Moodle XML Question Bank Translation QA

Suggested slug (not created): how-people-translate-quickly-moodle-xml-question-bank-qa Reader job: Translate a Moodle XML question-bank export and prove that the imported target-language questions grade the same learning decisions as the originals. Boundary: This is not the Moodle site-language customisation workflow, a general XML tutorial, or a QTI 3 package translation guide.

People looking for Moodle XML question bank translation usually have a practical deadline: a teacher has created a useful bank of quiz questions, learners need a second language, and nobody wants the translated quiz to award marks for the wrong answer. The job is not just to translate Moodle quiz questions into natural sentences. It is to preserve the mechanics by which Moodle presents a prompt, offers responses, records an attempt, applies grade weights and displays feedback after importing the revised XML.

A multilingual Moodle question bank can fail while looking convincing. A translated multiple-choice option might accidentally become a more persuasive distractor than the intended correct answer. A negative instruction can turn positive. A numerical response can change because a decimal separator was treated as ordinary punctuation. A Cloze item can lose the braces and separators that carry its answer logic. This guide explains how to localize the human-facing layer while maintaining assessment behavior.

The efficient workflow is deliberately small: export a representative bank, map translatable fields, preserve structural fields, translate by question type, import into a safe course, preview and grade realistic attempts, compare against a frozen source baseline, then release only after sign-off. Searchers also call this Moodle XML localization QA, quiz import/export validation, or Moodle question-bank translation. The terms vary, but the reader’s task is the same: the translated assessment must remain a fair assessment.

Quick answer: a question bank is an executable lesson, not a document

When a worksheet is translated, a teacher can often read the page and spot an error. In a Moodle quiz, the text sits inside an assessment engine. The question type controls the learner’s interaction; the answer records control credit; the feedback explains an outcome; and settings may determine whether answers are shuffled. If those components are treated as prose, the learner can receive misleading grades even when the target sentence reads fluently. Start with one unmodified reference export. Create a working copy. Restrict translation to approved human-facing content. Then compare the entire question record before and after translation, not only the text visible on screen.

An example shows why this matters. A science prompt asks, “Which change increases the rate of evaporation?” and the answer with a fraction value of 100 is “Increasing the exposed surface area.” A translator changes that to a target phrase meaning “Reducing the exposed surface area.” The XML remains valid, the imported quiz loads, and Moodle still grades that choice as correct. The file passed technical validation but failed the educational requirement. Every completed target question therefore needs two confirmations: is the language right? and is the machine still rewarding the same reasoning?

1. Begin with an ownership statement and a stable source export

Before anyone edits the file, write down which institution owns the question bank, the course and question category, the Moodle version, the target learner level, and the permitted translation changes. The translation team should know whether it may adapt examples, replace culturally unfamiliar objects, or only render the current material in another language. These choices are educational, not just stylistic. A mathematics item whose answer depends on the length of an English word cannot be safely adapted by a translator without teacher approval; the mechanism of the question depends on the wording itself.

Export the selected questions in Moodle XML format from the relevant question bank, recording the export time and question count. Do not mistake a question-bank export for a complete course or quiz configuration backup. Moodle XML is designed to represent question content; quiz scheduling, gradebook behavior and broader course settings require their own management. Retain the original as a read-only source and label the translated derivative with a language and revision. When a colleague later reports that one option is wrong, these records tell you whether the error arose in the source, translation, import or subsequent edit.

A useful project header might say: “Secondary science, 86 questions, source English, target Malay, preserve topic coverage, distractor function, mark allocation and feedback timing; no live learner attempts permitted in QA.” This short specification can save more time than a large glossary if the project has not yet settled what must remain invariant. Without it, every editor is making different assumptions about what “same quiz” means.

2. Classify the XML layers before translating a word

Moodle XML contains elements that are intended for users, elements used by Moodle, and elements whose treatment depends on question type. Typical learner-facing candidates include question text, displayed answer text, certain question names, general feedback and feedback associated with responses. Machine-facing items include XML element names, attributes identifying question types, grading fractions, format declarations, IDs, encoded attachments and many configuration values. The same spelling can appear in both contexts. A value of true inside a setting is not a translation target merely because the local language has a word meaning true.

Make a small extraction table with the XML location, content purpose, display context and rule. For <questiontext><text>, the rule may be “translate, preserve HTML semantics and embedded references.” For <answer fraction="100"><text>, the rule is “translate as the correct answer, preserve the fraction.” For <shuffleanswers>, the rule is “leave setting unchanged.” For a numerical tolerance field, the rule is “do not translate or reformat.” This prevents a file-wide replacement from touching fields whose literal values Moodle expects.

Do not rely solely on whether the string appears between <text> tags. That element name can wrap different concepts: a displayed phrase, a question name, an answer value or a technical fragment. Good extraction follows the ancestor path and question type. In a safe review sheet, two rows carrying identical English text can have different handling because one is a label and the other is part of a machine-evaluated matching rule.

3. Freeze the grade model before adapting the questions

A fast translation team writes down the answer model before it writes target options. For each question, record its type, source stem, intended correct response, distractor purpose, marks or weights, feedback and any special matching rules. The quickest route to an incorrect answer key is to begin translating the answer options without first understanding which idea each option represents. Distractors are not random wrong sentences; they are designed to expose specific misconceptions. “Mass increases when ice melts” tests a different misunderstanding from “temperature remains constant during melting.”

In Moodle XML multiple-choice exports, each <answer> can carry a fraction that expresses credit. A single-answer question commonly has one fully credited option, but there may be different structures and partial-credit configurations. The translation must not assume that every question is a simple four-choice item with exactly one 100 fraction. Check multi-select questions, negative fractions, penalties, partially correct responses and teacher-specific settings. Record what the source actually does rather than what the team thinks a quiz ought to do.

A useful invariant table gives every source question a local comparison identifier and a target status: grading weights unchanged, answer count unchanged, selected response logic equivalent, feedback refers to the same reasoning, and learner level comparable. The invariant table is not a second answer key to publish. It is a private QA instrument to detect drift. Reviewers should be able to identify a mismatch without rereading the whole file.

4. Work through one multiple-choice question from source to target

Suppose an English mathematics question asks, “A shop reduces a $80 item by 25%. What is the sale price?” The options are $20, $55, $60 and $75, with $60 marked for full credit. Before translation, identify the operation: a 25% reduction leaves 75% of the original, so 0.75 × 80 = 60. That reasoning is the protected meaning. A target-language version may change the natural word order, but it must not turn “reduces by 25%” into “reduces to 25%.” Those look similar in casual reading; mathematically they produce completely different results.

The reviewer then inspects the four response values and the feedback. The numerical values and grade fractions remain untouched unless an authorised assessment owner intentionally redesigns the example. A currency symbol can create unnecessary interpretation risk. If the local learners need a different currency, the teacher must approve a revised item and validate the calculation; a translator cannot silently substitute a currency at a different numerical amount and pretend the answer key still works. The safe alternative is to preserve the mathematical values and clarify the unit according to the course’s stated policy.

After translation, answer the question twice in the preview: once by choosing the mathematically correct option and once by choosing a distractor. Confirm the score and visible explanation. Repeat with shuffled order if enabled. This is an example of a strong test: it verifies learning and software behavior together, not simply file syntax.

5. Translate distractors as diagnoses, not decorative alternatives

Many quiz translations become weaker because reviewers focus on the correct answer while treating the wrong answers as disposable. A good distractor should be wrong for the same relevant reason in both languages. If a source English item contrasts “evaporation” and “condensation,” the target terminology must preserve that contrast. If the translation turns one distractor into a synonym of the correct answer, students could be penalised despite understanding the concept. If a distractor becomes absurd, students may guess the answer without the intended knowledge.

Build a one-line “misconception label” for each distractor: confuses numerator with denominator; assumes correlation proves causation; ignores an explicit not; chooses the wrong unit; uses an overly broad definition. Translate the option in a way that keeps its diagnostic function. For learners at Primary level, avoid adding highly technical words to a distractor that was originally a common misconception. For older learners, preserve the intended conceptual distinction even if target phrasing requires a longer sentence.

A useful classroom test asks a bilingual tutor to read only the target-language options and describe why each wrong one is wrong. Then compare the explanation with the source teacher’s diagnostic notes. If the tutor must invent a new misconception to explain an option, the distractor has drifted. This review is more effective than simply looking for typographical errors because it interrogates the function of the question.

6. Deal with negative stems and double conditions explicitly

“Which of the following is not a renewable energy source?” is a trap for translation teams because the crucial information occupies one short word. A target language may express negation by particles, verb forms, word order or constructions that do not line up neatly with English. The translator should preserve the question’s polarity and make it visible to the intended age group. Negation may require a full target sentence rather than a literal replacement of “not.”

The risk becomes greater when a question carries two conditions: “Which statements are true only if the switch is open?” It is not enough to recognize the words true, switch and open; the quantifier and condition define the answer set. For each condition-heavy question, write a logic paraphrase before translation: “Select the response that is false,” or “The statement is allowed only under condition A.” This technique avoids dependence on one grammatical form and exposes contradictions early.

Check polarity with a reverse-direction question, not only back-translation. Ask the reviewer to state the target question as a plain instruction: choose true, choose false, choose exception, choose all, choose one, or choose the best. If the instruction does not match the source, repair the stem before spending time polishing tone. A polite and elegant mistranslation is still a broken assessment.

7. Preserve all forms of feedback as part of the teaching system

Feedback after a Moodle attempt is sometimes more educationally important than the question itself. A teacher may write general feedback that explains the correct principle, response-specific feedback explaining a misconception, or distinct messages for correct, partially correct and incorrect performance. If the source gives a student a repair route, the translation should preserve that route. “Review conservation of energy” cannot casually become “review the energy chapter” if a broad chapter includes unrelated concepts and the teacher intended a precise next action.

The translation team should label feedback by trigger. A message attached to the fraction="0" option is not interchangeable with the message attached to the fully credited option. When a translator rearranges XML children or an import tool reorders entries, the team must confirm the same feedback appears for the same learner choice. That is another reason to avoid “translate and paste everything back” without preserving stable associations.

A worked check is simple. In the source, a student selects a distractor based on converting kilometres to metres incorrectly; the feedback says to multiply by 1,000. In the target, make that same error and inspect the response. If the imported quiz says “Good work” or recommends dividing by 1,000, the translation is unsuitable even if the correct option remains correct. The check is especially useful when the teacher wants feedback to support independent practice between tutorials.

8. Treat embedded HTML and CDATA as packaging, not prose

Moodle questions often contain HTML paragraphs, emphasis, tables, links, maths markup and images. An export may wrap HTML in a CDATA section so that markup survives within XML. The purpose of CDATA is to carry literal text safely, not to make every character inside it translatable. A translator can change the visible sentence while preserving tags such as <strong> and an image’s source reference. A generic translation tool that removes or rewrites closing tags may produce content that imports but displays unpredictably.

For example, <strong>not</strong> may visually emphasize the logical exception in a multiple-choice item. The target language may need emphasis on a different phrase. That adjustment is allowed only if the markup still accurately highlights the question’s decisive condition and does not change answer meaning. Similarly, a table cell containing a number should remain paired with the right row label after localization. Reflowing long target labels is safer than reconstructing the table structure without tests.

Use an XML parser to verify well-formedness, but do not stop there. A file can be well formed while its HTML is malformed, its embedded image no longer resolves, or its learner-facing text is wrong. Preview with the same Moodle version and theme as the intended course. On a narrow screen, confirm that long labels do not hide answer controls or overlap mathematical notation.

9. Make media and alternative text part of the item inventory

A diagram-based question might contain the entire difference between two possible answers in an arrow, a colour key or an image label. Export formats vary in how media are represented: files may be embedded, linked, or handled through associated content. The translator should identify every image-dependent item before the large batch begins. A quick inventory has question identifier, media reference, visible words inside the image, alternative text, caption and any assumption a learner needs to interpret the visual.

Suppose a science diagram shows a thermometer rising from 20 °C to 35 °C and asks which particle change is most plausible. Translating the question stem does not localize an English legend printed on the diagram. If the target learners cannot interpret the legend, the item becomes harder for reasons unrelated to science. But altering the temperature values or changing which arrow points to the correct region would corrupt the item. The media needs a separate controlled production decision, followed by a new visual QA check.

Do not make accessibility an afterthought. If an image alternative explains the answer rather than describing the necessary stimulus, a screen-reader user could receive an unfair hint. The target alternative should preserve the source’s intended information level. An accessible equivalent may require expert review where an image communicates relationships that are difficult to describe briefly. The correct question is whether both learners can do the same assessment task, not whether every language contains the same number of words.

10. Handle Cloze and embedded answers with extra caution

Embedded-answer or Cloze questions are a frequent source of damaged translations because their syntax can resemble punctuation mixed into a normal paragraph. Special delimiters carry question type, scores and possible responses. A word processor, text converter or machine translation system may reorder or replace the control characters while trying to improve readability. The safe approach is to extract the natural-language spans without passing the control syntax through uncontrolled rewriting.

Imagine a grammar exercise in which the learner fills a blank inside a sentence and the source answer is the English form “were.” A literal translation of the surrounding sentence into a language with a different verb system may make the original blank impossible or meaningless. This is not a case for preserving the literal source answer. It is a case for teacher-led assessment redesign because the construct being assessed may be English grammar itself. Translation can help explain an instruction, but it cannot promise an equivalent target-language grammar item by replacing every word mechanically.

For a knowledge-based Cloze item, draw a small map: learner-visible prompt, each blank’s semantic role, accepted answer forms, grading, feedback and syntax tokens. Translate the prompt first and decide whether each answer can remain valid. Then review syntax character by character and run graded attempts. If the accepted-answer configuration requires changes, flag a pedagogical adaptation and obtain approval rather than silently editing the answer parser.

11. Be precise about numerical answers, units and tolerances

Numerical questions contain a second language besides natural language: numbers, ranges, units and comparison logic. A question may accept an exact number, a tolerance interval, or an alternative value with a different unit. The target text must describe the same quantity and precision. Converting “round to one decimal place” into wording that means “round to the nearest ten” is not a stylistic issue; it changes what is graded. Likewise, converting centimetres to metres without changing the accepted response produces a mismatch between prompt and scoring.

Some locales display decimal separators differently, but the stored numeric values and parser behavior should be governed by the quiz engine and site configuration, not by a translator’s habit. Treat programmatic number fields as data. Test learner input on the target-language site to see what the engine actually accepts. If the source is 3.5, do not type 3,5 into an internal field merely because that is customary in the target language; investigate whether Moodle’s response handling supports locale-specific input for the item type and version.

A complete test includes a correct in-range response, a wrong response, and one value close to a tolerance boundary. If the question has units, try a response with the intended unit and another with an explicitly incorrect unit. The output should match the original educational policy. This is often the last place a translation team looks and one of the first places a learner notices unfair marks.

12. Matching, ordering and short-answer items have different equivalence rules

In matching questions, preserving the right-hand answer text is not sufficient; the relationship between every prompt and response has to survive. A translated pair may become ambiguous because two target terms collapse into one common word. A source exercise might distinguish “metamorphosis” and “development,” but the target language might require additional phrasing to retain the conceptual difference. Inspect every pair and each near-neighbour distractor. For shuffling configurations, test at least one reordered presentation so the answer does not depend on the original screen position.

Short-answer questions are even more delicate. An auto-graded short-answer item may compare a learner’s text with accepted values or patterns. Translating the instruction while leaving accepted responses in English can create a question that appears to require a target-language answer but rejects it. Conversely, adding target alternatives without considering casing, spelling variants or regular expressions can accidentally accept answers that were meant to be wrong. These are scoring design decisions, not purely linguistic substitutions.

Essay questions require a different QA approach because the scoring may involve a human marker and rubric. The translator should preserve task demands, the kind of evidence requested and response constraints such as word limits. A longer translation does not necessarily justify a larger student response; the teacher decides whether comparable difficulty requires changing permitted length. Never claim equivalent difficulty solely because both prompts have been proofread by a fluent bilingual speaker.

13. Control scientific vocabulary by concept, not word shape

A question bank can contain thousands of repeated words but comparatively few repeated concepts. Build a compact terminology sheet that lists the source term, target term, level, definition and a representative question identifier. Do not put every ordinary word into the termbase. Focus on concepts that affect answers: solute versus solvent, current versus voltage, mass versus weight, inference versus observation, median versus mean. Each of these pairs can create an unfair distractor if translated inconsistently.

For Singapore learners, curriculum level matters. A Secondary 1 explanation should not assume specialized JC vocabulary merely because a dictionary suggests a formal equivalent. The teacher can approve a simpler target phrase if it preserves the concept. On the other hand, changing an assessed scientific term to a casual synonym may deprive learners of practice with the curriculum language they are expected to recognise. The correct term is determined by the level and instructional aim, not by translation elegance alone.

Mark glossary entries as approved, provisional or unresolved. When one question reveals that a term has two possible senses, capture the context rather than forcing one target form across every instance. A source word such as “work” can mean physical work in mechanics, an assignment, or ordinary effort. A shared termbase accelerates safe decisions only when it records these distinctions. A fast workflow should reduce repeated thought without replacing necessary interpretation.

14. Keep question names useful without turning identifiers into text

Moodle question names are often visible to editors and may be present in exports. Some are human-readable labels; others encode course, chapter or revision identifiers. A translation project should agree which names will change. If a name is used as a stable matching key in a downstream workflow, rewriting it may complicate reconciliation, even if Moodle itself allows a human-readable name. Preserve technical identifiers and established revision references unless the course owner has explicitly planned a localized naming convention.

A useful naming scheme separates stable identity from display language. A question might have a controlled record ID such as SCI-S1-CH03-Q017, while its displayed subject label can be translated in an editorial register. The record ID allows reviewers to compare source and target without relying on the position of the question in an exported file. Position is a weak identifier because categories, sorting and import procedures can change order.

This becomes important after an educator corrects a conceptual error. If source question 17 was revised in English but the translator cannot identify its Malay counterpart, the two languages begin to diverge. A tiny change log with question ID, source revision, target revision, issue and approver is enough to preserve auditability. The aim is not bureaucratic complexity; it is preventing a corrected educational mistake from surviving in one language long after it was repaired in the other.

15. Why XML validation and import-preview are separate tests

An XML parser can tell you whether tags are balanced and whether the file is syntactically readable. It cannot tell you whether a multiple-choice distractor has become correct, an accepted answer has been omitted, or a science caption contradicts its image. Importing into Moodle catches another class of errors: unsupported attributes, unexpectedly ignored elements, invalid question syntax, missing resources, configuration mismatches and rendered problems. Even a successful import does not prove assessment equivalence.

Use three gates. At the file gate, parse the XML, check encoding, compare question counts and inspect machine-value invariants. At the Moodle gate, import into a sandbox category, preview different question types and verify visible media and text. At the learning gate, attempt selected questions as a learner, deliberately choose right and wrong answers and inspect feedback and marks. This progression costs time once but creates repeatable confidence for every subsequent batch.

Test a representative sample first, not a random handful. The sample should include one instance of every question type used, questions with images, negative stems, HTML, maths, Cloze syntax and high-stakes feedback. If the initial sample breaks, fix the extraction method before translating 300 more questions. At final release, use the agreed coverage level, including full checking of any item where a change could affect scoring. “No import error” is a technical observation, not a promise of fairness.

16. Build a mismatch report that tells an editor what to repair

A quality report should identify the actual failed invariant rather than simply mark the question “bad.” Consider a four-column table: question ID, evidence, consequence and repair owner. “SCI-S1-017: target negation missing; learner may choose the opposite response; translator + science reviewer” is actionable. “Wording feels odd” is not enough to triage safely. For a control-code problem, the report might say “MATH-S2-042: numerical tolerance differs; target grading accepts a response source rejects; technical editor.”

Prioritize errors by consequence. P0 failures change the answer key, accepted answer, marks, question identity or major safety-critical meaning. P1 failures make the item misleading, inaccessible or materially harder for the wrong reason. P2 failures create awkward language or noncritical presentation problems while preserving the construct. P3 issues are editorial preferences. These classes can be adapted to a school’s workflow, but the distinction prevents a cosmetic rewrite from displacing urgent grading repair.

For every P0 and P1 repair, rerun the test that exposed it. If a negative stem is repaired, have a reviewer paraphrase the polarity again. If media text is corrected, inspect the rendered image and the answer it supports. If scoring logic was restored, submit a right and wrong response. A fix is complete when the original failure cannot be reproduced, not merely when the editor has saved the file.

17. Protect fairness during cultural adaptation

Teachers sometimes want examples that feel natural to target-language learners. This may be appropriate, but there is an important difference between making a sentence familiar and changing the construct. Replacing the word “football” with the familiar local term for the same game can be a straightforward linguistic choice. Replacing a cricket scoring problem with a football scoring problem changes quantities, rules and potentially difficulty. That is assessment design and requires a subject specialist.

Even ordinary references can alter answer probability. A consumer-maths question based on a familiar local shop may be easier than an unfamiliar overseas example; a geography question about a named river may be much easier for learners who live beside it. Equivalence is not guaranteed by preserving sentence length, number of options or topic. Ask whether the same underlying skill is needed, and whether the new context provides or removes clues that the source did not contain.

If an adaptation is approved, label the target question as adapted and preserve the original. Explain the change, the educational rationale, any numerical recalculation, and the person who verified scoring. That trace is especially valuable when schools share multilingual materials or use them across academic years. Avoid silently replacing evidence, cultural references or data inside a bank presented as a straightforward translation. Students and teachers deserve to know when the item itself has been redesigned.

18. A complete classroom scenario: bilingual science quiz release

Consider a school that exports twelve lower-secondary science questions for a bilingual reinforcement activity. The bank has seven single-answer multiple-choice items, two multi-select items, one diagram-based item, one numerical item and one short-answer item. The teacher wants the same conceptual test in Malay and English so learners can practise science without having the assessment mechanics create unnecessary language barriers. The project starts with an English source export and a frozen teacher answer sheet.

The translator first inventories every stem, answer, feedback message and display asset. A target terminology table settles “force,” “mass,” “weight,” “heat transfer” and “condensation” at the appropriate class level. The editor translates stems and answer options in a working export while leaving grade fractions and numeric settings untouched. The team flags a diagram label that remains in English and a short-answer item whose target responses need explicit review by the teacher. Rather than guess, the technical editor blocks those two questions from release.

The first Moodle import loads all twelve records, but preview reveals that one long target option wraps beneath a nearby image and is difficult to tap on a phone. The team repairs the layout without changing the answer text. The teacher then tries every question, chooses at least one wrong answer per type and confirms the feedback. One multi-select option awards partial credit incorrectly because the source export had an old fraction value; that problem belongs to the original bank, not the translator. The project corrects the source through teacher approval, re-exports, and updates both languages. Good QA protects the source too.

After the repair, the release record includes question count, export revision, target revision, tested Moodle environment, known limitations and approval by the subject teacher. The school imports into the intended category, not over live questions whose learner attempts might have historical implications. It uses a scheduled, controlled release and keeps rollback material. The result is not merely a Malay version of twelve paragraphs. It is a second-language assessment whose learning decisions have been checked.

19. Design the fastest repeatable production route

The main speed gain comes from reducing uncertainty early. First, use the source export to classify question types and isolate all structural fields. Second, resolve terminology and adaptation rights before translators touch the large batch. Third, translate ordinary prose in stable groups while treating Cloze, numerical, matching and media-heavy items as specialist queues. Fourth, validate file structure automatically where possible. Fifth, import representative questions and perform learner-level tests. Sixth, require teacher sign-off on answer equivalence and any redesigned item. This is faster than translating everything and then discovering the extraction method was unsafe.

A practical workload board can track “ready for translation,” “needs subject clarification,” “translated,” “technical QA,” “educational QA,” “approved” and “blocked.” The board should also show question type and risk, not just percent complete. If 90% of ordinary multiple-choice questions are done but all four Cloze items remain unresolved, the bank is not 90% release-ready when those items are part of the required assessment. This keeps reporting honest.

Use automation for invariants, not for judgments it cannot make. A script can compare counts, fractions, question types and presence of media references. It cannot determine whether a target distractor has become correct by accident or whether an ambiguous translation changes the skill being measured. A bilingual subject reviewer can do that. A good workflow assigns mechanical checks to machines and semantic checks to responsible educators.

20. Practical checks before a question bank becomes live

Structure: Every required question appears once in the target bank; question types and category placement are expected; files are well formed and imported with no uninvestigated warnings. Scoring: Fraction values, tolerances, accepted-answer logic and total possible marks match the approved specification. Language: Stems ask for the same cognitive action; negation, quantifiers, conditions, units and names are unambiguous. Feedback: Every feedback message appears under the right trigger and explains the same mistake or principle. Presentation: Images, alternative descriptions, links, tables and mathematical notation remain usable. Governance: Changes are traceable to a source revision, and subject specialists have signed off on adaptations.

Run at least one wrong-answer test in each grading pathway. That means not only “Does the correct answer get a mark?” but also “Does a credible distractor fail appropriately?” and “Does partial credit behave as intended?” Test the learner view, including smartphone presentation if the course is used on phones. A question that works in a teacher preview but hides the submit button behind oversized localized content is operationally unready.

Keep the last known-good export and import notes. If a problem appears after release, the school needs to distinguish text correction, answer-key correction and a wider engine configuration problem. Avoid repairing an in-use assessed quiz without evaluating effects on attempts, grades and fairness. Where live scores are involved, use the institution’s assessment governance process before changing questions that learners have already answered.

21. Transfer the method to other digital assessments

The same principle applies beyond Moodle: separate the display language from the grading mechanism, then test both. QTI packages, learning apps and proprietary assessment systems each encode that separation differently. A workflow designed for Moodle XML should not be copied blindly into another format because field names, response processing and media packaging differ. What transfers is the reasoning: identify the protected construct, map translatable surfaces, keep machine invariants stable, perform a technical import or rendering test, and validate learner consequences.

For teachers, this method also improves original quiz writing. If it is difficult to explain why one distractor is wrong, the source item may already be ambiguous. If the answer depends on wording that cannot translate, perhaps it was testing language rather than science. If a question gives a misleading correction, the translated version simply makes that flaw visible. The benefit extends beyond multilingual deployment: disciplined translation QA can reveal weaknesses in the source bank.

Parents using bilingual practice should therefore ask a different question from “Are all the words in my home language?” Ask whether the learner is practising the same concept, receiving reliable feedback, and being awarded credit consistently. Those are the features that make translated questions a useful teaching instrument rather than a confusing second layer of difficulty.

22. Summary: preserve the student’s reasoning route

Moodle XML translation becomes efficient when the team stops treating a quiz as a word-processing document. The safe object to preserve is the learner’s task: the prompt, the intended reasoning, the distinction between correct and incorrect alternatives, the feedback, and the recorded result. Translation may change sentence structure, term order and natural explanations. It may not accidentally change which thought earns a mark. A structurally valid file, a successfully imported bank and a linguistically fluent item are three separate achievements. Release quality requires all three plus educational equivalence.

Frequently asked questions

Can I translate a Moodle XML export using machine translation?

Machine translation can support approved text extraction, but raw whole-file translation is unsafe. XML element names, question types, fraction attributes, embedded syntax, identifiers, numeric values and answer-matching structures must be protected. Any proposed target options need bilingual subject review, and the resulting export must be imported and graded in Moodle. The main risk is not that the text will sound unnatural; it is that a natural-looking mistranslation changes which answer receives credit.

Should I translate Moodle question names?

Only when their purpose is understood. A name intended solely for the teacher’s convenience may be localized, while a controlled identifier used for matching, revision or reporting should remain stable. Separate internal question identity from displayed editorial titles where your workflow allows it. This prevents release comparisons from breaking when question order changes.

Is Moodle XML the same as QTI?

No. Moodle XML is a Moodle-specific question interchange format. QTI is a separate specification family for interoperable assessment items. They may address similar educational questions, but their schemas, packaging assumptions and response handling are different. A QTI translation procedure is not automatically a valid Moodle XML procedure; verify the source format before choosing tools.

Why did an imported question lose its picture?

The media might have been linked rather than embedded, referenced by a path unavailable in the target environment, omitted during export or altered during editing. Inventory every media dependency and check the actual imported question. Do not assume that an XML file containing text also contains every image binary needed to display the quiz.

What is the most dangerous translation error in a multiple-choice item?

Any change that makes Moodle reward the wrong reasoning. Common causes include inverted negation, mismatched answer fractions, correct and incorrect options becoming synonymous, changed units, and misleading feedback tied to an otherwise valid answer. Those errors can survive syntax checks and ordinary proofreading, so they need graded test attempts.

Can teachers adapt examples to local culture?

Yes, with explicit approval and educational revalidation. Changing a setting, object, unit or numerical value may change the problem. Document the adaptation, verify the answer key and check that learners need comparable knowledge to solve it. Do not label an adapted assessment as if it were only a word-for-word translation.

How much should be reviewed by a subject specialist?

Every question with a possible scoring or construct consequence requires educational review. Risk-based triage can direct scarce time to negative stems, numerical items, Cloze syntax, matching items, multi-select choices and diagram-based questions, but the minimum release rule must still be defensible for the whole bank. The school or assessment owner decides the appropriate sampling or full-check policy based on stakes.

How do I know the target quiz is ready?

A release-ready bank has a documented source revision, question inventory, protected structural fields, correct and consistent translation, clean import, rendered-content inspection, demonstrated grading behavior and approval of adaptations. If students have already attempted the original quiz, include an assessment-governance review before replacing in-use questions. The final question is whether the target learner can demonstrate the same knowledge and receive appropriate credit.

How People Translate Quickly | Moodle XML Question Bank QA: Localize Quiz Questions Without Changing the Answer Key