When companies localize video for global audiences, subtitles are usually the first item on the list. The job looks simple: translate the source captions and add them to the video. In practice, the same message takes a different length, word order, and reading pace in every language. Subtitles can overflow the safe area, disappear before viewers finish reading, or describe a step that is no longer on screen. This is what sets subtitle translation apart from document translation.
A translated manual can always run onto another page. A subtitle cannot. It has a fixed amount of screen space and a fixed window of time, so a translation that is accurate on paper can still fail in the video.
Automatic speech recognition and machine translation can now produce a first draft of subtitles in minutes, and many teams already use them. What these tools do not settle on their own is where a sentence should break, what can be shortened without dropping a safety condition, and whether a line appears at the moment a button is pressed on screen. Those decisions determine whether a subtitle actually works.
This article walks through those decisions step by step and explains how a subtitle team with the right experience and tooling handles each one.
The Three Constraints of Subtitle Translation
Subtitle translation is defined by three constraints.
- Space
Subtitles should occupy only a small part of the frame, usually no more than two lines. If viewers have to concentrate on reading instead of watching, the subtitle is no longer doing its job. - Time
Each subtitle stays on screen for a limited time. However accurate the sentence is, it fails if viewers cannot read it within that window. - Picture
Subtitles have to follow the visuals as well as the audio. When the explanation drifts away from the moment a button is pressed or a graphic appears, viewers lose track of what is happening.
Because of these constraints, an accurate translation and a usable subtitle are two different things. The translation has to be adjusted to fit within these limits while keeping its meaning, and the quality of that adjustment determines the quality of the localized video.
How a Single Subtitle Line Gets Made
Multilingual subtitle production generally follows the sequence below. For each step, we look at the decision involved and how to manage it systematically.
1. Timecoding and the Master Template
Each line of the transcribed source dialogue receives in and out timecodes, and the text is divided into subtitle events that follow shot changes and the pacing of the video. The finalized source subtitle file becomes the master template for every target language. Once the template is locked, timecoding is done only once regardless of how many languages follow, and every language works from the same reference timecodes.
If the script is translated into each language first and subtitles are created afterward, each language can end up with its own set of timecodes. Decisions are then scattered across languages, and comparing the output against a common source becomes difficult.
When the master template is created after picture lock and all language work is based on it, decisions about segmentation, line breaks, and length can be managed against one consistent baseline.
2. Reading Speed and Line Length
For comfortable reading, the number of characters has to stay in proportion to how long the subtitle is on screen. For Latin-script languages, widely used guidelines set around 17 to 20 characters per second (CPS), 42 characters per line, and a maximum of two lines. Netflix’s English style guide, for example, allows 42 characters per line and 20 CPS for adult programs. Chinese, Japanese, and Korean use separate limits that reflect their writing systems. Actual values vary by platform and by project style guide.
The efficient approach is to agree on readability limits before work starts and then verify automatically that every subtitle meets them. When per-language CPS and line-length thresholds are registered as QA rules, the entire file can be checked in one pass and every violation flagged for revision. The same standard holds even when a project runs to thousands of subtitles.
3. Segmentation: Where to Split a Sentence
When a sentence does not fit into a single cue (one subtitle displayed under one timecode), the subtitler has to decide where to split it. A split that separates grammatically linked words forces viewers to hold the first half in mind while they wait for the second. Splits should follow units of meaning and grammatical structure.
This matters even more in training and procedural videos. If the description of one step spills into the next subtitle, viewers find it harder to follow the procedure.
Rather than keeping the source split points mechanically, each language should be segmented according to its own word order and sentence structure. When segmentation rules are managed in tool settings and rule files as well as in written guidelines, results stay consistent even when linguists change from one project to the next.
4. Condensation: Shortening Without Losing Meaning
When a translation exceeds the reading-speed limit, it has to be shortened while keeping its core meaning. This is called condensation, and it is one of the clearest differences between general translation and subtitle translation. It is the step that turns an accurate translation into a subtitle people can actually read.
Technical video does not allow the freedom of film subtitling, however. Product names, UI labels, numbers, procedural conditions, and warnings have to survive intact. If a single condition drops out of an installation step, a viewer may perform the wrong action.
The team should therefore decide in advance which elements may be condensed and which must stay, and then confirm that nothing critical was lost. Product names and UI labels can be checked automatically against the termbase, and numbers and units can be checked against the source with pattern rules. Organizations that already localize technical documentation can put their existing termbases and translation assets to work in this review.
5. Line Breaks: Where to Split Two Lines
Where a subtitle breaks across two lines changes how easily it can be read. The two subtitles below say the same thing, but the second one splits a UI label in the middle and makes the eye stop.
Case 1
Select the Network Settings
menu from the Home screen.
Case 2
Select the Network
Settings menu from the Home screen.
Line-break rules can be applied for each language and run as automated checks. Languages that do not separate words with spaces, such as Japanese and Chinese, need rules of their own. Automatically flagging subtitles that break at prohibited positions keeps quality consistent as the number of languages grows.
6. Timing: Matching Subtitles to the Picture
A subtitle that appears too early is as much of a problem as one that appears too late. In videos where an explanation, an on-screen action, and its result follow one another, timing directly affects comprehension. If the subtitle is already showing the next step at the moment the user on screen presses the Settings button, viewers will be confused.
After translation, the in and out times for each language should be checked again against the picture. Because condensation and segmentation differ from language to language, sharing the same timecodes does not guarantee sync in every language.
Issues that can be judged by numbers, such as subtitles shorter than the minimum duration, gaps between subtitles that are too small, and subtitles that cross a shot change, are best caught by automated checks first. Reviewers can then focus on what only playback reveals, such as whether each subtitle lines up with the action on screen.
Why the Adjustment Effort Differs by Language
Even with the same video and the same number of subtitles, the amount of additional work varies by language. Text expansion is the factor most often mentioned, but several others affect the real workload.
First, sentence length changes with the language. Translations from English into German, French, or Spanish often run longer, while Chinese, Japanese, and Korean are more compact in character count. When text grows against fixed timecodes, more condensation and timing adjustment may be needed.
Second, language pairs that differ widely in word order, word boundaries, or line-breaking conventions need extra attention. Keeping the source split points may produce unnatural sentences, and the segmentation has to be redone.
Third, some languages build sentences in an order that conflicts with the sequence in which UI elements appear on screen. When the order of actions on screen does not match the structure of the subtitle, the sentence has to be rebuilt.
This is why differences in quotes and schedules between languages cannot be explained by per-word rates alone. Knowing in advance which adjustments each language needs makes the workload easier to predict. A record of past expansion rates and rule-violation rates by language can be used to estimate the effort for a new video from its subtitle count and average line length.
Subtitles Need Two Rounds of Review
Subtitle review is not a single step. Linguistic QA, which checks accuracy and terminology in the subtitle file, and in-video QA, which checks the finished video during playback, catch different kinds of errors.
Some questions cannot be answered from the file alone. Do subtitles flash by too quickly? Do they cover important elements on screen? Does the explanation follow the on-screen action in the right order? Do the UI labels in the subtitles match what the screen actually shows?
The two reviews should therefore be separate steps with clearly defined roles. Rule-based issues such as reading-speed violations, line-length overruns, terminology mismatches, and number or unit mismatches can be caught by automated checks first. In-video QA can then concentrate on how the subtitles relate to the picture. A review limited to the subtitle file can miss problems that appear only in the video, so it is best to define the scope of in-video review up front, based on the purpose and quality requirements of the project.
What Technical Video Adds to the Checklist
Terminology Consistency
Product names, UI strings, and feature names in subtitles have to follow the same standard as the actual product screens and manuals. If the video uses a different term, viewers may not be able to find the item on screen. Processing subtitle text with existing translation memories (TMs) and termbases helps maintain this consistency. A workflow that converts subtitle files into a translation format, applies existing assets, and converts them back into a subtitle format links subtitle work to the assets built for technical documentation localization.
On-screen Text
Text already embedded in the video, such as slides, lower thirds, and UI captures, has to be managed separately from dialogue subtitles. It may need to be repositioned so it does not overlap with subtitles, or the video may need to be re-rendered. In product demos and training videos, localizing on-screen text can be a significant part of the work.
Accessibility: SDH and Closed Captions
Subtitles for the deaf and hard of hearing (SDH) and closed captions (CC) include speaker identification, sound effects, and descriptions of background audio, which makes them a separate deliverable. A request to “add English subtitles” can mean either plain English subtitles or SDH, and the scope and deliverables change accordingly. In the United States, captions for prerecorded video are also part of the Web Content Accessibility Guidelines (WCAG), which Section 508 and ADA compliance efforts reference, so it is worth confirming early which type the project needs.
Subtitle Formats
SRT is universally supported but carries limited formatting. WebVTT is used for web players, TTML/DFXP for OTT and broadcast, and ASS/SSA when position and styling have to be specified. The required format depends on where the video will be published, whether that is an LMS, YouTube, or a streaming platform, so confirm the final format when placing the order.
Questions to Ask a Subtitle Localization Vendor
When you evaluate subtitle localization vendors, the following questions help clarify the scope of work and how quality is controlled.
- How do you manage reading-speed and line-length limits? Ask which standards are applied, who reviews the output, and how automated checks are combined with human review.
- Which languages need the most adjustment, and why? Ask how differences in workload by language are estimated and reflected in the schedule and the quote.
- Do you review the finished video during playback, or only the text in the subtitle file? Confirm whether both reviews are included and what each one covers.
- Do you use our existing TMs and termbases? For projects where terms must match the product UI or manuals, and for training videos where term definitions matter, a link to existing translation assets is essential.
- How do you use AI in the workflow? Ask where speech recognition and machine translation are used and which steps are reviewed by people.
What Makes a Good Subtitle
A good subtitle goes beyond an accurate translation. It lets viewers follow the video without being aware that they are reading a translation. In product training and procedural videos especially, subtitles are part of the learning experience. Translation accuracy has to be managed together with reading speed, segmentation, and sync with the on-screen action for the video to deliver the information it was designed to deliver.
Hansem Global localizes subtitles so that the terminology in manuals and training videos is managed against a single standard. Our in-house subtitle platform pairs AI speech recognition and machine translation with automated checks for per-language readability limits and terminology consistency, while our linguists make the final calls on condensation and timing. Subtitle files are converted into a translation format so that existing TMs and termbases can be applied.
If you are evaluating manual development or translation and localization services, please get in touch. A specialist for each service will walk you through the scope and process your project needs.