Posted by: patenttranslator | August 26, 2015

The Impact of Machine Translation on Human Translators

The complex impact of machine translation on translators is not understood yet, not even by translators themselves.

Many or possibly most translators have been afraid until recently, some of them deathly afraid, that machine translation would eventually wipe most of them off the current list of occupations capable of surviving the Internet era, partly because they have been and still are inundated by a constant stream of press releases and other commercial propaganda generated mostly by merchants trying to sell “customized machine translation system” while promising to save large users of translation a lot of money if they make the switch from human to machine translation.

So-called translation industry would love to turn human translators into low paid post-processors of the machine translation detritus, or of machine pseudo translation, because there would be mucho dinero for operators of businesses based on the ingenious concept of humans who are forced to “assist” machines, if the concept can be put into practice.

Only time will tell, but I do not believe that it will work, mostly because the resulting product will be necessarily of very inferior quality.

For the most part, such a seismic shift in the “machine-human interface” has not happened for reasons that are obvious enough to anybody who understands what translation is and what it is not. We have been fed the line “although machine translation is not quite as good as human translation, within a few years it is expected to be just as good as ….” blah, blah, blah. While feeding of the same propagandistic pablum to the gullible public will no doubt continue for at least a few more decades, most translators and even some non-translators (“civilians”) have already figured out that this is just another extreme example of wishful thinking.

Sure, it’s possible that machine translation will replace human translators. How can I or anyone else say that it’s not possible? But it’s also possible that smallish hordes of purple, green and violet unicorns are roaming the deep forests of Virginia along the fabled Appalachian Trail. How can anybody say that this isn’t possible either?

Even the general public may be finally catching on and discovering the difference between human translation and machine pseudo-translation because so many people have now had direct experience with machine translation and know by now that while machine translation is a very useful tool, it is not really translation.

MT tools are everywhere, easily accessible for free from any computer, tablet or smartphone. But even people who don’t know anything about translation are beginning to understand that computers may never replace human translators.

Yet, machine translation has already had an important impact both on the supply of work available from our clients, and on the work of human translators, including the work of this patent translator.

Some Materials in Foreign Languages Are No Longer Translated by Humans

As I started predicting already in the last century, the impact of machine translation (more correctly referred to as machine pseudo-translation because only human brain is capable of translating the meaning of words rather than just words), on the work of human translators, has been negative in some respects, and positive in other respects.

As a result of the undeniable progress, despite certain equally undeniable limits of machine translation, some translators now probably have less work than they would have had if machine translation was not freely available on the Internet.

For example, it is very likely, in fact virtually guaranteed, that I and other patent translators are losing some work to machine translation. When relevant patent literature was cited in patent applications as evidence of prior art (existing technology) ten or fifteen years ago, the only way to find out what anything about what was described in the prior art was to have the sources in foreign languages translated.

This is no longer the case because machine translation is at this point good enough to give anybody a good idea of the content of sources in a foreign language. Machine translation is generally not sufficient to describe this content accurately, but the machine-generated descriptions are likely to be good enough to eliminate techniques, procedures and devices that are not directly relevant to a new patent application.

But Highly Relevant Materials Still Need to Be Translated by Humans

Some translation work that used to be done by humans is no longer needed thanks to machine translation. But sometime machine translation also uncovers descriptions of existing technology that are directly relevant to such an extent that a new patent application could be rejected due to a lack of “innovative step,” or so that even an issued patent could be invalidated.

Because it would be too risky to rely on the information contained in the millions of words supplied by machine translation, human translators still need to translate tons of prior art references from foreign languages for their clients. Some clients may decide to rely only on machine translation for prior art research, but they will do so at their own risk.

New patent applications also need to be translated from the original languages to file them for example in English and other languages. These translation are then generally reviewed and filed by patent lawyers here in United States and in other countries. That is why a significant amount of patent documents that I translate are translations of patent application for filing purposes. I haven’t seen any decrease in this type of translation.

Machine translation thus may have reduced or perhaps mostly eliminated translations of largely unnecessary materials, but it may also have increased the need for human translation of highly relevant and crucial materials that might not have been detected without machine translation.

Machine Translation Is Now The Best Friend of Human Translators of Patents

In addition to impacting the amount and type of translation work that is now available to human translators, machine translation has also changed the way some or possibly most human translators work. I now always try to first locate and print out machine translation of every patent document in every language that I am translating, whenever possible.

In some cases it’s not possible. For example, no machine translations are available for older Japanese patent applications or for Japanese utility models (a lower category than a patent application) and because the legibility of these documents available only in PDF format is often poor and sometime quite horrible, conversion to a format accessible to machine translation is not an option. Even if the legibility is good, complicated, similar characters in Japanese patent applications are often misread by software, especially in fields such as medicine, biotechnology and chemistry.

Whenever I translate one of these documents, I am on my own and I feel like I have stepped through a murky window back into 1980s or 90s.

Fortunately, only relatively recent patent documents in these field generally need to be translated, and these can be accessed with the machine translation function available for free on the websites of the Japan Patent Office (JPO), European Patent Office (EPO) and the World Intellectual Property Office (WIPO) in dozens of languages.

Post-Processing of Machine Translations of Patent Application Would Be Extremely Counterproductive

Although I print out machine translations first (if they are available,) I never “post-process” or “incorporate” them into my own translations because such a procedure would be extremely time consuming and counterproductive. I only use them as ad-hoc dictionaries, instead of and often in combination with online dictionaries and databases of English summaries of foreign patent applications, such as those available on the JPO, EPO and WIPO websites, and occasionally also with traditional dictionaries.

The only thing that I had at my disposal 25 years ago was an expensive printed dictionary, a book that was already obsolete at the printer’s shop. Fifteen years ago, I started using online dictionaries and English summaries of foreign patent applications. Now I also have machine translation at my disposal.

One implication of the availability of machine translations of patent applications to patent translators is that the threshold for entry into various fields of technical translation has been lowered. I remember that about 20 years, I started translating a medical patent about an ointment …. and gave up my valiant effort after the first 1,000 words or so. There were so many terms that I was unable to find or verify in that patent, Latin names of viruses transcribed into the Japanese alphabet called katakana and names of sea algae hiding somewhere in the Sea of Japan, that in the end I asked a colleague who translated only medical patents to translate it for me. “It’s not really worth your time chasing after these terms,” he told me slyly. And he was right.

But that was 20 years ago, and things have changed. Latin name of viruses, incomprehensible at first when transcribed into katakana, names of rare sea algae, or complicated anatomical nomenclature of human body organs, obscure medial testing methods named after the last names of their inventors, who may have been Dutch, Chinese, or Serbian, again completely unrecognizable once they are transcribed into the Japanese writing system, are instantly and correctly translated with machine translation into English.

And these translations are almost always correct, because unlike translations of common words that may have a dozen alternatives depending on context, there is only one possibility for translating a name, and only two translations of medical terms are usually possible: one in English and one in Latin.

Despite the many fears and apprehensions of human translators about the uncertain future of their profession due to availability of machine translation, the impact of machine translation on the work of this patent translator over the last 15 years has been definitely positive.



  1. “Machine translation […] may also have increased the need for human translation…”

    That is certainly my experience of the situation. People increase their communication exponentially, but the number of translators increases in a more linear manner. Thus, all that machine-reliant communication yields plenty of communication that requires actual, nuanced, human-style translation.

    The question is only how to be the human from whom the project is ordered, which is a marketing question. And marketing is storytelling. And storytelling is great fun.
    In other words: machine translation increases the amount of fun in the world, because it gets us translators into the storytelling tradition, and makes sure that LOTS of our work will be needed.

    As long as I’m not reworking that godawful junk it produces, this new situation suits me just fine.

    Liked by 3 people

  2. […] The complex impact of machine translation on translators is not understood yet, not even by translators themselves. Many or possibly most translators have been afraid until recently, some of…  […]


  3. Among other things, most people don’t appreciate simply the great variety of stuff that clients want translated.

    For example, I just had to turn down a potentially very lucrative job translating lab notebooks because the handwriting was just too scribbled for me to make out enough to make the job worthwhile for both me and the client (as I said to the agent, “I have a lot of trouble reading scribbled English, much less scribbled Japanese”). I imagine the agent will be able to find people who can read the handwriting, but whether they can also deal with the subject matter is another question.

    Obviously, for an MT apparatus to even start on such a project, someone would have to transcribe all the linguistic material into digital files. And since the notebook pages are a mishmash of chemical formulas and other symbols with the words interspersed, the machinery would have to be able to reproduce this structure–something that humans don’t have trouble with. Just the transcription would probably be costly and time-consuming enough that getting MT involved wouldn’t make any sense.

    Also obviously, eventually material like this will be ground out on computer systems, not handwritten. Even now, I’m sure, lab notes are being produced on computers, perhaps with as little input as possible from the human researchers, going directly from measurement instruments to computer, and perhaps MT’ed immediately into a variety of languages. But for the time being, there is still a huge amount of this kind of material that needs translating which only humans can deal with.

    But the basic fact is that language was “invented” by humans, somehow, and only humans can really deal with it fully.

    And your historical remarks on how it used to be to translate Japanese technical stuff is quite correct. In the 1980s, I was struggling with research reports on pesticide chemicals, and trying to track down the names of the chemicals, the plants they were being tested on (many of which didn’t have English names anyway), and the pests being exterminated without an Internet was a nightmare. The Internet alone has revolutionized the translation profession–forget about MT.


  4. Thanks for your comment, Zenner.

    I think I turned down the same Japanese handwritten test reports about two hours ago.

    I have translated a lot of handwritten Japanese test reports over the last 28 years, but this handwriting was particularly nasty.

    It came from phone area code 312, right?


    • Right — good luck to her trying to find a contractor.

      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 )

Facebook photo

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

Connecting to %s


%d bloggers like this: