• State Suggested Answer
  • Replies 1 reply
  • Answers 1 answer
  • Subscribers 182 subscribers
  • Views 628 views
  • Users 0 members are here
Options
Related

OpenAI Provider for Trados adding Simplified Chinese characters to Traditional Chinese translation

Eddie Shen

Hello

I'm currently testing the OpenAI provider for Trados in Trados 2024.and have multiple questions regarding the output. In all cases below we're pretranslating an entire file, and not checking individual AI suggestion for each segment.

  1. We've connected an sdltb to the project as well as enabled "terminology-aware translation suggestions". However, we've noticed that the output doesn't always adhere to the termbase. Is the OpenAI Provider for Trados currently able to leverage termbases when processing translations?
  2. Related to Question 1: When processing Traditional Chinese for Taiwan, we've noticed that the OpenAI provider is sometimes using Simplified Chinese characters in the translations, and this happens most frequently for termbase terms. We're using a multilingual termbase, so I'm wondering if the OpenAI Provider gets locales mixed up when checking TB terms? And if the provider is currently not actually able to leverage TB, then what could be causing Simplified characters to show up in the translation.
  3. When processing APAC languages, traditionally linguist would add spaces between the Asian characters and Western characters or numbers when translating. This is handled consistently in MT solutions like DeepL. But for the OpenAI provider we've found that its handling of spacing has been inconsistent, with it adding spaces in some cases and not in others. Has anyone run into a similar issue and is there a fix?

Thank you for your time and I'd appreciate any suggestions or input.

BR

Eddie
emoji
Rate translation
×
Suggest better translation
×
Moderator UI
×
  • Get AI Suggestion
  • 0

    Hi  , thank you for reporting this. 

    To address these issues, It would help us a great deail if you could provide some examples that reproduce the problems, including the Trados project or specific resources you're using. Please indicate the AI model, as well as the prompts and settings you’ve configured.  my e-mail: phartnett@rws.com

    1. Leveraging Termbases:
    The integration between the OpenAI Provider and termbases in Trados is designed to enhance translation quality by providing terminology-aware suggestions. However, it can vary depending on the AI model used. Some models are inherently more reliable in utilizing termbase data than others. If you notice that the output doesn't fully adhere to the termbase, it may be worth experimenting with different models or adjusting settings to improve adherence.

    2. Traditional vs. Simplified Chinese Characters:
    I would need sample data to help investigate further here...

    3. Inconsistent Spacing for APAC Languages:
    The inconsistency in handling spacing between Asian characters and Western characters or numbers is a known challenge when using AI translation models. Unlike MT solutions such as DeepL, which have specific training for these spaces, AI models might not consistently address this without specific tuning.  Again, it would useful to have sample data to help us review, address this if possible.

    If you or anyone else has encountered similar issues and found fixes or workarounds, sharing those would be incredibly beneficial.
    emoji