The Self-Revising Specialist

Every serious translation-quality standard says the same thing. A translation must be checked against its source by someone other than the person who produced it. ISO 17100:2015 calls it revision; the profession calls it the four-eyes principle. The logic is hard to argue with. A translator who has spent hours inside a text is the last person likely to notice what they have misread.

The difficulty is not the principle. It is the assumption underneath it: that a second qualified reviser exists, is available, and can be paid for.

In French-English legal and institutional work for African multilateral organisations, that assumption often fails. The pool of specialists who can revise a loan agreement, a board resolution or an evaluation report at full legal register is small. Deadlines are short. Budgets are fixed. And so a single specialist routinely delivers publication-ready work alone, which the standard does not contemplate and the literature has largely not examined.

I have spent the past year studying what actually happens in that situation, using my own practice as the case.

What the study looked at

The research draws on the complete, anonymised administrative record of a three-year engagement with a multilateral development-finance institution: 1.77 million words across 499 documents, from 2023 to 2026. Alongside the volume and turnaround data, I ran a small error-typology audit on one delivered text, scored independently by me and by three professional French-English colleagues working blind to who had produced the translation, using the categories consolidated in ISO 5060:2024.

The point of the audit was not to prove that the work was good. It was to find out whether a translator revising alone can see what an independent reader sees.

What it found

Three results matter.

The work held up on an absolute scale. No rater recorded a single critical error, and none found a major accuracy failure that would have compromised the document's use. The severity-weighted penalties the peers assigned ranged from 22.8 to 55.4 per thousand words, densities consistent with a publication-ready institutional text.

The revisers disagreed with each other as much as the research predicts. Those three peer figures span a more than twofold range on the same eleven segments. Reviser inconsistency is well documented in the literature, and this was a small, concrete instance of it. It is worth stating plainly, because it complicates the idea that adding a second reader automatically produces an objective verdict.

Self-revision has blind spots, and they are measurable. The peers and I flagged the same number of problems on average, four each, but not the same problems. Where two or more peers agreed a segment was faulty, I had agreed with them only half the time. I also rated my own choices less severely than they did: my weighted density was 26.1 against a peer mean of 39.1. The difference was not in how much I saw. It was in how gravely I judged what I saw.

That last finding is the uncomfortable one, and it is the reason the article does not conclude that self-revision is simply equivalent to independent revision.

The argument

What the record does show is that self-revision, in this setting, is not unaided introspection. It is an instrumented, multi-stage system: translation memories carrying prior institutional decisions, terminology bases, automated quality-assurance checks, and language models used under supervision, with human judgement at every decision point. The checking is distributed across those instruments rather than concentrated in one second person.

So the article proposes reading the second-reviser rule as one implementation of a broader principle rather than as the principle itself. What the standard is protecting is independent verification: the introduction of a check not anchored to the translator's own interpretive commitments. A second human reviser is the most familiar way to provide that. It is not self-evidently the only one, and, as the audit shows, it is not a perfectly reliable one either.

That reframing has consequences for how standards are written, how translators are trained to revise their own work, and how the economics of specialised, low-density language work are understood.

A note on where this stands

This is a summary of a study currently under peer review at Translation Spaces. I have set out the findings here in my own words rather than posting the manuscript, since the full text belongs with the journal until the review process has run its course. I will link to the published version when it appears.

I am also conscious of the obvious objection to a practitioner studying his own work, and the article addresses it directly: the audit was designed so that the peers could contradict me, and they did.