How to Translate Swearing and Sensitive Language in Subtitles

Each language swears about different things

Swear words draw on whatever a culture treats as taboo, and that differs from place to place. English leans on sex, bodily functions and religion. Dutch has a well-known set of curses based on diseases. Québec French swears with words from Catholic church vocabulary, which carry little force in France. Japanese has relatively few words that work like English swear words; much of its rudeness comes from rude pronouns and dropped politeness, which a word-for-word translation can't see.

Even within one language, the same word changes weight by region. The Spanish verb "coger" is an everyday word for "take" or "catch" in Spain and a vulgar word in much of Latin America. A translation aimed at the wrong region can turn a harmless line into an obscene one, or the reverse. Regional variants are covered more broadly in translating video for regional audiences.

So a literal translation of a swear word is usually either far too strong, oddly mild, or meaningless. The working unit isn't the word. It's the function and the strength.

Matching strength and function

Before deciding how to translate a swear word, ask what it is doing in the sentence. The common functions are emphasis ("that was damn good"), emotion (frustration, pain, surprise), aggression directed at someone, humor, and solidarity between friends. The same word can do any of these, and the translation should do the same job at a similar strength.

A simple strength ladder helps reviewers stay consistent across a whole video:

Mild

Exclamations most audiences accept; translate with a mild local equivalent

Moderate

Clearly rude but common in casual speech; keep the rudeness, pick a natural local word

Strong

The words broadcasters bleep; match the force, and expect platform and audience rules to apply

Slurs

Not swearing at all but attacks on a group; handle separately with a deliberate policy

If the target language has no equally strong word for a given context, the usual move is to compensate: put the force elsewhere in the sentence, or use a strong word slightly earlier or later where it fits naturally. That kind of adaptation is a judgment call for a fluent person, part of the wider work described in cultural adaptation in video.

What happens to swearing in an AI pipeline

Strong language passes through two automatic stages before you see it, and either can change it.

Speech recognition comes first. Swearing is often muttered, shouted, overlapped or cut off, all of which make recognition harder, so a quiet curse may be transcribed as a harmless word or missed entirely. If the original video already has bleeps, the recognizer hears a tone rather than a word, and the line may come out incomplete. Whisper-style models also tend toward clean transcripts that drop some disfluencies, so if you need a record of exactly what was said, the differences between styles in verbatim vs clean verbatim transcription matter.

Translation comes second. Machine translation handles profanity inconsistently: it may translate literally, choose a softer word, choose a stronger one, or drop the word, and it may do different things in different parts of the same video. None of that follows a policy. Your review supplies one.

Choosing a policy: preserve, soften or remove

Decide before the review starts, and write it down, ideally in a translation style guide. Three questions settle most cases.

• Who is watching? Adults who chose a stand-up special expect the language intact. A corporate training audience or a mixed-age audience does not.
• Where is it published? Platforms have their own rules on strong language, and some apply them to monetization or age restrictions. Check the platform's current guidelines rather than relying on memory, because they change.
• What is the video for? Documentary, journalism, research and legal material generally need fidelity: softening a quote changes the record. Entertainment allows more adaptation. Marketing usually wants none at all.

Two considerations often get overlooked. Many subtitlers find that a word reads stronger on screen than it sounds in speech, so subtitles are often toned down slightly even when the policy is "preserve". And captions are also used by deaf and hard-of-hearing viewers, who should get the same content hearing viewers get; censoring subtitles while the audio stays uncensored gives them a different, poorer version.

Slurs, identity terms and other sensitive language

Slurs need their own policy because getting them wrong does real harm. When a speaker uses a slur in a documentary or interview, the usual choices are to translate it with a term of matching offensiveness, or to translate it neutrally and mark that the original was a slur, for example in square brackets. Either way, a fluent reviewer should make the call. Words reclaimed within a community can be affectionate between members and hateful from outsiders, and a translation can lose that distinction.

The opposite error is just as important. Machine translation sometimes chooses an outdated or offensive term where the speaker used a neutral one, especially for disability, ethnicity, religion, sexuality and gender identity. Ask reviewers to check every identity term against the respectful current usage in the target language and region. Other sensitive areas worth a pass include descriptions of violence, suicide and self-harm, where many publishers follow careful wording guidelines, and medical or legal claims.

Hearing it in a dubbed voice

In a dub, the synthetic voice speaks the translated text exactly as written, in the same tone it uses for everything else. A strong curse delivered in a calm narrator voice can sound comic or jarring, and an insult loses the anger that made it mean something. For heavily emotional or confrontational scenes, subtitles often serve better than a single-voice dub.

The voice track also has no bleeps unless you add them. If your policy is to bleep, you do it in an editor on the downloaded audio, at the same moments you censored in the subtitles. Keeping the subtitles and the audio consistent is the most common thing teams forget.

Editing the subtitle file before you publish

1. Make a search list for each language: the strong words, slurs and sensitive terms you expect, plus the source-language words to look for in the original transcript.
2. Search the source transcript first and note every timestamp where strong language occurs, including lines where recognition may have softened or missed a word.
3. Open the translated SRT and go to each timestamp. Compare the translated line with the original strength and function.
4. Apply your policy: keep, swap for a better local word, soften one step, mask with asterisks, or remove. Use one masking style throughout.
5. Save the file as UTF-8 so accented and non-Latin characters survive. The mechanics are covered in how to edit an SRT file.
6. If you also publish a dub, listen to the same timestamps in the translated audio, and bleep or replace lines in an editor so the voice matches the subtitles.
7. Have a fluent reviewer read the edited lines in context with the video before sign-off.

Pitfalls and risks

• Over-censoring changes characters. A furious person who speaks politely in the subtitles is a different person.
• Inconsistency across a video, such as a word translated strongly in minute 3 and softly in minute 30, undermines trust in the whole translation.
• Using the wrong regional variant turns an innocent word obscene, or the other way around.
• Subtitles and dubbed audio disagree, so viewers who use both see one thing and hear another.
• Broadcast standards and rules on offensive content differ by country and platform. If you publish in a regulated context, check the applicable rules or ask someone qualified; this isn't legal advice.

Strong language in mydubly outputs

mydubly has no profanity filter and no bleep option. It transcribes the speech it recognizes, translates it, and gives you files to edit: a timestamped transcript, subtitles as SRT and VTT in the chosen language, an optional translated transcript, and for dubbing, a translated voice track and video. Each subtitle job produces one track in one language, so you edit one file per language.

Because mydubly doesn't import an edited SRT, changes you make to the subtitle file don't flow into the voice track. Bleeping or replacing a line in the dub happens in your editor, using the downloaded M4A. For a 20-minute interview, subtitles in one language cost 20 credits (2¢) and a dubbed version costs 1,000 credits ($1.00). The video translator and subtitle generator pages describe both outputs.

Agree the policy before the first file

Write a three-line policy for each language, covering what to keep, what to soften and how to handle slurs, and give it to your reviewer with the SRT and the video. Applying it is then a fast, mechanical pass instead of a debate on every line. If you publish the subtitle file yourself, how to add subtitles to a video covers the upload step.