Fora Translations

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.

How it works

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.

  1. 01

    Brief & file prep

    We take in your materials, requirements and reference texts and prepare the files for translation.

  2. 02

    Term extraction

    Key terms and recurring concepts are pulled from the text automatically.

  3. 03

    In-context term translation

    Terms are translated in the subject matter and context of the project.

  4. 04

    Terminology sign-off

    A native-speaker linguist reviews and approves the glossary before translation starts.

  5. 05

    Translation memory

    The system applies exact and fuzzy matches from your translation memory.

  6. 06

    Machine translation

    For new segments, neural machine translation provides a first draft.

  7. 07

    AI post-editing

    AI merges memory matches, the MT draft and the approved glossary into a coherent version.

  8. 08

    Native-speaker editing

    A native speaker refines meaning, tone and style; the depth depends on the service level.

  9. 09

    Quality checks

    Automated checks for numbers, tags, terms, typography and consistency.

  10. 10

    File assembly

    The translation is reassembled into the original format with markup preserved.

  11. 11

    Final check & delivery

    A named specialist approves release and delivers on the agreed deadline.

  12. 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 speakers on our team

Native-speaker linguists, not just a model, are accountable for the final meaning.

Consistent terminology

Glossaries and translation memory keep terms consistent across the whole project.

ISO 17100 and ISO 18587

Mandatory revision by a second linguist and controlled machine-translation post-editing.

Your data stays yours

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.