Governed AI · Terminology · Human review

AI translation services with professional human review

We combine AI, brand context, terminology and professional oversight to translate faster while retaining editorial control, traceability and the quality appropriate to each use case.

Specialists in the destination marketWorkflow matched to risk and volumeQA linguistic and editorial
MULTILINGUAL CONTENTSOURCE → MARKET
01 · DiagnoseGoal, audience, channel and marketbrief · context · terminology
02 · AdaptAI translationlanguage · intent · brand voice
03 · ValidateReviewed and ready to publishQA · consistency · delivery
LANGUAGE EXPERTISEVALIDATED CONTENT ✓
01 / AI translation

Translation quality depends on the workflow, not only the model

An AI system can produce fluent copy while inventing attributes, changing terminology or diluting brand voice. We configure authorised sources, prompts, glossaries and translation memories, then apply automated checks and human review according to content risk and market exposure.

We configure prompts, terminology, reference material and validated source content before generation, then apply automated checks and human review at the level the content requires.

Tell us about your project →
01

Authorised context

Validated sources, non-inference rules, terminology and data boundaries.

02

Calibrated prompts

Instructions by brand, channel, language, format and communication goal.

03

Automated QA

Checks for numbers, omissions, terminology, consistency, tags and variables.

04

Professional review

Market specialists validate meaning, naturalness and suitability for use.

02 / How we work

Pilot, measure, correct—then scale

The workflow defines what AI may transform, what it must preserve and which claims require validation, with traceability from source to final multilingual version.

  1. 01

    Content and risk classification

    Classify content by use, visibility, regulation and commercial risk.

    Risk classified
  2. 02

    Approved sources and terminology

    Prepare authorised sources, glossaries, memories, prompts and QA rules.

    Sources approved
  3. 03

    Controlled pilot and evaluation

    Run a representative pilot and have it reviewed by destination-market specialists.

    Pilot evaluated
  4. 04

    Production with risk-based review

    Measure recurring errors and adjust the system before increasing volume.

    Review completed
  5. 05

    Delivery, traceability and learning

    Scale with the agreed level of review and retain traceability of decisions.

    Feedback retained
03 / Scope

Controls that turn AI output into accountable translation

01

Sources, risk and terminology

The service can include sample analysis, content and risk classification, source preparation, terminology and prompt configuration, pilot translation, automated QA, agreed human review and traceable delivery.

02

Controlled translation and review

We can work with TMX memories, TBX glossaries, XLIFF files and CMS or PIM workflows.

04 / Evidence

AI translation · Ontranslation

Branni Pets: recovering market fit from machine-translated copy

For Branni Pets, the starting point was machine-translated ecommerce content for Germany. Cultural analysis, international SEO and native-specialist oversight restored brand coherence and commercial usefulness. The case illustrates the principle behind our workflow: fluency is not the acceptance criterion; meaning, terminology, market fit and intended use are.

Discuss the project →
05 / Choosing the right scope

When to choose governed AI translation

Choose this workflow when multilingual content is recurring, scalable and starts from approved source material. Use post-editing when machine-generated translations already exist. Keep human-led translation or full specialist review for legal, medical, regulated, highly creative or high-exposure content where the cost of a subtle error is unacceptable.

06 / Delivery criteria

Controls that make AI translation usable

Before production, we agree where generated content may be published, which sources are authoritative and who can approve exceptions. The pilot is assessed against explicit error categories rather than general impressions of fluency. We also separate model configuration from editorial policy: changing a tool must not erase terminology, risk decisions or the evidence behind approved output. This makes the workflow portable, measurable and easier for internal teams to govern.

01

Source boundaries

The system receives authorised reference material and explicit non-inference rules. Product attributes, figures, legal wording and named entities are checked against the source rather than accepted because the generated sentence sounds fluent.

02

Terminology governance

Glossaries and translation memories record approved choices, prohibited variants and market-specific language. These resources are maintained as working knowledge, not uploaded once and forgotten when the content or product changes.

03

Risk-based review

Public, regulated and brand-sensitive content receives stronger human oversight than low-risk internal material. The review model is agreed before production so speed never silently replaces the required level of accountability.

04

Traceability and learning

Recurring corrections are classified and fed back into prompts, rules and QA. This creates an auditable improvement loop and helps the client understand which problems originate in the source, the model or the publishing workflow.

07 / Frequently asked questions

Questions teams ask before using AI for translation

Is this the same as machine translation?

No. The service adds authorised context, terminology, governance, QA and professional oversight.

Is every translation reviewed by a person?

We agree the review level according to risk, visibility, regulation and commercial impact.

Can you use our glossaries and memories?

Yes. They are essential for consistency and for reusing validated knowledge.

How do you reduce hallucinations?

We restrict sources, define non-inference rules and validate attributes, numbers and claims with automated and human controls.

Which content is suitable?

Catalogues, web content, support and recurring documentation can be suitable. Sensitive or regulated content requires stronger controls.

How do you evaluate quality before scaling?

We review a representative pilot, classify recurring errors and, where useful, combine specialist assessment with automated checks or metrics such as COMET.

How do you protect confidential content?

Tools, access, data processing and retention are agreed before production. We do not assume that every public AI system is acceptable for client material.

Can you improve AI translations we already have?

Yes. We first assess whether post-editing is efficient. If correcting the output would cost more than translating again, we say so before scaling.

Next step

Choose the right human–AI workflow

Tell us about your project

What do you want your content to achieve?

Tell us about the scope, languages and timeline. We’ll respond with a clear proposal.

    Or send us your docs for a quote

    I've read and accept the Disclaimer and Privacy policy

    I consent to receive the information Ontranslation & Interpretation, SLU considers appropriate to send me by email (you can unsubscribe at any time).

    Privacy Overview
    Ontranslation

    This website uses cookies so that we can provide you with the best user experience possible. Cookie information is stored in your browser and performs functions such as recognising you when you return to our website and helping our team to understand which sections of the website you find most interesting and useful.

    Strictly Necessary Cookies

    Strictly Necessary Cookie should be enabled at all times so that we can save your preferences for cookie settings.

    Analytics

    This website uses Google Analytics to collect anonymous information such as the number of visitors to the site, and the most popular pages.

    Keeping this cookie enabled helps us to improve our website.