Subtitle translation: what has to fit inside the speaker's time

What separates subtitle translation from translating text

When you translate text, the reader has time. In a subtitle the speaker has already spent the time and you are only filling the gap they left. The unit of work is no longer the sentence either: it is the event, a block that arrives on one exact second and leaves on another, and every decision about it has to make sense inside that interval.

Three constraints press on that one block at once. The clock says how many seconds this text stays on the picture. The box says how many characters fit on a line and how many lines are allowed. And the reader, whose actual job is watching rather than reading; if your line takes all of their attention, they have lost the film and you have lost the job.

There is one more difference, and it is rarely stated. A subtitle is the only translation whose source is still audible. The viewer hears the original, sees your line, and compares them without meaning to. Somebody who knows ten words of the source language will look for exactly those ten words, and if they cannot find them they feel something went missing, even when the translation is entirely right. That is why a free but correct rendering smells of error in subtitling more than anywhere else.

The numbers that actually tie your hands

Two published guides carry numbers you can actually read off, and both were written for delivery into European languages. The BBC subtitle guidelines take a reading rate of 160 to 180 words per minute and conclude from it that you should allow around 0.3 seconds per word, so a four word subtitle stays on screen for at least 1.2 seconds. Netflix gives a floor and a ceiling for the event instead of a rate: a minimum of five sixths of a second and a maximum of 7 seconds.

One thing the BBC says plainly and almost nobody quotes: in a proportional font, counting characters is not the right measure at all, because letters differ in width and what limits you is the width of the region. To keep the number usable it gives an equivalence: 37 characters in a 75 percent wide region of a landscape video is about 25 characters in a 90 percent wide region of a 9:16 vertical video. Same sentence, same font, a third less room, purely because the video turned vertical.

Now the honest boundary: neither number is Persian. Netflix keeps a separate guide per language and its Farsi one sits behind a partner login; we opened the public address today and it redirected to a sign-in page, so we quote no number from it. What a professional does instead is take the number from the platform that will receive the file, and where that platform publishes none, measure once and make that the house rule. The technical layer of the file itself, the difference between SRT and VTT and between burned and soft subtitles, is explained elsewhere: the subtitles and captions lesson.

Persian fits the box; the clock is another matter

The practical question for every subtitle translator is whether the text fits the line. That is measurable, and we measured it on our own text. This section of the site holds 152 lesson files, each written in four languages side by side, so every Persian passage has an English counterpart. Across the 2,827 long passages in that corpus the Persian came out shorter than the English in 98.4 percent of cases, and took 87 percent of the English characters overall. Arabic took 75 percent and Turkish 95 percent.

Now count the same passages in words and the direction reverses: Persian holds 104 percent of the English word count and uses more words in 69.4 percent of the passages. The reason sits in one more number, the average characters per word: 4.61 in Persian and 5.5 in English. Persian pours the same content into shorter and more numerous words.

For subtitling those two numbers say two different things, and this is exactly where translators get it wrong. The box counts characters, so a Persian line usually fits. The clock, though, is measured with something the guides wrote in words, and 160 to 180 words per minute means every extra word costs extra time. Time your Persian with an English rule of thumb and you produce a line that fits the box and not the interval: it does not overflow and it raises no warning, the reader simply never reaches the end of it.

So for Persian the binding constraint comes from time rather than from width. There is a boundary worth stating plainly: these texts are instructional prose rather than dialogue, and they are not translations either, since the four languages are written separately. The numbers show a direction and not the real size of speech.

Condensing: what goes and what never does

Condensing has a bad name because it is usually explained badly. Condensing is not a style, it is what the clock forces. The BBC guidelines put it from the other side: if there is time for verbatim speech, do not edit unnecessarily, and the aim is to give the viewer as much access to the soundtrack as you can. Do not automatically cut small words like "but" or "too", because they carry meaning. And it rejects simplification for deaf viewers outright: it is condescending, and it frustrates anyone who lip-reads.

So the working rule is this: keeping is the default, and cutting is imposed on you by the clock rather than chosen by taste. When it is imposed, the order is not random. First to go is whatever the picture is already saying or the ear has just heard: repetition, hesitation, addressing somebody who is in frame. Then the markers of ordinary speech, carefully, because those are what carry tone. Then a second explanatory sentence, if the first one did its work.

And the things that are never sacrificed to condensing: negation, numbers, names, and the single word the joke or the threat or the plot turn is sitting on. The BBC is explicit about names too, saying not to edit out names used to address people, because they are easy targets and following the plot depends on them. Cutting a negation is the worst of the set, because it produces an error the reader cannot even feel: the sentence stays sound and its meaning turns over.

In practice the test is simple. Read the line at real speed while watching the picture. If you had to read it twice, the line is long; if you lost the picture, the line is long; if you got to the end and something was missing, you cut too much.

Where to break the line, and a trap that only happens in Persian

Both guides share one logic: break the line where the language already breaks. The best place is at punctuation, and after that before a conjunction or a preposition. The list of things not to separate is nearly the same in both: an article from its noun, an adjective from what it describes, a first name from a surname, a pronoun from its verb, and the parts of a complex verb.

In Persian that same logic has two cases the English lists do not carry. The first is the ezafe, the linking vowel: "project manager" or "the original sound of the film" is one unit in Persian, and breaking the line inside it is exactly what those guides forbid between an article and a noun. The second is the Persian compound verb, whose nominal part and light verb are one verb together; putting "decision" at the end of one line and "took" at the start of the next leaves the reader hanging inside a verb.

Then a trap that only visits right-to-left text. A subtitle file is plain text and has no attribute for declaring direction; a web page has dir and an SRT has nothing. So when a Persian line ends in a Latin name or a number, the Unicode bidirectional algorithm decides where the final full stop sits, and in some players it appears at the other end of the line. The common fix is one of the invisible direction control characters such as U+200F, from the same family this site bans in its own copy, because on the web the dir attribute does that job and an invisible character is only a machine's fingerprint. In a subtitle file that attribute does not exist, so this is the exception and it should be used knowingly.

There is a simpler route, and we prefer it: arrange the line so that it does not end on a Latin name or a number. One Persian word after it is enough and the problem never arises. If you cannot avoid it, open the file in the very player your audience uses and look, because this is the class of thing that looks right in the editor and breaks in someone's hands.

The things a subtitle does not translate at all

The Netflix general guide carries several rules that are rarely quoted in Persian, and each of them takes one decision out of the translator's hands, to the benefit of the work.

Currency is not converted. Any amount of money spoken in the dialogue stays in the original currency. A hundred dollars remains a hundred dollars and does not become the local equivalent, not only because the rate moves, but because that line belongs to a scene and the character said dollars.

A brand name has three routes and only these three: the same English name where it is widely known in that territory, the name the brand is known by in that territory, or a generic term for the product. And one explicit prohibition: do not swap one company's brand for another company's trademarked item.

Quotations are the delicate case. The guide says it is best practice to originate a new translation for any quoted text, because a new translation is free of rights issues, and that an existing translation may be used only if it is in the public domain or documented permission has been granted and payment made. A translator who lifts a line of poetry or a sentence from a published novel off the internet is making a legal decision, not a linguistic one.

And one rule that comes back to the translator: the translator credit is written as the last event of the file, in the target language, over the copyright card at the end of the programme, with a duration of up to 5 seconds. Your professional credit has a defined place in the file, and it stays empty only if you leave it empty.