Services / MT Quality Review

Machine translation review

Translating with DeepL or AI? I review samples, score quality measurably and show where to improve.

Typical errors

Machine translation reads fluently today. Errors therefore often surface only when a customer or technician trips over them.

  • Wrong terminology

    A component has one name in chapter 2, another in chapter 7 and a third in your software. The translation sounds right but is not consistent.

  • Shifted meaning

    Negations, units, limits and conditions get lost or flipped. In safety messages, that can have serious consequences.

  • Wrong tone

    Informal instead of formal address, German German instead of Austrian, a brand voice that isn't yours. It costs trust even when every sentence is grammatically correct.

Process

  1. Sample

    Each month, you give me access to a selection of source texts and machine translations, as an export or directly in your system.

  2. Scoring

    I check every segment against the source and classify errors by type and severity, based on the MQM error typology.

  3. Report

    You receive a quality score, the distribution of error types, every critical error with a correction and the trend versus the previous month.

  4. Improvement

    Plus concrete recommendations: glossary entries, style rules or prompt changes, and which content is safer with human translation.

Scope of services

It starts with a one-off audit. If you want to track quality over time, the review continues monthly.

ItemServiceScopeBilling
1

Audit

A thorough assessment of your machine translation output.

  • Sample from up to three content types
  • Report with quality score and all critical errors
  • Recommendation on which content can stay machine-translated
  • Wrap-up video call
Fixed price
2

Quality monitor

Ongoing review for companies that machine-translate regularly.

  • Monthly sample of an agreed size
  • Monthly report with trend
  • Upkeep of your glossary and style rules
  • Cancel monthly after the first three months
Monthly fee
3

Quality monitor Plus

For several products, markets or systems.

  • Larger sample and several content types
  • Comparison of different systems or prompts
  • Correction of critical segments included
  • Quarterly call with your team
Monthly fee

Post-editing

If the audit shows that revising is worthwhile, I post-edit machine translations based on ISO 18587. For safety-relevant content, I will tell you openly when a new translation is faster and safer.

  • Full post-editing

    The result matches a human technical translation: accurate, with consistent terminology and natural to read.

  • Light post-editing

    Understandable and accurate, without stylistic polish. For internal or less critical content in large volumes.

Background

I have translated and revised technical content, software and marketing into German from English and Dutch since 2015. As an electrical engineer, I spot technical errors that a purely linguistic check misses.

I also work as a product owner in data and analytics. Making quality measurable and turning it into decisions is what I do every day.

Languages
English → German, Dutch → German
Locales
German for Germany, Austria and Switzerland
Content
Technical documentation, software and UI, support, e-commerce, marketing
Systems
DeepL, Google, language models, output from Phrase, Smartling, Lokalise and others
Confidentiality
NDA on request, nothing goes into public online services

Frequently asked questions

Do you also correct the translations?

The report includes every critical error with a correction. Full post-editing of larger volumes is available if the review shows it is needed.

What is MQM?

MQM stands for Multidimensional Quality Metrics, an open framework for assessing translation quality that originated in an EU research project. Each error is given a category, such as accuracy, terminology, language or style, and a severity: minor, major or critical. The number and weight of errors produce a quality score. This lets you compare months, content types or translation systems instead of just “sounds fine” or “sounds wrong”.

How large is a sample?

It depends on how much you translate and how varied your content is. We set the size together in the audit, so the results are meaningful and the effort stays predictable.

Do you review other languages?

I review translations into German. For other target languages, I can recommend colleagues who work to the same scheme.

We translate with a language model and our own prompts. Does that work?

Yes. That is where a review pays off most, because output can shift with every prompt or model change. On request, I compare two variants side by side.

Request an audit

Start with a fixed-price audit. Afterwards you know where you stand and whether ongoing review is worth it for you.

Or email me directly at ch_translations@outlook.com.