writing articles and such very occasionally nowadays) what I see in all this is a trend of disregard for human language in general. I see bad writing (bad grammar, bad spelling, bad punctuation), which has not been translated, everywhere, as if ‘less is enough’ in order to communicate (and I don’t mean ‘in a succinct way’), as if ‘as long as they get what this is about, that’s fine, it doesn’t have to be perfect.’ And from the translation point of view I see the same when I do editing, proof-reading (whatever they call sorting out somebody else’s mess sometimes): the PMs don’t like me to point out all the mistakes, they want me to say the translation was fine, to just do “basic proofreading and spotting translation mistakes.” If people don’t care about non-translated texts why would they bother too much about translating, which is perceived by many as an inferior writing task (“anyone who knows two or more languages can do it”)?
    Another funny thing is that now and then I take some short test to participate in bigger projects (I never really took part in any of those projects, have no idea what they pay, but I always take them in consideration for the steady income they might offer for some months as long as they are compatible with my general schedule and pay is ok – which I know is unlikely but I just give it a go if the contact has been made in a professional way). I never pass. They are not very difficult and the fact that I am pretty sure I didn’t do any major mistakes in my translations intrigues me. Then I think – maybe it’s because I don’t always do ‘segment’ translation, that if they check the order of the sentences, the syntax, it will not look like a “correct translation” (I say that because one of the agencies I work for sometimes query on those things, because the translation don’t look in the same order as the original, so they double-check). It is puzzling because they don’t even say what the problem was, they just say “you didn’t pass and we cannot give you any further feedback on this”. Good riddance. I almost feel that trying to do a well written translation instead of a ‘mechanical one’ is a kind of crime.
    The other day I saw something a bit scary on a LinkedIn ad, by the people who sell that CAT tool so worshipped by lots of translators. They were coming up with a ‘improved post-editing functionality.’ From what I’ve gathered even that function, that lots of translators now accept will be the job that will be left for them to do, can be better aided by the machine (what again possibly means less work and less pay for the translators who are content to do mostly post-editing). I may be wrong and this is not scary at all and it is just some inconsequential improvement but…


    • “Another funny thing is that now and then I take some short test to participate in bigger projects (I never really took part in any of those projects, have no idea what they pay …..”

      This is just a good way to determine whether you are going to be one of the countless minions willing to work for a pittance.

      My standard response is that I would be happy to do any test as long as it paid for.

      Liked by 1 person

      • Hi again. I never really did get the chance to see how much they would pay (sometimes they send those impersonal emails and say how much they are going to pay, these I discard straightaway
        …) but I have a feeling that my stated minimum rate for this kind of job is another factor of me not passing the tests 🙂
        I don’t do long tests for free, no way. The only long ones I did were for books which I ended up translating and therefore getting paid for doing them (they were excerpts of those books). But I do the occasional short test if it’s a project that interests me and sometimes I get paid for the tests I do (mainly for agencies I have worked for).


  6. Jo, alt magt til ordblindere!
    And let me be absolutely clear. We don’t need PE-functionality!


  7. “We as translators should adapt and come up with our own disruptivity — if I may say so — to prevent being turned into new-born 19th century journeyman by the market forces unleashed onto us by unscrupulous business people.” ….. and aided and abetted by our so-called professional association, I would add.

    What is the position of your “professional association” on post-editing of the machine translation detritus?

    If they say that it is a valuable tool that translators should add to their set of skills, who are they really working for?

    Not for us, translators, that’s for sure.


    • “….. and aided and abetted by our so-called professional association, I would add.”

      My professional association has de-facto frustrated many attempts to support their members’ professional status and opportunities, in the past. For me it is clear that they have no intention to facilitate working for direct clients. And as far as I know they do not have any apparent opinion on MT and the whip of PE.

      PE is not at all any kind of skill that one should wish to acquire, if any skill at all. PE negatively affects your ability to improve fluency in translation. This is especially very harmful for translators going trough their first phases of building on their skill-sets through experience (freely quoted after Kevin Lossner’s remarks at the IAPTI conference in Athens.
      I did a lot of PE from 2004 through 2010 for one of the dominant globally present agencies. I stopped after they started to involve substandard translators that did not have what it takes to understand technical content. Before that their MT projects were well planned and staffed.
      Just to avoid confusion; Slate Desktop does not rely on PE. After the initial training with YOUR personal TMs on YOUR PC — and nothing remotely in the proverbial cloud — SD suggests MT on a segment by segment basis. Whether you take the suggestion and adapt it or reject it, is up to you. You as a translator keep full control.
      If you want to know more for your specific language combinations please contact Tom Hoar.


    • Steve, I agree that translators need to adapt (see above). In the phenomenon’s jargon, this is called a competitive response, not disruptivity. Post-editing is not the answer. Get this… I make software that uses an SMT kernel and I disavow post-editing. Our translator customers make engines with the potential to create 20%, 30%, or even 50% correct suggestions. I teach customers, “if the suggestion isn’t correct, just trash it and do it yourself.” Hence, not editing or post-editing involved.

      Please don’t attack me for saying this. This is not supposed to turn into a sales pitch. I will gladly explain those percentages if someone asks. There are many caveats and not all customers experience the potential.

      My point is supposed to be about post-editing. If the translator can make engines in the privacy of their own PC (i.e. no 3rd party service involved) and those engines generate a sufficient number of acceptable suggestions, then traditional post-editing “best practices” become obsolete. Heck, those best practices date back to the early 1960’s.

      Steve, I’ll share one customer experience that I believe you can relate to. Igor Goldfarb is an EN-RU translator in Moscow. His average EN sentence length is 65 tokens. His patent engine that he made from 4 years of his personal TMs and uses on his PC changed his entire work experience. He’s refined his work habits in the privacy of his own workspace (i.e. no cloud or oversight) such that now he can read, comprehend and fix 35% of his engine’s suggestions within 60 seconds. If he judges he can’t do it that fast, he simply scraps the suggestion and falls-back to his pre-engine habits. So, he scraps 65% of what his engine spits out. In the end, he now meets deadlines with time to spare. I’ll put him in touch with you if you like. Or, maybe you’re reading this… Igor?


  8. [Here is my response to Tom Hoar copied from FB — nothing revolutionary or necessarily insightful — just to provide some clarity on my position.]

    The impact of GT is already melting the foundations of the bulk market, and driving down rates. The instances of translators being asked to post-edit rather than translate have skyrocketed, and the GT going on behind translators back (by agencies, mostly) is important to recognize.

    Dismissing GT’s ability to leverage the work of tens of thousands of our colleagues into very clean copy is a huge mistake. This is not even about statistical machine translation, it’s that GT has become a massive real-time CAT tool delivering compelling formalized texts (like NDAs, or basic contracts, or even P/L statements, for example) for free.

    More than one former ATA President has declared “the gig is up.” I don’t happen to agree with that position at all — the “gig” is just evolving, and fairly rapidly.

    The purpose of my article was in fact to point out how translators could offer a competitive response to GT that would effectively shield them from this massive leveraging.

    There is an appropriate tool for every job, and the closer-than-you-think near future of “good enough to understand” will likely exclude humans (except in rare language pairs — there will always be exceptions).

    But then, humans will be required for other jobs GT can’t handle. It just requires humans to dramatically upgrade their skills. So that will be their competitive response.

    I agree with you on Intel, but they missed their first attempt when confronted with competition in the smartphone industry, a market that totally blindsided them (Qualcomm killed them) so they had to learn the second time.

    My article cites Kodak as another company blind to creative disruptions, but there are other meltdowns, like poor Nokia, which almost disappeared overnight.

    All these companies have an internal narrative that their product is different, or the disruption is not real. Kodak’s response was to try to hide from their own invention! Intel got smart by asking for advice the second time (but not the first).

    It’s important that translators be aware that these market forces are out there, and they need to start making changes in their own training and working routines now to remain competitive.


  9. Everybody knows what the position of the translation industry is, namely the same as your position, Kevin:

    “Work harder, faster, and lower your rates, translators, or you will die, die, die, because GT is out there and it is going to kill all of you if you don’t do as we tell you!”

    Based on my experience of a specialized patent translator of 30 years who does not need to work for the translation industry, it’s total nonsense. So much so that I will not waste my time trying to discuss it with you here.

    I would appreciate if you saved your opinions for Facebook and your blog, which I do not intend to read.

    Is that so much to ask? After all, this is my blog, not yours.

    Thank you so much in advance!


    • Steve, just for the record, your description of what the market purportedly demands of translators: “Work harder, faster, and lower your rates, translators, or you will die, die, die, because GT is out there and it is going to kill all of you if you don’t do as we tell you!” — happens to be the opposite of my position.

      Diametrically the opposite.

      I’m arguing for how translators can in fact AVOID that trap. To tell people who are demanding that translators behave like crack-smoking hamsters running like crazy on the hamster wheel to (figuratively) go to Hell.

      It’s a pity that you are passing judgment on a blog post you don’t intend to read!

      I think you might be pleasantly surprised about how my emphasis has always been on skills and subject-matter expertise — about moving above that crazy fray you describe for good — and I sure would never, ever encourage translators to jump on the perpetually faster-turning hamster wheel.


  10. For some slightly different views on this MT armageddon, you may wish to look at which depending on how you look at it either extend your thesis or provide ways to deal with the scenario.


    • What do you think will happen if everyone tries to occupy upmarket niches and even the premium market?
      The premium market is occupied by creative translators and writers who do not necessarily excel at their jobs. They are there because they know the right people. They are well connected. So, what do you think they will do when you knock on the door with a talent and a skill-set that totally outreaches theirs?


      • I am not sure I follow you, Wouter.

        I am not well connected, unless you mean by that being connected directly to direct customers and ignoring the translation industry.

        And I consider myself being being in the top tier of translators, but only because I spent much of the last thirty years trying to figure out how to get there and how to stay there. It is a continuing process.

        The key is figuring out how to bypass the translation industry, or at least the worst corporate forms of this industry.

        It’s not easy, but I believe that just about anybody can do that.

        But most translators are too lazy, they don’t even try. All they do is just bitch online.

        Misery loves company.


  11. Translating is writing, so as long as writing can be automated, translating can, too. Automated writing might be of help in definite settings (some were mentioned above), but automated translation – even of automatically generated texts – is prone to errors (let alone it doesn’t inspire to read).

    The cause of failure lies in the inability to interpret the written text or rather the message behind. Interpreting, on the other hand (along with a talent to write), is something that MT is (will be?) always lacking. For this reason, I feel kinda immune to the dangers. I have been always interpreting the written text, rather than rendering it in another language, be it adapting it to the needs of a target audience or modifying it to bring the message across in a more convincing manner. “Better than the original” is hardly a fit slogan for our clients, but in the end, it is what it boils down to.

    As for the analysis of MT, PE, and the general state of the “industry”, I am mostly agree with Wouter. Well said!

    Liked by 1 person

    • MT is recycling of translations done by humans in a much smarter way than the CAT-tool way. That’s what it is and there really is nothing more to it. If you feed an MT-process with shitty translations as a basis to translate a disaster of a source text then the result is even shittier than you could ever imagine it to be. If however a skilled and talented human is asked to translate a disaster of a source text — let’s say a maintenance manual written by a technician who neither had sufficient time nor any proper writing skills — then people like me and you, and Steve are able to transcreate that disaster into a correct and maybe even inspiring piece of assistance for operators in doing things the right way 🙂


    • Valerij, Whether you interpret or rerender the written text, you use the language’s rules to construct the translation. The software can’t do that, but it can discover and mimic some of those rules some of the time. The problem with shared cloud-based system is that it learned to discover and mimic an amalgamation of junk from thousands of translators. If it learned to discover and mimic your work, it does a much better job.

      I agree that the software lacks the ability to interpret the written text or the message behind it, but I disagree that’s cause of failure. I believe failure is due to corporate greed that brainwashed gullible (thank you, Steve, for the word) customers and translators that they can expect more from the software than the ability to discover and mimic some of the language rules some of the time.


  12. Inability to think would be another way to put it. Ability to think is an obvious prerequisite for writing or translating.

    Machines can’t think, they can only rehash the thoughts of other people in a way that simulates thinking.

    But the machine does not know whether the rehashing was done properly, and it cannot make conclusions about the accuracy of the content of what it has just translated.

    So no matter how much processing power is available to “language technology tools”, the result will always be just a tool for human translators, not a real translation.

    But of course, the merchants of “language technology tools” can’t admit this simple fact if they want to be able to keep peddling their merchandise to gullible customers.


    • Steve, I agree with all but one thing. You said it “can only rehash the thoughts”. I’ll dumb down the SMT and NMT software a few steps from that. It “can only rehash the tokens”. It doesn’t respect words or sentences or ideas or thoughts. Everything is a token of one or more characters separated by spaces. AI really is a very stupid.

      So, let me ask you this. If you were were talking to one, would you recognize a “merchant of ‘language technology tools’” that admits the tools’ results are always just for a human translator, that admits the results are not a real translations and that is searching for intelligent customers capable of critical thinking?


  13. Reblogged this on Translator Power.


  14. Reblogged this on Make Quality Translations Easier and commented:
    If you’d asked me whether Google Translate was disruptive innovation before I read this blog, I would have answered, “Not at all.” Now, everything is different. I see Google Translate as potentially disruptive innovation. Why “potentially?” Because I believe Slate Desktop is the solution to stop the Machine Translation Armageddon.


  15. “What if big corporations and publishing houses decided to save money by switching to machine translation, lightly edited by poorly-paid humans called post-editors?”:

    it’s already happening, at Amazon for example. And at some EU institution. So some extremely large organisations are already doing it.

    That being said, the guy in charge at Amazon has admitted (cf a recent article in Slator) that machine translation was far from perfect/satisfactory…

    I guess that what is happening is the following: organisations’ upper bosses overhear something about saving “huge amounts of money” on translation, in the era of Big Data (huge volumes to translate).

    So their Communication directors etc feel compelled to at least try the beast (machine translation). And underpay translators in the meantime.

    I can testify that I have been treated like shit by the Amazon contact person.

    From university-trained linguists were are becoming faceless sub-slaves that worthless and brainless contact people feel entitled to treat like shit…

    We should only accept to deal with university-trained people and not secretaries or people like that.

    The more brainless they are, the worse they act…

    I think I will add that requirement to my next mailing: “I will only accept to deal with your university-trained personnel (if any) (except your accountants 🙂 )”.

    99% intermediaries have never translated a single line in their whole lives…

    As you once wrote, translation “agencies” are “growing like mushrooms after rain”: it’s exactly that, just mushrooms…

    The other day, some new mushroom said she was interested in working with me, she said she wanted to test me (for free, of course) and asked “what tools” I was using (for the fuzzy match rebates, of course).

    So I increased my base rate (by only 2 eurocents compared to my usual Belgian agency rate), to compensate for what she wanted to extort from me. No more news ever since. French rates are much lower than Belgian rates already, so…

    I could have tried to contact her again. But for one thing, most of these agencies have names nobody ever remembers (the only one I remember was Genius something and a few other striking names. Their names often start with Trad- / Trans- / Word- which denotes a total lack of imagination and their not belonging to the translation world) and on the other hand I hate her already, so…

    I have reached total saturation point with those unprofessional/amateur intermediaries who want everything for free.

    They don’t realise that it’s OUR TIME they are paying for.

    It has ALWAYS been like that, with higher per-line rates for translations that required more research/formatting work…

    SDL & co make them believe that they are BUYING WORDS! Which is TOTALLY SILLY!!

    “How many dozens eggs, or rather words, Ma’am?” “Shall I wrap them up?”… 🙂


    The ONLY translation intermediaries who work professionally (most of the time) are trained and seasoned translators.

    The rest are crooks.

    And they most probably cheat end-customers just as much as they cheat freelance translators, since it’s in their genes…

    So, customers, beware of those Trans- / Trad- / Word- / Text- things!

    They are growing like mushrooms after rain and do not have the faintest idea of what translation is all about.


    • “The more brainless they are, the worse they act…”

      Tell me. I have a lifetime experience with such people. Only 3 years ago at the age of 55 I have come to the conclusion that I finally have learned how to handle the the crooks. They do however have one single talent which they have nurtured into a skillset from hell. It’s called sociopathic behaviour.

      And yes, most large scale MT-projects border con-operations.


  16. Yes, so many of those agencies springing up like mushrooms after rain are poisonous mushrooms these days.

    So glad I helped you to get all of that off your chest, Isabelle!


  17. * at the age of 55 I could conclude ….


  18. I received this private email from a translator who I respect and support. Thank you, friend, for your returned moral support as you read/listen to these exchanges.

    “The people you engaged are extremely vehement and damaging with their opinions, especially the two Kevins. Other translators and companies have also suffered in their hands.”

    Good luck with your post-M Armageddon. I choose not to die on this hill.


  19. “Basically, I agree with Kirti – all those who still translate the conventional way (text in text out), risk to become obsolete”

    Kirti is just selling a product, what else do you expect him to say?

    As are some of the other commenters here.

    I translate text in, text out.

    It does not matter how you translate, as long as you do it well.

    The important thing is what you translate and who you translate it for.


  20. “Kirti is just selling a product, what else do you expect him to say?”
    Please allow me to disagree. Kirti is selling consultancy services.


  21. “It does not matter how you translate, as long as you do it well.”
    I concur. Whatever tools I feel come in handy, I will use.


  22. @ Steve

    “I am not sure I follow you, Wouter.I am not well connected, …”

    I was referring to the adage that translators should enter the artist niche to become creative translators. Being a proficient translator, and if needed transcreator — of poor or even disastrous source content, in the technical and scientific domains is a bit different from being a celebrated translator and transcreator of marketing content for let’s say a new brand of chocolate pastilles or working as a translator of non-technical and non-scientific prose.
    Basically our work is good to excellent when we succeed in rendering content correctly in another language. For manuals I’d like to add + producing very readable and easy to understand content in the target language, while for patents the + is to produce a translation well harnessed against infringement. You may add to that for I did some patents but a do not consider myself to be proficient in that domain. I actually like to thank you for pointing out that splitting up long winding claims in order to improve readability is a NO NO in patent translation. I didn’t know that until you made a remark to that extent on this blog 🙂


  23. Every Jack has his Jill. For each post-editing project, there are certain translators (graduates?) who are unable to translate patent projects or cope with website localization projects. Agencies (or clients) don’t care much who you are, they offer the project to everyone until one accepts it at the lowest possible rates.

    I guess this is just some kind of another niche in the translation industry post-editing. And it will exist as long as there is a demand in the market. I hope, the end customers are notified of the process so that they could see the reasons why the translation service is cheap.

    I work mostly in the localization industry. I am pretty sure, machine translation will never replace translators and localization specialists because the main goal of website translation is to bring new customers through Google search results. And that involves keyword research and analysis – machine translation tools doesn’t do that.


  24. […] this I mistakenly thought started from some Twitter banter, when I presumed that Kevin saw this post, which sees MT driving translators (and the translation industry) into an Armageddon scenario. […]


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: