NMT is completely changing numbers from source to target!

Is anyone else having this issue? the NMT suggestions often completely change numbers, especially when percentages are involved.

Screenshot of a text comparison in Trados Studio showing a mistranslation where numbers are incorrectly changed in the NMT suggestion. Original text mentions '83.25%' and '12.93%', while the suggestion shows '83 3.83%' and '12%'.

Is anyone else having this problem? It's not just a one off, I am constantly having to change numbers back. Is there anything I can do about this?

Thanks

Anna



Generated Image Alt-Text
[edited by: RWS Community AI at 11:23 AM (GMT 0) on 26 Mar 2025]
emoji
Parents
  • "Probability score": 0.7

    "Question improvement suggestion": Please provide more specific examples of the issue you're experiencing, including the source and target languages, and the exact version of Trados Studio you're using. Also, it would be helpful if you could share any error messages you've received, if any.

    In the meantime, based on the information you have provided, here is a suggested answer to your question:

    Hello Anna,

    It seems like you're experiencing issues with the Neural Machine Translation (NMT) feature in Trados Studio, particularly with numbers and percentages. This could be due to a variety of factors, including the specific language pair you're working with, the complexity of the text, or even a potential bug in the software.

    One potential solution could be to use the TermInjector application from the RWS AppStore. This tool can help you handle numbers and currencies that are not automatically recognized or localized by Trados Studio. You can set up specific rules in TermInjector using regular expressions (regex) to handle different formats of numbers and currencies.

    Here's a basic step-by-step guide on how to use TermInjector:

    Step 1: Download and install the TermInjector application from the RWS AppStore.

    Step 2: Open your project in Trados Studio.

    Step 3: Set up a rule in TermInjector using regex to handle the specific number or currency format you're dealing with. For example, you could use a rule like this to handle different currency symbols followed by numbers: ((¥|£|€|$))([0-9,. ]+)

    Step 4: Run a Translation Memory lookup for each segment after only translating one segment. The TermInjector should automatically transform the numbers according to your rule and insert them into the Translation Memory match.

    Please note that this is a general guide and you might need to adjust the steps based on your specific situation. Also, remember to always back up your work before trying new tools or settings.

    I hope this answer addressed your question. If not, please reject the answer, add the question improvement suggestion information, and click the RWS Community AI checkbox on the right to try again.

    RWS Community AI

    emoji
  • The language pair is Spanish to English, and it's not a problem with most numbers, only percentages within statistical information where it seems to be swapping numbers around.

    emoji
  •   

    It’s probably due to the wrong decimal separator used in Spanish (Spain). If it’s Spanish Mexican it’s right, but the MT engine might not understand it. Which Spanish version are you using? I did a quick text in the free DeepL version availavle at www.DeepL.com and use the wrong decimal separator, and the result was OK. Which MT engine are you using?

    I’d pre-process the source file and replace a couple of dot decimal separators (.) with commas (,), and then re-open the prepared document in Trados Studio to check if this time everything was OK. If it’s you know the reason and then you may want to replace all decimal separators before the percentage symbol.

    emoji
Reply Children