VIEW THIS AS

Auto mode follows the Route Engine until you choose a viewpoint.

YOU ARE HERE

ROUTE CHECK

CONNECTED TO

WHAT NEXT

Use the canonical route for this room, or HELP if you are unsure.

Top Ways to Translate Correctly | Software Strings, Placeholders and Plural Rules

A correct software translation must preserve the meaning of a button or notification and the interface behaviour that displays it. People searching for how to translate software UI strings, software localisation best practices, placeholder translation and plural rules in translation need methods that keep variables intact, show natural target-language grammar and prevent broken screens. Translating isolated words without understanding the application’s state can produce a fluent but unusable interface.

This practical guide teaches app localisation, interface string translation, dynamic message translation, plural-aware grammar and layout testing. It covers buttons, error messages, names, counters, permissions, billing notices and accessible labels. Each example is invented for education and has a clearly stated failure mode, repair and observable test. The subject is the live behaviour of software text, not general dictionary meaning or unrelated translation genres.

The short answer: identify the user’s task; mark tokens that software controls; translate the complete action or claim; ask for the necessary plural and grammar variants; test ordinary and edge-case substituted values; review the rendered target-language screen; and document a reproducible correction. The twenty-two scenarios below show how to apply that sequence across products.

What makes a translated software string correct?

A resource string can be a statement, a control label, a warning, a navigation landmark or a fragment assembled with other fragments. Its correctness depends on the user being able to infer the right consequence at the point of interaction. “Apply” illustrates this: it may activate a filter, submit an application or apply new configuration. A dictionary lists possibilities, but the product screen selects which one is true. Thus context is not a cosmetic preference; it is the evidence required to translate accurately.

Correctness also includes referential integrity. A placeholder identifies a runtime value, a link marker identifies an interactive destination, and a plural selector chooses message variants. Those machine-owned structures should not be rewritten as if they were ordinary words. When natural grammar cannot be expressed through the current resource format, the right answer is a small engineering request with a concrete example, not a knowingly unnatural target sentence.

The MDN guide to Intl.PluralRules explains the reason count categories matter: languages may group numeric quantities differently. The W3C guidance on labelling controls shows why interface names must identify functions. These are foundational references rather than a substitute for the rules of any particular localisation framework.

First principles: meaning, data and observable action

Begin with three questions: What is the system telling the person? Which portion of the message will be supplied by code? What should the user be able to do next? An attractive translation that answers the first but destroys the second fails; an exact token-safe translation that hides the third also fails. Review all three together. The goal is a coherent interface experience, not a row of individually polished equivalents.

Software content also has high-consequence pockets. A short confirmation can authorise access to a private folder, begin billing after a free trial or permanently delete information. Such strings deserve extra scrutiny because changing one qualifier can affect a user’s decision. Treat source ambiguity as a question for the product owner rather than a licence to make the strongest or safest-sounding guess.

A repeatable nine-stage method

1. Define the user action

Record the screen, state and consequence before translating any short button label.

At this stage, record a test that another person can repeat. Ask which user action or data value would expose a mistake, and which observed result would confirm the translation. Do not rely on how natural the text sounds when removed from the screen. A working string must survive substitution, navigation and the actual product state. That test should be specific enough to distinguish a linguistic defect from a rendering defect.

A second review should challenge the assumptions behind the first. What happens if the value is empty? If the interface is narrower? If a different person is named? If the product is in a failure rather than a successful state? Not every question applies to every string, but choosing the relevant stress test is part of translating it correctly.

2. Mark machine-controlled tokens

List braces, variable names, markup, escape sequences and data formats that must not be changed.

At this stage, record a test that another person can repeat. Ask which user action or data value would expose a mistake, and which observed result would confirm the translation. Do not rely on how natural the text sounds when removed from the screen. A working string must survive substitution, navigation and the actual product state. That test should be specific enough to distinguish a linguistic defect from a rendering defect.

A second review should challenge the assumptions behind the first. What happens if the value is empty? If the interface is narrower? If a different person is named? If the product is in a failure rather than a successful state? Not every question applies to every string, but choosing the relevant stress test is part of translating it correctly.

3. Request runtime context

Obtain sample values, empty values, long strings and meaningful plural counts.

At this stage, record a test that another person can repeat. Ask which user action or data value would expose a mistake, and which observed result would confirm the translation. Do not rely on how natural the text sounds when removed from the screen. A working string must survive substitution, navigation and the actual product state. That test should be specific enough to distinguish a linguistic defect from a rendering defect.

A second review should challenge the assumptions behind the first. What happens if the value is empty? If the interface is narrower? If a different person is named? If the product is in a failure rather than a successful state? Not every question applies to every string, but choosing the relevant stress test is part of translating it correctly.

4. Write target-language grammar

Allow word order and agreement to change without corrupting the data contract.

At this stage, record a test that another person can repeat. Ask which user action or data value would expose a mistake, and which observed result would confirm the translation. Do not rely on how natural the text sounds when removed from the screen. A working string must survive substitution, navigation and the actual product state. That test should be specific enough to distinguish a linguistic defect from a rendering defect.

A second review should challenge the assumptions behind the first. What happens if the value is empty? If the interface is narrower? If a different person is named? If the product is in a failure rather than a successful state? Not every question applies to every string, but choosing the relevant stress test is part of translating it correctly.

5. Handle plural selection

Ask which locale rules and message syntax the actual product supports.

At this stage, record a test that another person can repeat. Ask which user action or data value would expose a mistake, and which observed result would confirm the translation. Do not rely on how natural the text sounds when removed from the screen. A working string must survive substitution, navigation and the actual product state. That test should be specific enough to distinguish a linguistic defect from a rendering defect.

A second review should challenge the assumptions behind the first. What happens if the value is empty? If the interface is narrower? If a different person is named? If the product is in a failure rather than a successful state? Not every question applies to every string, but choosing the relevant stress test is part of translating it correctly.

6. Protect important consequences

Review deletion, payment, permission, consent, health and recovery text carefully.

At this stage, record a test that another person can repeat. Ask which user action or data value would expose a mistake, and which observed result would confirm the translation. Do not rely on how natural the text sounds when removed from the screen. A working string must survive substitution, navigation and the actual product state. That test should be specific enough to distinguish a linguistic defect from a rendering defect.

A second review should challenge the assumptions behind the first. What happens if the value is empty? If the interface is narrower? If a different person is named? If the product is in a failure rather than a successful state? Not every question applies to every string, but choosing the relevant stress test is part of translating it correctly.

7. Test interface and accessibility

Inspect responsive layouts, keyboard actions, screen readers and mixed scripts.

At this stage, record a test that another person can repeat. Ask which user action or data value would expose a mistake, and which observed result would confirm the translation. Do not rely on how natural the text sounds when removed from the screen. A working string must survive substitution, navigation and the actual product state. That test should be specific enough to distinguish a linguistic defect from a rendering defect.

A second review should challenge the assumptions behind the first. What happens if the value is empty? If the interface is narrower? If a different person is named? If the product is in a failure rather than a successful state? Not every question applies to every string, but choosing the relevant stress test is part of translating it correctly.

8. Log reproducible defects

Record key, locale, app version, input values, expected outcome and actual result.

At this stage, record a test that another person can repeat. Ask which user action or data value would expose a mistake, and which observed result would confirm the translation. Do not rely on how natural the text sounds when removed from the screen. A working string must survive substitution, navigation and the actual product state. That test should be specific enough to distinguish a linguistic defect from a rendering defect.

A second review should challenge the assumptions behind the first. What happens if the value is empty? If the interface is narrower? If a different person is named? If the product is in a failure rather than a successful state? Not every question applies to every string, but choosing the relevant stress test is part of translating it correctly.

9. Verify regression after changes

Test other screens reusing the resource to prevent a local fix causing global damage.

At this stage, record a test that another person can repeat. Ask which user action or data value would expose a mistake, and which observed result would confirm the translation. Do not rely on how natural the text sounds when removed from the screen. A working string must survive substitution, navigation and the actual product state. That test should be specific enough to distinguish a linguistic defect from a rendering defect.

A second review should challenge the assumptions behind the first. What happens if the value is empty? If the interface is narrower? If a different person is named? If the product is in a failure rather than a successful state? Not every question applies to every string, but choosing the relevant stress test is part of translating it correctly.

Twenty-two worked cases in software localisation

Case 1: File counts

Situation: The app reports {count} files uploaded, including zero and one.

Visible risk: A literal English plural ending misses languages with more count categories.

Translation method: Preserve {count} and provide locale-specific variants through the product’s chosen plural framework.

Acceptance test: Test values 0, 1, 2, 5, 21 and 101 in the rendered interface.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 2: Personal greetings

Situation: Welcome back, {first_name} appears on an account dashboard.

Visible risk: Renaming the token or assuming every user supplies a name breaks display.

Translation method: Keep the identifier unchanged and request a grammatically complete fallback for missing names.

Acceptance test: Test long names, empty names and scripts other than Latin.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 3: Argument order

Situation: {user} shared {file} with {group} describes an activity.

Visible risk: Copying source word order can misstate who shared what with whom.

Translation method: Move named placeholders where permitted without exchanging their roles.

Acceptance test: Substitute unmistakably different names for user, file and group.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 4: Apply button

Situation: A single resource called Apply is used for a filter action.

Visible risk: A dictionary equivalent might suggest applying for employment instead.

Translation method: Identify the actual action and translate the control’s consequence.

Acceptance test: Activate the button and confirm what changes on screen.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 5: Delete confirmation

Situation: Delete workspace? has Cancel and Delete controls.

Visible risk: A vague translation hides permanent deletion or suggests archiving.

Translation method: Preserve action, scope and reversibility according to actual product behaviour.

Acceptance test: Ask a reviewer which data will disappear after tapping the button.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 6: Retry error

Situation: Connection failed ({error_code}). Try again appears after a network error.

Visible risk: Changing the placeholder or treating every failure as retryable creates confusion.

Translation method: Preserve the code and distinguish retry from contact-support states.

Acceptance test: Trigger a test failure and verify code, help link and action.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 7: Inline hyperlink

Situation: A privacy notice includes a linked phrase inside translated grammar.

Visible risk: Unbalanced markup or a misplaced link changes what can be opened.

Translation method: Use supported rich-text markers and link the correct target phrase.

Acceptance test: Inspect rendered click area and keyboard navigation.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 8: Long mobile tabs

Situation: Notifications appears in a narrow bottom navigation bar.

Visible risk: Hard truncation removes essential distinctions.

Translation method: Choose concise natural wording and test responsive layout instead of deleting meaning.

Acceptance test: Test small devices and large text settings.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 9: Time variables

Situation: Your appointment is on {date} at {time} is dynamically formatted.

Visible risk: Hardcoded punctuation may duplicate date elements or conflict with locale formatting.

Translation method: Clarify whether tokens already supply local date and time representations.

Acceptance test: Test two regions and a time-zone transition.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 10: Mixed scripts

Situation: An Arabic notification contains an English URL and a number.

Visible risk: Bidirectional ordering may become visually confusing.

Translation method: Use supported direction-aware components rather than undocumented invisible characters.

Acceptance test: Check copy-paste, cursor movement and reading order in an RTL build.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 11: Keyboard shortcut

Situation: Press Ctrl+S to save appears in keyboard help.

Visible risk: The shortcut may be wrong on another operating system.

Translation method: Preserve platform-specific key tokens and translate only the actual instruction.

Acceptance test: Test advertised shortcuts on supported platforms.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 12: Gender assumptions

Situation: {name} invited you appears without identity or grammatical-gender information.

Visible risk: A translated construction may impose an unsupported assumption about the sender.

Translation method: Use a natural neutral sentence or ask for supported grammatical variants.

Acceptance test: Test with a mixture of names and an organisation sender.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 13: Percent progress

Situation: The upload label reads {percent}% complete.

Visible risk: A formatter may already include the percent sign, creating duplication.

Translation method: Document which component owns symbols and which owns text.

Acceptance test: Render at 0, 25, 99 and 100 percent.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 14: Shared key collision

Situation: Close names both a dismiss button and geographic proximity.

Visible risk: Forcing one translated word onto both changes one meaning.

Translation method: Split the resource keys or supply clear context metadata.

Acceptance test: Search every runtime use and distinguish the actions.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 15: Trial billing

Situation: Free for seven days, then billed monthly appears before payment.

Visible risk: Shortening can erase the transition to recurring charges.

Translation method: Keep the exact commercial consequences and obtain appropriate review.

Acceptance test: Compare target text with checkout terms and billing configuration.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 16: Accessible icon

Situation: An icon-only delete control has the label trash can.

Visible risk: Describing the graphic does not tell screen-reader users what it does.

Translation method: Translate the action, such as Delete saved item, not the image shape.

Acceptance test: Navigate the control with a screen reader.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 17: Missing fallback

Situation: Welcome, {name}! displays even when name is absent.

Visible risk: Punctuation and grammar remain stranded after an empty substitution.

Translation method: Build a neutral fallback message as a separate supported state.

Acceptance test: Test null, empty and whitespace-only values.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 18: Filter count

Situation: {count} filters selected changes as filters are checked.

Visible risk: A single translated string may fail in zero or other plural categories.

Translation method: Use supported plural categories and preserve the dynamic variable.

Acceptance test: Test each category with native reviewers.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 19: Forced newline

Situation: A desktop banner contains a source-language hard line break.

Visible risk: Copying it can split a target-language phrase on mobile.

Translation method: Allow responsive wrapping where possible and document true semantic breaks.

Acceptance test: Test desktop, mobile and enlarged font sizes.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 20: Permission prompt

Situation: Allow {app} to access {folder}? asks for authorisation.

Visible risk: Swapping placeholder referents can misstate who is gaining access.

Translation method: Verify referents and keep each named placeholder associated with its role.

Acceptance test: Test with contrasting names and explicit permission scopes.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 21: Search states

Situation: No results found appears while the query is still loading.

Visible risk: The user can mistake a temporary state for permanent absence.

Translation method: Separate loading, empty-result, failure and completed states in copy and code.

Acceptance test: Simulate all four states and check available next actions.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Case 22: Renamed feature

Situation: A product release replaces an old feature label, but help text lags.

Visible risk: A consistent old translation can direct users to a nonexistent button.

Translation method: Update translation memory with versioned context and correct the new release’s strings.

Acceptance test: Inspect onboarding, tooltips, settings and support routes together.

Why this matters: the decision is not settled by a bilingual word list. The same sentence can become wrong if it appears in a different product state, contains substituted data or gives a user the wrong available action. Write a one-line explanation naming the decisive context, then identify the smallest reliable correction. If engineering support is needed, describe the necessary runtime behaviour rather than disguising it as a stylistic concern.

Practice variation: recreate this problem using a different target language, screen size or sample value. Ask a second reviewer what the screen means and what action comes next. Compare their response with the intended product behaviour and revise only when there is a demonstrable gap. A good learning record includes the original message, contextual meaning, translated message, rendered state and a note on which assumption was tested.

Failure modes to diagnose instead of simply proofreading again

1. Wrong token name

A machine-controlled identifier is translated as a natural word and stops being replaced.

Reproduce the problem with one exact locale and input value. Compare the source resource, translation resource, substitution layer and screen output. Correct the first place where intended meaning or machine syntax diverges. Recheck all screens using that key: a local improvement can create a regression in another state if resources are reused without context.

An effective bug report describes the user-visible consequence. Saying that a word is unnatural gives less guidance than explaining that the button’s label suggests archiving while the action deletes data. Use the review to teach the distinction between a linguistic ambiguity, an engineering format issue and an accessibility or layout defect.

2. No grammar variants

A count-driven message renders the wrong plural or surrounding agreement.

Reproduce the problem with one exact locale and input value. Compare the source resource, translation resource, substitution layer and screen output. Correct the first place where intended meaning or machine syntax diverges. Recheck all screens using that key: a local improvement can create a regression in another state if resources are reused without context.

An effective bug report describes the user-visible consequence. Saying that a word is unnatural gives less guidance than explaining that the button’s label suggests archiving while the action deletes data. Use the review to teach the distinction between a linguistic ambiguity, an engineering format issue and an accessibility or layout defect.

3. Incorrect action word

An ambiguous source label is translated without the screen or consequence.

Reproduce the problem with one exact locale and input value. Compare the source resource, translation resource, substitution layer and screen output. Correct the first place where intended meaning or machine syntax diverges. Recheck all screens using that key: a local improvement can create a regression in another state if resources are reused without context.

An effective bug report describes the user-visible consequence. Saying that a word is unnatural gives less guidance than explaining that the button’s label suggests archiving while the action deletes data. Use the review to teach the distinction between a linguistic ambiguity, an engineering format issue and an accessibility or layout defect.

4. Truncated meaning

A warning becomes shorter but loses what is being authorised, deleted or charged.

Reproduce the problem with one exact locale and input value. Compare the source resource, translation resource, substitution layer and screen output. Correct the first place where intended meaning or machine syntax diverges. Recheck all screens using that key: a local improvement can create a regression in another state if resources are reused without context.

An effective bug report describes the user-visible consequence. Saying that a word is unnatural gives less guidance than explaining that the button’s label suggests archiving while the action deletes data. Use the review to teach the distinction between a linguistic ambiguity, an engineering format issue and an accessibility or layout defect.

5. Broken accessibility

The visible button works but its accessible label states the wrong purpose.

Reproduce the problem with one exact locale and input value. Compare the source resource, translation resource, substitution layer and screen output. Correct the first place where intended meaning or machine syntax diverges. Recheck all screens using that key: a local improvement can create a regression in another state if resources are reused without context.

An effective bug report describes the user-visible consequence. Saying that a word is unnatural gives less guidance than explaining that the button’s label suggests archiving while the action deletes data. Use the review to teach the distinction between a linguistic ambiguity, an engineering format issue and an accessibility or layout defect.

6. Mismatched locale formatting

A number, date or unit is reformatted by both code and text.

Reproduce the problem with one exact locale and input value. Compare the source resource, translation resource, substitution layer and screen output. Correct the first place where intended meaning or machine syntax diverges. Recheck all screens using that key: a local improvement can create a regression in another state if resources are reused without context.

An effective bug report describes the user-visible consequence. Saying that a word is unnatural gives less guidance than explaining that the button’s label suggests archiving while the action deletes data. Use the review to teach the distinction between a linguistic ambiguity, an engineering format issue and an accessibility or layout defect.

7. Reused obsolete label

A translation remains tied to a renamed feature in a newer release.

Reproduce the problem with one exact locale and input value. Compare the source resource, translation resource, substitution layer and screen output. Correct the first place where intended meaning or machine syntax diverges. Recheck all screens using that key: a local improvement can create a regression in another state if resources are reused without context.

An effective bug report describes the user-visible consequence. Saying that a word is unnatural gives less guidance than explaining that the button’s label suggests archiving while the action deletes data. Use the review to teach the distinction between a linguistic ambiguity, an engineering format issue and an accessibility or layout defect.

8. Unavailable message state

The wrong string appears during loading, failure, empty result or success.

Reproduce the problem with one exact locale and input value. Compare the source resource, translation resource, substitution layer and screen output. Correct the first place where intended meaning or machine syntax diverges. Recheck all screens using that key: a local improvement can create a regression in another state if resources are reused without context.

An effective bug report describes the user-visible consequence. Saying that a word is unnatural gives less guidance than explaining that the button’s label suggests archiving while the action deletes data. Use the review to teach the distinction between a linguistic ambiguity, an engineering format issue and an accessibility or layout defect.

Practice and transfer: a six-session learning sequence

Practice 1: Actions

Build a small settings screen with Save, Apply, Cancel and Reset. Write the real outcome for every control, then translate. Ask a reader to predict consequences from target labels.

Record your first version, the rendered evidence and the correction. Then repeat the task with a different number, name, or interface state. Real transfer means identifying the governing problem in a fresh example without needing a heading that announces which rule to use.

Practice 2: Dynamic names

Prepare welcome, invitation and sharing messages with named placeholders. Substitute long, short and missing names. Look for grammar and punctuation that survive each substitution.

Record your first version, the rendered evidence and the correction. Then repeat the task with a different number, name, or interface state. Real transfer means identifying the governing problem in a fresh example without needing a heading that announces which rule to use.

Practice 3: Counts

Translate a file-count message into a language with different plural categories. Use the application’s supported plural format and test representative numbers, including zero.

Record your first version, the rendered evidence and the correction. Then repeat the task with a different number, name, or interface state. Real transfer means identifying the governing problem in a fresh example without needing a heading that announces which rule to use.

Practice 4: Responsiveness

Place translated controls into desktop and narrow mobile layouts. Increase text size and inspect clipping. Fix layout before removing essential meaning.

Record your first version, the rendered evidence and the correction. Then repeat the task with a different number, name, or interface state. Real transfer means identifying the governing problem in a fresh example without needing a heading that announces which rule to use.

Practice 5: Security and billing

Translate a permission prompt and a trial-to-paid renewal notice. Underline every affected object, duration and consequence before revising the wording.

Record your first version, the rendered evidence and the correction. Then repeat the task with a different number, name, or interface state. Real transfer means identifying the governing problem in a fresh example without needing a heading that announces which rule to use.

Practice 6: Release review

Rename a mock product feature, then inspect settings, tooltips, onboarding and support messages for obsolete references. Test the current build rather than trusting older screenshots.

Record your first version, the rendered evidence and the correction. Then repeat the task with a different number, name, or interface state. Real transfer means identifying the governing problem in a fresh example without needing a heading that announces which rule to use.

Frequently asked questions

Why is software localisation different from ordinary translation?

A string may be assembled with dynamic values, shown in a specific interface state or chosen by a plural rule. Wording must be accurate and render correctly under those constraints.

To verify this guidance on a real project, identify the exact string resource, the supported locale and a test value. Examine the rendered result and ask which decision the user can now make. If the interface behaves differently from the sentence’s promise, record a defect even if the wording initially reads smoothly.

Do I translate {count} or {first_name}?

Usually not. These are machine-controlled identifiers. Preserve them exactly and translate the surrounding human-readable language unless the project’s string syntax explicitly says otherwise.

To verify this guidance on a real project, identify the exact string resource, the supported locale and a test value. Examine the rendered result and ask which decision the user can now make. If the interface behaves differently from the sentence’s promise, record a defect even if the wording initially reads smoothly.

Why are plural categories important?

English has relatively simple cardinal plural patterns. Other languages can require multiple categories and changes to nearby words, so a single English-style suffix is unsafe.

To verify this guidance on a real project, identify the exact string resource, the supported locale and a test value. Examine the rendered result and ask which decision the user can now make. If the interface behaves differently from the sentence’s promise, record a defect even if the wording initially reads smoothly.

Can an AI tool handle interface strings?

It can suggest language, but missing screen context and dynamic token rules make unreviewed output risky. Always inspect the rendered state and preserve identifiers.

To verify this guidance on a real project, identify the exact string resource, the supported locale and a test value. Examine the rendered result and ask which decision the user can now make. If the interface behaves differently from the sentence’s promise, record a defect even if the wording initially reads smoothly.

What is the safest way to handle a long translation?

Try meaningful concise wording and layout adaptation, then test actual screen widths and enlarged text. Never hide a payment or security consequence merely to save space.

To verify this guidance on a real project, identify the exact string resource, the supported locale and a test value. Examine the rendered result and ask which decision the user can now make. If the interface behaves differently from the sentence’s promise, record a defect even if the wording initially reads smoothly.

Should a translator edit code to solve grammar?

Changes to resource schemas, plural selectors and runtime fallbacks belong with product engineering. Identify the linguistic requirement clearly and agree on a supported implementation.

To verify this guidance on a real project, identify the exact string resource, the supported locale and a test value. Examine the rendered result and ask which decision the user can now make. If the interface behaves differently from the sentence’s promise, record a defect even if the wording initially reads smoothly.

What are good sample counts for testing?

Use values that cover all plural categories in the chosen locale and include zero and large numbers. Do not assume that the English singular/plural pair covers every target.

To verify this guidance on a real project, identify the exact string resource, the supported locale and a test value. Examine the rendered result and ask which decision the user can now make. If the interface behaves differently from the sentence’s promise, record a defect even if the wording initially reads smoothly.

How should a label for an icon be translated?

Describe the control’s function, such as opening settings or deleting an item. A visual description alone may not communicate the action to a screen-reader user.

To verify this guidance on a real project, identify the exact string resource, the supported locale and a test value. Examine the rendered result and ask which decision the user can now make. If the interface behaves differently from the sentence’s promise, record a defect even if the wording initially reads smoothly.

What if different screens use the same English word?

Check their meanings. One resource can be shared when the user action is truly the same; otherwise use separate keys and context annotations.

To verify this guidance on a real project, identify the exact string resource, the supported locale and a test value. Examine the rendered result and ask which decision the user can now make. If the interface behaves differently from the sentence’s promise, record a defect even if the wording initially reads smoothly.

How do I learn this skill as a beginner?

Build a mock interface containing a counter, name placeholder, deletion warning, error state and form controls. Translate it, substitute sample data and test the resulting actions.

To verify this guidance on a real project, identify the exact string resource, the supported locale and a test value. Examine the rendered result and ask which decision the user can now make. If the interface behaves differently from the sentence’s promise, record a defect even if the wording initially reads smoothly.

What should a localisation defect report include?

Include locale, key, app version, screen, input values, current rendering, expected result, screenshot and reproduction steps. This makes the issue actionable.

To verify this guidance on a real project, identify the exact string resource, the supported locale and a test value. Examine the rendered result and ask which decision the user can now make. If the interface behaves differently from the sentence’s promise, record a defect even if the wording initially reads smoothly.

Which external language API can help?

Modern internationalisation libraries offer locale-aware plural and formatting facilities. Inspect the specific stack used by the product and test its current supported locales.

To verify this guidance on a real project, identify the exact string resource, the supported locale and a test value. Examine the rendered result and ask which decision the user can now make. If the interface behaves differently from the sentence’s promise, record a defect even if the wording initially reads smoothly.

Further routes: grammar, vocabulary and interface reading

For careful lexical decisions, read the translation-series owners on false friends and natural collocations. The earlier numbers, units and dates article owns general human-readable format choices; this one explains the software runtime contract. For target-language study, continue to the Vocabulary Learning Hub and How English Works.

The final test is functional equivalence: can a target-language user recognise a control, understand a warning, interpret substituted data and complete the same task without new ambiguity? That is the practical standard for software string translation. When the answer depends on unavailable context or framework behaviour, document what must be verified rather than presenting a guess as finished work.