December 15, 2002
The honest answer is usually both. What decides the split is not budget but the risk profile of each piece of content.
Machine translation post-editing works well on product manuals, release notes and datasheets. The characteristics are repetitive wording, fixed sentence patterns and concentrated terminology, so the raw machine output is usually decent and the human effort goes into terminology alignment, awkward phrasing and formatting. Against fully human translation, the schedule can compress to a fraction of the original, and a thirty to fifty percent cost reduction is a common figure.
What needs noting is that the price of MTPE itself depends heavily on source quality. If the source document is disorganised and inconsistent in terminology, editing hours can exceed those of translating from scratch. Many people assume machine translation will absorb a bad document; in fact the opposite — it distributes the bad document's problems evenly across every sentence.
So before choosing MTPE, review the source: is the terminology consistent, are the sentence structures clear, is there a lot of colloquial expression? Without cleaning those up, talking about efficiency is meaningless.
Marketing copy, legal contracts, patient-facing medical information and any text that speaks for the brand are better served by fully human translation. The reason is not only quality but responsibility and nuance. Machine translation chooses words on statistical grounds, not on grounds of context; it does not know who the sentence is addressed to or what effect it is meant to produce.
Legal and medical content also carries compliance liability. A dispute caused by a mistranslated contract clause costs far more than the translation saved; ambiguity in patient-facing safety information has consequences beyond money. What these documents need is a person making the judgement — and a named person standing behind it.
The test is simple: if this goes wrong, does it cause personal injury, financial loss or legal liability? If yes, go human, without hesitation.
Most mature teams run both in parallel: high-volume content through MTPE, key content through human translation. The point is not how to split them but that both lines share the same termbase and style guide — otherwise users sense that one company speaks in two voices.
In practice, sort content by risk level first: safety and regulatory through human translation, product documentation through MTPE, marketing and after-sales layered by market importance. The sorting standard should be set at project start rather than left to each translator to judge.
Running the two lines has a further benefit that is often missed: the good output from human translation flows back into the translation memory and raises the quality of later MTPE. The longer you use the combination, the higher its efficiency.
If the choice has to become an executable standard, look at four dimensions. Cost: MTPE clearly wins, though the gap narrows when source quality is poor. Schedule: for large volumes on a tight deadline, MTPE is close to the only option. Quality: where tone matters and legal liability attaches, human translation is irreplaceable; for technical content with concentrated terminology, the gap between the two is small. Risk: anything touching personal safety or compliance goes human, because there has to be a named party accountable for the outcome.
At bottom this is not an either-or question but a content-classification one. Rank the content by risk, let MTPE and human translation each own the range they are good at, and let the termbase keep the output consistent. FangService runs a content classification pass at project start and prices accordingly — sorting out which content is worth the human effort is better value than negotiating the total down.
Machine translation plus human post-editing looks like good value, but it is extremely sensitive to source quality. Where the source text is clean, structured and consistent in terminology, post-editing really does remove thirty to forty percent of the work. Where the source is cobbled together, the editor first has to guess the intent, then rewrite it, then check it against the glossary — and the hours spent exceed a straight translation, while the result still carries a machine's fingerprints. Our experience is that MTPE clearly pays on well-structured technical documentation, but for marketing copy, legal clauses and survey reports, where tone and precision are both critical, full human translation is more controllable.
The test is simple. Give the editor a sample to post-edit, record hours per thousand words, and compare it against full human translation while allowing for rework probability. FangService quotes both MTPE and fully human workflows, and clients can trial one on a small batch before committing a whole contract, deciding on measured hours rather than a bet.
The choice between MTPE and full human translation does not have to be a judgement call; source quality decides it. Score the source on three indicators: is terminology consistent (does one concept have only one name), is sentence structure complete (many subjectless clauses, broken sentences, nested clauses), and has the content been edited (obvious colloquialisms, repetition, contradictions). With all three favourable, post-editing saves thirty to forty percent of hours; with two or worse, it is not worth it.
The follow-up action is equally clear. Good source text goes to MTPE, with human effort on QC and final review. Poor source text gets a revision pass first and then translation. Skipping source revision and going straight to MTPE asks the editor to do harder work in less time, and the quality slide is close to inevitable.
FangService performs this source assessment at project start and returns a process recommendation with an hours estimate. Experience says a week spent revising the source saves two weeks of post-editing and a round of rework, and few clients decline that arithmetic.
Tell us your project scenario and requirements, and we will match the right solution and provide a quote within 24 hours.
Get a Free Quote