In-depth practical guides

Build a multilingual journey acceptance plan

Test from discovery to decision: an acceptance plan connecting navigation, search, language switching, tools, forms and error recovery.

Updated :

Laptop, phone and test matrix for comparing multilingual form states. Illustrative scene.

AI-generated illustration

Can a user complete their task in every language?

A translated page is insufficient if its menu returns to French, search ignores its words or the form produces an unexplained error. Test observable tasks and complete states. This complements text and URL reviews by examining what users can actually accomplish. The following situations are preparation examples, not claimed assignments.

  1. Inventory tasks

    List page families and actions: find a definition, compare sources, open a profile, compose a reference, calculate or make contact. Give each task a starting point, destination and expected result. Include secondary pages, not only primary menu entries.

  2. Describe covered states

    Include initial state, long content, active filter, no result, invalid input and recovery. Use clearly labelled synthetic data. For contact, separate local validation, safe server rejection and explicitly authorised delivery testing: do not send real messages by default.

  3. Follow language and context

    Enter through a deep profile and switch language. Check preservation of the entity, filter or selected source where relevant. Inspect lang, direction, labels and messages. Canonicals and hreflang describe equivalent versions; they do not establish that the switcher works.

  4. Test actual discovery

    Write representative queries for each script, with accents, variants and multiple words. Check relevance, result links, pagination and no-result states. A sitemap URL may still be hard to reach through navigation. Test related links and the route back to the section.

  5. Perform actions

    Test search, filters, selectors and forms with normal and edge cases. Check expected results, public context, keyboard use, field errors and confirmation. Also test readable content and fallbacks when JavaScript is unavailable.

  6. Decide and retest

    Record expected and observed results, evidence and severity relative to the task. Fix the shared cause, repeat the failing case and affected languages and components. Passing acceptance means the defined matrix passes; it does not mean every possible use and device has been tested.

  7. Check what the reader takes away

    After an action, inspect copied text, downloaded files and shared links. Reader-facing explanations should follow the announced language; stable technical keys may remain identical. Compare copied and displayed values and state a source’s reference country. Translating an interface does not change the underlying document’s scope.

Reading workshop

Six planned cases are not six successful cases

This fictional matrix keeps failures, blockers and unexecuted attempts visible instead of dropping them from the acceptance report.

Fictional data and records for learning: none of these results comes from a live service query.

Campaign DEMO-T06

Six identifiers cover one journey per language here. Their invented states teach follow-up reading; they do not describe checks performed on this website.

If the table exceeds the screen, scroll horizontally; arrow keys work when the region is focused.

Campaign DEMO-T06
State and countAffected casesFictional observationFollow-up decision
PASS · 2 casesFRTOP, ENFORMThe French deep link reaches its section; the English form error remains associated with its field.Retain expected result, observation and evidence for these two tasks only.
FAIL · 1 caseESSEARCHA Spanish search result opens a French page despite the Spanish expectation.Fix language preservation, then rerun the case and related journeys.
BLOCKED · 1 caseRULINKThe required destination is unavailable in the demonstration environment.Identify the missing prerequisite; infer neither success nor a journey defect.
NOT_RUN · 2 casesARRTL, FIMENUArabic bidirectional reading and the Finnish menu have not been tried yet.Schedule execution; being listed in the matrix is not a result.

Read a test matrix without turning missing tests into successes

From 1 to 40 cases, without a header: identifier;state. Unique ASCII identifiers of 1–32 characters (letters, digits, dot, hyphen, colon, underscore). Exact states: PASS, FAIL, BLOCKED, NOT_RUN. UTF-8 text, at most 8192 bytes.

Results for the supplied set

Supplied cases
6
Cases with PASS or FAIL verdict
3
PASS
2
FAIL
1
BLOCKED
1
NOT_RUN
2
Share of cases with a verdict (%)
50.00
PASS among cases with a verdict (%)
66.67

This tool defines a completed verdict as PASS or FAIL. BLOCKED remains blocked, possibly after a partial attempt; NOT_RUN was not executed. Success rate uses PASS + FAIL, never only PASS. Without a verdict it remains undefined.

The supplied plan contains failures, blocked or unexecuted cases. Retain these states in the assessment.

States retained for each case
CaseExact codeMeaning
FRTOPPASSExpected criterion declared satisfied
ENFORMPASSExpected criterion declared satisfied
ESSEARCHFAILExpected criterion declared unsatisfied
RULINKBLOCKEDBlocked attempt without completed verdict
ARRTLNOT_RUNUnexecuted test
FIMENUNOT_RUNUnexecuted test

A declared matrix does not test the site. Scope, evidence, languages and acceptance criteria still need checking. High success among completed cases can coexist with low coverage.

Which conclusion respects all six tracked cases?

Open each proposition to compare its diagnosis. More checks may still be needed.

Every language is validated because each is listed.

Incorrect. The list shows intended scope, not execution or success in each language.

Two successes out of three are enough to close.

Too quick. That subset omits one blocked and two unexecuted cases; the remaining failure also needs a decision.

Two PASS, one FAIL, one BLOCKED and two NOT_RUN; campaign open.

That is the complete exercise summary. It retains all six cases and separates an observed defect, a missing prerequisite and outstanding work.

Fictional six-case tracking extract

Campaign: DEMO-T06; invented data
FRTOP: PASS; expected section reached
ENFORM: PASS; error associated with field
ESSEARCH: FAIL; wrong-language destination
RULINK: BLOCKED; prerequisite unavailable
ARRTL: NOT_RUN; visual order and copying to try
FIMENU: NOT_RUN; keyboard to try

Identifiers and state codes stay stable; explanations follow the reading language. This log is a demonstration, not the website’s test history.

Documentation of the rules, separate from fictional examples

Compare other situations and exercises

Language changes but the profile is lost

From a selected source, the English button opens a default profile. Preserve the public entity parameter, then compare all six destinations. Do not put private notes or drafts in these URLs.

Search exists without being useful

An accented query misses an existing profile. Check normalisation and the language index, then multiple words, results and pagination. Equal record counts do not establish relevance.

A form error disappears

After rejection the message is off-screen or unassociated with the field. Check localisation, announcement, focus and retention of non-sensitive entries. Retest using a deliberately invalid submission that cannot send a message.

An export retains another language’s instructions

Illustrative case: the form is Spanish but its exported template has French instructions. Open the file using the intended characters and delimiter, then inspect headers, instructions and values. Separate intentionally stable technical identifiers from explanations requiring translation; download completion alone does not validate the journey.

A Latin identifier reads poorly in Arabic text

Test example: REF-12/AB appears beside a bracket and number in an Arabic sentence. Check visual order and copied value. Isolate the opposite-direction string with appropriate markup; do not reverse its characters to make one screenshot look correct.

Worked exercise

Six loads do not test eighteen states

Fictional example: the values and documents teach the method and do not describe a real organisation.

Testing covers three states per language. Only the initial view was examined, with no data submitted.

If the table exceeds the screen, scroll horizontally; arrow keys work when the region is focused.

Starting information
LanguageInitialErrorCorrection ready
FRYesNot checkedNot checked
ENYesNot checkedNot checked
ESYesNot checkedNot checked
RUYesNot checkedNot checked
ARYesNot checkedNot checked
FIYesNot checkedNot checked

Your checks

  1. Build the language × state matrix.
  2. Identify cells actually checked.
  3. Test recovery and messages with fictional data.
Read the answer

Answer and decision

There are 6 × 3 = 18 cells; only 6 are covered, one third. The remaining 12 specifically include errors and recovery. Equal numbers of translated pages do not prove translated, usable interactions.

If the case changes

With two planned widths, the matrix becomes 36 cells. Document scope rather than treating two screenshots as general coverage. After correcting a message, replay the error and recovery in the affected language; a screenshot of the initial state does not close that defect. Keep the tested version, announced wording and focus destination so another reviewer can reproduce the result.

Record to retain

Keep matrix, observed state, message, focus and outcome; use an authorised test destination.

Record to retain

  • Case identifier, task, starting URL, language, state and test data.
  • Expected result, observed result and useful screenshot or HTTP response.
  • Defect, responsible component, correction and review date.
  • Matrix coverage and limits: delivery, physical devices and external services not tested.
  • Output language, reference country, copied values and bidirectional-string checks.

Official references

Related protocols

Checks before reaching a conclusion

  • Does changing language preserve the selected record or public object without transferring a private form draft?
  • Does an error explain the problem, identify the affected field and allow recovery without losing useful information?
  • Can the journey still be completed by keyboard and without JavaScript for content that does not require it?
  • Does copied or downloaded output preserve the expected meaning, language and identifier?

Frequently asked questions

Must every page be tested?

Inventory every route and language, then cover each template’s tasks and states. Add targeted checks for unusual profiles and parameters.

Should language switching preserve input?

Keep useful public context where intended. Avoid placing drafts or sensitive information in URLs; explain when an action resets input.

Is automated testing enough?

It detects reproducible regressions. Visual review and keyboard journeys remain necessary to assess readability, understanding and action order.

When is acceptance complete?

When the defined criteria pass and remaining defects are explicitly documented. Fix and retest defects blocking the priority task before approval.

Why do you test errors in each language?

They are part of the path and can use different translations or components. A correct page to read does not demonstrate the consistency of the messages displayed after interaction.

DEEPER READING / PRACTICE

Check the final action in each language

A translated page seems correct, but a filter, error or an expected email interrupts the path. A functioning test should perform interactions instead of limiting itself to visual reading.

Updated on .

Priority checks

Testing entry, navigation, search, valid entry, invalid entry and confirmation. Keep the selected language and check the messages effectively presented at each step.

Interpretation pitfalls

A screenshot does not demonstrate the functioning of an interaction. Shared elements can introduce a regression in several languages after an apparently local correction.

Working record to retain

The operation test sheet document cases, expected outcome, obtained outcome and anomaly. The closure requires a new execution of the corrected case and pathways that use the same component.