Subtitles, captions and translation workflows

Subtitles are usually the first and cheapest step into a new language market. A well-made subtitle track lets viewers who do not speak the original language follow a video, improves discovery because platforms can index the text, and serves deaf and hard-of-hearing viewers in the original language. A poor subtitle track does the opposite: it confuses viewers, misrepresents the creator and signals that the market is an afterthought. This lesson covers the difference between captions and subtitles, the technical formats, reading-speed and line-length conventions, and a translation workflow with glossaries, style guides and quality checks. CRE-410 covered the AI tools that transcribe and translate; this lesson focuses on the process and standards that make the output good.

Captions, subtitles and translated metadata

The terms are often used loosely, but the distinctions matter for planning. Captions transcribe the spoken audio in the same language and are designed primarily for viewers who are deaf or hard of hearing; good captions also describe relevant sounds, such as music or laughter, and identify speakers when needed. Subtitles translate the dialogue into another language for viewers who can hear the audio but do not understand it. Subtitles for the deaf and hard of hearing (SDH) combine both: a translation that also includes sound descriptions. Metadata translation covers titles, descriptions, chapter names and tags, which drive discovery in search and recommendations.

Translated metadata is often the highest-value, lowest-cost step. A title and description in the viewer's language can make the difference between a video being offered to a viewer at all and being ignored. Platforms such as YouTube let creators add titles and descriptions in many languages, shown automatically to viewers whose interface language matches. Where platforms auto-translate titles, creators can usually replace the machine version with their own, and should, because a machine-translated title can be awkward, misleading or unintentionally funny.

File formats and technical basics

Subtitle files are plain text with timing information. The two most common formats are SubRip (SRT) and WebVTT (VTT). Both list numbered or sequential cues, each with a start time, an end time and one or two lines of text. WebVTT adds options for positioning and styling and is the web standard used by HTML5 video. Most platforms accept both, and editing software can export either. Burned-in (open) subtitles are part of the video image and cannot be turned off; they suit short-form platforms where viewers watch without sound, but they lock a video to one language and are invisible to search indexing.

• Timing: cues should appear when speech starts and disappear shortly after it ends, without flashing on screen too briefly or lingering over the next shot.
• Line breaks: break lines at natural linguistic points, keeping phrases together, rather than wherever the character limit falls.
• Shot changes: avoid cues that cross a cut by a few frames, which viewers notice as flicker.
• Encoding: save files in UTF-8 so accented letters and non-Latin scripts display correctly.
• Positioning: move subtitles when they would cover on-screen text or faces, which WebVTT supports.

Reading speed and line length

Viewers need time to read. Professional subtitling guidelines express this as characters per second (CPS), counting all characters including spaces and punctuation. Common professional targets for adult content are around 15 to 20 characters per second, with lower limits for children's content; many major streaming services publish style guides with their own limits for each language, and platforms that serve children recommend slower speeds. Line length is typically limited to about 37 to 42 characters for Latin scripts, with no more than two lines per cue. These are conventions, not laws, but following them makes subtitles comfortable to read.

Condensing is the skill that separates professional subtitles from transcripts. Fast speakers can produce 180 words a minute or more, which no one can read in full. A good subtitler keeps the meaning, tone and key words while dropping repetitions, fillers and redundant phrases.

A translation workflow

A dependable workflow has five stages, whether the work is done by people, machines or both.

• Source transcript: start from an accurate same-language transcript, corrected by a human, with speaker names, timing and terminology checked. Errors in the source multiply in every translation.
• Translation: translate from the corrected transcript, either by a professional translator, by machine translation followed by human post-editing, or by a bilingual team member.
• Glossary and style guide: give translators a list of key terms and how to render them, plus guidance on tone, formality, names, units and recurring jokes.
• Review: a second fluent person checks meaning, tone and timing in the video, not just in the text file. Watching in context catches errors that reading misses.
• Publish and monitor: upload the files and translated metadata, then watch comments from the target market for confusion or corrections.

Machine translation has become very good for many language pairs, but it still makes errors with idioms, humour, technical terms, names and context. Machine translation plus human post-editing is now common and cost-effective; fully unreviewed machine output is risky for anything that represents your brand. As CRE-410 noted, review effort should match the stakes: a casual vlog subtitle may need a light check, while a sponsored segment, a health claim or a safety instruction needs careful review because errors there can mislead viewers or breach advertising rules in the target market.

Glossaries and style guides

A glossary lists the terms that recur in your content, with an approved translation for each and notes on when not to translate. It typically includes your channel name and catchphrases, product names, technical terms in your niche, names of recurring people, and terms that must stay in the original language. A style guide records decisions about tone, formality, regional variety, numbers and units, and how to treat jokes and wordplay.

Regional variety is a real choice. Spanish in Mexico, Argentina and Spain differs in vocabulary, slang and formality; Portuguese in Brazil and Portugal differs noticeably; French in Quebec and France differs in vocabulary and idiom. Choose the variety that matches your primary market, or a neutral variety if you serve several, and record the choice so every translator follows it.

Costs and budgeting

Subtitle costs depend on language, turnaround, quality and method. Human transcription and translation are often priced per minute of video; machine translation with human post-editing costs less; fully automated tracks cost almost nothing but carry quality risk. Prices vary widely by provider and language, so the figures below are illustrative only.

Accessibility as well as reach

Accurate same-language captions are an accessibility measure as well as a growth tactic. Deaf and hard-of-hearing viewers, viewers with auditory processing differences, people learning the language and people watching in quiet places all rely on them. In some contexts captions are a legal requirement, for example for broadcasters and for some public-sector and business content under accessibility laws, as CRE-450 described. Even where they are not required, good captions widen the audience at home and make translation easier, because they provide the corrected source transcript every translation depends on.

Measuring subtitle quality

Quality is easier to manage when it is measured. A simple scorecard for each subtitled video can record the number of meaning errors (mistranslations that change what the speaker said), the number of style errors (awkward phrasing, wrong formality or inconsistent terms), the share of cues above the reading-speed target, and the number of timing problems noticed in review. Tracking these numbers over a few months shows whether a translator, a machine-translation setup or a post-editing workflow is improving or slipping, and gives concrete feedback rather than vague impressions. Many creators set a simple rule, such as no meaning errors in sponsored segments and fewer than one style error per minute elsewhere.