Translate.Wonk
Machine translation in your company voice: glossaries, existing translations, and CMS data as training input, reliable, secure, and continuously improving.
Translate in your company language
Maintaining glossaries, long approval chains, and hand-offs between email, Word, CMS, and external tools can stop here.
We use company knowledge and existing translations on websites, in documents, translation memory systems, and other sources to train models for your terminology and tone. The result is machine translation in real time at high quality.
What Translate covers
Translate combines the economics of machine translation with the quality bar of human review.
| Area | Benefit |
|---|---|
| Websites & CMS | Format-faithful .xml, .xliff, and other exports |
| Glossaries & stop words | Control terminology and proper nouns |
| Office | Word, Excel, PowerPoint with layout preserved |
| PDF & print | Finished PDFs in target languages |
| IDML & InDesign | Editable output in your usual workflow (see IDML-Translate) |
| Learning | Proofreading and API feedback improve models over time |
Privacy: APIs and data centres in Europe or Germany, industry-standard encryption. Customer data is not used to train unrelated models.
Better than the status quo
Many translation processes no longer match speed and quality needs:
- Time-to-market: weeks of waiting
- Quality: generic output misses brand and domain language
- Efficiency: manual steps and media breaks
- Cost: traditional per-word pricing is high
With models trained on your language, projects often see faster rollouts, lower cost, and better semantic fit than generic services. Exact numbers depend on data volume and language pairs.
How it works
Training machine translation on your company language:
- Collect language data (sites, glossaries, TMS, documents)
- Extract sentence pairs
- Process: validate, clean, prepare
- Train models
- Evaluate with your subject-matter teams, including blind tests
Independent quality review
After training, stakeholders get access to a dedicated evaluation environment. You compare output to past human translation or generic services without knowing the source system.
If scores are not yet positive, that is transparent, often with limited training data. Continuous collection and enrichment improves the model over time.
Phases from data to production
| Phase | Focus |
|---|---|
| 01 Data exchange | Language list and access to sources |
| 02 Data checkup | Quality and achievable pairs |
| 03 Testers | Internal reviewers |
| 04 Training | Extraction, cleaning, model training |
| 05 Evaluation | Review in your environment |
| 06 Go-live | Web UI and API in your landscape |
Typical duration: two to four weeks.
FAQ (short)
On-premise? Yes. Default is Microsoft Azure in Germany; private cloud and on-premise with GPUs are possible after analysis.
Languages? Dozens of languages in past projects; 100+ pairs technically feasible. We review your list before kick-off.
Systems? API integration for CMS, commerce, PIM, Office, and more, plus a web UI.
Training data? Sentence pairs from sites, TMS, glossaries, and many file formats.
Still need a TMS? Trained models combine translation memory-style knowledge with proofreading feedback. For many use cases that replaces a classic TMS workflow.
Running cost? Depends on models and annual word volume, typically well below classic translation spend.