More Reflections of a Human Translator on Machine Translation

in The Field of Patent Translation

leven years ago I wrote an article for the Translation Journal titled "Reflections of a Human Translator on Machine Translation." The subtitle of the article was "Will MT Become the 'Deus Ex Machina' Rendering Humans Obsolete in an Age When Deus Est Machina?"

A lot has changed since I wrote that article. Computers have become much smaller and much more powerful. As Michio Kaku writes in his new book "The Physics of the Future," the Sony Playstation today, which costs 300 dollars, has the power of a military supercomputer of 1997 which cost millions of dollars. We used to measure the space available for storage in our computers first in megabytes, then in gigabytes, and now we measure it in terabytes. What is coming next after terabytes? What will the universe of data available to a computer the size of a cell phone be used for 10 years from now?

On the battlefield of human versus machine translation, human translators are the winners so far in the second decade of the 21st century.

Machine translation (MT) has become ubiquitous during the last decade. This may seem like an exaggeration to people who live in English-speaking countries, but everywhere else people use MT all the time on the Internet. I see articles that I write for my blog (www.patenttranslator.wordpress.com) translated into Chinese, Arabic, Russian, Portuguese, Polish and many other languages just about every week. The Japan Patent Office (JPO) has had for many years a free MT tool on its website for translations of all Japanese patent applications published since 1994. Google Translate is now available on the European Patent Office (EPO) and World Industrial Property Organization (WIPO) website for translations of patent applications between at least 28 languages, including many European languages as well as languages such as Japanese, Chinese, and Korean. I am sure that many patent agents, inventors, investors, and other people are using these MT programs all the time and find them very useful.

So how did machine translation, which is mostly free, change the battlefield of commercial translation, which is provided by human translators, usually for a lot of money?

I can only speak about my own experience, of course, but as a human translator who has been translating patents from Japanese, German, French and other languages since 1987, long before cheap or free MT became omnipresent, I believe that I do have relevant experience. I think that it is obvious that MT has not achieved what many people thought it would have achieved by now, namely to put human translators like me out of business. In fact, I think that MT has had precisely the opposite effect: instead of taking work away from experienced human translators, it created more work for them. Because most patents available in complicated languages such as German or Chinese or Japanese can now be translated into English in a few seconds with a few mouse clicks for example with Google Translate, many more patent agents are now aware of what is in these foreign patents than a decade ago, and some of them then in fact do decide to order a human translation based on the information which is provided by Google Translate or another machine translation tool.

What I wrote in my article for Translation Journal 11 years ago, namely that: "Regardless of the speed with which computer chips can process information, and in spite of the fact that this speed has increased about 500 times in the last few years, the quantum leap that is a characteristic of human thinking, when we reach a new conclusion based on the information presented to us and based on our human experience, is something that we will never be able to program into a machine. If we could do that, machines could replace not only human translators, but also their clients—patent lawyers, medical doctors, judges, and engineers" is as true today as it was in the year 2000 when I wrote those words.

Machine translation will probably never replace human translators. I use machine translation several times a week, usually when I translate a fairly recent patent from Japanese to English, basically because I am too lazy to look up technical terms in a dictionary. When I translate a Japanese patent published after 1994, I automatically print out a short English summary, about two hundred words written by a human translator, and a hard copy of the MT product which I use instead of a dictionary and also to make sure that I did not skip a line or a paragraph, which can easily happen to a human translator who deals with highly repetitive passages. Incidentally, I think that machine translation and general and specialized dictionaries available on the Web must have put a dent in the profit of publishers of specialized dictionaries because I basically stopped buying dictionaries about 10 years ago. And I also think that Wikipedia probably did or will do to Encyclopedia Britannica what Netflix did to Blockbuster.

There is a school of thought that maintains that as machine translation is being constantly improved, many if not most human translators will eventually become just post-editors of the MT product because this is a more efficient and, most importantly, much cheaper way to provide translation. Translators talk about this subject on their blogs incessantly and often with much trepidation. Some translators are violently opposed to this idea, and I can understand why. I personally don't think that human translators will become post-editors of machine translations because machine translation of patents will never become as good and sophisticated as human translation.

Even a translation that was done by a non-native speaker who does not really understand the technical subject very much will be generally much better than machine translation for a simple reason: even this hypothetical, unqualified human translator is likely to understand the most important concepts described in the original text and his translation will reflect this understanding. But a computer has no understanding of concepts or of the meaning of the words at all, it simply replaces words in one language by words in another language according to a software program. And you cannot program understanding of the highly complex concepts that we can express with our beautiful and complicated human languages into a machine. It is simply not possible to create a software program that can do that, regardless of the power and speed of the computer and how many millions of words are accessible to the machine. Not even the best machine translation tools available now can do that, and this barrier to what can be achieved with machine translation will still be in place 10 years from now or 20 years from now, or even 200 years from now.

If it were possible, I would probably have noticed it by now. Computers are really powerful now, they have plenty of data and talented and highly motivated developers have been working on machine translation software while trying different angles for more than half a century now. But the improvement in the results of machine translation achieved in the last decade or so, although quite significant, has been only incremental, not revolutionary.

I am a frequent user of machine translation. But I only use the MT software basically as a dictionary. I could simply cut and paste the MT product into my word processor and then edit it. But I don't do it because it would take me longer, much longer than if I translate the patent application from scratch, while occasionally glancing at the MT product to confirm my interpretation of a Japanese technical term. Post-editing of the MT product by an experienced translator would probably create a better product, but it would still not be a good translation. It would simply take too long to try to create a good translation in this manner and it would be quite an unpleasant job. It is not cost-effective to work in this manner. Perhaps translation can be done in this manner for some limited applications if the input is pre-edited and the output is post-edited, but based on my experience, post-editing of machine translations of Japanese patents, for example, would be extremely time-consuming and therefore much more expensive than real human translation. Obviously, what I am saying in this paragraph is based on the assumption that the quality of the translation is important. If the quality is not that important, perhaps post-editing of the MT product does make economical sense and there may be applications for post-editing of machine translations in some fields.

But in the patent translation field, which is where I work, quality is important. Sometimes I am queried by patent lawyers about terms and formulations that I used in my translations months or even years ago. There may be also mistakes in the original text of the patent application, and patent translators can approach such mistakes with different strategies. These mistakes in the original text are sometime corrected in a "Procedural Amendment," but more often than not they remain undetected. These are just some of the decisions that I have to make as a patent translator and I have to be able to defend every choice I make. Patent translations are not discarded after use (although they do get lost somehow quite often). They are often used for a long time and by many people during patent litigation, or when patent attorneys are writing new patent applications for their clients.

Is a "Translation Triage" (1. MT, 2. MT with Post-Editing and 3. Certified Translations Performed by Experienced Patent Translators) On the Horizon?

On the other hand, I think it is quite possible, perhaps even likely, that in view of the undeniable progress in the application of machine translation technology to translation of patents, the following three types of patent translations will be used, (a sort of triage if you will), in the near future for different purposes:

1. A machine translation will be ordered first to determine whether a more accurate translation is required. Since machine translations are free, it really is only due diligence to have all documents that are potentially relevant translated with MT software. 
    
2. Some of these machine translations may then be "cleaned up" by a human translator, such as a bilingual paralegal employee of a patent law firm or an independent contractor, perhaps a beginning translator. The human translator in this case does not have to be an experienced patent translator. For example, while MT software may not be able to distinguish when the German word "Schweiss" should be translated as "welding" and when it really means "sweat, perspiration", a human translator can usually tell quite easily from the context what is the likely correct translation.
    
3. Real translations of patents by human translators will be ordered for important documents that are clearly relevant for litigation purposes or for drafting of new patent applications to be filed in English based on certified translations provided by experienced patent translators.
   
   Contrary to popular belief, this "triage" system will probably not create a new, permanent class of somewhat pitiful human translators who simply edit the MT product for a living at low hourly rates, although some people may eventually indeed perform such tasks. However, since documents obtained from these two types of translations (using MT or edited MT) are not reliable enough, for example to be used in court, and probably never will be, the amount of work that will be available to experienced patent translators is likely to be increased rather than decreased as a result of new advances in machine translation.

On the battlefield of human versus machine translation, human translators are the winners so far in the second decade of the 21st century, at least in the field of patent translation, and I think that they will be winning for many more decades or even centuries, especially if they realize that machine translation is a friend rather than a foe.