How do you translate e-learning courses and training modules correctly when the content is not just text but a sequence of instruction, narration, interactions, quizzes, buttons, media and LMS behaviour? The reliable method is to translate the learning path as a functioning system. Identify what the learner must understand or do at each stage, preserve the instructional objective, keep terminology stable, adapt narration and on-screen text to timing constraints, protect variables and technical tags, and test the localized module inside the actual learning platform. E-learning translation succeeds when the learner can complete the same learning journey and demonstrate the same competence in the target language.
People searching for e-learning localization, eLearning translation, training course translation, LMS localization, SCORM translation, xAPI localization, online course translation, multilingual training and localize training modules are dealing with a hybrid task. The language may live in slides, voice-over scripts, subtitles, interactive scenarios, branching decisions, knowledge checks, downloadable job aids and software packages that report completion or scores to a learning management system.
This guide explains how to localize e-learning without breaking the instruction, timing or LMS functionality. It covers first principles, learning objectives, course architecture, terminology, narration, subtitles, interactive buttons, branching, quizzes, feedback, SCORM and xAPI packages, screenshots, downloadable resources, accessibility, cultural adaptation, quality assurance, testing and release. The central idea is simple: the translated course must teach, not merely read well.
The visible problem: a course is a sequence of learning decisions
An ordinary document can sometimes survive small inconsistencies because the reader can reread. An e-learning course behaves more like an interactive route. The learner receives an explanation, sees an example, makes a decision, receives feedback, completes an activity and moves to the next concept. Translation can damage that route even when every sentence is grammatically correct.
A button label may no longer match the instruction. A voice-over may become too long for the animation. A quiz option may accidentally reveal the correct answer. A localized screenshot may show an interface different from the learner’s actual software. A translated variable may break a SCORM package. A cultural example may stop illustrating the intended concept.
The first rule is therefore: translate the instructional function of every element, not the string in isolation.
First principle: preserve the learning objective
Every module should answer a question: what should the learner know, recognize, decide or perform after completing it? Write that objective down before translating the content.
If the objective is “identify phishing indicators,” the target course must still train discrimination between safe and suspicious messages. If the objective is “perform a lockout procedure,” the translated sequence must preserve the order and conditions of the procedure. If the objective is “apply a customer-service policy,” the scenario choices must still measure policy application rather than vocabulary difficulty.
Learning objectives are the anchor that prevents localization from becoming stylistic rewriting.
Step 1: inventory the complete course package
Do not begin with the visible slides alone. Inventory all translatable components: course title, module names, slide text, narration, subtitles, captions, buttons, menus, instructions, quiz questions, answer options, feedback, glossary, references, certificates, downloadable PDFs, image text, alt text, transcripts, scenario dialogue, resource links and help messages.
Then inventory the technical layer: SCORM manifest, xAPI statements, variables, JavaScript, file names, media references, placeholders, IDs and tracking fields. Many of these should not be translated.
A complete inventory prevents the classic failure in which the course is ninety percent localized but the remaining ten percent breaks the learner’s experience.
Step 2: separate content from code and control strings
E-learning authoring tools often export text together with technical identifiers. A string such as {learner_name}, %SCORE% or module_03_complete may look like language but function as a variable or system key.
Mark protected strings before translation. Translators should know which text is visible to learners and which text controls behaviour. If a visible label contains a variable, translate around the variable without altering its syntax.
The same rule applies to HTML tags, XML elements and placeholders. Language can change; control syntax usually cannot.
Step 3: map the course architecture
Understand the learning sequence: introduction, explanation, demonstration, guided practice, independent practice, assessment and closure. Not every course follows this exact pattern, but most have an internal instructional logic.
Record prerequisites and branches. Does Module 3 assume completion of Module 2? Does an incorrect answer trigger remediation? Does a scenario branch return the learner to a decision point? Does the course unlock a certificate only after a passing score?
Translation should preserve these relationships. A linguistically correct module that sends learners down the wrong branch is functionally incorrect.
Step 4: build a course terminology base
Extract recurring terms before translation begins. Include subject-specific concepts, product names, policy terms, job roles, software labels, compliance vocabulary and assessment language.
Define each important term in plain language and record the approved target equivalent. If a course teaches a distinction between “hazard” and “risk,” the translation must not collapse both into one word. If “incident,” “near miss” and “accident” are separate categories, preserve them.
The terminology, glossaries and quality-checks owner provides the deeper system for maintaining these decisions.
Step 5: identify the learner profile
A course for subject-matter experts can use terminology that would overwhelm new employees. A course for school-age learners may need simpler syntax and more explicit examples. A compliance module may need exact policy wording. A consumer tutorial may prioritize direct action.
Translate for the learner the source course was designed to teach. Do not raise or lower the assumed knowledge unless the localization brief explicitly changes the audience.
Audience also affects tone. “You will learn” may be natural in one market but overly direct or informal in another. Adapt voice without changing instructional responsibility.
Step 6: preserve the difference between instruction and explanation
E-learning frequently alternates between telling the learner what to do and explaining why. “Select all that apply” is an instruction. “Multiple indicators may appear in one message” is an explanation.
If translation softens the instruction or turns explanation into a command, the activity can become confusing. Label content types during review: instruction, explanation, example, feedback, warning and assessment.
This simple classification makes it easier to choose appropriate grammar and register.
Step 7: translate slide text for scanning
On-screen text competes with narration, animation and visuals. Dense target text can overload the learner even if it is accurate. Translate meaning fully, then edit for on-screen readability.
Preserve headings, key terms and logical structure. Avoid squeezing a paragraph into a small text box simply because the source was shorter. If the target expands, adjust layout or split the content rather than reducing important instruction.
Screen text should support learning, not become a transcription of everything the narrator says.
Step 8: manage text expansion deliberately
Some target languages require more characters than the source. Expansion affects buttons, tabs, speech bubbles, timeline labels and mobile screens.
Create character-limit notes for constrained strings. Decide whether approved abbreviations exist. Where expansion is unavoidable, modify the interface or responsive layout rather than inventing cryptic language.
Localization engineers and linguists should share responsibility for fit. A translator should not silently remove meaning to solve a design problem.
Step 9: translate narration for speaking time
Voice-over translation has two constraints: meaning and duration. A target script may be twenty percent longer when spoken. If narration is synchronized to animation, the timing can drift.
First create a faithful script. Then measure reading time and revise naturally. Remove redundancy before removing meaning. Reorder clauses where the target language allows more efficient delivery.
Do not force narrators to speak unnaturally fast just to match the source waveform. The learning experience matters more than frame-perfect imitation unless timing is technically fixed.
Step 10: maintain synchronization with animation
If the narrator says “Now select the Settings tab” exactly when the tab is highlighted, the target script must reach that instruction at the same logical moment.
Mark synchronization points in the script. Long target phrases may need the animation extended, or the narration may need careful restructuring. Do not let the audio describe an element after it has disappeared.
Review the final localized video, not only the script and audio files separately.
Step 11: subtitles are not the same as narration transcripts
Subtitles may need to be shorter than the spoken target because learners must read while watching. Preserve the instructional meaning, but divide text into readable units and keep timing aligned with speech.
Do not place a technical term in one subtitle and its modifier in the next if that separation makes interpretation harder. Keep key phrases together where possible.
The dedicated subtitles, captions and audiovisual translation owner develops the timing and segmentation problem further.
Step 12: translate buttons by action
Buttons such as Continue, Submit, Retry, Review, Start, Exit, Resume and Download are action controls. Translate the action the learner will perform.
Keep button terminology consistent throughout the course. If “Submit” becomes one target verb on quizzes and another on forms, learners may wonder whether the actions differ.
Check the visible button against instructions. “Select Continue” should use the exact same target label the learner sees on screen.
Step 13: distinguish navigation from learning actions
“Next” may move to the next slide. “Continue” may advance after an interaction. “Submit” may commit an answer. “Check” may reveal correctness without final submission.
Do not collapse these controls into one generic target word. The learner uses labels to infer what will happen after clicking.
Interface consistency is part of instructional clarity.
Step 14: localize branching scenarios as decision problems
Scenario-based learning asks the learner to choose an action in context. Translate the situation, available evidence and options so the same decision remains difficult for the same reason.
A careless translation can make one option obviously wrong because of awkward grammar, while the correct option sounds polished. It can also add moral judgment not present in the source.
After translation, ask whether the learner is still choosing between the same conceptual alternatives.
Step 15: preserve character roles and relationships
Training scenarios often include managers, customers, patients, colleagues or team members. Titles, pronouns and forms of address can encode hierarchy and social distance.
Adapt these naturally to the target culture without changing who has authority, who is responsible or what relationship the scenario is testing.
If the learning objective depends on challenging a supervisor, reporting to a manager or helping a customer, the translated social relationship must remain legible.
Step 16: translate examples by function
An example exists because it demonstrates a concept. A culturally unfamiliar example may fail even if translated perfectly.
Ask what the example teaches. If it illustrates probability, workplace conduct, cybersecurity, measurement or customer communication, preserve the underlying mechanism. Replace surface details only when needed and when adaptation is authorized.
Do not substitute a culturally familiar example that changes the concept being demonstrated.
Step 17: adapt names, dates and everyday references carefully
Names, currencies, addresses, holidays, foods and social practices may require localization, but not every course benefits from full domestication.
If the source course is explicitly set in a real jurisdiction, keep that setting and explain where necessary. If the scenario is intentionally generic, localized names may make it more natural.
Document adaptation decisions so one module does not localize heavily while another remains source-bound without reason.
Step 18: preserve the exact task in knowledge checks
Assessment instructions such as choose one, select all, drag each item, rank from highest to lowest, enter a number or match pairs define the task.
Translate these instructions before the question itself. If “select all that apply” becomes a single-answer instruction, the learner may fail for interface reasons rather than knowledge.
Assessment validity depends on task equivalence as well as semantic equivalence.
Step 19: keep answer options parallel
Multiple-choice options should be similar in grammatical structure and plausibility. Translation can accidentally make the correct answer longer, more formal or more precise than distractors.
Review all options as a set. Keep parallel syntax where the source does. Avoid introducing clues through gender agreement, article choice or vocabulary that appears only in the lesson’s correct explanation.
The dedicated assessment-item translation owner offers the deeper measurement framework.
Step 20: feedback must teach the same correction
Incorrect-answer feedback often explains why a choice fails and redirects the learner to the principle. Preserve that diagnostic function.
Do not translate all negative feedback as “Incorrect. Try again.” if the source gives specific guidance. Conversely, do not add hints that make the next attempt easier than intended.
Feedback is part of instruction, not a decorative message after assessment.
Step 21: preserve scoring logic
Some courses award partial credit, weight questions differently, require minimum scores or limit attempts. Translation should not alter numeric thresholds or labels such as pass, fail, complete or incomplete.
Check whether localized decimal separators or percentage formatting affect displayed scores. Verify completion messages against actual LMS states.
Learners should receive the same outcome for the same performance.
Step 22: protect SCORM structure
SCORM packages connect course content to an LMS through defined files and tracking behaviour. Translation may involve text inside package resources, but manifest structure, identifiers and technical references must remain valid.
Do not rename files or IDs casually. After localization, repackage and test launch, bookmarking, completion, score reporting and resume behaviour in the target LMS.
A beautiful translation that no longer reports completion is not a successful localized course.
Step 23: understand xAPI statements before editing them
xAPI can record learning experiences as structured statements. Some human-readable labels may be localizable, while activity IDs and technical identifiers may need to remain stable.
Coordinate with the learning-technology team. Determine which values are display strings and which are data keys. Test reporting after localization.
Language work should not corrupt analytics.
Step 24: keep LMS navigation separate from course navigation
The course may have its own Continue button while the LMS has separate controls for modules, transcripts, resources or completion. These layers can use different terminology.
Localize them coherently. A learner should understand when they are navigating inside a module versus returning to the course catalogue.
If the LMS already has an official localized interface, match its terminology rather than inventing new labels inside the course.
Step 25: localize course metadata
Course title, description, duration, prerequisites, objectives, audience, tags and completion criteria may appear in the LMS catalogue before the learner launches the module.
Translate this metadata with the same terminology as the course. Search terms should reflect how target-language learners and administrators describe the topic.
Metadata affects discoverability and learner expectations.
Step 26: screenshots must match the learner’s interface
Software training frequently relies on screenshots. If the target audience uses a localized application, source-language screenshots can make instructions harder to follow.
Capture the target-language interface where possible. Ensure version, menu structure and labels match the learner’s environment. If software remains in the source language, keep the visible labels as the learner will see them and translate the explanation around them.
Screenshot accuracy is operational accuracy.
Step 27: image text needs its own extraction pass
Diagrams, infographics and screenshots may contain text not present in the translation file. Extract it deliberately.
When rebuilding graphics, preserve arrows, numbering, callouts and relationships. Longer target labels may require redesigned callout boxes.
Visual content should be reviewed after rendering, not only in a spreadsheet.
Step 28: charts and data visualizations must preserve data
Translate axis labels, legends, titles and notes without changing numbers or categories. Check decimal conventions and date formats.
If a chart teaches a concept, ensure that translated labels still allow the learner to interpret the pattern. Do not reorder categories merely to fit text unless the visual meaning remains intact.
Data is part of the lesson and deserves separate QA.
Step 29: downloadable job aids should align with the course
A checklist or reference sheet may be used long after the learner finishes the module. Its terminology must match the training exactly.
Do not localize the course one way and the job aid another. If the job aid contains steps, numbers or policy rules, perform the same action and numerical QA used for technical procedures.
Transfer from training to work depends on this continuity.
Step 30: certificates and completion messages need identity accuracy
Certificates may include learner name, course title, completion date, score, organization and credential wording. Protect variables and formatting.
Translate credential language carefully. “Certificate of completion” may not mean professional certification. Do not make the localized document imply a qualification the course does not award.
The learner’s achievement should be represented accurately across languages.
Step 31: accessibility is part of localization
Localized courses should preserve captions, transcripts, alt text, keyboard navigation, focus order and readable contrast where these features exist.
Translate alt text by describing the informational purpose of the image, not by mechanically translating every visible detail. Check that screen-reader labels match the target interface.
Accessibility text is user-facing instructional content and belongs in the localization scope.
Step 32: right-to-left languages may change the interface
Right-to-left localization can affect text alignment, navigation arrows, progress indicators and layout. Do not assume mirroring everything is correct: media controls, charts, timelines and directional icons may have semantic meanings independent of reading direction.
Review the course visually with native target-language users. Interaction logic must remain intuitive.
Bidirectional text and mixed numbers or product names need special testing.
Step 33: voice casting affects meaning
Voice-over is not merely a linguistic artifact. Age, authority, warmth, pace and formality shape how instruction is received.
Select target voices appropriate to the learner profile and course purpose. A safety course may need calm authority; a children’s module may need energy without sounding patronizing.
Pronunciation of technical terms and names should be documented before recording.
Step 34: pronunciation guides prevent inconsistent audio
Build a pronunciation list for product names, acronyms, people, places and technical terms. Decide whether acronyms are spoken as letters or words.
If several narrators record different modules, the guide prevents the same term from changing pronunciation throughout the course.
Audio consistency helps learners build stable vocabulary.
Step 35: timing is not only an audio problem
Animations, countdowns, timed questions and auto-advancing slides may have fixed durations. A target-language learner may need more time to read expanded text.
Review timing after localization. Extend display duration where technically and pedagogically possible. Do not force learners to race because the source language happened to be shorter.
Instruction should adapt to human reading time.
Step 36: compliance training requires exact source governance
Policies, legal statements and mandatory wording may not be freely adaptable. Mark protected text and verify against the authoritative policy or regulation.
If local law differs, the project may require a local legal review rather than simple translation. Do not silently replace source requirements with assumptions about the target country.
The translator’s role is to preserve the approved obligation and flag jurisdictional questions.
Step 37: version control prevents mixed courses
Training content changes often. A localized module can become obsolete if the source is revised while translation is underway.
Record source version, localization version and change date. When a source update arrives, perform delta review so changed content is traced into narration, subtitles, quiz feedback, job aids and metadata.
One changed policy sentence may affect several assets.
Step 38: use translation memory with instructional caution
Repeated sentences can be reused efficiently, but context still matters. “Continue” may be navigation in one place and instruction in another. “Complete” may be a verb or a status.
Review reused segments in context and maintain termbase control. High match percentages should speed review, not remove it.
Consistency is valuable only when the repeated meaning is genuinely the same.
Step 39: use comparable target-language courses
Original target-language training materials reveal how instructions, feedback, learning objectives and navigation are normally expressed.
Use comparable courses as evidence for style and terminology without copying their pedagogy blindly. A familiar target convention can make the localized course feel native rather than translated.
The dictionaries, corpora and parallel-texts method helps evaluate these sources.
Worked example 1: “Select all that apply”
Suppose the source quiz allows three correct answers. A target instruction that implies one answer changes the task. The learner may know the content and still receive a low score because the interface instruction was mistranslated.
Translate the task type first, then review answer options as a set. Test the final interaction in the module.
Assessment language must describe the actual behaviour of the software.
Worked example 2: synchronized software demo
The narrator says, “Open Settings, select Security, then enable two-factor authentication” while the screen highlights each control.
The target script must preserve the sequence and match the localized interface labels. If the target wording is longer, adjust animation timing or restructure the sentence so each action aligns with the highlight.
Text, audio and interface are one instructional unit.
Worked example 3: branching customer scenario
A learner chooses how to respond to an upset customer. The correct option demonstrates empathy while following policy. An incorrect option sounds polite but promises something the policy does not allow.
The translation must preserve both dimensions. If the incorrect option becomes obviously rude, the question becomes easier. If the correct option weakens the policy limit, the course teaches the wrong behaviour.
Scenario translation should preserve the decision boundary.
Worked example 4: safety procedure animation
A course shows four lockout steps in sequence. The target language naturally places a condition at the beginning of the sentence, while the source places it at the end.
Reorder the wording if needed, but keep the animation synchronized with the correct action. Verify warnings and negation separately.
The technical instructions owner provides the deeper procedural translation route.
Worked example 5: localized certificate
The source certificate states that the learner completed a course. A target phrase commonly used for professional certification sounds more prestigious but changes the credential.
Choose wording that preserves the actual achievement. Keep course title, learner name and completion date correct.
Formal-looking language should not upgrade status.
A repeatable e-learning localization workflow
- Freeze the source version.
- Inventory every visible and technical asset.
- Define learning objectives and learner profile.
- Extract terminology and protected strings.
- Translate screen text, scripts, interactions and metadata in context.
- Adapt timing, layout and examples where authorized.
- Rebuild audio, subtitles, screenshots and graphics.
- Repackage SCORM or xAPI content.
- Test launch, navigation, branching, scoring, completion and resume behaviour.
- Run linguistic, visual, accessibility and instructional QA.
This workflow treats localization as a course rebuild rather than a document export.
Failure mode: translating strings out of context
A translator sees “Review,” “Complete,” “Next” and “Submit” in a spreadsheet without knowing whether they are nouns, verbs, statuses or buttons.
Provide screenshots, string IDs, neighbouring text and interface function. Context reduces ambiguity and keeps controls consistent.
String-level translation becomes reliable when the course-level system is visible.
Failure mode: translating only the visible lesson
The main slides may be perfect while alt text, quiz feedback, certificate text and downloadable resources remain in the source language.
Use an asset inventory and completion checklist. Search the final package for source-language residue.
Partial localization creates a fragmented learner experience.
Failure mode: shortening until the learning point disappears
Design constraints can pressure translators to remove qualifiers or explanations. A shorter screen may look cleaner while teaching less.
Protect the instructional objective first. Reduce duplication, redesign the layout or move detail into narration before deleting meaning.
Space is a design constraint, not permission to weaken instruction.
Failure mode: changing difficulty through language
A translated quiz can become easier if distractors sound unnatural or harder if the target vocabulary is more advanced than the lesson.
Compare item difficulty conceptually. Keep options parallel and use vocabulary already taught unless the item intentionally tests new language.
The assessment should measure the same learning, not translation skill.
Failure mode: breaking placeholders
Variables and tags can disappear, move or acquire translated characters. The result may be a blank learner name, broken score or failed package.
Protect placeholders with non-translatable markers and automated checks. Then test the actual course.
Technical syntax is part of localization quality.
Failure mode: voice-over that no longer matches the screen
The target narration may be accurate but arrive too late for the highlighted element. Learners hear one instruction while seeing another.
Use synchronization markers and review the rendered module. Scripts cannot be approved independently from animation when timing carries meaning.
Multimedia localization is temporal as well as linguistic.
Failure mode: inconsistent terminology between course and workplace
If employees learn one target term in training but encounter another on real forms, equipment or software, transfer becomes harder.
Align the course termbase with operational documents and user interfaces. The Vocabulary Learning Hub helps explain why stable word meaning and retrieval matter to learning.
Training should prepare the learner for the language of the real task.
Testing: perform linguistic QA in the live course
Reviewers should click through the localized module, not only read exported text. Check context, truncation, labels, line breaks, subtitles, audio and feedback.
Record issues by screen or interaction ID. Fixes should be traceable back to the translation source so the same error does not reappear in future builds.
The rendered experience is the real target.
Testing: perform functional LMS QA
Launch the course through the intended LMS. Test first launch, return visit, bookmarking, completion, scoring, failed attempt, passed attempt, certificate generation and reporting.
Where the project uses SCORM or xAPI, confirm that the LMS receives the expected states. Localization can alter files without obvious visual symptoms.
Learning technology must be tested as technology.
Testing: use target-language learners
A native reviewer can judge grammar. A representative learner can reveal whether instructions are clear, examples make sense and interactions behave as expected.
Ask learners to complete tasks without coaching. Note hesitation, misclicks and wrong interpretations. If many users misunderstand the same screen, investigate the interaction rather than assuming they were careless.
Usability testing converts learner behaviour into localization evidence.
Practice exercise: localize one micro-course
Create a five-screen lesson with objective, explanation, example, quiz and feedback. Translate it into a target language you know.
Track every element separately: title, screen text, narration, button, quiz instruction, answer options and feedback. Then test whether another learner can complete it without seeing the source.
This exercise teaches that a course is a network of dependent strings.
Practice exercise: rewrite for timing
Translate a sixty-second narration faithfully. Read it aloud and time it. If it runs long, revise for natural compression while listing the ideas that must remain.
Repeat until the script fits a realistic pace. The exercise trains the difference between semantic loss and stylistic economy.
Good audio adaptation is deliberate, not rushed.
Practice exercise: inspect one SCORM package conceptually
You do not need to edit code to learn the principle. List which elements are learner-facing text and which are structural: IDs, manifest references, file names, variables and tracking fields.
Ask what would happen if each were translated. This builds respect for protected technical content.
Localization requires knowing what language is allowed to move.
How English knowledge supports e-learning translation
Instructional language depends on command verbs, sequencing, reference, conditionals and cohesion. “Before,” “after,” “unless,” “select,” “identify,” “compare” and “evaluate” carry different learner actions.
The How English Works route helps learners see how grammar organizes those functions. Strong structural knowledge makes it easier to rebuild instructions naturally in another language.
Translation improves when the learner can see the instructional grammar beneath the sentence.
How AI can help with e-learning localization
AI can draft translations, extract terminology, generate timing alternatives and compare source and target interactions. It can also help identify repeated strings across a large course.
But it may not know which strings are controls, variables or scoring logic. It can make distractors unequal or simplify a compliance statement. Use AI inside a controlled workflow with context and human review.
Fluency is not proof that the course still teaches the same thing.
Release sign-off
Before publication, record source version, localized version, language, termbase version, voice-over build, subtitle build and LMS package build. Confirm that open linguistic and technical issues are closed.
For regulated or safety-related training, include qualified subject-matter review. For public-facing courses, include accessibility and usability review.
Release control prevents a final package from mixing assets from different revisions.
Frequently asked questions
What is e-learning localization?
E-learning localization adapts online learning content for another language and audience while preserving learning objectives, instructional sequence, interactions, media, assessments and technical LMS behaviour.
What is the difference between e-learning translation and localization?
Translation changes language. Localization may also adapt layout, narration timing, screenshots, examples, interface labels, date formats and technical packaging so the course works naturally in the target environment.
Can SCORM packages be translated?
Yes, but translators must distinguish learner-facing content from technical identifiers and then test the repackaged module in the LMS for launch, completion, scoring and resume behaviour.
Should quiz questions be translated literally?
No. They should preserve the same construct, task type, difficulty and answer logic. Literal wording can change what the item measures.
How should voice-over translation be handled?
Translate accurately first, then adapt for natural spoken timing and synchronization. Do not force unnatural speech or remove essential instruction merely to match source duration.
Can AI localize an entire course automatically?
AI can accelerate drafting, but a complete localized course still needs terminology control, functional testing, assessment review, multimedia synchronization and target-language usability checks.
Contextual next routes
For the semantic foundation, continue with Translate Meaning, Not Just Words. For contextual decisions, use Context, Tone and Register. For research and terminology, use Dictionaries, Corpora and Parallel Texts. For interface strings, continue to Software, Apps and UI Strings.
The principle to keep
An e-learning translation is correct when a target-language learner can enter the course, understand the same concepts, perform the same interactions, make the same decisions, receive the same feedback, demonstrate the same competence and have the LMS record the same outcome.
Translate the learning path, not only the lesson text. The true unit of localization is the complete learner experience: objective, explanation, example, action, feedback, assessment, technology and transfer.