There is a moment, maybe twenty segments into a revision job, when you can feel the temptation arrive. The translation in front of you is fine. It is accurate, it reads naturally, the terminology is consistent. And something in you — call it the reviser's ego, call it the need to justify the line on the invoice — starts itching to improve it. To swap a perfectly good word for your preferred synonym. To restructure a sentence that was never wrong, only differently right.
This is the pitfall nobody warns you about. We train ourselves obsessively to catch errors. We build checklists, we run QA tools, we sharpen our eyes for mistranslations and missing negatives. But the more dangerous failure mode in revision is not under-correcting. It's over-correcting. It's the reviser who treats every divergence from their own idiolect as a defect.
Error versus preference
The single most useful distinction I've internalized over fifteen years is the line between an error and a preference. An error is something that is objectively wrong: a mistranslation, a number transposed, a term that violates the glossary, a grammatical breach, a sentence that misleads the reader. A preference is a choice you would have made differently but which is not wrong.
The entire craft of revision lives in respecting that line. ISO 17100 builds bilingual revision into the standard workflow precisely because a second pair of eyes catches what the translator's brain auto-corrected past. But the standard says to revise against the source for accuracy and suitability — it does not license you to rewrite a competent colleague into your own voice. When you change "begler" to "jenaplar" because you happen to prefer it, with no register justification, you have not improved the file. You have introduced churn, you have insulted a capable translator, and — if the original translator gets the corrected file back — you have taught them nothing, because there was nothing to learn.
For Turkmen this matters more than for many languages, and here's why. We work in a language with thin standardization in technical and software domains. There's often no single "settled" term, which means two skilled translators can land on two defensible solutions. If the reviser treats their own gut as the dictionary, every job becomes a coin flip on whose preferences happen to be in the chair that day. The client sees instability. The translator sees arbitrariness. Nobody sees quality, because quality was never the thing being measured.
The cost of a tracked change
Every edit you make has a price, and revisers routinely ignore it because the price is paid by other people.
The project manager pays first. A file that comes back drowning in track-changes looks like a file that needed heavy intervention — which triggers escalation, a query to the translator, possibly a downgraded quality score, possibly a lost relationship with a good linguist. The PM cannot easily tell, from a wall of red, which changes were necessary and which were vanity. The signal is buried in the noise. A clean revision with eight surgical, correct changes communicates far more than a bloodbath of forty edits, half of them lateral moves between equally valid phrasings.
The translator pays next. There is no faster way to demoralize a strong Turkmen translator — and there are not many of us — than to hand back their work shredded for reasons that amount to taste. They stop trusting the process. The good ones quietly decline the next job from that chain.
And the text itself pays, because over-editing introduces error. This is the part people forget. Every time you touch a sentence you didn't need to touch, you create a fresh opportunity for a typo, a broken agreement, a dropped tag, a vowel-harmony slip in an attached suffix. I have seen revisers introduce more defects than they removed. A change you didn't need to make is pure downside risk.
How I keep myself honest
A few working rules, hard-won.
First: for every change, name the reason in one word. Mistranslation. Glossary. Grammar. Punctuation. Register. Ambiguity. If you cannot name it — if the honest answer is "I just like mine better" — revert it. This single habit cuts preferential edits by more than half.
Second: read the target as a reader before you read it against the source. If a sentence stops you, flows wrong, or makes you reread, that's a real signal worth acting on. If it reads smoothly and you only object once you're cross-checking word by word, be suspicious of your own objection.
Third: separate the two passes. A correction pass for errors. A flagging pass for genuine improvement suggestions, which I leave as comments, not changes — "consider X for consistency with the UI string." That hands the call back to where it belongs and keeps the tracked changes clean.
Fourth, and this is the one that takes maturity: assume the translator is competent until the text proves otherwise. Revision is not a contest. The job is to make the deliverable correct and consistent, not to win.
The best revisers I know are not the ones who change the most. They're the ones who can sit on their hands in front of a sentence that is merely not how they'd have done it, and let it stand. Restraint reads like laziness from the outside, which is why it's underrated and underpaid. But a light, precise hand is the harder skill, and on a language as under-resourced as ours, it's the one that actually protects quality.