Fluent machine translation can be convincing—and still be wrong. What does that mean for quality, pricing, and the specialists responsible for the final result?
There is a familiar assumption in the translation industry: if a machine has already produced the text, a linguist only needs to check it. Surely that must be quicker, easier, and therefore cheaper than translating from scratch?
Sometimes it is quicker. But machine translation post-editing (MTPE) is not just proofreading. A professional post-editor must compare the output with the source, identify what has been misunderstood, and decide whether to correct a phrase, rewrite a sentence, or discard the machine translation altogether.
The difficulty is that the output can sound completely natural even when its meaning is wrong.
Fluent doesn't always mean accurate
A machine-translated sentence can read beautifully while it:
- leaves out an important detail or adds information that was never in the source;
- chooses the wrong meaning of an ambiguous word;
- changes the level of obligation or certainty;
- misinterprets the relationship between sentences;
- uses plausible but incorrect technical terminology;
- mishandles warnings, instructions, cultural references, or tone.
A spelling error stands out. A fluent mistranslation often doesn't.
This is why reliable post-editing requires more than excellent target-language writing skills. It requires an understanding of the source, the subject area, the intended audience, the client's terminology and style, and the ways in which machine translation can fail.
The key question is not whether a sentence sounds good. It is: does it actually say what the source says?
Why the economics are more complicated than they look
Machine translation can improve productivity significantly when the source content, language pair and MT system are suitable for the task. A lower post-editing price may therefore be reasonable in some projects.
But there is no fixed relationship between how few words a linguist changes and how little work the linguist performs.
Even an unchanged sentence must be read, understood and checked against its source. A segment with two visible edits may have needed careful research and evaluation. Another segment might be so misleading that starting again is faster than repairing it.
The original article discussed research involving 90 million words and 879 linguists, which found substantial variation in post-editing productivity and cautioned against using edit distance as a simple measure of effort. Those figures are reproduced from the original article; a primary research citation should be added before presenting them as independently verified evidence.
The practical lesson remains important: the number of edits is not the same as the amount of expert judgment involved.
What happens when expertise is priced out?
Pressure to reduce MTPE rates can create a difficult situation. Skilled linguists who have spent years developing domain knowledge may decline work when the compensation no longer reflects the responsibility involved.
This does not mean that less experienced linguists cannot do good work. It means that demanding quality standards and increasingly low budgets are not always compatible—particularly for technical documents, software interfaces, healthcare information or safety-related content.
If a project is assessed primarily on speed and cost, subtle errors may receive less attention. And as machine output becomes more fluent, these errors may be harder, not easier, to detect.
The industry should therefore be careful not to describe full post-editing as a task that requires little expertise simply because the first draft was produced automatically.
The professional's role is changing, not disappearing
Machine translation is a powerful tool. It can reduce repetitive work, speed up suitable projects and allow specialists to focus on difficult passages.
But someone still needs to decide whether the result is accurate, complete, consistent and appropriate for its purpose. In high-impact content, that decision can matter more than the number of keystrokes used to make a correction.
A useful way to think about the process is that machine translation offers a proposed solution; the professional evaluates whether that solution is right.
Sometimes the proposal is excellent. Sometimes it needs small adjustments. And sometimes the fluent output conceals a fundamental misunderstanding.
At MITO Solutions, we see professional linguistic review and quality evaluation as essential parts of responsible AI-assisted localization. The appropriate level of review should reflect the content's purpose and risk, not merely the fact that machine translation was used.
What clients and localization teams should ask
Before setting a post-editing workflow or rate, it is worth considering:
- What quality level is required? Internal, low-risk informational content may have different needs from customer-facing instructions or safety-related materials.
- How good is the actual MT output? Evaluate representative samples rather than assuming a fixed productivity gain.
- What expertise does the work require? Domain knowledge, terminology, source-language comprehension and editorial judgment all matter.
- How will quality be measured? Review accuracy and meaning, not only the amount of text changed.
- Who is responsible for the final result? The person approving the translation needs the time, information and authority to do the job properly.
These questions help clients benefit from automation without confusing faster first drafts with guaranteed final quality.
The question behind the price
The real issue is not how much of the first draft was generated by a machine. It is how much professional responsibility remains with the person who approves the final text.
Machine translation can be highly effective. But productivity gains should not be mistaken for the disappearance of linguistic expertise.
If we expect professional-quality results, we still need to value professional judgment.
About MITO Solutions
MITO Solutions combines specialist localization experience with linguistic QA, AI/MT evaluation and practical workflow expertise. Our core language specialization is Hungarian, and we work with trusted professionals and partners on wider multilingual needs.
For a discussion about quality-focused MTPE or AI-assisted localization, visit MITO Solutions or contact contact@mitosolutions.com.
Adapted for MITO Solutions from an article by founder Miklós Tóth, originally published on LinkedIn on 11 August 2026. Read the original article.