Somebody else reads it first
The second translator is not checking your spelling. They are checking your decisions.
Revision by a second translator is not proofreading; it is a comparison against the source, line by line.
What revision actually is
When a translation leaves one professional's hands and reaches another's, the handover is not for a final polish. Under ISO 17100, revision is defined as a bilingual activity: the reviser reads the target text against the source text, segment by segment, and evaluates whether the meaning transferred correctly, whether register was maintained, whether any omissions crept in, and whether the terminology is consistent. Proofreading — reading the target text alone for surface errors — is a different, lesser task. Revision subsumes it but goes far beyond it.
The instructive failures are the ones where a translated procedure inverted two steps, which is why this trade has a standard at all. A step in the wrong order
The distinction matters because the errors that damage a technical manual, a regulatory filing or a court submission are rarely typographic. They are the paragraph that sounds fluent in the target language but has quietly dropped a qualifying clause from the source. They are the sentence where two near-synonyms were treated as interchangeable in contexts where the subject matter requires them to be kept apart. A monolingual read catches neither.
How a reviser works
In practice the reviser opens both texts in a translation environment — usually the same computer-assisted translation tool the translator used, so the segment pairing is already structured — and moves through them in order. For each segment they are asking a cluster of questions at once: Is the meaning equivalent? Is anything missing or added that the source does not sanction? Is the terminology consistent with the rest of the document and with any approved term base? Does the register match the source's register? Only after those questions are satisfied does the reviser attend to fluency.
The reviser may accept a segment, suggest a change, or flag it for discussion with the translator. At better-resourced organisations — the Directorate-General for Translation is a widely studied example — the workflow builds in a dialogue at that stage: the translator sees the reviser's interventions and can respond before the text is finalised. This matters because a reviser who simply overwrites a translator's choices, without visibility, removes the accountability that makes the process useful. The translator learns nothing; the reviser may be wrong; and the decision trail disappears.
Post-editing is a different job from translating, with a different failure mode: the errors are fluent. What machine output changes
There is also the question of scope. ISO 17100 requires revision of the whole text, not a sample. Selective revision — spot-checking a proportion of segments — belongs to quality assurance at a different tier and cannot substitute for a full bilingual comparison on a document where the stakes are high. The standard itself lists revision as a mandatory step, not an optional enhancement, and distinguishes it explicitly from checking and proofreading in its process requirements.
What revision finds that nothing else does
The failure modes that full bilingual revision reliably catches cluster around a few recurring types. Omission is the most common: a sentence or phrase present in the source simply does not appear in the target, usually because a long source sentence was mentally parsed as shorter than it was. Addition is the inverse and is less common but more insidious — the translator has elaborated on the source, adding a nuance or softening a claim that the source does not contain. Mistranslation of specialised terms is the third major category, and here the reviser's own subject knowledge matters: a reviser without domain familiarity may pass a term that is technically wrong while being linguistically plausible.
Register drift is subtler still. A source document pitched at a formal regulatory audience can gradually shift toward a more colloquial target-language register through a hundred small, individually defensible choices — a synonym here, a contracted construction there — and the cumulative effect is a document that misrepresents its own standing. This is the kind of error invisible to anyone who has not read both texts, in order, from beginning to end.
None of this is a critique of translators. Revision exists because a single reader of their own work — in any language, in any discipline — will miss things that a second reader catches. The cognitive load of producing the translation is itself the reason a fresh eye is needed. The second translator is not evidence of distrust. They are the structural acknowledgement that accuracy at this level is a two-person job.