Translate CAPTCHA & Verification Prompts Accessibly

How should you translate CAPTCHA challenges, human-verification prompts and bot-check errors without breaking accessibility or the security workflow? Correct CAPTCHA translation keeps the challenge instruction, expected response, alternative access route, expiry state and result message intact. It also distinguishes proving that a person has completed a bot check from authenticating a known account or consenting to tracking.

One of the smallest mistakes is translating ‘Verification expired. Try again’ as ‘Your account is blocked.’ Another is converting ‘Select all images containing bicycles’ into ‘Select one bicycle picture.’ The second prompt changes how the challenge is answered, while the first invents a security sanction. Accessible localisation requires testing instructions, labels, retry states and help paths in their real interface, not translating each button without context.

This guide addresses searches for CAPTCHA translation, ‘I’m not a robot’ meaning, human verification instructions, audio CAPTCHA alternatives, accessible bot checks, verification failed, challenge expired, refresh challenge and WCAG CAPTCHA accessibility. It teaches practical language decisions using fictional examples and W3C guidance. It does not describe bypass techniques or promise that any particular anti-bot implementation meets an accessibility standard.

The visible problem: why a fluent translation can be wrong

A site presents a visual challenge and an alternative audio method. The English interface names both, but the target-language version hides the audio-option label behind an untranslated icon and gives no explanation of how to change modes. The challenge may still technically run. Yet users who cannot perceive the image may have no functional route. This is not a mere cosmetic translation error.

The interface can include several independent states: challenge available, challenge running, challenge expired, answer rejected, technical service unavailable, account login pending and form submission blocked. Translators sometimes collapse all of these into ‘verification failed’. That phrase can remove the user’s next possible action and make the product appear to have reached a conclusion it has not reached.

WCAG 2.2 includes requirements concerning non-text CAPTCHA and accessible authentication. W3C also explains that an audio transcription challenge is not automatically a sufficient alternative for every cognitive-authentication concern. Translators cannot solve a missing accessible workflow by replacing a label, but they can avoid hiding an existing alternative and can document defects to the product team. Accessibility belongs to the whole interaction, not just the sentence.

The task in this article is narrow by design: CAPTCHA challenges and human-verification interface text. This is not a general guide to translating idioms, changing tone, choosing broad cultural equivalents or building a universal glossary. Those are different translation jobs. Here we inspect the actual objects, states and choices so that the translated material communicates the same limited evidence and permits the same justified action.

First principles: the distinctions that protect meaning

1. A CAPTCHA is a task, not a declaration of identity

A bot check aims to distinguish automated activity from user activity in a particular workflow. Completing it is not proof of a real-world legal identity or account ownership.

Do not translate a successful challenge as ‘your identity is verified’ unless the interface truly performs that separate check. The practical question is not whether a bilingual reader likes the wording; it is whether a reader in either language would understand the same source fact, its qualification and the boundary of the available evidence. Where context does not answer that question, the translation needs a documented query instead of a confident guess.

2. Instruction scope matters

Select one tile, select every matching tile and select images in sequence are different tasks. The translator must preserve number, inclusion rule and stopping condition.

Test the translated prompt against the visible challenge and count possible selections. The practical question is not whether a bilingual reader likes the wording; it is whether a reader in either language would understand the same source fact, its qualification and the boundary of the available evidence. Where context does not answer that question, the translation needs a documented query instead of a confident guess.

3. Challenge status must not be exaggerated

Expired, failed, unavailable, skipped and passed indicate different conditions and next actions.

Keep the specific state and avoid inventing account suspension, ban or permanent rejection. The practical question is not whether a bilingual reader likes the wording; it is whether a reader in either language would understand the same source fact, its qualification and the boundary of the available evidence. Where context does not answer that question, the translation needs a documented query instead of a confident guess.

4. Accessibility requires real alternatives

A label for audio, another modality or a support route is useful only if reachable, understandable and operational. Translating a nonexistent control is not remediation.

Inspect keyboard, screen reader and mobile flows with the product team as appropriate. The practical question is not whether a bilingual reader likes the wording; it is whether a reader in either language would understand the same source fact, its qualification and the boundary of the available evidence. Where context does not answer that question, the translation needs a documented query instead of a confident guess.

5. Authentication and bot detection are separate

Login requirements, multi-factor authentication, CAPTCHA and consent to data processing may appear on the same page but have different meanings.

Retain the separate labels and do not make one control appear to authorise another. The practical question is not whether a bilingual reader likes the wording; it is whether a reader in either language would understand the same source fact, its qualification and the boundary of the available evidence. Where context does not answer that question, the translation needs a documented query instead of a confident guess.

6. Security strings should preserve exact actions

Retry, refresh, resend, verify and continue can produce different operations. Incorrect substitution can cause a user to discard progress or submit an incomplete form.

Match each label with the control’s actual function and success state. The practical question is not whether a bilingual reader likes the wording; it is whether a reader in either language would understand the same source fact, its qualification and the boundary of the available evidence. Where context does not answer that question, the translation needs a documented query instead of a confident guess.

7. Avoid culturespecific assumptions in examples

Visual object categories, idioms, signs and number conventions may be differently understood. A grammatically correct prompt can still be ambiguous when the displayed category is unfamiliar.

Prefer controlled descriptions grounded in the actual task and test target-language comprehension. The practical question is not whether a bilingual reader likes the wording; it is whether a reader in either language would understand the same source fact, its qualification and the boundary of the available evidence. Where context does not answer that question, the translation needs a documented query instead of a confident guess.

8. Do not expose protected implementation details

A translator needs legitimate UI context, test screens and authorised documentation, not bypass procedures or secret scoring rules.

Localise the user-visible message while keeping internal tokens and security values protected. The practical question is not whether a bilingual reader likes the wording; it is whether a reader in either language would understand the same source fact, its qualification and the boundary of the available evidence. Where context does not answer that question, the translation needs a documented query instead of a confident guess.

9. Timeout is not lockout

A time-limited challenge can expire while a user’s account remains active. Reuse of generic ‘blocked’ wording misstates risk.

Translate time-limit and retry instructions separately from any true account access restriction. The practical question is not whether a bilingual reader likes the wording; it is whether a reader in either language would understand the same source fact, its qualification and the boundary of the available evidence. Where context does not answer that question, the translation needs a documented query instead of a confident guess.

10. Placeholder tokens are functional

Dynamic values such as {seconds}, {triesRemaining} or an accessible link label are inserted by software and may have grammar constraints.

Keep variables intact, document plural forms and test singular/zero/multiple values. The practical question is not whether a bilingual reader likes the wording; it is whether a reader in either language would understand the same source fact, its qualification and the boundary of the available evidence. Where context does not answer that question, the translation needs a documented query instead of a confident guess.

An icon may need an accessible name explaining refresh challenge or play audio. Replacing it with a generic image label breaks navigation.

Review visible and accessible text together, including focus order and feedback announcements. The practical question is not whether a bilingual reader likes the wording; it is whether a reader in either language would understand the same source fact, its qualification and the boundary of the available evidence. Where context does not answer that question, the translation needs a documented query instead of a confident guess.

12. Unclear instructions are a product defect to escalate

If the source omits which items to select or the alternative route is inaccessible, the translator cannot safely invent missing behaviour.

Flag the source issue, request a product decision and keep translation notes separate from published instructions. The practical question is not whether a bilingual reader likes the wording; it is whether a reader in either language would understand the same source fact, its qualification and the boundary of the available evidence. Where context does not answer that question, the translation needs a documented query instead of a confident guess.

A repeatable twelve-stage method

Use the method in order on a short source extract, then apply it to the full document or interaction. A second person should be able to reproduce your decisions from the version record and the cited source, rather than relying only on your fluency or personal preference.

Step 1: Map the entire verification journey

Collect screens for loading, challenge available, alternate mode, retry, expiry, failure, success and post-verification continuation. Capture browser and mobile versions.

Create a string-to-state map before translating isolated snippets. Record what was checked, which source supported the choice and what remains unresolved. This evidence trail is particularly important when the translated text could change a reader’s action. A concise and accurate unresolved question is preferable to a beautifully phrased invention.

Step 2: Define the purpose of each control

Determine whether the control starts a challenge, submits an answer, refreshes the image, plays audio, requests assistance or continues to another form step.

Record expected system behaviour and user action next to each source string. Record what was checked, which source supported the choice and what remains unresolved. This evidence trail is particularly important when the translated text could change a reader’s action. A concise and accurate unresolved question is preferable to a beautifully phrased invention.

Step 3: Extract task grammar

Underline quantifiers, objects, exceptions, direction and order. A small grammatical change such as all to one can break the user task.

Draft target wording and test it against sample challenges. Record what was checked, which source supported the choice and what remains unresolved. This evidence trail is particularly important when the translated text could change a reader’s action. A concise and accurate unresolved question is preferable to a beautifully phrased invention.

A privacy or tracking choice cannot be silently merged into ‘I am not a robot’. If the interface has consent choices, translate them as their own decisions.

Escalate any product copy that falsely bundles unrelated permissions. Record what was checked, which source supported the choice and what remains unresolved. This evidence trail is particularly important when the translated text could change a reader’s action. A concise and accurate unresolved question is preferable to a beautifully phrased invention.

Step 5: Review alternate modalities

Identify labels for audio challenge, new challenge, text alternatives, help, contact and accessible verification routes.

A localised interface must make available options discoverable. Record what was checked, which source supported the choice and what remains unresolved. This evidence trail is particularly important when the translated text could change a reader’s action. A concise and accurate unresolved question is preferable to a beautifully phrased invention.

Step 6: Protect variable structure

Preserve machine placeholders, reference tokens and security codes. Adapt grammatical agreement with tested plural variants rather than changing variable syntax.

Inspect both generated text and right-to-left or narrow-screen rendering as applicable. Record what was checked, which source supported the choice and what remains unresolved. This evidence trail is particularly important when the translated text could change a reader’s action. A concise and accurate unresolved question is preferable to a beautifully phrased invention.

Step 7: Translate outcome messages precisely

Make passed, wrong answer, temporary service failure, expired token, unavailable challenge and rate-limit conditions distinguishable.

Each outcome should tell the reader the next supported action without asserting a sanction not recorded. Record what was checked, which source supported the choice and what remains unresolved. This evidence trail is particularly important when the translated text could change a reader’s action. A concise and accurate unresolved question is preferable to a beautifully phrased invention.

Step 8: Use plain language without removing detail

Short prompts should be direct, familiar and unambiguous. If space constraints make correct wording impossible, seek interface redesign or expansion.

A small button is not permission to sacrifice task meaning. Record what was checked, which source supported the choice and what remains unresolved. This evidence trail is particularly important when the translated text could change a reader’s action. A concise and accurate unresolved question is preferable to a beautifully phrased invention.

Step 9: Run an accessibility-language review

Ask target-language users to find alternate methods, understand the task and recover from errors. Include people using different input and assistive methods when feasible.

Record observations by interaction state rather than only collecting stylistic opinions. Record what was checked, which source supported the choice and what remains unresolved. This evidence trail is particularly important when the translated text could change a reader’s action. A concise and accurate unresolved question is preferable to a beautifully phrased invention.

Step 10: Review against W3C guidance

Consult WCAG 2.2’s non-text content and accessible-authentication provisions and their explanatory materials for product requirements. Avoid declaring compliance solely because a string was translated.

Distinguish editorial accuracy from accessibility conformance testing. Record what was checked, which source supported the choice and what remains unresolved. This evidence trail is particularly important when the translated text could change a reader’s action. A concise and accurate unresolved question is preferable to a beautifully phrased invention.

Step 11: Verify security-neutrality

Check that localized success messages do not claim account ownership or identity proof and that failure messages do not reveal sensitive diagnostics.

Test public copy against the designed security disclosure policy. Record what was checked, which source supported the choice and what remains unresolved. This evidence trail is particularly important when the translated text could change a reader’s action. A concise and accurate unresolved question is preferable to a beautifully phrased invention.

Step 12: Version and release responsibly

Maintain screenshots, language files, approved strings, UI states and defect tickets. Re-test after changing CAPTCHA provider or authentication flow.

A translation that worked yesterday may become misleading if the workflow changes. Record what was checked, which source supported the choice and what remains unresolved. This evidence trail is particularly important when the translated text could change a reader’s action. A concise and accurate unresolved question is preferable to a beautifully phrased invention.

Twenty worked translation cases

The following miniature sources and mistakes are fictional teaching examples, not genuine institutional records or interface screenshots. Their purpose is to train the reader to preserve a specific distinction. In a real project, the authorised original and applicable rules or product documentation always control.

Case 1: Select all

Source scenario: “Select all images with traffic lights.” A plausible but unsafe rendering would be “’Choose a traffic light image.’” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “Select every image that contains a traffic light.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: The plural quantifier determines the correct interaction. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Transfer question: if the same line appeared in a different document, under a different authority, interface or version, what would still be true and what would require a fresh check? The transferable skill is to preserve the boundary between the recorded fact and the translator’s inference. When that boundary is visible, even a very short translation can be reliable.

Case 2: Select one

Source scenario: “Select the one image that differs from the others.” A plausible but unsafe rendering would be “’Select all different images.’” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “Choose the single image that is different from the rest.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: A mistaken all instruction can make an answer impossible. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Case 3: Audio alternative

Source scenario: “Choose an audio challenge instead.” A plausible but unsafe rendering would be “’Turn up the sound.’” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “Switch to the available audio-based challenge.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: Do not imply an audio mode is already playing. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Case 4: Refresh challenge

Source scenario: “Get a new challenge.” A plausible but unsafe rendering would be “’Repeat the same image.’” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “Request another challenge.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: It may regenerate content rather than replay the previous test. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Case 5: Expired challenge

Source scenario: “Challenge expired. Please try again.” A plausible but unsafe rendering would be “’Your account is locked.’” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “The challenge has expired; start a new attempt.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: Expiry is not a permanent security restriction. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Case 6: Incorrect answer

Source scenario: “The answer was not accepted. Try again.” A plausible but unsafe rendering would be “’This user is not human.’” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “The challenge response was not accepted; retry as instructed.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: Avoid making an inference about the person. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Case 7: Unavailable service

Source scenario: “Verification service temporarily unavailable.” A plausible but unsafe rendering would be “’You have failed verification.’” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “The service is temporarily unavailable; no personal failure is established.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: Keep infrastructure failure separate from wrong responses. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Case 8: Not a robot

Source scenario: “I’m not a robot.” A plausible but unsafe rendering would be “’I confirm my legal identity.’” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “A short user-facing statement that participates in the bot check, not an identity attestation.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: Do not silently strengthen the claim. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Case 9: Loading state

Source scenario: “Checking your response…” A plausible but unsafe rendering would be “’Verification successful.’” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “Your response is still being checked.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: In-progress and completed states need different text. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Case 10: Retry countdown

Source scenario: “Try again in {seconds} seconds.” A plausible but unsafe rendering would be “’One second remaining’ for all numbers.” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “Retain {seconds} and implement target-language plural forms correctly.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: Test zero, one and many without replacing the token. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Case 11: Help link

Source scenario: “Can’t complete the challenge? Get help.” A plausible but unsafe rendering would be “’You must solve it before receiving help.’” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “Provide an accessible route to assistance for someone unable to complete the task.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: Check whether the link is operable without solving the challenge. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Case 12: Read-aloud option

Source scenario: “Play the audio challenge.” A plausible but unsafe rendering would be “’Open the image gallery.’” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “Play the audio version of the verification task.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: The accessible name should match the control’s behaviour. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Case 13: Form continuation

Source scenario: “Verification passed; continue to submit the form.” A plausible but unsafe rendering would be “’Your application was accepted.’” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “The bot check succeeded; form submission is still a separate step.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: Do not confuse gate completion with final application acceptance. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Case 14: Token invalid

Source scenario: “Verification token is invalid or expired.” A plausible but unsafe rendering would be “’Password is incorrect.’” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “The token cannot be used in its current state.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: A token error is not necessarily a credential error. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Case 15: Challenge changed

Source scenario: “New images have been shown; select again.” A plausible but unsafe rendering would be “’Previous answer is still saved.’” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “The challenge content changed and requires a new selection.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: Dynamic challenge state must be explained accurately. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Case 16: Privacy label

Source scenario: “Privacy information about this verification service.” A plausible but unsafe rendering would be “’Accept all tracking.’” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “A link to information on the verification provider’s data practices.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: Viewing information and granting consent are different actions. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Case 17: Keyboard alternative

Source scenario: “Use the keyboard-accessible verification method.” A plausible but unsafe rendering would be “’Touch the screen.’” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “Use the alternative method available from the keyboard.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: Do not prescribe an input method excluded for the reader. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Case 18: Attempts remaining

Source scenario: “Two attempts remaining.” A plausible but unsafe rendering would be “’You are banned after this attempt.’” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “The system reports two remaining attempts, without an unsupported ban claim.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: Keep counters and consequence claims separate. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Case 19: Challenge complete

Source scenario: “Verification complete.” A plausible but unsafe rendering would be “’You may now access every service.’” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “The current verification step is complete.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: A successful local step does not create global access rights. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Case 20: Image description

Source scenario: “Image challenge: select squares with bridges.” A plausible but unsafe rendering would be “’Decorative picture.’” The problem is not just word choice. It changes the state or relationship that the reader would rely on when choosing what to do next.

Safer rendering: “Accessible text identifies the challenge’s purpose and its required action.” This version does not claim more knowledge or permission than the source provides. It keeps the relevant qualifier attached to the relevant object or event rather than hiding it in a separate footnote.

Verification test: An inaccessible object cannot be rescued by a generic image label. Now imagine a reader who sees only the target-language version. Ask them to state the document’s claim and the next action it supports. Compare their account against the source scenario, not against how polished the sentence sounds.

Failure modes and how to repair them

1. Removing ‘all’ from a selection instruction

The challenge outcome can depend on the quantifier. Cross-check images and target text before release. Repair it by returning to the smallest source span that controls the decision, writing down what that span does and does not establish, and choosing a target expression that keeps the distinction. If the source is inconsistent, open a query rather than quietly rewriting reality.

2. Translating expiry as a ban

This invents a sanction and may frighten legitimate visitors. Name the temporary state precisely. Repair it by returning to the smallest source span that controls the decision, writing down what that span does and does not establish, and choosing a target expression that keeps the distinction. If the source is inconsistent, open a query rather than quietly rewriting reality.

Different choices have different purposes. Preserve separate controls and permissions. Repair it by returning to the smallest source span that controls the decision, writing down what that span does and does not establish, and choosing a target expression that keeps the distinction. If the source is inconsistent, open a query rather than quietly rewriting reality.

4. Failing to translate the accessible name of an icon

An unlabeled refresh or audio button may disappear from a screen-reader user’s practical workflow. Repair it by returning to the smallest source span that controls the decision, writing down what that span does and does not establish, and choosing a target expression that keeps the distinction. If the source is inconsistent, open a query rather than quietly rewriting reality.

5. Assuming audio CAPTCHA is a complete accessibility solution

W3C guidance raises specific concerns about transcription and cognitive demands. Evaluate the whole accessible path. Repair it by returning to the smallest source span that controls the decision, writing down what that span does and does not establish, and choosing a target expression that keeps the distinction. If the source is inconsistent, open a query rather than quietly rewriting reality.

6. Using unexplained technical jargon

Choose plain action wording for readers while retaining necessary technical meaning in supporting details. Repair it by returning to the smallest source span that controls the decision, writing down what that span does and does not establish, and choosing a target expression that keeps the distinction. If the source is inconsistent, open a query rather than quietly rewriting reality.

7. Publishing a label without live UI context

An accurate standalone sentence may point to a nonexistent button after interface changes. Test in the released workflow. Repair it by returning to the smallest source span that controls the decision, writing down what that span does and does not establish, and choosing a target expression that keeps the distinction. If the source is inconsistent, open a query rather than quietly rewriting reality.

8. Changing a variable token

Removing braces, changing variable spelling or ignoring plurals can break generated messages. Preserve and test token integrity. Repair it by returning to the smallest source span that controls the decision, writing down what that span does and does not establish, and choosing a target expression that keeps the distinction. If the source is inconsistent, open a query rather than quietly rewriting reality.

9. Reporting a service outage as user failure

A temporary provider error should not be described as a wrong answer or suspicious account. Repair it by returning to the smallest source span that controls the decision, writing down what that span does and does not establish, and choosing a target expression that keeps the distinction. If the source is inconsistent, open a query rather than quietly rewriting reality.

10. Declaring WCAG compliance from translated strings

Accessible language is one component of a product. Conformance requires testing the implemented interface and applicable criteria. Repair it by returning to the smallest source span that controls the decision, writing down what that span does and does not establish, and choosing a target expression that keeps the distinction. If the source is inconsistent, open a query rather than quietly rewriting reality.

Practice and transfer: four weeks of deliberate work

The programme uses brief extracts, error diagnosis and independent retrieval. Each week moves beyond recognizing a correct answer and asks whether you can preserve the source’s structure in an unfamiliar situation. Avoid practising only with a fixed phrase list; new objects and new decision states are what reveal false confidence.

Week 1: map the journey

Create a state diagram of a fictional sign-in page: load, challenge, alternate route, retry, timeout, completion and continuation. Translate each screen without confusing verification with form submission.

End-of-week check: hide your first translation and reconstruct the source action, evidence, exception and uncertainty from memory. Then make a second translation and compare the two for missed qualifiers, overconfident claims and lost identifiers. Log one concrete rule you can carry into a different topic.

Week 2: test quantifiers and controls

Translate twenty sample selection prompts using one, all, none, next and try again. Test with example images and accessible control labels. Record any prompt where the target language invites a different selection.

Week 3: accessibility review

Compare visible text, accessible names and help paths in a test interface. Consult W3C’s relevant criteria. Flag product-level defects rather than promising that textual fixes alone guarantee access.

Week 4: error-state audit

Use fake tokens and simulated timeouts to compare messages for expired, rejected, unavailable and completed. Check placeholders, plurals, screen sizes and the next permissible action before issuing a release report.

Frequently asked questions

What does CAPTCHA mean?

It denotes an automated challenge intended to distinguish automated from human interaction in a particular context. It is not, by itself, proof of real-world identity.

Is ‘I’m not a robot’ a legal declaration?

It is generally interface text for a bot-check workflow. Translate its functional intent without turning it into an account or legal identity guarantee.

Do all visual CAPTCHAs need alternatives?

W3C’s non-text content guidance addresses identifying CAPTCHA purpose and alternative modalities, while accessible-authentication criteria add further considerations. Review the specific application.

Can an audio challenge solve every accessibility problem?

No. Depending on the test, audio transcription can still impose cognitive or sensory barriers. Evaluate functional alternatives and user needs.

Should translators localise error codes?

Usually retain machine identifiers while translating accompanying human explanations. Ask the development team about any token that appears to be user-facing text.

What is the distinction between expired and failed?

Expired describes a time or validity condition. Failed generally describes an unsuccessful operation or answer. Neither necessarily means a permanent account restriction.

How do we translate challenge objects?

Keep the exact selection category and quantity. Test that target-language users identify the same intended set of images.

Can a translation team repair an inaccessible CAPTCHA?

It can surface and sometimes improve discoverability of an existing option. It cannot create missing functionality through copy alone.

Only where the product truly offers or requests a separate consent choice. Do not infer a consent grant from completing a challenge.

What is the best test?

Run an end-to-end task in each language, including unsuccessful and alternate-mode paths, and compare the action the reader believes they should take.

A practical final quality checklist

What exact object, person, document, control or record is being described?

Which words report a fact, which request an action, and which express uncertainty or a condition?

Have any numbers, units, dates, identifiers, roles, statuses or source-versus-destination relationships changed?

Does the target version grant approval, guarantee success or assert completion without evidence?

Would a target-language reader take the same justified next step as a source-language reader?

What should remain untranslated because it is a machine-readable identifier, code or controlled label?

What would a specialist need to verify before the text is used in a consequential setting?

Is there a documented source version, review record and clear path to correct newly discovered errors?

Use this checklist to test CAPTCHA challenges and human-verification interface text before release. A reader who answers differently in another language has revealed a functional translation risk. The repair should be local and precise: restore the missing distinction, check the interface or record again and confirm the corrected target text.

Sources and further reading

W3C: WCAG 2.2 accessibility requirements — https://www.w3.org/TR/wcag/

W3C: Understanding Accessible Authentication (Minimum) — https://www.w3.org/WAI/WCAG22/Understanding/accessible-authentication-minimum.html

W3C: CAPTCHA text alternatives and purpose — https://www.w3.org/WAI/WCAG22/Techniques/general/G143

These references provide definitions, standards or official guidance within their respective scopes. They do not validate a fictional example or replace the governing rules, user documentation or local professional review for a real case. Check dates and versions when relying on a standard or a product guide.

Continue learning with eduKateSG

The next route depends on the problem you actually have. For general lexical precision, learn how word meaning, register and use develop. For language structure, study how English represents instructions and evidence. For a neighbouring specialist task, use the dedicated translation article rather than folding two different search intentions into one answer.

Accessible alt-text translation — https://edukatesg.com/2026/10/31/top-ways-to-translate-correctly-alt-text-screen-reader-labels/

Support ticket status translation — https://edukatesg.com/2026/10/31/top-ways-to-translate-correctly-support-tickets-escalation-response-deadlines/

How English Works — https://edukatesg.com/how-english-works-the-complete-map-and-learning-routes/

Vocabulary Learning Hub — https://edukatesg.com/vocabulary-learning-hub/

The central principle is simple: a translation of CAPTCHA challenges and human-verification interface text should preserve the same decisions, distinctions and uncertainty as the source. Accurate prose is useful only when the reader can still tell what happened, what has not happened, and what to do next.

Top Ways to Translate Correctly | Translate CAPTCHA Challenges and Verification Prompts Without Blocking Accessibility