Posted by: patenttranslator | March 6, 2017

Interplanetary Travel and Perfect Machine Translations – Two Currently Unattainable Dreams

In 1969 when Neil Armstrong and Buzz Aldrin became the first humans to land their lunar landing module on the Moon, it appeared that humankind has entered a new era, the era of interplanetary travel. Surely, if we could make it to the Moon, we would soon be landing on Mars and then also on other celestial bodies, which should be more interesting than the rocky Moon landscape. Perhaps it would take a decade or two, but definitely not longer than that, most people thought. Maybe we’ll even meet some cute and friendly aliens along the way if we get lucky.

Some people maintain to this day that the Moon landing never really happened and that the whole thing was a hoax, just a movie, not unlike those that are made in a Hollywood basement, as Red Hot Chili Peppers put it in one of their songs.

I tend to think that this conspiracy theory is probably incorrect – although who knows these days.

Other people say that the Moon landing did happen, but that it was only a tour de force designed after the sneaky Russians managed to launch the first human into space, and that the United States did the Moon landing largely for propagandistic reasons to prove the continued superiority of American technology. Skeptics say that this feat, spectacular as it was, did not bring back to Earth anything that would be particularly valuable for science and R&D. In other words, that just like launching the first human into space, the Moon landing was also designed mostly for propagandistic reasons.

I tend to think that this conspiracy theory is probably correct.

The fact is that almost half a century later, interplanetary travel, a dream that seemed so close that one could almost touch it in 1969, still remains what it was almost half a century ago – a beautiful, but elusive, improbable and unrealizable dream.

Unlike the Moon, those planets that humans would dearly love to reach now are so far and such an incredible amount of energy would be needed to power a space craft for interplanetary travel with humans on board, not to mention the years that it would take to get there, that at our present stage of technological development, we are simply unable to do anything about interplanetary travel, except to write sci-fi novels and make movies about it.

Vendors of various customized machine translation packages will never admit that the other dream mentioned in the title of my post today, machine translation that is as good as human translation, another popular feature of sci-fi books and movies, is also just a dream that remains unrealizable at our present stage of technological development.

Of course they can’t say something like that out loud, although most of them must know it by now. Because if they dared to say what they know in their hearts must be true, they would be instantly out of a job, just like the alchemists from the 15th century would have been immediately out of a job had they dared to reveal to their rich and greedy sponsors that, unlike Rumpelstiltskin, they cannot turn straw or lead into gold.

The alchemists of old made sure to wring out as much money as possible from their credulous sponsors before admitting the obvious. Some three hundred years later, machine translation alchemists of our era are merely following in their tracks when they say that machine translation may not be perfect yet, but if you wait just a few more years, perfect translations, or at least translations that near perfection, are just around the corner.

Legend has it that when a primitive machine translation system was put to a test decades ago with a structurally very simple sentence (although not really so simple when it comes to content), namely a sentence from the New Testament, (Matthew 26:40-43, New International Version) that says “the spirit is willing, but the flesh is weak”, it was allegedly “translated” into Russian as “the vodka was good, but the meat was rotten”.

It’s possible that this particularly revealing mistranslation is just a myth, and maybe another famous example of an alleged mistranslation, when “out of sight, out of mind” was allegedly translated from English into Chinese with machine translation as “invisible insanity”, is also only a myth.

But translators who use machine translation on a daily basis know that not only both mythical mistranslations, which may or may not be real, but even much, much more hilariously nonsensical mistranslations are indeed quite plausible because they are produced now by our present-day machine translation system every day.

When I tried to put machine translation to a test a few months ago on a translation of a Japanese document which you can see it for yourself right here in this post to counter the constant drumbeat of machine translation propagandists, the result was pretty unbelievable…

“But what about GoogleTranslate and Google’s new “neural” statistical machine translation model?” some people are bound to object. “Isn’t it already sometime almost as good as human translation?”

Well, no, it’s not, not by a long shot. It looks in parts as good as human translations because it is based in part on translations that were done by real human translators. But the software sometime incorporates unrelated human translations, and sometime aggregates different pieces of different translations together the wrong way, which means that the machine translation cannot be trusted. If it could be trusted, I would already be out of job, but it so happens that I am so busy that I’ve hardly had any free weekends since November and it is already March.

This weekend, for example, I have been translating nine sets of claims from nine examined Japanese patent applications. These patent applications were filed in Japan in Japanese, but originally they were filed in different countries and in different languages – in English, French and Portuguese.

I found a lot of information about other versions of these patents that exist in other languages than Japanese, most of which I can read, and when I was looking at this information, I was thinking, it must be very difficult for a patent lawyer who does know Japanese to figure out what really was filed in those Japanese patents in Japan.

I don’t know how many good translators are there who can translate French and Portuguese patents to Japanese. I doubt that there are many translators like that in Japan, but even if there are some or quite a few, my guess would be that in a pinch, a patent law firm would simply use a translation from English for translation to Japanese if they can’t find a translator for the original language. Which would mean that I may have been translating to English something that was translated from French or Portuguese to English, and then from English to Japanese.

Combined with the fact that there are many differences between different versions of the same invention depending on whether these documents were filed as unexamined patent applications, as examined patent applications, or as issued patents in different countries, and depending on in which language these different documents were filed, there is basically no telling what really is in the Japanese version unless one can read Japanese, or unless one can pay somebody to translate the Japanese version, although the Japanese version might be the result of two other translations, for example from Portuguese to English, and then from English to Japanese.

And how does one know which differences are due to the different rules for filing patents in different countries with different patent systems, and which ones are due to potential misunderstandings and mistranslations? If a text was translated several times between different languages, there must have been a mistranslation or two involved somewhere in there, right?

I did print out machine translations of the Japanese texts, which I did find confusing. Basically, only machine translations of very short sentences made sense, the rest of the machine-translated output was usable basically only by a translator as a dictionary.

Which means that the machine translations were very helpful to me, but only because I do understand Japanese.

There is no denying that machine translation technology has come a long way since the times of good vodka, rotten meat and invisible insanity. It is much, much better than what it was 50, 20, 30 or even 10 or 5 years ago.

But the amazing and very significant developments in machine translation technology over the last few decades have basically only covered a distance that is approximately equivalent to landing on the Moon. This was a major accomplishment, but an accomplishment that, contrary to expectations, was not followed by traveling to faraway planets for a simple reason – this is presently impossible at our level of technological development.

Creating software capable of producing machine translation that is as good as human translation is about as easy creating a robot that would be indistinguishable from a beautiful woman who can play a hauntingly beautiful melody on a piano, with all of the feeling that only a human being who happens to be an experienced and accomplished concert pianist can put into a command performance.

Yet, I don’t think that practical magic of this  type is impossible, and I even think that I know when something like will become a reality.

It will become a reality one day when humans finally meet cute and chummy aliens who will land on our planet and generously share with us technology that will enable us, humans, to do the same things that have been possible for centuries or millennia already on faraway plants that are much more technologically advanced than what we have here on planet Earth.


  1. In 2000 and 2010 you said that excellent machine translation wouldn’t be possible for ceenturies. Then in 2013 you wrote it would take 200 years and later wrote excellent MT won’t be available for 20 years.

    Do you see where this is going?

    Excellent MT doesn’t have to be perfect to reduce high end tranlator’s rates by 50% within a couple of years as it keeps improving. I wonder how many high end American J>E translators will be willing to translate at $40,000 a year which is below the median income.

    Liked by 1 person

  2. Wow, you remember what I said on my silly blog 10 years ago. I am flattered, Ghost. I don’t remember so far back myself.

    I think that MT did reduce the rates of some translators, but mostly of those on the low end, not on the high end.

    I also agree with a comment of a patent lawyer on my blog a few months ago who said that it is mostly the fact that English equivalents of foreign patents are now clearly identifiable on internet, not machine translation, that causes less of demand for patent translations from Japanese.

    Machine translation keeps getting better, but only incrementally, and from my perspective, the increments are very small.

    I think that the greed of the “translation industry” is a much more important factor behind falling rates. The “translation industry” has been waging a bloody war on our profession and the results are not pretty. The war has been so bloody that the industry is in fact killing itself without realizing it, just like the “telephone marketing” industry is killing itself too. The result of its attacks on all of us is that people no longer answer the phone unless they recognize the caller, and the result of attacks of the “translation industry” on rates being paid to translators is that most of our customers now think that all translations are likely to be of horrible quality.

    We have to be smart and flexible if we want to be able to survive this bloody war. Unlike 20 years ago, when I was mostly translating Japanese patents for prior art research, I am now translating mostly German patents for filing. There is a lot of demand in this area right now, so that is what I have been doing mostly for the last few years. I still translate Japanese patents, but not nearly as many as I used to.

    Tempora mutatur et nos mutamur cum illis (Times change and we change with them), as Ovidius put it.

    I think that we basically have two choices is: 1. we change with times, or 2. we allow the “translation industry” to kill our profession if we cooperate with it.


  3. I think your esssay in Translator’s Journal in 2001 was pretty widely read by translators.

    I disagree that MT improvements have been very small as you wrote. This is for Japanese > English, one of the hardest pairs for MT:

    2003 MT produced incomprehensible text that included many Japanese kanji.

    2006 Google Translate appeared and the kanji were gone but all you could fo is maybe make out the topic but even that was difficult.

    2007 Google Translate jumped a level as simple news stories began to make sense although with many errors. More difficlt topics made almost no sense.

    2010 There was a jump in GT with easier topics but something harder made some sense but often not.

    2013 GT for Japanese>Engilsh was now good enough that simple stories made a lot of sense and was a boon to less skilled translators since GT could give clues to the correct grammar.

    2016 The new GT nueral system is yet another jump just as many said GT wouldn’t get better.

    So huge gains from 2003 to 2016 where 2003 was no better than 1993 for J>E.


  4. Well, you are certainly entitled to your opinion.


  5. Um perfekte maschinelle Übersetzungen zu erhalten, bräuchten wir erst einmal fehlerfreie Ausgangstexte und die habe ich selten. Da ich ein Mensch bin, kann ich alles nebenbei ausbügeln. GT und Co könnten das nicht.


  6. I agree.

    In other words, to obtain perfect machine translation, the input for machine translation needs to be pre-processed, and the output needs to be post-processed, otherwise the result may be incorrect.

    This is was true when machine translation was in its infancy 50 years ago, and it is still true now.

    That is why I call perfect machine translation an unattainable dream at this stage of our technology, just like interplanetary travel.

    (Unless some space aliens at a much higher level of technological development decide to share with us their much more advanced technology).

    Liked by 1 person

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: