How subtitles get made, and why translations disagree

Anyone who has watched a film with subtitles in a language they partly understand has had the experience of noticing that the subtitle says something other than the dialogue. The usual conclusion is that the translator made an error. Occasionally true; far more often the subtitle is doing a different job, under constraints that make literal translation impossible.

The constraint that governs everything

A subtitle must be read in the time it is on screen, and reading is slower than listening. Professional practice therefore works to a reading speed — a maximum number of characters per second — alongside limits on line length and number of lines, typically two.

The consequence is unavoidable: a fast speaker cannot be fully transcribed. A line delivered in two seconds may permit around thirty characters, which will not accommodate a long sentence.

So subtitling is not translation into another language. It is translation into another language and compression into a smaller space, and the two goals frequently conflict.

What gets cut first

Compression follows a rough hierarchy. Redundancy goes first — repetitions, filler, false starts, names when the referent is visible on screen. Then information conveyed by other means: if a character is visibly angry, the words marking anger can be reduced.

What is preserved is meaning that advances plot or character. This is why subtitles frequently feel cleaner than the speech: politeness formulae, hesitations and social lubrication are exactly the material that compresses away.

It also means subtitles systematically make dialogue look more efficient than it was, which is a real distortion, just an unavoidable one.

Subtitling versus dubbing

Dubbing operates under a different constraint — the translated line must fit the time of the original delivery and, ideally, match visible lip movements. This can push a dubbed script further from the literal meaning than a subtitle, because it must satisfy a physical fit rather than a reading one.

Neither is more faithful in principle. They are faithful to different things: subtitles preserve the original performance and sacrifice completeness of text; dubbing preserves the flow of watching and sacrifices the original voices.

Why two subtitle tracks disagree

Several reasons, most of them legitimate.

Different source material. Translators may work from a transcript, from a translated intermediate script, or from the film itself. The pivot-language route — translating from an existing English version rather than the original — introduces a second layer of interpretation and is common for less widely spoken languages.

Different target reading speeds. A track built for a slower reading speed compresses more aggressively.

Different registers. Slang, dialect and humour rarely have equivalents. A translator must choose: keep the foreignness and lose the immediacy, or find a local equivalent and lose the specificity. Two competent translators will make that trade differently.

Different clients. A festival subtitle track, a streaming track and a broadcast track may have different guidelines on profanity, on-screen text, and how much to explain culturally specific references.

Timing is a craft of its own

Beyond translation, subtitles must be spotted — assigned in and out points. Good spotting respects the edit, so a subtitle does not persist across a cut, and gives the eye time to travel.

Poor spotting is invisible as a specific fault and produces a general sense that a film is hard to follow. Viewers rarely say the timing was bad; they say they found the film tiring.

Subtitles for the deaf and hard of hearing

A distinct discipline, often confused with the translation kind. These carry speaker identification, relevant sound effects and music cues, and are usually in the same language as the dialogue.

The judgement is about relevance: a door closing matters if someone left, and not otherwise. Done well it is close to invisible; done badly it either floods the screen or omits information a hearing viewer receives.

The conditions the work is done under

Worth knowing, because it explains variation in quality more than talent does. Subtitling is frequently paid per minute of runtime rather than per hour worked, which means careful compression of dense dialogue earns the same as straightforward material taking a fraction of the time.

Turnaround has also compressed, particularly for simultaneous global releases where a film must appear in dozens of languages on the same day. Some of that work is now machine-translated and post-edited rather than translated outright, and the results vary considerably with how much time the editing pass is given.

None of this is visible to a viewer, who simply experiences a track as good or bad. But when a subtitle track is noticeably poor, the cause is more often the schedule and the rate than the translator.

Watching with this in mind

Once you know a subtitle is a compression rather than a transcript, the mismatch stops reading as error. What you are seeing is a translator’s judgement about which meaning survives when only thirty characters fit — a genuine editorial decision, made hundreds of times per film, by someone whose name usually appears nowhere.

Sources

• BBC Subtitle Guidelines — the reading-speed and line-length limits that govern what a subtitle can contain, and the rules for subtitling for deaf and hard-of-hearing viewers
• Netflix, timed text style guide: general requirements — a streaming platform’s specification for character limits, duration and line treatment, applied across every language it ships
• Netflix, English (USA) timed text style guide — how one language’s rules differ from the general spec, which is why two subtitle tracks of the same film disagree
• W3C, Timed Text Markup Language (TTML2) — the specification that subtitle files are authored against, and which defines timing and positioning