A compliance course is built in English, approved internally, ready to launch. Then the requirement changes: the same course must now reach employees across Europe, Asia and Latin America. It sounds simple — translate and publish. Then text expands and breaks the layout, the voice-over drifts out of step with the animation, and examples that felt natural in one region feel foreign in another. What began as a translation job has become something larger. This guide explains what that larger thing is.
Translation aims for a near word-for-word reflection in the new language. It builds basic comprehension.
Localization takes a specific place into account. Not “Spanish” but Mexican, Colombian or Chilean Spanish — each with its own forms of address and register. Speakers understand each other, but the difference is felt, the way an unfamiliar regional accent takes a moment of extra effort.
Transcreation takes the spirit of the content and rebuilds it for the audience — judging how much cultural context to add, which terms need explaining, and what the easiest path to understanding actually is.
A brochure is read once. A course is worked through — the learner clicks, answers, fails, retries. Every word carries a task, not only a meaning. Translate the words but miss the task, and the course still runs; the learner just stops learning.
AI translation works well on short, unambiguous text — two or three sentences, clear instructions, low context. Good prompting genuinely improves the output. It is still an automatic translation.
It struggles wherever understanding of context matters: technical subjects, regulated language, anything where a phrase that must not be said can slip through despite instructions.
The working rule in practice is AI-assisted, human-finished. A native-speaking editor makes the last pass, checking not only grammar but whether the teaching lands for that audience. And for headline or marketing copy, what looks like translation is really copywriting — AI can produce a shortlist, but a human still chooses.
A typical course is built so the learner must read it, see it, and hear it. All three need work.
Slides, labels, instructions, buttons, player text.
Reading speed matters more than word count.
Re-recorded, dubbed, or synthesised.
Text burned into images; characters, dress, setting.
Distractors must stay equally plausible, or the question gives away free marks.
Menus, progress messages, certificates, completion emails.
Visuals carry as much as words. A hero image showing one kind of family, or statistics that only apply in one country, tells a learner elsewhere that the course was not made for them.
Audience is not only language, either. A sales team, a frontline worker, a manager and a new hire meet the same training differently. Examples and tone should reflect each group’s actual responsibilities — while the learning objectives stay fixed.
Much of the cost of localization is decided before a single word is translated.
Two linguistic assets do the quiet work across a whole programme. Translation memory reuses what has already been translated, so cost and turnaround fall as the library grows. A termbase locks product names, technical terms and preferred phrasing so they stay identical in module one and module fifty. Without them, every course starts from zero and every translator makes their own decisions.
Check the source for idioms, slang, humour, cultural assumptions and region-specific examples. The cheapest moment to change anything.
Pull text out of images. Allow room for expansion. Set text boxes to shrink or resize on overflow. Choose fonts covering the target scripts.
Written in the target language first, then set beside English for review. It fixes the form of address, the terminology, what is in bounds and what is out. The best clients mark it up and say no, this is how we do it.
Matching the level to the content.
Native-speaker validators check meaning in context, seeing how each correction will look on screen. Two rounds is a reasonable norm.
Voice-over, generated speech, re-synced captions and video.
Translated text typically runs 20–30% longer than English; right-to-left scripts reverse the layout entirely. A DTP specialist works alongside the translator to find the wording that actually fits the page — left to one person, either meaning or layout always loses.
LMS, mobile, desktop — tested on the running course, not on the spreadsheet.
Roll out by priority. Afterwards, track completion by language and keep every version in step as the source changes. Version control is what stops the set drifting apart.
First decide whether you need voice at all. Voice-over is more immersive and carries far better in scenario-based training — and it is markedly more expensive, needing scripting, casting, recording and re-timing against the animation. Subtitling costs a fraction of it, but competes for the learner’s eye with whatever is on screen. Many programmes settle it per course: voice-over for flagship and customer-facing material, subtitles for the long tail.
Reproduces the original voice and pace in the new language. It protects objects already timed to the audio — but the voice may speed up or slow down unnaturally.
Builds a fresh voice from the translated script. You gain control of the wording; you lose the timing, and every timed object must be re-synced.
Whichever route, supply a pronunciation guide for product names, acronyms and technical terms. Its absence is one of the most common causes of re-recording. Voice-over is the most expensive element to change — which is why the source script is frozen before recording begins.
Subtitles are no longer optional in many markets. They are how e-learning meets accessibility standards such as the European Accessibility Act, and the audience is far wider than the requirement suggests: a large majority of viewers now use captions by choice. Subtitled content is watched for longer and understood better by non-native speakers — the same people a localized course exists for.
Authoring tools have moved past the old XLIFF round-trip of exporting text, translating it, and re-importing a stack of separate files. Current localization features translate on-screen text and subtitles in one pass, hold every language inside a single project file, flag layout issues where text has overflowed into scroll bars, push source-language edits out to all child languages, and send review links to validators who need no account.
Published as one SCORM package, the learner chooses their language — and progress is tracked as a whole rather than per language, so a learner who switches mid-course keeps their completion.
What these features still do not handle: audio narration, video, text burned into images, and attachments. That is where the budget goes.
Text baked into an image cannot be edited by a translator. You are left choosing between redesigning the graphic or shipping a page with one language stranded inside another. Learners read that correctly: as a sign nobody cared enough to finish.
Machine output with no human validator, a translator who is not a native expert, no termbase behind them. The saving is real and short-lived — poorly translated courses are the ones learners abandon, and an abandoned course wastes the entire cost of building it.
The cheapest mistake to prevent, and the most visible one when it survives to launch.
Two more worth naming: starting before the source content is final, and sending voice talent to the studio without a pronunciation guide.
At scale the problem changes shape. Localizing one course is a task; localizing a hundred is a system. What breaks first is rarely the language — it is review bottlenecks, version control, timelines and stakeholder coordination.
The packaging standard that lets a course talk to a learning management system — reporting who started, who finished, and what they scored.
A newer standard that records learning activity beyond the LMS, including offline and on-the-job actions.
Learning Management System — the platform that hosts courses, enrols learners and stores their records.
An exchange file format that carries translatable text out of an authoring tool and back in again.
Computer-assisted translation software — the translator’s working environment, holding memory, termbase and quality checks in one place.
A store of previously approved translations, reused automatically when the same sentence appears again.
An approved list of terms and their required translations, so product names and technical vocabulary never drift.
The growth in length when text is translated — commonly 20–30% for English into Thai or German.
Rebuilding a message for a new audience rather than translating it, keeping the intent instead of the wording.
Validation by a native speaker in the target market, checking that the content reads correctly in context.
In-house and partner native translators, ISO 9001 quality control, in-house DTP, and voice-over and subtitling in over 100 languages.
Global reach, local voice.
CMYK, bleed, and binding — the language of print, in plain terms.
Read the guide → Topic 02Page layout, typography, and the craft of multilingual documents.
Read the guide → Topic 04Manuals people can actually follow — with zero confusion.
Read the guide → Topic 05Layers, dielines, and finishes — three jobs in one box.
Read the guide → Topic 06Design thinking, the Double Diamond, and how projects run.
Read the guide → Topic 07Warehouse, assembly, and delivery — from press to doorstep.
Read the guide → Topic 03Going global by going local — the essentials, in plain terms.
Read the guide →