Technology · our own TilBridge TMS
AI in translation — under human control
TilBridge is our own translation management system. AI speeds up the work where it's reliable — suggesting from translation memory, drafting, aligning terminology. But no output reaches the client unchecked — automated quality control on every delivery and a required final release check by a responsible specialist. The depth of human involvement is agreed to the task and its risk.
Not a black box — a process you can steer
The three principles behind how AI works in TilBridge.
Predictable, not magic
Code drives the workflow, not the model. AI is used only for bounded sub-tasks and never decides what happens next — the sequence of steps stays with us.
The human decides
AI suggestions land in translation memory as a draft and are never reused until a linguist approves them. Nothing reaches the client without a check.
Everything on the record
Every AI and machine-translation call is recorded — we can always explain exactly how any segment was produced.
A translation is assembled step by step, not "one button"
The whole pipeline runs inside one system — TilBridge TMS. The automated steps, the AI and the linguists' work all happen in one environment: what used to be a set of scattered, semi-manual scripts is now a single integrated process.
- 01
Brief & file prep
We take in your materials, requirements and reference texts and prepare the files for translation.
- 02
Term extraction
Key terms and recurring concepts are pulled from the text automatically.
- 03
In-context term translation
Terms are translated in the subject matter and context of the project.
- 04
Terminology sign-off
A native-speaker linguist reviews and approves the glossary before translation starts.
- 05
Translation memory
The system applies exact and fuzzy matches from your translation memory.
- 06
Machine translation
For new segments, neural machine translation provides a first draft.
- 07
AI post-editing
AI merges memory matches, the MT draft and the approved glossary into a coherent version.
- 08
Native-speaker editing
A native speaker refines meaning, tone and style; the depth depends on the service level.
- 09
Quality checks
Automated checks for numbers, tags, terms, typography and consistency.
- 10
File assembly
The translation is reassembled into the original format with markup preserved.
- 11
Final check & delivery
A named specialist approves release and delivers on the agreed deadline.
- 12
Memory update
Approved segments return to the translation memory for future projects.
Three service levels — matched to the task and its risk
The depth of human involvement is agreed before work starts. No output reaches the client unchecked.
AI-assisted translation with controlled QA
MT- and AI-based translation with automated quality control on every delivery, human linguistic sampling, and a required final release check by a responsible specialist. For high volumes and fast turnarounds.
Machine-translation post-editing (MTPE)
A machine draft with full post-editing by a native-speaker linguist — working to ISO 18587. A balance of speed and accuracy.
Human translation with independent review
Human translation with independent review by a second linguist — working to ISO 17100. For high-stakes and public-facing material.
Where machine translation falls short
Kazakh, Uzbek, Kyrgyz and other languages of the region are considered low-resource in natural-language processing: little training data, few reliable machine-translation models. But it isn't only about data — each language carries its own register, cultural context and audience expectations that an algorithm can't capture.
So for us AI doesn't replace the translator — it amplifies them: it takes on the routine (matches, drafts, terminology consistency), while meaning, tone and appropriateness stay with a native speaker.
Native-speaker linguists, not just a model, are accountable for the final meaning.
Glossaries and translation memory keep terms consistent across the whole project.
Mandatory revision by a second linguist and controlled machine-translation post-editing.
Your content stays confidential, with access limited to the project team.
What TilBridge does
A full translation environment — from memory to delivery in the original format.
Translation memory & glossaries
Exact and fuzzy matches, terminology management, reuse of everything you've built.
MT post-editing (MTPE)
A fast machine draft plus native-speaker editing to ISO 18587 — speed without losing quality.
Quality assurance
Automatic checks for numbers, tags, punctuation, typography, terms and consistency.
Document and localization formats
Word, XLIFF, SDL packages and other common formats — with markup preserved and re-merged back to the source format.
Continuous localization
We fit into your release process and work with regularly updated content.
Flexible choice of engines
We match machine translation to the language pair and subject matter — DeepL and neural models.
See how it fits your process
Tell us about your task and languages — we'll set up the process and match a team to your workflow.