Translating Subtitles Without Losing Timing or Meaning

You finished an English caption file, ran it through a translation tool, and now the Spanish lines flash past too fast to read and break in the middle of phrases. Translating subtitles is not the same job as translating a document, and treating it like one produces files that are technically accurate and practically unwatchable.

I have managed caption and subtitle work for archive projects in six languages, and nearly every problem I see comes from the same place: the translation was done line by line, with no thought for reading speed, line length or the fact that languages expand and contract. Here is the workflow that fixes it.

Why translating subtitles is harder than translating text

A subtitle is a timed unit with hard limits. A typical cue holds two lines of about 42 characters each and stays on screen for one to seven seconds. The translation has to fit inside those limits while still matching the speech the viewer hears.

Languages do not cooperate. English text translated into German or Spanish commonly runs 20 to 30 percent longer. Translated into Chinese or Japanese, the character count drops sharply, but each character carries more reading load, so the limits change rather than disappear. A line that fits comfortably in English can become a three line block in German that nobody can finish reading before it vanishes.

Then there is meaning. Spoken dialogue is full of idioms, half sentences and jokes that depend on timing. A literal translation of "break a leg" helps no one, and a punchline that appears two cues before the speaker delivers it ruins the moment.

Start from the right source file

Before you translate anything, clean the source. If the English captions came from automatic speech recognition, fix them first. Translating errors multiplies them, and a misheard name becomes a misheard name in five languages. Our guide to fixing auto generated captions before you publish covers that pass.

Next, decide what you are translating from. Captions for deaf and hard of hearing viewers include sound descriptions like [door slams] and speaker labels. Subtitles for foreign language viewers usually drop most sound tags because the audience can hear. If you need both, create a dedicated subtitle master in the source language first, with sound tags removed and lines condensed where possible, then translate that.

Finally, keep the format portable. SRT and WebVTT are the safest choices for translation work because almost every tool reads them. If you are unsure which to standardise on, read about choosing a caption file format that survives re-uploads.

A step by step workflow for translating subtitles

1. Lock the timing. Finalise the source cue timings before translation begins. Changing timings later means every language file needs the same fix.
2. Write a short glossary. List names, product terms, place names and recurring phrases with their approved translations. Ten to thirty entries is normal for a one hour programme.
3. Translate with the video playing. The translator needs to hear tone, see who is speaking and notice visual jokes. Translating a bare text file loses all of that.
4. Condense, then translate. If a source line is already long, cut it to its meaning before translating. "Well, I mean, I think we should probably go" becomes "I think we should go."
5. Check reading speed per cue. Most subtitle editors show characters per second. Aim for around 15 to 17 for adult viewers and 12 to 13 for children's content.
6. Re-break the lines. Break at natural phrase boundaries in the target language, not where the source broke.
7. Review with a fresh viewer. A native speaker who has not seen the script watches the full video once. Every place they pause or rewind gets flagged.

Reading speed and line length targets by language

Reading speed rules apply in every language, but the numbers shift. These are the working targets I use, and they line up with our piece on caption reading speed and line length rules.

When a translated cue goes over the limit, you have three honest options: shorten the wording, extend the cue into a gap of silence, or split it into two cues. Shrinking the font is not one of them.

The advice I disagree with: keep it word for word

Clients often ask for subtitles that are "as faithful as possible," meaning every word translated. I push back every time. A viewer reading subtitles is also watching faces, action and on screen text. If you ask them to read every filler word, they miss the picture, and that is a worse loss of meaning than trimming an "actually" or a repeated "you know."

The professional standard is to translate meaning and intent, not word count. Cutting 20 to 30 percent of spoken words is normal for subtitles. The exception is legal, medical or testimonial content where exact wording matters. There, extend timings and split cues rather than cutting.

Machine translation as a first draft only

Machine translation tools are now good enough to save real time on a first pass, especially for European language pairs. On a 45 minute interview, a decent draft can cut a translator's work from roughly six hours to three. But the output needs a full human edit. Tools still mistranslate idioms, flatten formal and informal address, and ignore line breaks entirely.

Budget accordingly. Professional subtitle translation commonly runs $5 to $15 per video minute for common language pairs, and post editing a machine draft usually lands at the lower end. Never publish a raw machine file as a finished subtitle track. A single wrong word in a name or a number can undo the trust your whole channel has built with a new language audience, and fixing it after upload means re-exporting and replacing every copy.

Frequently asked questions

Can I use automatic translation for YouTube subtitles?

You can, but treat it as a draft. Automatic translation ignores reading speed and line breaks, and it often mistranslates names and idioms. Download the translated file, edit it against the video, then upload the corrected version as a separate track.

How many characters should a translated subtitle line have?

For most languages using the Latin alphabet, keep each line to about 42 characters and no more than two lines per cue. Japanese and Chinese use much shorter lines, usually 13 to 18 characters, because each character carries more meaning.

Should subtitles match the audio exactly?

Not for foreign language subtitles. They should match the meaning and timing of the speech, and condensing is expected so viewers can read and still watch the picture. Same language captions for accessibility should stay much closer to verbatim.