Posted by: patenttranslator | October 20, 2011

Is Machine Translation Making Experienced Patent Translators Redundant?

I received the following e-mail some time ago from a German patent translator who has been sending me Japanese patents for translation for about the last seven years:


 I hope this finds you well and busy. 

 At my end the working situation is pretty grim.  I was told by a client who used to send me about 2 patents a month – now 3 a year – that they are now using MT.  Apparently the software is good enough, at least in German, to be used with patents.  I am glad this is happening at this time in my working life since I am not much inclined to compete with a machine.

Is machine translation now good enough to make experienced patent translators redundant? It seems to be doing that in the case of this German translator who translates mostly chemical patents from German to English. Incidentally, this translator, who is not much inclined to compete with a machine, has a PhD in chemistry.

But the working situation at my end is very good, and the interesting thing is, I have been probably translating at least as many German as Japanese patents this year. So what is happening here?

I can think of several trends that may be taking place at the same time. It is true that there have been improvements in machine translation (MT), improvements that are incremental rather that revolutionary, but still tangible. But the main change that I can see now is not in the quality of the MT software but rather in its availability. Omnipresence would be probably a better word to describe what happened to MT in the last few years.

While MT used to be available only for Japanese patents on the JPO website a few years ago, it is now easily accessible also on the EPO (European Patent Office) website and on the WIPO (World Intellectual Property Office) website. I can see on my dashboard that even my blog posts are frequently translated into languages such as Chinese, Korean, Croatian, or Portuguese. So it is no wonder that some clients are probably tempted to try to save money by using free MT rather than expensive and slow humans.

I think that this will be probably true in particular in certain fields and for certain languages. For example, the most important part of a chemical patent is the description of which components were reacted together, with what ratios and at what temperatures. This would be hard to mistranslate even with MT. You can get a basic understanding of a German chemical patent with MT and save thousands of dollars if all you need is some basic understanding of chemical reactions. It might not quite work that way for example with Japanese chemical patents because MT works much better with European languages that are similar to English, such as German or French, than for example with Japanese or Chinese.

There are basically two reasons why patents are translated: Either to determine existing technology, which is referred to as prior art, or to file a patent which was originally filed abroad in a foreign language in English.

A large percentage of patents that I have been translating from German this year were for filing rather than for research of prior art. I may be wrong here, but I don’t think that too many patent lawyers would be willing to risk using MT for filing purposes. A simple understanding of the gist of a patent would not be sufficient. If you want to file a version of a German patent application in English, you’d better have everything translated correctly and only an experienced human translator can do that.

I think that another reason why I have been translating so many German patents for filing this year has been the collapse of the US dollar over the last 9 years. When the actual physical notes of a brand new currency called Euro were introduced in 2002, 1 dollar was worth about 80 Euro cents and I remember quite a few snide comments about this “funny money” in editorials of US newspapers at that time. Today, 1 Euro is worth about 1 US dollar and 37 cents and even major upheavals such as the Greek crisis and other problems in the Euro zone have so far not revived the value of the legal tender on this side of the pond at all. Which is probably why thrifty German intellectual property managers are looking for savings by having patents translated and filed through translators and patent lawyers who are based in United States.

Although one result of MT could be that some translators may have seen their workload significantly diminished, I believe that another result of easy access to MT will be more work for other translators. I don’t think that I have lost much work so far, if any, to machines. Even when I translate patent applications for information about prior art (conventional technology) which is still the majority of my work, I think that while translations of some prior art references may be cancelled based on information obtained with machine translation, new prior art literature may be also discovered through MT when it is listed in foreign patent applications, and if these documents appear to be important, a real translation will be ordered from a human translator.

A number of documents that are not patents per se but are highly relevant to patents, such as legal briefs and lawsuit records, so called fire wrapper documentation about procedures such as application rejecting notices from national patent offices and relevant articles from technical journals will also need to be translated by experienced human translators. I don’t think that it will be possible to use machine translation for these purposes in the foreseeable future.

So to answer the question in the title of my post today, I think that some human translators may be made redundant by machine translation, while on the other hand, the same machine translations may result in more work for other human translators.

It all depends on what it is that we translate. Machines equipped with MT software are suitable for translations that are relatively simple and not very important.

Call me crazy but I believe that for everything else, human translators will be needed for a few more centuries.


  1. One thing I’ve noticed over the past year or two is that a higher proportion of the Japanese patents I’m being asked to translate for reference purposes are old Showa era patents as opposed to newer Heisei era ones.
    I have assumed that this is at least partly because these older patents are generally only available as poor quality scanned pdf files, not as readily text-selectable files, and are therefore too much hassle to OCR for use in MT.
    In terms of the actual volume of work coming in, I haven’t seen any reduction, so I don’t think we J-E patent translators need to look for an alternative career just yet, at least not in my field (also chemicals, like your German friend).


  2. That’s interesting.

    Most of the patents that I translate are from the Heisei era and some from the last year of Showa, relatively few are old patents prior to Showa 50.

    But I get a lot of very old chemical articles from German from 1930 or so and from German Reich.

    I wonder why.


  3. Steve, Matt: in the good old days when I was an employee, one of my tasks was to obtain English machine translations of Japanese patents via the Japanese Patent Office website. The gisting was enough for the patent agent to be able to realise that something was totally irrelevant, or conversely highly relevant, in which case we’d commission a “proper” translation, or was potentially relevant, in which case we’d get a few chunks translated and examine those to see if more was needed. I can’t imagine patent agents, if they appreciate the potential for error in machine translation, being willing to risk *not* getting a proper translation done, because I think they’d likely end up paying for it by missing a vital piece of prior art, and then having to go through costly opposition proceedings or the like as a result.

    I did once see a machine translation of the claims “gracing” the pages of a B1 specification at the European Patent Office, though. It was really, really bad.


  4. Hi Alison:

    But the fact is that this friend of mine who has been translating patents for decades and has a PhD in chemistry did lose a major customer as a result of MT, at least for the time being.

    MT has gotten slightly better. I usually print out an MT version before I start translating and then look at it occasionally.

    Machine translations of Japanese patents usually make no sense unless you can read Japanese, but machine translations of German patents occasionally do make sense.


  5. […] The Same Language In The Kitchen Real voices: What translators do and why we need to keep doing it Is Machine Translation Making Experienced Patent Translators Redundant? 10 things to do when starting up a translation service business Quality Assurance for the […]


  6. I think MT engines that are built with a specific task and domain in mind can produce unbelievable results, and are a real productivity increaser. The problem is customized MT has been so expensive so far and broad domain engines like Google and Bing aren’t customizable. You might want to have a look at where even an individual translator can build their own engine, with their own data.


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: