The pairs where the standard post-editing discount does not hold
In the largest study of post-editing productivity ever published, Swedish post-editing was 7% slower than translating from scratch. French was 130% faster. The same discount cannot be right for both — and the Nordics are where the industry's default pricing quietly breaks.
The finding
The largest study of post-editing productivity that has been published covered 90 million words, 879 linguists and two and a half years of real production data. Average productivity gain across language pairs: 66%.
The range behind that average is the part nobody talks about. French: 130% faster than translating from scratch. Swedish: 7% slower.
Sit with the Swedish number. Post-editing took longer than translating the content fresh. Every per-word post-editing discount applied to Swedish in that period destroyed value for whoever accepted it — and post-editing discounts in this industry are typically 20% at the company level and 40 to 60% at the freelance level, applied uniformly across pairs.
Why this happens
Nordic languages are not exotic. They are well-resourced, Latin-script, structurally close to English, and engines produce fluent-looking output in all of them. That is precisely the problem.
Fluent-but-off output is slower to fix than obviously bad output. When a draft is 85% right, a post-editor has to read every sentence carefully to find the 15%, and then decide whether each fix is worth making. When a draft is clearly wrong, the decision is instant. Effort does not scale linearly with error rate, and in the band the Nordic engines occupy, it inverts.
Compound formation. Swedish, Danish and Norwegian build compounds the way German does. Engines split them incorrectly, and a wrongly split compound in Swedish can change meaning entirely while remaining perfectly readable.
Definiteness is a suffix, not an article, so errors are inside words rather than between them — harder to spot, harder to fix cleanly.
Norwegian is two written standards. Bokmål and Nynorsk are separate deliverables with separate conventions, and public sector work in Norway may require both.
Finnish is a different problem entirely. It is not a Scandinavian language and it does not behave like one: extensive case morphology, agglutination, and far less training data behind it than its neighbours. It is genuinely closer to the difficult-language category than to the Nordic one, and it should not be priced as though it sits with Swedish.
What we do
- We measure our own throughput by pair and content type, and we will show you the numbers rather than assert them. If our data says post-editing genuinely saves time on your content in Danish, you should pay less, and we will tell you that.
- We price by revision depth, not a flat per-word discount — which is the only pricing model that survives this evidence.
- We ask for a quality floor. Where engine output falls below an agreed threshold, the segment is treated and priced as translation. In the Nordics this clause is not a negotiating tactic; it is the difference between a sustainable rate and a loss.
- We treat Bokmål, Nynorsk and Finnish as separate jobs with separate reviewers.
The thing nobody has data on
That productivity study — and every comparable one — covers only European languages. Nobody has published post-editing productivity data for Arabic, Japanese, Korean, Chinese or Thai.
If post-editing can be slower than translating for Swedish, the probability that a uniform 40% discount is economically justified for Arabic or Japanese is low. But no one has the data to prove it, which means no one can disprove it either.
We log hours, word counts, language pair and content type on every post-editing job we run. Over time that produces evidence the industry does not currently have — and we would rather negotiate rates from numbers than from opinions, in either direction.