Regulatory Open Forum

 View Only
  • 1.  Machine translations

    Posted 05-Nov-2020 15:42

    I have a client who is working on translations. (In my MDR classes I tell people to buy stock in a good translation company.)

    We are looking at machine translation as an option for small jobs such as a sentence or two in an IFU. Today, the problem came up with the free-form fields in the Eudamed UDI database. MDCG 2018-7 says that in general any free-text field should be in English as well as in the languages of those countries where the device is available. A data field will be available for each relevant language.

    Translation companies may use machine translation following ISO 17100:2015 Translation services – Requirements for translation services.

    Another area with translation is the SSCP. Simply, the process in MDCG 2019-9 has the manufacturer submit the draft SSCP in English. The NB reviews it and, when acceptable, posts it to Eudamed. The manufacturer has it translated and "should ensure, through their quality management system, that the translations are correct". This opens the possibility of the manufacturer being the translator through machine translation.

    Two (at least) on-line services have relevance – Google translate and Microsoft. There is a Microsoft web page that discusses breakthroughs in machine translation using AI. Also, the article says that Microsoft uses its own translation software, available in MS Office, for all of its manuals and technical documentation.

    I want to explore the possibility of a small manufacturer using machine translation as part of the QMS. The QMS would have the associated processes, work instructions, methods, etc. An example process would write a short statement in English, such as for UDI, translate it into the target language using MS Word, then translate the output back to English and see if it is reasonable.

    Of course, there are problems. Many years ago, as I recall, a machine translation of "The flesh is willing but spirit is weak" came back as "The meat is good, but the wine in awful".

    One of my favorites for ambiguity is "Time flies like an arrow" which obviously means there is a kind of fly, perhaps like a house fly, that prefer to eat arrows – an ever present danger at archery ranges.

    More seriously, I would like to see if anybody else is working in this area. Also, I would appreciate any thoughts about feasibility.






    Dan O'Leary CQA, CQE
    Swanzey NH
    United States

  • 2.  RE: Machine translations

    Posted 05-Nov-2020 20:55
    Risk assessment. Risk assessment. Risk assessment.

    Machine translation is nowhere near as awful as it used to be. For relatively simple texts in similar and simple languages, it is sometimes very good. A weak tool on a complex text in a difficult language can still produce spectacularly bad results.

    What's the worst that could happen? If it's only the national competent authorities reading the possibly bad translation, maybe they'll get a good laugh, or maybe they'll ask for clarification. If a patient's life depends on someone being able to understand the information, you had better use appropriate methods to ensure they can use it as expected.

    Anne LeBlanc
    Manager, Regulatory Affairs
    United States

  • 3.  RE: Machine translations

    Posted 06-Nov-2020 08:56
    Hi Dan,
    Hi Anne,

    quick thoughs:
    (1) About regulatory / QMS records: I may use machine translations. If used, I would add a statement to the translation like: Maschine translation by >tool< from the  the original version in  >language<. The original version shall prevail in case of translation issues.
    (2) About IFU's: I would still validate any translation by a competent and experienced (in language and the product application) human.

    Uwe Zeller | Regulatory Affairs / Risk Management Consultant
    Biberach an der Riß, Germany

  • 4.  RE: Machine translations

    Posted 06-Nov-2020 09:18



    Hi Dan,


    I enjoyed your post quite a lot.  The version of the "time flying" bit as I've heard it (I expect most have) is, "Time flies like an arrow, but fruit flies like a banana."


    Many years ago I took an intensive German class and we were at one point paired up with a group of students visiting from Austria.  My sense of humor often includes wordplay, but they were not really amused by my literal translation of a common English idiom: "Heute est regnet katzen und hundes."  :-/


    Now, more serious.  I agree with Anne's observation that translation software has become markedly better than it used to be.  In fact, I have used Google Translate for some regulatory work (e.g., filling out private import forms for Japan), *but* am in no way going to depend on it entirely--I always ask someone fluent in the other language to review and revise the output as needed.  Ideally, this would be someone who is a native speaker, because some of the language nuances are subtle.


    My understanding is that at least some of the translation services use software to make an initial translation that's subsequently reviewed and revised by professional translators.  This can allow for some reduction in time and labor than translating from scratch (though those who've had to review and revise really poor writing know from hard experience that it's sometimes more work than a complete write-up would have been).


    Not even close to a complete answer, but perhaps of some interest.


    Best regards,





    Theodore (Ted) Heise, PHD, RAC

    Vice President Regulatory and Clinical Services


    MED Institute Inc.

    1330 Win Hentschel Blvd.

    West Lafayette, IN  47906-4149 USA

    765.463.1633 ext. 4444






  • 5.  RE: Machine translations

    Posted 07-Nov-2020 07:03
    Thanks to all of you.

    I have a path forward: use the technical for non-safety areas such as the UDI database or some parts of the IFU. Also, to keep good records of each translation, the software, etc.

    I'm put it to the QMS with procedures, etc.

    Dan O'Leary CQA, CQE
    Swanzey NH
    United States

  • 6.  RE: Machine translations

    Posted 08-Nov-2020 19:19
    Edited by Soichiro Iida 09-Nov-2020 03:19

    Dear Dan,

    Your strategy seems to be valid and appropriate. 

    Through my experience working in a translation company, machine translation (MT) has characteristics below:

    1. The quality of MT result varies a lot between the pair of language or depending on the document type.  
    E.g. English-Chinese MT works very well. Japanese has few good partner but Japanese-Korean MT works well.
    English-Japanese MT improved very much but still some translation results in a reverse meaning. 
    I am not sure because of the lack of experience but the MT between the most of European language should work very well. 

    2. Some words or sentences may disappears.
    It is a nature of AI translation. (I don't know the latest situation, but it was in a few years before.)
    At the time when the MT improved greatly, we started to see some sentences are missing after the translation. 
    There are some web article on the reason of this phenomenon. 
    It seems there is no simple rule where to disappear and we need to cautious about it. 

    By the way, there might be a case which open MT like Google translation shall be avoided in consideration of confidentiality, though I don' know about the detail. 

    Soichiro Iida

  • 7.  RE: Machine translations

    Posted 09-Nov-2020 05:01
    Edited by Stephanie Grassmann 09-Nov-2020 05:04

    Dear Dan

    During the 15 years writing technical files and clinical evaluations for medical device, I have been requested - now and again - to check the language, the German to English translation, in terms of regulatory compliance of informational materials and regulatory files for international market submissions, i.e. EU, USA FDA.  


    One firm, in which I was the RA manager, had used a translation firm having ISO 13485 certification. ( I do not remember if they also had ISO 17100:2015.) Unfortunately, translations of the risk management and the engineering technical file had to be redone since they did not reflect the meaning of the German text. The words had been correctly translated but it was obvious that the person did not have a technical and regulatory background. The company understood that submitting these files to the notified body would have resulted in deficiencies.


    Machine translations/ translators, I have only been using to understand foreign publications or web sites.


    Based on my experience thus far, I would recommend to get an individual who is a subject matter expert and who is fluent (best would be mother tongue) for any documents which will be submitted or reviewed by regulatory authorities. The costs of a delay in market approval (possible in EU to be several months) may far outweigh the cost of the translation.

    Best Regards,

    Stephanie Grassmann
    Managing Director of MedTechXperts Ltd

  • 8.  RE: Machine translations

    Posted 09-Nov-2020 23:59
    Hi Dan,

    Professional translators these days uniformly start with a machine translation and then fine tune by a native speaker.  

    We use machine translation routinely for review of foreign language guidance documents.  As others have said, the quality varies a bit with the translator and language.  For example we find that the inbuilt translation in Microsoft office and via Bing seems to deal with Mandarin much better than does Google translate.  Sometimes it's worth using both and cross checking.

    Using this approach you can, say,  take a Mandarin language guidance from NMPA and produce a meaningful English language copy that's certainly good enough to understand the overall intent and most of the detail of a guidance, and this typically takes at most an hour or two for a guidance of say 10-15 pages.

    But in the end - always have a native speaker check the result.  The machine first step makes it a whole lot less labour intensive than doing a complete human translate.


    Arthur Brandwood PhD FRAPS
    Director and Principal Consultant
    Brandwood CKC
    Sydney, Australia