AI Assistant not working

Hi,

I have just startet using AI assistant (OpenAI gpt-4o model) and it is not working as it should. F. eks. in the example below it should automatically sett "lege" in stead of "dr." because this is how this term is translated in the Termbase "lek.med = lege". AI assistant is enabled with Refine translation prompt. Should it not have insertet "lege" automatically? Or have i misunderstood something?

Please, look at the printscreens below.

Screenshot of a translation software interface showing a term 'lek.med Anna Klima' incorrectly translated as 'dr. Anna Klima'. Term recognition window highlights 'lek.med' with a correct term 'lege' suggested but not applied in the translation.

AI assistant is not showing that anything is wrong with the sentence, though the word dr. is an incorrect translation of lek.med and "Refine Translation" prompt is on.

Screenshot of an AI Assistant interface with a message 'No matches found or no lookup performed.' and an animated character holding a phone, indicating no translation suggestions are available.

lek.med exists in the activ Termbase, see: Term recognition in the first pic.

My settings are as follows: Is there anything wrong in the settings? or should I do something more than just to tick the right boxes?

Screenshot of settings dialog for AI translation software. Options for 'Ignore translated segments' and 'Enable Terminology-aware Translation Suggestions' are checked. Connection to 'gpt-4o' is established.

Can anyone help to do this the right way?

Or paste a link to the documentation that shows the right way to do it? Thank you in advance.!



Generated Image Alt-Text
[edited by: RWS Community AI at 3:44 AM (GMT 0) on 15 Nov 2024]
emoji
  • Hi  , yes, you will be able to uninstall the old version of the plug-in, as described here: https://gateway.sdl.com/apex/communityknowledge?articleName=000010617. Please let me know if you encounter any issues.

    emoji
  • Hi again,

    I am testing the updated plugin but it is still not working for the polish & norwegian languages, unfortunately. As you can se the term "poradnia urazowo ortopedyczna" has been properly recognized and is available in the Term Recognition window, the value is 89 because the term does not have a hyphen but it is such a small difference that OpenAI should be able to recognize it as the same or at least a very similar term. In stead it suggests its own translation. Example nr 2. "bez kolizji z korzeniami nerwów rdzeniowych" THIS one has been translated wrong so many times that I have saved it as a term, it is perfectly recognized with 80 % value but still not used by the OpenAI. I suspect this is because these terms have a value less than 100 and then they are ignored by the plugin and the Copilot. I suggest to update the plugin so that it takes into consideration also terms with less than 100% value or it will not be usable for languages with a more complex language system than f. eks English, German, French etc. Slavic languages like Polish have a number of casus which means that the same word, be it an object, verb or an adjectiv can be written in 7 different forms, and just one change in one letter results in unproper recognition of the word which get a lower value than 100. It would be literally impossible to save each termn in 7 different cases (forms) so that it is recognized properly, so the plug in needs to understand that even if the word is written in a different casus it is still the same word. 

    SO my question is> Does OpenAI make use of terms that have a value less than 100 ? In my opinion it does not, at least I have not seen one example of it using a term from the Termbase with the value less than 100. This must be fixed, or the plugin will be useless for us who translate from a language which has a more complicated structure than English. 

    I strongly recommend to do this. Pray 
    Screenshot of a translation software interface showing term recognition for 'poradnia urazowo ortopedyczna' with a 89% match in Norwegian, suggesting 'Skadeortopedisk poliklinikk' as a translation.

    Screenshot of a translation software interface highlighting a Polish medical term 'Na poziomie L2L3 obecna koncentryczna wypuklina krazka miedzykregowego na 4,5mm z modelowaniem worka oponowego, bez kolizji z korzeniami nerwow rdzeniowych' with a Norwegian translation suggestion.

    emoji


    Generated Image Alt-Text
    [edited by: RWS Community AI at 3:34 PM (GMT 0) on 3 Dec 2024]
  • Hi again,

    I am testing the updated plugin but it is still not working for the polish & norwegian languages, unfortunately. As you can se the term "poradnia urazowo ortopedyczna" has been properly recognized and is available in the Term Recognition window, the value is 89 because the term does not have a hyphen but it is such a small difference that OpenAI should be able to recognize it as the same or at least a very similar term. In stead it suggests its own translation. Example nr 2. "bez kolizji z korzeniami nerwów rdzeniowych" THIS one has been translated wrong so many times that I have saved it as a term, it is perfectly recognized with 80 % value but still not used by the OpenAI. I suspect this is because these terms have a value less than 100 and then they are ignored by the plugin and the Copilot. I suggest to update the plugin so that it takes into consideration also terms with less than 100% value or it will not be usable for languages with a more complex language system than f. eks English, German, French etc. Slavic languages like Polish have a number of casus which means that the same word, be it an object, verb or an adjectiv can be written in 7 different forms, and just one change in one letter results in unproper recognition of the word which get a lower value than 100. It would be literally impossible to save each termn in 7 different cases (forms) so that it is recognized properly, so the plug in needs to understand that even if the word is written in a different casus it is still the same word. 

    SO my question is> Does OpenAI make use of terms that have a value less than 100 ? In my opinion it does not, at least I have not seen one example of it using a term from the Termbase with the value less than 100. This must be fixed, or the plugin will be useless for us who translate from a language which has a more complicated structure than English. 
    Please address the issue stated below:
    "

    In Trados Studio 2024, the OpenAI plugin's Terminology-aware Translation Suggestions feature is designed to incorporate terms from your termbase into AI-generated translations. However, there are known issues where the plugin may not utilize termbase entries with a match value less than 100%, even when the "Enable Terminology-aware Translation Suggestions" option is activated.

    Possible Causes:

    1. Plugin Limitations: The current implementation may not effectively handle fuzzy matches (i.e., matches below 100%) from the termbase. This limitation can result in the AI-generated translations not incorporating these terms."

    I strongly recommend to do this.  It will make our work much easier and the limitations described above make this plugin in many ways unusable. 

    Please, send me a message or reply to this post when the new plugin addressing this issue is available!. Thank you. Relaxed  ♥ 
    Screenshot of a translation software interface showing term recognition for 'poradnia urazowo ortopedyczna' with a 89% match in Norwegian, suggesting 'Skadeortopedisk poliklinikk' as a translation.

    Screenshot of a translation software interface highlighting a Polish medical term 'Na poziomie L2L3 obecna koncentryczna wypuklina krazka miedzykregowego na 4,5mm z modelowaniem worka oponowego, bez kolizji z korzeniami nerwow rdzeniowych' with a Norwegian translation suggestion.

    emoji


    Generated Image Alt-Text
    [edited by: RWS Community AI at 3:34 PM (GMT 0) on 3 Dec 2024]
  • We will be pleased to help you with this problem so have logged a support case on your behalf. We do prioritise customers who have a support contract so please keep in mind it could take up to 48 hours, especially over a weekend, but rest assured we will contact you as soon as we can. You will be contacted at the email address used for this Community. Support Case Id : 00879686 .

    emoji
  • Thank you for your very fast response. Please, look at the screenshot below. Here the term has 100 value but it was not taken into consideration, even not by the AI Assistant. 

    In stead of conus medullaris which is a proper term in latin it used something that does not even exist in the medical language Speak no evil 

    Anyway thank you so much for looking into this. I am sure with time we will work out the perfect plugin Relaxed 
    Screenshot of a translation software interface with highlighted text showing a mismatch in term recognition. The term 'conus medullaris' is incorrectly replaced by an unrecognized term, despite having a 100 value in the term base.

    emoji


    Generated Image Alt-Text
    [edited by: RWS Community AI at 3:55 PM (GMT 0) on 3 Dec 2024]