Skip to content


Review of AI-translated content

Review of AI-Translated Content

A growing number of companies have started to send AI-generated translations to us for control. The source text and the AI target arrive together, and the request is simple: confirm that the translation is right before it is published, delivered to a client or filed with an authority.

Two things are usually lost along the way. The first is the translation environment. A text produced in a chat window or through an API has no translation memory, no termbase and no automated QA, so the checks a CAT tool applies as a matter of course were never run. The second is visibility. Nobody downstream knows which model produced the text, how it was prompted, or whether a glossary and reference material were supplied. A fluent result can hide an omitted warning, a changed figure, or a term rendered three different ways in the same document.

Alafranga reviews AI-generated, machine-translated and human translation output in Turkish and more than 40 other languages. We align the source and the AI target in memoQ or Trados, review the text against the source with your terminology and our QA checks, and return a corrected file, a QA report, and the translation memory and termbase the review produced.
 


Need the translated document typeset and print-ready?
We offer DTP as an integrated service alongside document translation and support InDesign, FrameMaker, Word, and structured documentation formats. The translated file is returned in the original layout, with text expansion handled, fonts verified for target language character sets, and formatting consistent with the source.

Contact us to include DTP in your project. read more..

 

▮How We Review AI Output
  • Information before we start
    We ask which model or engine produced the text, the prompt or instructions it was given, and any glossary, style guide or reference documents that were supplied. Where this information is not available, we review a sample section first and tell you whether the file can be corrected or whether retranslation is the cheaper route.

  • Alignment in a CAT tool
    The source and the AI target are aligned segment by segment in memoQ or Trados. This brings back what the AI workflow left out: your termbase, any existing translation memory, and the automated checks a translation project normally passes through.

  • Review against the source
    A specialist in the subject compares each segment with the source for meaning, completeness, terminology and register. AI errors are often fluent, so the review is read against the source, never against the target alone.

  • Automated QA
    Numbers, units, dates, tags, placeholders, omissions and terminology are checked across the whole file, which catches inconsistencies a reader would not see one page at a time.

  • Delivery
    You receive the corrected file in its original format, a QA report grouped by error type, and the aligned translation memory and termbase, so the next AI or human translation of the same content starts from approved terminology.

▮What We Review
  • Linguistic accuracy
    The translation is checked against the source text for meaning, completeness and fluency. Mistranslations, omissions, dropped qualifiers and unnatural phrasing are identified and corrected.

  • Terminology consistency
    Content is aligned with client-specific glossaries and previously approved terminology, so terms stay consistent across documents, teams and language versions.

  • Formatting and structural integrity
    Tags, numbers, dates, units and formatting are checked against the source. Structural issues introduced during translation or file conversion are corrected.

  • Cultural and tonal appropriateness
    Content is reviewed for cultural fit in the target market: tone, register, idiomatic expressions and any phrasing that may carry an unintended meaning.

  • Compliance-critical content
    For regulated content, a stricter protocol applies: the text is checked against the source line by line, with a full QA report on request.

▮Questions We Hear Before Every Project
 
Can you review a translation produced by another agency or an AI tool?
Yes. We review output from other agencies, freelancers, machine translation engines and AI tools. The source text is the reference, and the translation is checked against it for accuracy errors, terminology inconsistencies, formatting issues and compliance gaps.
 
What do you need from us besides the source and the AI output?
Ideally, the name of the model or engine, the prompt or instructions used, and any glossary, style guide or reference documents. None of this is mandatory, but it shows us where errors are likely to sit. Without it, we start with a sample review.
 
What does the QA report contain?
A record of every change, grouped by error type, with the source and target segments shown side by side. It shows where the problems were concentrated, which is usually more useful than the corrections themselves. It is available on request for any review level.
 
When is post-editing more expensive than translating from scratch?
When the AI or MT output is poor enough that the reviewer rewrites more than they correct. We assess the output before committing, and we will tell you if retranslation is the cheaper route.
 
Can you handle confidential documents?
Yes. Project data is held in our own ]Project Open[ installation, self-hosted in the EU since 2009, not on third-party services. NDAs are signed at company and individual specialist level.