Who uses Amazon's NMT or SDL's NMT? Looking for feedback.

My coworkers and I mainly use Microsoft and Textra because of their easy integration into Trados and decent quality Japanese to English translation. The best though was Google's NMT which we used until our free trial ran out. We are currently looking to evaluate different providers for a subscription including Amazon's new NMT and SDL's NMT.

Does anyone have any experience with the quality and integration of Amazon's NMT or SDL's NMT? How do the termbase/dictionary functions work?

Thank you.

Parents
  • Hello  

    In the absence of another response...

    In the Jan 2019 Intento report the Beglobal system received very good rating for the EN > JA  system – apparently we are in the Top 5%  of all the systems they have tested. Though this is not true for JA > EN.

    Amazon has done well in raw scores but do not really allow the dictionary or terminology control that we allow.  Also they make it VERY CLEAR that they will use your data to “improve the service”.

    Our latest terminology control is the most flexible and powerful capability we know of in NMT systems as each user could technically do this.  The Russian breakthrough blog post has a good write up on this:

    blog.sdl.com/.../

    Paul Filkin | RWS

    Design your own training!
    You've done the courses and still need to go a little further, or still not clear? 
    Tell us what you need in our Community Solutions Hub

  • Hi Paul,

    AWS Translate has recently introduced "Custom Terminology":

    AWS Translate interface showing 'Uebersetzung in Echtzeit' with options for source language and target language, and a section for 'Automatische Terminologie' and 'Terminologie Uebersetzungsregeln'.

    I have not yet tested it, but apart from SDL and Systran, AWS is one of the 3 NMT providers who are offering dictionary integration (this was mentioned in the TAUS MT User Group onJan 24).

    Kind regards

    Christine

    emoji


    Generated Image Alt-Text
    [edited by: Trados AI at 5:13 AM (GMT 0) on 5 Mar 2024]
  • Hi

    Yes, this is a very recent update and will no doubt evolve as they iron out the kinks SDL and Systran have worked on for some years. It's interesting watching the various players entering this area and will be even more interesting to see what happens over the next year or so. Security of data is likely to play a big part in how customers, LSPs, choose to make use of NMT going forward:

    aws.amazon.com/.../

    Paul Filkin | RWS

    Design your own training!
    You've done the courses and still need to go a little further, or still not clear? 
    Tell us what you need in our Community Solutions Hub

  • Hi and
    Tim Cook put it this way: "A few years ago, users of Internet services began to realize that when an online service is free, you're not the customer. You're the product." I am neither an Apple fan nor an Apple user, but there is not much you could add to Cook's statement.
    Daniel
  • Hi Paul,
    I fully agree with you - and you know that I am one of those who are "preaching" about data security and privacy issues assoicated with online MT ;-))
    As for terminology in NMT: I met Mark Post from Amazon (formerly John Hopkins University) in January 2018 and this was the first time when I heard about their "lexically constrained decoding" research. So it took them one year to go productive - which does not mean that they (or any of the other players) have fully solved the challenge of consistent terminology use in NMT.
    Kind regards
    Christine
Reply
  • Hi Paul,
    I fully agree with you - and you know that I am one of those who are "preaching" about data security and privacy issues assoicated with online MT ;-))
    As for terminology in NMT: I met Mark Post from Amazon (formerly John Hopkins University) in January 2018 and this was the first time when I heard about their "lexically constrained decoding" research. So it took them one year to go productive - which does not mean that they (or any of the other players) have fully solved the challenge of consistent terminology use in NMT.
    Kind regards
    Christine
Children
No Data