How best to handle differing client terminology when working with neural machine translation?

Good morning,

In the case of clients who require the translation of the same term in different ways, is there any best practice when using NMT?

For example, on one project, I have the French term "chiffre d'affaires" that should be translated as "turnover". In another project, it should be translated as "sales". In another project, it should be translated as "revenue".

These different translations are to comply with client preferences and to be consistent with existing translations.

1) When post-editing NMT, does the fact that I change the machine-translated term get fed back to the NMT engine to affect future translations?

2) If so, can I prevent this by unchecking "update" in the machine-translation memory settings?

3) If I leave "update" checked, does this mean that the post-edited term will be applied until I change it again by post-editing the term again?

Is there any best practice for this situation?

Thank you.

Neil

Parents Reply
  • In the meantime, when using NMT, is it better to un-check the "update" checkbox in the memory dialogue as I cannot see what good any project-specific feedback would do for the NMT engine?

    Good question.  In fact that update won't be used at all when you use NMT but because we have not properly amended the UI for a solution that's mix and match the update has been left enabled.  You could uncheck it for your sanity... but in practice it's simply not used.  You'll note that if you do use the plugin this box is not enabled at all.

    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

Children