Medical translation has a reputation for being difficult, and it deserves it — but not always for the reasons people assume. The terminology is the easy part. Terminology can be looked up, verified, and stored in a termbase. What actually separates a usable medical translation from a dangerous one is a set of habits that nobody teaches you in a certification course. Here are the ones that cost me the most to learn.

1. Decide who is reading before you translate a single word

The same source sentence can require two completely different translations depending on the document type. A discharge summary written by a cardiologist for another cardiologist keeps "myocardial infarction". A patient information sheet for the same condition needs the everyday phrase your target reader would actually say out loud to a nurse.

This is not a stylistic preference. Many ethics committees and sponsors impose a readability ceiling on informed consent forms—often around a sixth- to eighth-grade reading level in the source. If you translate a consent form in the register of a journal article, you have not produced a stylistically conservative version. You have produced a non-compliant one, and it will come back.

Ask for the document type before you quote, not after you deliver.

2. Abbreviations are the most reliable way to injure someone

"MS" is multiple sclerosis, mitral stenosis, morphine sulfate, or mental status, depending entirely on the surrounding chart. "PT" is physical therapy, prothrombin time, or patient. "DC" is discharge or discontinue — two instructions that mean nearly opposite things at the bedside.

The rule I follow now is simple: never expand an abbreviation unless you can justify it from the document itself. If the surrounding text does not disambiguate it, that is a query, not a judgment call. Handwritten records make this worse, because the abbreviation may not even be standard — hospitals develop local shorthand that no reference will contain.

3. Run a separate pass for numbers, units, and separators

Do this as its own pass, with the terminology part of your brain switched off. Decimal commas and decimal points swap between locales. A misplaced separator turns 1.000 mg into 1,000 mg. Microgram symbols get mangled by encoding, OCR, or an over-helpful autocorrect, and "µg" rendered as "mg" is a thousandfold dosing error sitting quietly in an otherwise clean file.

Also: do not convert units unless the client has explicitly asked you to. Converting pounds to kilograms in a clinical record without instruction introduces a number that does not appear in the original — which, in a sworn or certified context, is a defect regardless of how correct your arithmetic was.

4. Write for the reader, not for the back-translation

If you work on patient-reported outcome instruments, your text will be back-translated and compared to the source. Translators who know this sometimes start calquing — choosing stiff, structurally source-shaped wording so the back-translation looks reassuringly identical.

Resist it. The back-translation exists to detect meaning shifts, not to reward mirror-image syntax. A reviewer reading a clean back-translation of an unnatural target has learned nothing, and the patient completing the questionnaire is the one who pays. Translate naturally, then defend your choices in the reconciliation notes. That is what the notes are for.

5. Build a source hierarchy and actually follow it

Mine, in order: client glossary and style guide, then approved regulatory templates and product information for the target market, then national pharmacopeia and health ministry publications, then reputable specialist dictionaries. General web search sits at the bottom, and translator forums are for orientation only — never as a citable authority.

The point of a fixed hierarchy is to remove negotiation from tired, late-night decisions. When you are eleven hours into a file, you do not want to be relitigating which source wins.

6. Learn to write a query that gets answered

A bad query is "Please clarify the highlighted segment." It gets ignored or answered three days after your deadline.

A good query does the work for the person answering it. Quote the segment and give its location. State precisely what is ambiguous. Offer two or three defensible readings. Say which one you recommend and why, and state what you will do if you receive no reply. Most project managers are not clinicians; they forward your question to someone who is, and that person will answer a specific question in thirty seconds, but never an open-ended one.

7. Do not silently repair the source

Medical documents contain errors: a wrong laterality, a date that cannot be right, a drug name misspelled by whoever typed the discharge note. It is tempting to quietly fix these. Do not.

Translate what is there, and flag any discrepancies in a translator's note or query. In certified and sworn work, this is not optional — your attestation covers fidelity to the source, and a helpful correction is still a deviation. Silent repairs also destroy the paper trail that a clinician or lawyer may later need.

8. Treat illegible handwriting as data, not as a puzzle to solve

Scanned records defeat OCR constantly. When a word genuinely cannot be read, mark it with a consistent convention — `[illegible]` — and keep a note of where. Guessing a drug name from three ambiguous letters is not diligence. It is an invention with good intentions.

9. Build QA passes that look at different things

One pass for meaning against the source. One-pass reading of the target alone, without the source visible, to check whether it makes sense as a document in its own language. One pass for numbers and units. One pass for terminology consistency against the glossary. Four narrow passes catch more than one heroic pass, because each one lets you ignore everything else.

Reading the target aloud, at least in patient-facing material, catches an embarrassing number of errors. Your ear notices unnatural phrasing that your eye has already forgiven.

10. Know the edge of your competence and say so

Medical translation is not one field. Oncology protocols, medical device instructions for use, psychiatric assessments, and dental records demand genuinely different background knowledge. Accepting an oncology trial protocol because you have done cardiology well is how good translators produce their worst files.

Declining work is not a failure of professionalism. It is professionalism, and in my experience, clients remember the person who said, "This one is outside my area, but I can recommend someone," far more warmly than the person who delivered something that looked competent but was slightly wrong.

A worked example: where the time actually went

The most instructive job I have had recently was a hospital medical record translated into English for use outside the issuing country. I cannot describe what was in it, and I would not want to. What I can describe is the *shape* of the problem, because the shape is what generalizes — and because almost none of the difficulty lived where a client would expect it to.

The file was not really a document. It was a composite: institutional letterhead blocks, a laboratory history listing, several detailed result panels, and a number of screens from a hospital information system that someone had reconstructed inside a word processor. Roughly a dozen separate tables, nearly all of them using merged cells, plus multiple header and footer sets and an embedded raster image. Flowing prose accounted for a small minority of the visible text.

That matters because a CAT tool treats every table cell as an isolated segment. You lose the sentence-level context you normally rely on, and the segments arrive in an order that has nothing to do with how a reader traverses the page. Rebuilding the reading order in your head, repeatedly, is slow and invisible work that no word count captures.

Formatting fidelity fights legibility. Column widths in the source were sized to the source language. English ran longer in some cells and shorter in others, so preserving the grid exactly meant wrapped, cramped cells; letting the grid breathe meant a page that no longer resembled the original. For a document whose whole purpose is to be recognizable as a faithful counterpart of the original, I lean toward preserving the layout and adjusting typography rather than geometry — but it is a judgment call, and it is worth stating the choice in your delivery note rather than letting the client discover it.

Deciding what not to translate. Institution names, street addresses, registration and license numbers, and the identifiers printed across letterheads and footers are not really text. They are how a receiving institution verifies that the document is genuine. I keep them verbatim, and where a reader genuinely needs the sense, I add a bracketed gloss beside the original rather than replacing it. Translating an institutional name outright can quietly make a document unverifiable.

Reference ranges are provided by the issuing laboratory. They are method-dependent and instrument-dependent, not universal. Substituting the target country's conventional ranges or converting units to what the destination reader is used to would have produced a cleaner-looking document, but one that would be materially false. Same rule as always: carry it across, flag it, let the clinician interpret it.

Standards versus what was actually written. Clinical shorthand is often locally conventional rather than internationally standardized, and it does not always have a tidy equivalent in the target language. The temptation is to upgrade the source — to render everything in polished, standards-aligned English. In a certified translation that is not an improvement, it is a substitution that hides from the reader the fact that the original was less precise than the translation appears to be.

Interface strings have no context at all. Short labels lifted from a hospital system arrive without surrounding sentences to disambiguate them and with a fixed cell width to fit into. They need to be consistent across the whole file and short enough not to break the layout, and the only way to get them right is to look at where they sit on the page rather than at the segment in front of you.

Adding it up: the actual translation was the smallest component of the job. Layout reconstruction, boundary decisions about what stays in the source language, and the notes documenting both took considerably longer — and they are the parts that determine whether the document is usable when it reaches the desk it was made for.

None of this is glamorous. Medical translation rewards the same unremarkable virtues every time: check the number twice, ask the question, resist the helpful fix, know what you do not know. The terminology gets easier every year. The habits are what keep the work safe.