Hello,
I am currently working on translating XLIFF files used in Microsoft Dynamics 365 Business Central. I am using the GPT-OSS 120B model for this task. The model produces reasonable results, but the translation accuracy is inconsistent, especially when dealing with Business Central terminology, field captions, system labels, and technical context.In our internal review, we identified that the issue is not with the model itself — but with the prompt structure. Therefore, I am looking to improve the prompt to make the translations more accurate while preserving the original XLIFF structure.
Key Requirements:
-
The XLIFF tags and placeholders must remain unchanged.
-
Only the text content should be translated.
-
Business Central terminology consistency should be maintained (e.g., “Customer Ledger Entry”, “General Journal”, “Purchase Order” etc.).
-
Style should remain professional and aligned with standard BC localization guidelines
What I need help with:
-
Best practice prompts for translating structured files (especially
.xlffiles for Business Central). -
Techniques to ensure GPT does not modify the tags, IDs, placeholders, or XML formatting.
-
Example prompt templates that maintain technical and product-specific terminology.
-
Any recommended strategies for handling domain-specific terms in Business Central translations
If anyone has experience with prompt engineering for Dynamics 365 Business Central localization workflows, or if you can share a sample prompt pattern that has worked for you, it would be extremely helpful.
Thank you .


Report
All responses (
Answers (