Behind the scenes of subtitle translation

Subtitling is one of the least visible crafts in media production. Translation subtitling is even less visible. Most viewers never think about the person who converted dialogue from one language to another, working within constraints that make the job much harder than it looks.

This is about the work.

The job

A subtitle translator receives a source subtitle file — dialogue with timings — and produces a target-language subtitle file that fits the same timing constraints. It is not a translation task that happens to be subtitle-shaped. It is a subtitle task that happens to involve translation.

The translator must:

• Translate the dialogue accurately.
• Fit it within the same cue timings as the source.
• Keep line length within the target language's guidelines.
• Keep reading speed within the target language's targets.
• Preserve speaker labels, formatting and tag structure.
• Adapt cultural references and wordplay to the target audience.
• Maintain consistency across cues (character names, terminology).

That is six or seven constraints that must all be satisfied simultaneously. Anyone who has translated film knows the feeling of needing to lose a crucial word to fit the line.

The invisible negotiations

Every translated cue involves negotiation. The dialogue in the source language takes some amount of time and conveys some amount of meaning. The translated dialogue must fit in the same amount of time and convey enough meaning that the viewer understands what is happening.

Three variables: time, characters, meaning. Only two can be pushed.

• Increase time — not possible in most cases, because the timing is fixed by the audio.
• Increase characters — not possible beyond the reading-speed limit.
• Decrease meaning — possible but distorts the story.

The translator's skill is to find the point where these three constraints converge: the shortest, clearest translation that preserves the essential meaning.

Some cues have room. A five-second cue with a short line leaves time and space for expansion. Other cues have no room at all — a dense, fast-paced exchange where every character counts.

The things that do not translate

Some things resist translation:

Wordplay and puns

A pun works because of the double meaning of a word or phrase. That double meaning almost never survives translation. The translator has two choices: translate the joke literally (losing the wordplay) or invent a new joke in the target language (possibly losing the meaning). Neither is perfect.

Cultural references

A reference to a specific national TV show, historical event or celebrity may not carry meaning in another culture. The translator either adds a brief explanation (adding length) or drops the reference (losing meaning).

Slang and idioms

Idioms rarely translate word for word. The translator must find an equivalent expression in the target language. If none exists, they must translate the meaning — which usually requires more words than the original.

Register and formality

Some languages have formal and informal pronouns (tu/vous in French, tú/usted in Spanish, ты/вы in Russian). Dialogue in the source language may not signal register clearly. The translator must infer register from context — who is speaking to whom, with what relationship — and choose correctly.

Rhythm and mouth shape

For dubbing, the translation must match the on-screen actor's mouth shape. For subtitling, this matters less, but the translator still tries to preserve the sense of speech rhythm, which is not always the same as word-for-word translation.

When machine translation is enough

Machine translation has improved dramatically. For certain content, machine output is now serviceable:

• Technical content with standard vocabulary.
• News and informational content where literal accuracy matters more than style.
• Content where the audience speaks both languages and can fill gaps.

But for entertainment content — dramas, comedies, documentaries with strong voice — machine translation is a starting point, not a final deliverable.

When human translation is essential

• Comedy. Machine translation rarely handles jokes. Wordplay, irony and timing collapse.
• Dramatic dialogue. Tone, register and subtext require human interpretation.
• Content with cultural references. Localisation of cultural touchpoints is human work.
• Languages with limited training data. Machine translation quality drops sharply for less-resourced language pairs.
• Accessibility content. SDH subtitles with sound effects require understanding the visual context, which machine translation cannot do.

The economics

Subtitle translation is not a high-margin business. Rates vary enormously by market. Professional subtitle translators in high-income markets charge $2–5 per subtitle minute. In lower-income markets, rates can be 10–20% of that.

Streaming platforms have driven down per-minute rates by demanding volume. They have also driven up quality requirements by publishing detailed style guides and QC processes. The result is pressure on translators to work faster and more accurately at the same time.

Machine translation has reduced rates further. Post-editing machine translation — where a translator corrects AI output rather than translating from scratch — is now a common workflow. The rates are lower but the work is faster, and the overall economics for translators are mixed.

The invisibility problem

Subtitle translators are almost never credited in the film. Their names appear in subtitle file metadata sometimes, in streaming platform credits rarely, in printed programs almost never. The viewer experiences the translator's work without knowing it exists.

That invisibility is, in one sense, the point. Good subtitle translation disappears. The viewer does not think "this is a translation" — they think "this is the film". But it also means the craft is undervalued, and the people who do it are rarely recognized.

What the work actually looks like

A translator working on a two-hour film does not simply translate two hours of dialogue. They:

• Read the source subtitle file end to end to understand the film.
• Research character names, terminology and cultural references.
• Translate cue by cue, in context.
• Check timings and adjust where the target language needs different cue lengths.
• Review consistency — same names, same terms, same register.
• Watch the translated file against the video to see if it works.
• Fix what does not work.

A careful translator spends 2–4 hours of work per hour of finished film. Fast translators can do 1–2 hours per finished hour, with lower quality.

The tools that respect the craft

The best subtitle tools respect the translator's judgment. They do not try to translate for them. They handle the mechanical parts — file format conversion, cue splitting, character counting, validation — so the translator can focus on the linguistic work.

That is what the tools on this site try to do. They are not translation tools. They are the practical, format-handling infrastructure that makes translation work easier.

And that is why the invisibility of subtitle translation is a compliment. When the work is done well, nobody notices. They just watch the film, and understand it.