Forum Discussion
Solution for Scalable Multilingual Power BI Reports (ENG to FRE) - UI, Metadata & Data Translation
Hi KD_Infocepts
I don't think AI should be the primary source of translations in an enterprise solution. It's great for generating an initial translation, but I'd still recommend having a centrally governed translation repository that is reviewed by the business. As you mentioned, terms like Order can have completely different meanings depending on the context.
A pattern I've seen work well is to separate the problem into two parts. I have worked on this previously and I agree to the fact that scaling it may still be tedious:
- Metadata translations (measures, columns, hierarchies, display folders, etc.) using the semantic model's translation capabilities.
- Business data translations (Country, Segment, Product, Status, etc.) using a centralized translation table or master data repository that's shared across reports rather than maintaining report-specific mappings.
AI can then be used to suggest new translations whenever new business terms are introduced, but those suggestions should go through an approval workflow before being added to the central repository. That keeps the solution scalable while ensuring business terminology remains consistent across reports.
You've actually built the right primitives. The gap is that they're all maintained by hand. The enterprise pattern is the same pieces, but sourced from one governed place and pushed out automatically.
Metadata (measure names, column names, captions): don't do these with text boxes or mapping tables. Power BI semantic models support culture/metadata translations natively, and Tabular Editor is the standard tool to author and bulk-manage them. You export the caption set to a file, get it translated, reimport. They kick in when the report is consumed with the language set (the app language, the ?language= parameter, or the embedding locale), so test that path early.
Report UI (titles, buttons, tooltips): there's no native feature here, so a central Labels table (key + language + text) feeding SELECTEDVALUE measures is the accepted approach. That's what you're doing, just centralise it into one table instead of per report.
Data values (Country, Segment): push the translated columns up into your gold-layer dimensions rather than separate mapping tables with relationships. One dimension, an English and a French column, switched by the language context. Cleaner to maintain and faster at query time.
Where to master it: one governed table in your Lakehouse or a SQL database, not SharePoint. That's the single source. From there a script (Tabular Editor CLI / TOM) generates the model translations on deploy, and the gold layer picks up the dimension translations. Everything downstream is generated, not hand-edited.
On AI and context: people do use Azure AI Translator or Azure OpenAI for a first pass, but not unattended. The "Order = Sales Order vs Sort Order" problem is handled with a glossary/term base (Azure Custom Translator supports one), so approved business terms always translate the same way. The workflow that scales: detect new untranslated keys, AI first-pass through the glossary, route to a human for validation, write the approved value back to the master table. Automated draft, human gate. Fully manual doesn't scale, fully automated gets business vocabulary wrong.
Net: keep your model, but move translations to one governed table, generate the metadata translations with Tabular Editor, translate data values in the gold layer, and put an AI-draft-plus-review pipeline in front of new terms.
Hi ,
I tried to do the transaltions in the power bi file, please find the attached file. In the past, I used SSAS for langugae metadata transaltions, now Tabular Editor does support for language transaltions similar to SSAS. One Tabular/Semantic Model will support for that particular domain language transaltions. If you want to do it for other domain, you need a separate Tabular Model/Sematic Model.
1. Deploy the Power BI file into workspace
2. Test with French culture
Below are my suggestions to test this.
After adding translations, save the PBIX and publish it. Translations can be tested in Power BI Service by appending a culture parameter such as: ?language=fr-FR to the report URL.Page names
Power BI page-tab names are not handled as dynamically as visual titles. Many multilingual implementations use:
- Navigation buttons with translated labels
- Hidden page tabs
I am not sure about Automation process for this implementation. May be below enterprise pattern may help:
Native metadata translations
+ Localized label measures
+ Translated gold-layer dimensions
+ Field parameters
+ Central TranslationMaster repository
+ AI Usage (May be if possible)If this reply helped solve your problem, please consider clicking "Accept as Solution" so others can benefit too. And if you found it useful, a quick "Kudos" is always appreciated, thanks!
Best Regards,
Maruthi
Super UserYou can user-context-aware calculated column to automate the switching betwen different languages based on the USERPRINCIPALNAME or USERCULTURE (based on browser settings). Note: the translations must exist in the model but selecting which language to use for a column can be automated.
https://learn.microsoft.com/en-us/power-bi/transform-model/desktop-calculated-columns