AdaptiveMT vs. NMT


I'd like to receive more information about the implementation of NMT for those using AdaptiveMT with a subscription.

Will NMT appear on the list of available engines?

Does AdaptiveMT needs to be manually disabled before adding NMT?

Is the existing data from SMT going to be used for the NMT?

A bit unrelated, will the PEMT certification guide be updated with more NMT information?



  • Hi Claudia,

    To answer your questions, please allow me to first take some time to clarify SDL's machine translation offerings. 

    SDL has been evolving our own in-house machine translation offering to our customers for over 15 years.  As the technology evolves and advances, SDL has made the MT capabilities available to various product suites (such as Trados Studio, Language Cloud, etc) while also having the MT capabilities as standalone products (BeGlobal and ETS).

    AdaptiveMT is a unique and innovative capability that SDL developed on our Statistical Machine Translation (SMT) technology and made available to Trados Studio, with subscription managed through Language Cloud.

    In the meanwhile, SDL continue to develop our core MT products (BeGlobal and ETS), now branded under "SDL Machine Translation Cloud and Edge", respectively.  Our neural machine translation (NMT) breakthroughs are introduced and delivered in this product line.  Future advancement and new capabilities around machine translation will always be made to this product line first, unlike what happened with AdaptiveMT where it was only integrated for Trados Studio.

    So now getting toward your questions --

    As machine translation capabilities sit outside of the Trados Studio tool, what MT you can use depends on the capabilities exposed by either 1) an embedded connection or 2) a plugin.

    1. Embedded connection as "SDL Language Cloud translation resources" - If you have been using AdaptiveMT, then you are using an embedded connection.  Please see the release note "Cumulative Updates for SDL Trados Studio 2019 SR2" for more information on the offer for Studio users in regards to access to using SDL MT.  And here is the help guide on how to use that embedded connection.  
      1. The embedded connection is a unique path to using SDL Machine Translation that we offer to our Trados Studio users. 
      2. From the users perspective, they interface with Studio.  Behind the scene, there is a connection from Studio to Language Cloud to SDL Machine Translation Cloud (BeGlobal). I understand that it is confusing with all the different product hops.  However, this is the connection path that will allow you to choose the use of an AdaptiveMT model or an NMT model.
    2. Plugin - This is a plugin that you would need to download from the SDL App Store and install.  This is what our customers with standalone subscriptions/licenses to SDL Machine Translation products will use (SDL MT Cloud plugin and SDL MT Edge plugin).
      1. Plugin allows the products (Studio and SDL MT) more independence from each other.  New capabilities introduced on SDL MT products could be more quickly updated to the plugin, which can be release on an as-needed-basis.

    To your specific questions:

    Will NMT appear on the list of available engines?

     It should if you do the update described in #1 Embedded connection above.

    Does AdaptiveMT needs to be manually disabled before adding NMT?

    Same as above, and the settings are exposed in the embedded connection "SDL Language Cloud translation resources".  Alternatively, if you have a subscription to SDL MT and can use the plugin, then the plugin will just be another resource and you shouldn't have to disable another resource.

    Is the existing data from SMT going to be used for the NMT?

    No, AdaptiveMT and the NMT provided through SDL MT product are independent.


    Quite a long response, but I do hope that help answer your questions.

  • Hi . I'm so sorry for the late reply, I put it on my to-do list and completely forgot about it. I really appreciate your detailed answer and now everything is super clear. 

Reply Children
No Data