When translations grow or shrink: what text expansion does to subtitles and dubbed audio

Why translations change length

Languages encode the same meaning with different machinery. English is relatively compact: it drops articles in many places, stacks nouns ("user account settings page") and uses short function words. Romance languages often need prepositions and articles where English uses word order, so "user account settings" becomes Spanish "configuración de la cuenta de usuario". German packs nouns into compounds ("Benutzerkontoeinstellungen"), which cuts the word count but not the character count, and its formal address adds words ("Sie", "Ihre") that English expresses with nothing.

Translation itself adds length too. Translators, human or machine, tend to make implicit things explicit: a vague English "it" becomes a named noun, a compressed phrase becomes a clause. Going the other way, from a wordier language into English, translations often contract. And languages written in Chinese characters or Japanese kana and kanji use far fewer characters for the same content, because each character carries a syllable or a whole morpheme.

Three kinds of length: characters, words and speaking time

"Longer" means different things depending on the medium, and mixing them up leads to wrong conclusions.

• Characters matter for subtitles and on-screen layouts, because space and reading speed are measured in characters.
• Words matter mainly for pricing in traditional translation, and are a poor guide across languages with compounds or no spaces.
• Speaking time matters for dubbing and voice-over, because the translated speech has to fit the original's timing.

These do not move together. A Chinese sentence can have a quarter of the English character count yet take a similar time to say, because each character is a spoken syllable. Languages also differ in how fast they are typically spoken and how much each syllable carries; fast-sounding languages tend to pack less information into each syllable, so spoken durations differ less than character counts suggest. They still differ, and for dubbing that remaining difference is what has to be absorbed.

Which directions tend to expand and which contract

Rather than quote percentages, which vary wildly with the text, here is one short interface sentence in several languages with its character count:

English (44 characters)

Save your changes before closing the window.

Spanish (46)

Guarde los cambios antes de cerrar la ventana.

French (57)

Enregistrez vos modifications avant de fermer la fenêtre.

German (63)

Speichern Sie Ihre Änderungen, bevor Sie das Fenster schließen.

Japanese (23)

ウィンドウを閉じる前に変更を保存してください。

Chinese (11)

关闭窗口前请保存更改。

The pattern is typical but not a rule. Translations from English into German, French, Russian or Polish usually grow, Spanish and Portuguese often grow a little, and Chinese, Japanese and Korean shrink sharply in characters. The Japanese line, though, is long to say: roughly "uindō o tojiru mae ni henkō o hozon shite kudasai", with its polite ending adding syllables English doesn't have. Short sentences vary most; over a whole script the differences even out somewhat.

What expansion does to subtitles

Subtitle timing comes from the original speech: each subtitle appears when the line starts and disappears when it ends. Professional subtitle guidelines cap reading speed in characters per second and limit each subtitle to two lines of around forty Latin characters, with different limits for Chinese, Japanese and Korean; check the style guide for your platform or client.

A translated line that grows has the same time on screen but more text, so reading speed rises. Human subtitlers deal with this by condensing: they cut redundancy and paraphrase so the meaning fits the time. Machine translation translates fully and does not condense, so translated subtitles in expanding languages often run faster than comfortable, especially for fast speakers. The fix is editing, covered in the post-editing guide, or choosing a dubbed version for content where viewers can't keep up. The trade-offs between the two are laid out in subtitles vs dubbing.

What expansion does to dubbing timing

In dubbing, the original speaker's timing defines a slot for each line. If the translated line takes longer to say, something has to give. There are only a few levers: start a little early into the preceding pause, run a little late into the following one, speak faster, or shorten the text. Each has a cost. Overlapping too far makes lines collide with the next speaker or drift out of step with the picture; speeding up too much sounds rushed and robotic; shortening requires editing the translation.

Contraction is the easier case. A line that is too short can be slowed slightly or left with a little extra silence, which viewers rarely notice. That is why English to Chinese dubbing tends to fit comfortably while English to German needs more adjustment.

How mydubly fits translated speech into the original timing

mydubly's AI dubbing step is built around exactly this problem. Before synthesis, short translated fragments are merged into whole sentences of up to about 220 characters, because the voice model, Chatterbox Multilingual by default, sounds more natural on complete sentences. Each generated clip is normalized (mono, 24 kHz, edge silence trimmed, consistent loudness) and then placed elastically against the original timing.

• A line may start up to 0.3 seconds early or run up to 0.6 seconds late.
• Speeding up by up to about 1.08 times is treated as inaudible.
• Beyond that, tempo rises up to 1.15 times by default, with a little extra allowed only when a chunk cannot otherwise fit.
• Lines that come out too short are slowed slightly, never below 0.9 times.
• The chunks are then joined gaplessly into one AAC track that matches the video's length.

Where elastic timing helps, and its limits

Elastic placement absorbs the moderate expansion that most language pairs produce, and contraction costs almost nothing. It has limits all the same:

• Fast speakers with no pauses leave no room to start early or run late, so expansion turns directly into speed-up.
• The picture is never altered and there is no lip-sync, so on-camera speakers' mouths won't match any language; timing fit keeps lines aligned with when people speak, not with lip movements.
• One chosen voice reads the whole video, and the original voices are removed rather than kept underneath; only the music and effects stay, lowered while the new voice speaks.
• Subtitles are timed to the original lines and are not condensed automatically, so reading speed in expanding languages still needs a human check.
• On-screen text in the picture is not translated, so expansion in labels and slides is outside this process entirely.

Writing source content that survives expansion

If you are scripting or recording content you plan to translate, a few habits leave room for longer languages:

1. Leave a short natural pause between sentences; that pause is where expanded lines borrow time.
2. Prefer two short sentences to one long, clause-heavy one.
3. Slow down slightly for dense instructions; rushed source speech leaves no slack in any language.
4. Avoid packing a key instruction into the last second before a cut or a music sting.
5. After translating, read the subtitles in the target language at playback speed and condense any line you can't finish before it disappears.

Next step

To hear how your language pair behaves, dub a short clip with AI dubbing into one expanding language, such as English to German, and one contracting one, such as English to Chinese. A ten-minute video costs 500 credits (50 cents) per language, and listening to both side by side shows exactly where expansion bites.