What professional translation is, and how it differs from word for word

What translation actually carries across

Translation is rewriting a text in another language so that the new reader takes away what the reader of the original took away. Words are one of the things that move across, and as it happens the least important of them.

Four things have to arrive. Meaning, which is the most obvious and the least trouble. Tone, meaning whether the writer is serious or playful, distant or close. Register, meaning whether the text is formal, spoken or technical. And function, meaning what the text is supposed to do: sell, warn, teach, or simply stand as a record.

One short example is enough. In a message between friends, "see you later" becomes something casual; in an official letter it never becomes a ceremonial formula. Both readings defend themselves on meaning alone, and one of them, in the wrong place, makes the text ridiculous. Being wrong is not the problem here. Being out of place is.

That is where the line between professional translation and word for word translation gets drawn. Word for word delivers the meaning and leaves the other three to luck. A professional decides all four on purpose, and when one of them has to be sacrificed, knows which one and why.

What has to be settled before the first sentence

A beginner opens the file and starts at the first sentence. A professional has four answers before that, and will not start without them.

• Who reads this and how much do they already know. A technical guide for an engineer and the same guide for a customer are two different texts, even when the source is one file.
• What should the reader be able to do afterwards. If they are meant to click a button, the sentence has to be short; if they are meant to sign a document, it has to be exact even when that makes it heavy.
• Where the register sits. Decide it once for the whole text and write it down somewhere you can see, because register is the thing that slides without permission on page two.
• What must not be translated at all. Brand names, product names, a legal term with no matching institution in the target language, a code fragment, a file name, a unit that is standard. This list is the shortest part of the job and the one that causes the most argument.

Take that fourth item seriously. Whenever a translation falls apart halfway, it is usually because someone translated a product name, or rendered one term differently in chapter three. Writing the list before you start takes ten minutes and prevents a full rewrite.

How translation and localisation differ

Translation moves the text. Localisation changes the text so it fits where it lands.

Things localisation touches and translation must not: units, currency and dates, an example that only means something to the source reader, the shape of a phone number and an address, the order of given name and family name, the direction of the page, and anything the destination country requires by law that the source does not. These are not translations. They are decisions, and the owner of the text has to know about them.

Most sites that call themselves multilingual have only been translated, and that is where it stops. The result is a text that is correct and reads foreign: a price in dollars for a reader who thinks in another currency, or a form asking for the family name first from someone who has always written their given name first.

An example from this site. The Arabic version of this section is not the Persian page in Arabic letters: its digits are the Arabic-Indic set rather than the Persian one, and its letters are the Arabic yeh and kaf rather than the Persian pair. Those forms are wrong in Persian and right in Arabic, and the other way round. Had we only translated, the Arabic would have been correct and slightly off for an Arabic reader. Google publishes its own guidance on localised versions of a site, and it says the same thing: each version has to be built for the reader of that version, not merely rendered into it.

Where Persian and English catch on each other

Every language pair has its own catches, and this pair has two large ones.

First, politeness and distance. Persian has an informal and a formal second person and a whole system of verb forms and suffixes that tunes the distance between speaker and listener. English has one "you" and does the same job through word choice and sentence shape. So coming from English into Persian forces a decision the source never made, and getting it wrong is immediately audible to the reader. Most translated texts that read as off on Iranian sites broke exactly there, not on meaning.

Second, sentence length. Persian tolerates long chains and a long sentence sits naturally in it. English punishes the same chain. A good Persian to English translation therefore tends to break sentences up, and a good English to Persian one sometimes joins two into one.

Now a number, from this section's own data. Each of the 148 lessons here is written in four languages, and we counted the Persian and English of every text field pair by pair: 7,152 pairs where both sides ran to at least eight words, on 9 September 2026. The result was not what we expected. By word count Persian came out only slightly longer than English; by character count it came out 13 percent shorter. Persian takes more words, and each of its words is shorter.

The average is not the point, though. In 82 percent of the pairs the two sides stayed within twenty percent of each other, in 12 percent Persian ran more than twenty percent longer, and in 6 percent more than twenty percent shorter. Roughly one field in six swings hard one way. A layout built around the average breaks on exactly that sixth. Space for a button or a heading has to be designed for the spread, not for the mean.

One honest caveat: these texts are not translations, they are written separately in all four languages. So the figure above describes parallel writing, and on a real translation, which is obliged to follow the structure of its source, the spread gets wider rather than narrower.

Where a professional translator says no

Three places, and all three are marks of professionalism rather than weakness.

The first is direction. You write into the language you are strongest in and read from the language you are merely competent in. Someone whose first language is Persian and whose English is good will usually translate better into Persian than out of it, and that is not a flaw. Turning down a Persian to English job when the text is going to be published is the right call. No model vendor publishes separate numbers for the two directions either, so not even the tools will tell you which direction of yours is weaker.

The second is the field. Medical, legal and financial texts carry terms whose correct equivalent is known only to someone who has worked in that field. One displaced word in a contract inverts the meaning. If you have a text of that kind and you are not its translator, go to someone who has worked in the field; our own translation service is organised along exactly that split.

The third gets said less often: when the source sentence is itself wrong. A translator does not quietly fix it. They translate it, then tell the owner of the text where the problem is. Fixing it quietly means the reader of the target text reads something the author never wrote, and when the two versions are eventually laid side by side, it is the translator who has to explain.

The three together say one thing: a translator does not own the text, they hold it in trust. Wherever that gets forgotten, the work slides from translation into rewriting, and nobody notices until it is late.