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 | ActivityPub Multilingual Posts and Actors

ActivityPub and ActivityStreams 2.0 human-language values in federated social objects means preserving the meaning of human-readable content while keeping the technical structure that lets the system identify, display and update it. This guide answers practical questions such as how to translate ActivityPub and ActivityStreams 2.0 human-language values in federated social objects, which values are safe to localize, how to prevent identifier damage, and which language, validation and quality assurance checks should happen before a multilingual release. It is written for federated-community moderators, documentation writers, accessibility reviewers, translators and social-software localization teams. A useful result is not merely elegant target-language wording: it must continue to behave correctly when a real reader or application uses the translated content.

The visible problem is specific: A multilingual post can be translated accurately yet be published to the wrong audience, detach from its original author, lose a link or represent a translated copy as a new unrelated original. Searchers may call this multilingual content localization, technical translation, translation quality assurance or localization testing. Those expressions point to the same practical demand in this chapter: make a clear boundary between language meant for people and data meant for software, then test what the reader actually sees. Translators should begin with the source’s communicative job, not with the first word in a file, and keep a record of what they did not have permission to alter.

The central rule is to preserve actor and object IDs, recipient lists, content HTML, language-tag mappings, attribution, publication and update relationships, attachments and moderation context. This article provides a repeatable first-principles method, worked field-level examples, predictable failure modes, practical exercises and answers to common questions. It complements, rather than replaces, the series owners on meaning, context, idioms and terminology. The examples are original teaching illustrations and not claims that a particular commercial product or institution uses exactly the strings shown. In production, a qualified target-language reviewer should approve localized wording and the owner of the data or software should approve technical behaviour.

  • Know the exact translation boundary before editing any field.
  • Use the official data model and original evidence, not guesswork based on familiar English words.
  • Preserve identifiers, permissions, conditions and source meaning.
  • Check real outputs with target-language and structural tests.
  • Document uncertainties, corrections and next actions so later editors can reproduce the decision.

The problem: a correct sentence can still be a wrong translation

Ordinary proofreading asks whether the target sentence is grammatical, clear and faithful. Structured-content proofreading must also ask whether the string lives in the correct field, whether it remains associated with the right object, whether a client can select it for a reader, and whether translating it accidentally changes an operation. A quick translation that ignores these questions may be praised in isolation and still fail during publication. The learner must therefore separate three layers: the source’s claim, the representation of that claim, and the way a consuming interface turns it into an action or visible record.

Consider a source that contains a human-readable title, a coded identifier and a conditional notice. The title may have several good linguistic renderings. The identifier typically has a single stable machine meaning, not a target-language synonym. The notice may contain an important permission or limitation that must survive every version. A translator who handles all three as free text has already lost the structure of the problem. Conversely, a developer who freezes every string can preserve the machine’s behaviour but make the content inaccessible to non-English readers.

A strong review is deliberately asymmetric. It gives translators freedom to reorganize natural-language expressions while protecting fields that are not prose. It makes reviewers explain their choices, yet it does not ask them to invent expertise they do not possess. The aim is not a universal word-for-word rule. It is an explicit set of constraints for this content type, supported by traceable evidence and a small real-world test.

First principles: understand the system before choosing words

1. Identity is not the human name

An actor’s id points to an identity in the federation; a display name is human-language content. The two may be shown together but serve entirely different functions.

In practice, first identify which part of the record carries the reader-facing meaning. Compare it with its neighbouring fields, trace how those fields are consumed, and ask whether a change to the visible wording would alter a name, condition, reference or action. For ActivityPub and ActivityStreams 2.0 human-language values in federated social objects, the defensible decision is the one that preserves actor and object IDs, recipient lists, content HTML, language-tag mappings, attribution, publication and update relationships, attachments and moderation context without making the target copy unnatural. Do not conceal a missing source explanation by guessing. Instead add an editorial question, seek the appropriate owner and keep the unverified point separate from approved text.

2. ActivityStreams separates words from wire structure

content, name and summary can have language-map forms contentMap, nameMap and summaryMap. Translating them must not overwrite type, id, actor or object routing.

In practice, first identify which part of the record carries the reader-facing meaning. Compare it with its neighbouring fields, trace how those fields are consumed, and ask whether a change to the visible wording would alter a name, condition, reference or action. For ActivityPub and ActivityStreams 2.0 human-language values in federated social objects, the defensible decision is the one that preserves actor and object IDs, recipient lists, content HTML, language-tag mappings, attribution, publication and update relationships, attachments and moderation context without making the target copy unnatural. Do not conceal a missing source explanation by guessing. Instead add an editorial question, seek the appropriate owner and keep the unverified point separate from approved text.

3. Language tags communicate evidence

A map key is a BCP 47 language tag, and und can represent undetermined language. Do not label a guessed machine translation as reviewed source-language content.

In practice, first identify which part of the record carries the reader-facing meaning. Compare it with its neighbouring fields, trace how those fields are consumed, and ask whether a change to the visible wording would alter a name, condition, reference or action. For ActivityPub and ActivityStreams 2.0 human-language values in federated social objects, the defensible decision is the one that preserves actor and object IDs, recipient lists, content HTML, language-tag mappings, attribution, publication and update relationships, attachments and moderation context without making the target copy unnatural. Do not conceal a missing source explanation by guessing. Instead add an editorial question, seek the appropriate owner and keep the unverified point separate from approved text.

4. A translated post is not a new permission grant

To, cc and audience semantics control distribution. The translation job does not authorize expanding who receives a private or limited post.

In practice, first identify which part of the record carries the reader-facing meaning. Compare it with its neighbouring fields, trace how those fields are consumed, and ask whether a change to the visible wording would alter a name, condition, reference or action. For ActivityPub and ActivityStreams 2.0 human-language values in federated social objects, the defensible decision is the one that preserves actor and object IDs, recipient lists, content HTML, language-tag mappings, attribution, publication and update relationships, attachments and moderation context without making the target copy unnatural. Do not conceal a missing source explanation by guessing. Instead add an editorial question, seek the appropriate owner and keep the unverified point separate from approved text.

5. HTML changes meaning and can create security risks

Human-readable content can include markup. A safe translation preserves intended links and emphasis while respecting sanitization and the consuming software’s accepted format.

In practice, first identify which part of the record carries the reader-facing meaning. Compare it with its neighbouring fields, trace how those fields are consumed, and ask whether a change to the visible wording would alter a name, condition, reference or action. For ActivityPub and ActivityStreams 2.0 human-language values in federated social objects, the defensible decision is the one that preserves actor and object IDs, recipient lists, content HTML, language-tag mappings, attribution, publication and update relationships, attachments and moderation context without making the target copy unnatural. Do not conceal a missing source explanation by guessing. Instead add an editorial question, seek the appropriate owner and keep the unverified point separate from approved text.

6. Replies and edits have a history

inReplyTo, Update and Delete describe relationships and events. Translating the visible sentence must not detach a reply or present an old correction as the current source.

In practice, first identify which part of the record carries the reader-facing meaning. Compare it with its neighbouring fields, trace how those fields are consumed, and ask whether a change to the visible wording would alter a name, condition, reference or action. For ActivityPub and ActivityStreams 2.0 human-language values in federated social objects, the defensible decision is the one that preserves actor and object IDs, recipient lists, content HTML, language-tag mappings, attribution, publication and update relationships, attachments and moderation context without making the target copy unnatural. Do not conceal a missing source explanation by guessing. Instead add an editorial question, seek the appropriate owner and keep the unverified point separate from approved text.

7. Implementations have variable coverage

ActivityStreams language maps are standards-based, but receiving applications may display only default values. Publication should include a verified interoperability and fallback plan.

In practice, first identify which part of the record carries the reader-facing meaning. Compare it with its neighbouring fields, trace how those fields are consumed, and ask whether a change to the visible wording would alter a name, condition, reference or action. For ActivityPub and ActivityStreams 2.0 human-language values in federated social objects, the defensible decision is the one that preserves actor and object IDs, recipient lists, content HTML, language-tag mappings, attribution, publication and update relationships, attachments and moderation context without making the target copy unnatural. Do not conceal a missing source explanation by guessing. Instead add an editorial question, seek the appropriate owner and keep the unverified point separate from approved text.

8. Moderation context travels with wording

Warnings, reports, quotations and consent-related descriptions cannot be softened, heightened or detached from their source circumstances just to sound natural.

In practice, first identify which part of the record carries the reader-facing meaning. Compare it with its neighbouring fields, trace how those fields are consumed, and ask whether a change to the visible wording would alter a name, condition, reference or action. For ActivityPub and ActivityStreams 2.0 human-language values in federated social objects, the defensible decision is the one that preserves actor and object IDs, recipient lists, content HTML, language-tag mappings, attribution, publication and update relationships, attachments and moderation context without making the target copy unnatural. Do not conceal a missing source explanation by guessing. Instead add an editorial question, seek the appropriate owner and keep the unverified point separate from approved text.

A twelve-stage translation workflow that can be repeated

The following method can be taught as a sequence and then compressed into a checklist for experienced editors. Each stage has a separate observable outcome. A student can perform the first stages with a small sample, while a production team can attach versions, reviewers and automated checks. The stages are not twelve synonyms for proofreading. They move from understanding the representation to drafting the human message, checking structural safety and demonstrating the result in context.

Stage 1: Identify the object and activity

Read whether this is a Note, Question, Create, Update, Announce or another object or activity; decide which text is actually editable.

The checkpoint for this stage is an artifact that another reviewer can inspect: a field inventory, source excerpt, translation note, comparison table, serialized sample, error log or screenshot as appropriate. Ask what evidence would persuade a colleague that the decision was correct rather than merely plausible. In the ActivityPub and ActivityStreams 2.0 human-language values in federated social objects workflow, it is especially important to document exceptions: a field whose language is unknown, a target-language ambiguity, a client limitation, or a statement that requires the institution’s approval. If the evidence is incomplete, mark the step open instead of silently promoting it to complete.

Stage 2: Record authorship and scope

Capture actor, attributedTo when present, to, cc, audience, inReplyTo and object id before the translation begins.

The checkpoint for this stage is an artifact that another reviewer can inspect: a field inventory, source excerpt, translation note, comparison table, serialized sample, error log or screenshot as appropriate. Ask what evidence would persuade a colleague that the decision was correct rather than merely plausible. In the ActivityPub and ActivityStreams 2.0 human-language values in federated social objects workflow, it is especially important to document exceptions: a field whose language is unknown, a target-language ambiguity, a client limitation, or a statement that requires the institution’s approval. If the evidence is incomplete, mark the step open instead of silently promoting it to complete.

Stage 3: Separate display strings from routing

Mark name, summary, content and their maps as language-bearing; mark id, inbox, outbox and recipients as structural.

The checkpoint for this stage is an artifact that another reviewer can inspect: a field inventory, source excerpt, translation note, comparison table, serialized sample, error log or screenshot as appropriate. Ask what evidence would persuade a colleague that the decision was correct rather than merely plausible. In the ActivityPub and ActivityStreams 2.0 human-language values in federated social objects workflow, it is especially important to document exceptions: a field whose language is unknown, a target-language ambiguity, a client limitation, or a statement that requires the institution’s approval. If the evidence is incomplete, mark the step open instead of silently promoting it to complete.

Stage 4: Confirm source and target languages

Inspect existing maps and default strings; use language tags supported by the publishing application.

The checkpoint for this stage is an artifact that another reviewer can inspect: a field inventory, source excerpt, translation note, comparison table, serialized sample, error log or screenshot as appropriate. Ask what evidence would persuade a colleague that the decision was correct rather than merely plausible. In the ActivityPub and ActivityStreams 2.0 human-language values in federated social objects workflow, it is especially important to document exceptions: a field whose language is unknown, a target-language ambiguity, a client limitation, or a statement that requires the institution’s approval. If the evidence is incomplete, mark the step open instead of silently promoting it to complete.

Stage 5: Draft equivalent natural language

Translate factual claims, conditions, quoted words and social intent; resist turning a private message into a public invitation.

The checkpoint for this stage is an artifact that another reviewer can inspect: a field inventory, source excerpt, translation note, comparison table, serialized sample, error log or screenshot as appropriate. Ask what evidence would persuade a colleague that the decision was correct rather than merely plausible. In the ActivityPub and ActivityStreams 2.0 human-language values in federated social objects workflow, it is especially important to document exceptions: a field whose language is unknown, a target-language ambiguity, a client limitation, or a statement that requires the institution’s approval. If the evidence is incomplete, mark the step open instead of silently promoting it to complete.

Stage 6: Protect markup and links

Preserve intended URLs, mentions, link text relationships and safe HTML structure; check for unsafe markup introduced by editing.

The checkpoint for this stage is an artifact that another reviewer can inspect: a field inventory, source excerpt, translation note, comparison table, serialized sample, error log or screenshot as appropriate. Ask what evidence would persuade a colleague that the decision was correct rather than merely plausible. In the ActivityPub and ActivityStreams 2.0 human-language values in federated social objects workflow, it is especially important to document exceptions: a field whose language is unknown, a target-language ambiguity, a client limitation, or a statement that requires the institution’s approval. If the evidence is incomplete, mark the step open instead of silently promoting it to complete.

Stage 7: Build language-map values

Use standards-based nameMap, summaryMap or contentMap with string values keyed by appropriate BCP 47 tags.

The checkpoint for this stage is an artifact that another reviewer can inspect: a field inventory, source excerpt, translation note, comparison table, serialized sample, error log or screenshot as appropriate. Ask what evidence would persuade a colleague that the decision was correct rather than merely plausible. In the ActivityPub and ActivityStreams 2.0 human-language values in federated social objects workflow, it is especially important to document exceptions: a field whose language is unknown, a target-language ambiguity, a client limitation, or a statement that requires the institution’s approval. If the evidence is incomplete, mark the step open instead of silently promoting it to complete.

Stage 8: Define tested fallback

Where consumers ignore maps, decide which default human-readable value they should show without mislabeling it as universal.

The checkpoint for this stage is an artifact that another reviewer can inspect: a field inventory, source excerpt, translation note, comparison table, serialized sample, error log or screenshot as appropriate. Ask what evidence would persuade a colleague that the decision was correct rather than merely plausible. In the ActivityPub and ActivityStreams 2.0 human-language values in federated social objects workflow, it is especially important to document exceptions: a field whose language is unknown, a target-language ambiguity, a client limitation, or a statement that requires the institution’s approval. If the evidence is incomplete, mark the step open instead of silently promoting it to complete.

Stage 9: Review recipients without mutation

Check that the localization layer does not expand to, cc or other audience fields as a side effect of editing copy.

The checkpoint for this stage is an artifact that another reviewer can inspect: a field inventory, source excerpt, translation note, comparison table, serialized sample, error log or screenshot as appropriate. Ask what evidence would persuade a colleague that the decision was correct rather than merely plausible. In the ActivityPub and ActivityStreams 2.0 human-language values in federated social objects workflow, it is especially important to document exceptions: a field whose language is unknown, a target-language ambiguity, a client limitation, or a statement that requires the institution’s approval. If the evidence is incomplete, mark the step open instead of silently promoting it to complete.

Stage 10: Inspect thread and revision links

Keep replies attached to correct parent and verify whether an update replaces previous content without losing attribution.

The checkpoint for this stage is an artifact that another reviewer can inspect: a field inventory, source excerpt, translation note, comparison table, serialized sample, error log or screenshot as appropriate. Ask what evidence would persuade a colleague that the decision was correct rather than merely plausible. In the ActivityPub and ActivityStreams 2.0 human-language values in federated social objects workflow, it is especially important to document exceptions: a field whose language is unknown, a target-language ambiguity, a client limitation, or a statement that requires the institution’s approval. If the evidence is incomplete, mark the step open instead of silently promoting it to complete.

Stage 11: Test on receiving implementations

Check the emitted JSON, network logs where permitted, displayed language, readability, links and accessible rendering.

The checkpoint for this stage is an artifact that another reviewer can inspect: a field inventory, source excerpt, translation note, comparison table, serialized sample, error log or screenshot as appropriate. Ask what evidence would persuade a colleague that the decision was correct rather than merely plausible. In the ActivityPub and ActivityStreams 2.0 human-language values in federated social objects workflow, it is especially important to document exceptions: a field whose language is unknown, a target-language ambiguity, a client limitation, or a statement that requires the institution’s approval. If the evidence is incomplete, mark the step open instead of silently promoting it to complete.

Stage 12: Archive decisions and correction route

Keep source-to-target evidence, approval state and an update plan when an original claim or moderation warning changes.

The checkpoint for this stage is an artifact that another reviewer can inspect: a field inventory, source excerpt, translation note, comparison table, serialized sample, error log or screenshot as appropriate. Ask what evidence would persuade a colleague that the decision was correct rather than merely plausible. In the ActivityPub and ActivityStreams 2.0 human-language values in federated social objects workflow, it is especially important to document exceptions: a field whose language is unknown, a target-language ambiguity, a client limitation, or a statement that requires the institution’s approval. If the evidence is incomplete, mark the step open instead of silently promoting it to complete.

Worked examples: source, unsafe shortcut, safer decision and verification

These cases are intentionally granular. Each isolates a different failure point rather than supplying twenty generic examples of translate and proofread. The source descriptions are miniature test fixtures. For each, identify the role of the field, diagnose the unsafe shortcut, perform the smaller authorized change, and finally test the consequence. The reason for this discipline is simple: the most damaging translation errors often leave the sentence sounding entirely natural. The model checks ask whether the underlying action or evidence remains what it was.

Worked example 1: nameMap title

Source situation: An event Note has English and French names. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Translate actor id instead of human name. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Use language-tagged map values for the title. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Inspect receiving client language choice. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from nameMap title is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 2: contentMap post body

Source situation: A post is available in two languages. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Put an object of strings into content as plain text. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Use the defined language-map property for natural-language values. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Validate JSON and compare target rendering. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from contentMap post body is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 3: summaryMap warning

Source situation: A content summary includes a sensitive warning. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Change it into a cheerful preview. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Preserve caution and uncertainty in equivalent words. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Compare summary with source under moderation review. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from summaryMap warning is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 4: Unknown source language

Source situation: A legacy post language cannot be established. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Label guessed language en. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Use und where appropriate and document uncertainty. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Inspect independent language-identification evidence. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from Unknown source language is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 5: Actor identifier

Source situation: actor id is an HTTPS URI. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Translate the domain or URL slug. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Keep actor id unchanged; localize display name separately. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Resolve profile before and after text edit. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from Actor identifier is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 6: Inbox/outbox

Source situation: Actor advertises federation endpoints. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Translate endpoint paths for readability. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Keep endpoint URIs as machine identifiers. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Compare emitted actor object fields. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from Inbox/outbox is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 7: to recipients

Source situation: A post addresses followers rather than everyone. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Change recipients to public for discovery. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Keep original audience fields intact. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Inspect serialized to and cc lists. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from to recipients is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 8: cc recipients

Source situation: A message includes additional distribution. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Drop cc as if it were an email footnote. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Preserve recipient semantics outside language text. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Compare delivery rules and source object. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from cc recipients is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 9: attributedTo

Source situation: Object has author attribution. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Present translator as original creator. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Keep origin attribution and separately record translator credit. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Open originating actor and object identity. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from attributedTo is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 10: inReplyTo

Source situation: A translated reply references a parent post. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Delete parent link and show isolated quote. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Preserve thread relation and translated context. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: View conversation tree in another client. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from inReplyTo is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 11: Create activity

Source situation: One activity introduces a Note. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Treat its translation as permission to create a second public activity. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Localize content within authorized publication plan. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Compare activity count and audience. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from Create activity is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 12: Update activity

Source situation: Source author corrects a number. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Translate prior text without checking revision. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Review corrected source and update localized content appropriately. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Inspect updated timestamp and object revision. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from Update activity is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 13: Delete activity

Source situation: A source post is withdrawn. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Keep translated copy live without checking authority. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Follow deletion policy and preserve audit trace where appropriate. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Check downstream visibility. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from Delete activity is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 14: Announce activity

Source situation: An account boosts an existing object. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Translate booster as original author. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Keep distinction between amplifier and creator. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Inspect actor and object references. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from Announce activity is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 15: HTML emphasis

Source situation: Content uses strong emphasis for a warning. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Remove markup and weaken emphasis or insert unsafe HTML. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Keep safe markup and meaningful stress. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Render in target client with sanitization. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from HTML emphasis is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 16: Link title

Source situation: Post says See report with hyperlink. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Invent translated URL without checking destination. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Preserve working destination while translating link label. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Click with a safe preview and compare intent. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from Link title is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 17: Media attachment description

Source situation: Image is accompanied by description text. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Translate only visible caption, omit access description. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Review language-bearing attachment text and platform support. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Test accessibility presentation in receiving client. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from Media attachment description is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 18: Moderation notice

Source situation: A note includes a community safety warning. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Replace specific rule with vague encouragement. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Keep the rule and consequence exact. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Ask moderator to review before distribution. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from Moderation notice is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 19: Poll question

Source situation: Question object offers two options. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Change the choices while translating question tone. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Preserve options, scope and any temporal bounds. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Run test poll without publishing real votes. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from Poll question is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 20: Mixed-script display

Source situation: A right-to-left sentence includes Latin usernames. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Allow names and punctuation to reorder confusingly. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Test proper bidirectional rendering without altering IDs. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Review rendered posts in mobile and desktop. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from Mixed-script display is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 21: Fallback string

Source situation: Consumer ignores contentMap but reads content. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Leave an empty or misleading fallback. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Provide an approved fallback that does not falsely claim all languages are represented. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Test a client without map display support. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from Fallback string is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Worked example 22: Quoted claim

Source situation: A Note quotes a third party. The translator’s first responsibility is to decide which words describe the user’s experience and which values belong to the record’s technical identity. The source situation must be examined in its actual interface rather than converted into a free-floating bilingual flashcard.

Tempting but unsafe shortcut: Rewrite attribution to the translator’s viewpoint. That move may look tidy in a spreadsheet and even produce a polished sentence, but it leaves an important relationship untested. A reviewer must explicitly ask what a reader would now believe, what a consuming application would parse, and whether the original owner authorized the implied change.

Safer translation decision: Preserve quotation status and speaker distinctions. Preserve the relevant structure, then draft or revise only the natural-language span that the reader is meant to understand. If the target language requires reordering, explain the reordering as a linguistic choice rather than using it as an excuse to rewrite machine identifiers or policy conditions.

Evidence and acceptance test: Compare source citation and translated line. A good acceptance test specifies an expected result as well as a test action. Compare the same source object before and after localization; record the field values that should change and those that should remain identical. Where a second person is available, ask that reviewer to explain the translated message independently without seeing the proposed answer. Agreement about purpose and consequence matters more than identical word choice.

The transferable lesson from Quoted claim is that linguistic fidelity and operational integrity are two separate checks. It would be possible to pass one while failing the other. In a student practice task, write the unsafe assumption in one column and the specific piece of evidence that rejects it in another. In professional use, retain that comparison as a short release note. It protects the next editor from repeating the mistake after the current reviewer has moved on.

Failure modes: how fluent text becomes misleading

Failure mode 1: Identity confusion

Localized names must not overwrite actor IDs, object IDs or origin attribution.

Diagnosis begins by reproducing the exact input and comparing source, target and real output. Repair should be scoped to the smallest faulty decision; replacing a whole document because one field is wrong increases the number of unreviewed changes. After correction, introduce a related but different test case to check that the underlying rule was learned. Maintain a record of what the test did not establish: for example, syntax validation may pass while target-language meaning still needs human review.

Failure mode 2: Audience expansion

A translation system that changes to or cc can unintentionally expose content beyond its original recipients.

Diagnosis begins by reproducing the exact input and comparing source, target and real output. Repair should be scoped to the smallest faulty decision; replacing a whole document because one field is wrong increases the number of unreviewed changes. After correction, introduce a related but different test case to check that the underlying rule was learned. Maintain a record of what the test did not establish: for example, syntax validation may pass while target-language meaning still needs human review.

Failure mode 3: Noninteroperable map assumption

A standards-valid contentMap is not automatically displayed by every receiver; test fallback behaviour.

Diagnosis begins by reproducing the exact input and comparing source, target and real output. Repair should be scoped to the smallest faulty decision; replacing a whole document because one field is wrong increases the number of unreviewed changes. After correction, introduce a related but different test case to check that the underlying rule was learned. Maintain a record of what the test did not establish: for example, syntax validation may pass while target-language meaning still needs human review.

Failure mode 4: False source ownership

Publishing a translated version without provenance can make readers think it was newly authored by the translator.

Diagnosis begins by reproducing the exact input and comparing source, target and real output. Repair should be scoped to the smallest faulty decision; replacing a whole document because one field is wrong increases the number of unreviewed changes. After correction, introduce a related but different test case to check that the underlying rule was learned. Maintain a record of what the test did not establish: for example, syntax validation may pass while target-language meaning still needs human review.

Failure mode 5: Broken reply links

Removing inReplyTo may turn a contextual response into a disconnected claim.

Diagnosis begins by reproducing the exact input and comparing source, target and real output. Repair should be scoped to the smallest faulty decision; replacing a whole document because one field is wrong increases the number of unreviewed changes. After correction, introduce a related but different test case to check that the underlying rule was learned. Maintain a record of what the test did not establish: for example, syntax validation may pass while target-language meaning still needs human review.

Failure mode 6: Unsafe markup

An innocuous-looking translation edit can inject invalid HTML or disrupt a meaningful link.

Diagnosis begins by reproducing the exact input and comparing source, target and real output. Repair should be scoped to the smallest faulty decision; replacing a whole document because one field is wrong increases the number of unreviewed changes. After correction, introduce a related but different test case to check that the underlying rule was learned. Maintain a record of what the test did not establish: for example, syntax validation may pass while target-language meaning still needs human review.

Failure mode 7: Untracked corrections

The source may be edited or deleted while a stale translated copy continues circulating.

Diagnosis begins by reproducing the exact input and comparing source, target and real output. Repair should be scoped to the smallest faulty decision; replacing a whole document because one field is wrong increases the number of unreviewed changes. After correction, introduce a related but different test case to check that the underlying rule was learned. Maintain a record of what the test did not establish: for example, syntax validation may pass while target-language meaning still needs human review.

Failure mode 8: Changed warning force

A translated moderation or sensitive-content notice that understates a restriction can mislead users.

Diagnosis begins by reproducing the exact input and comparing source, target and real output. Repair should be scoped to the smallest faulty decision; replacing a whole document because one field is wrong increases the number of unreviewed changes. After correction, introduce a related but different test case to check that the underlying rule was learned. Maintain a record of what the test did not establish: for example, syntax validation may pass while target-language meaning still needs human review.

Practice and transfer: six tasks with model reasoning

The exercises are designed for readers with no access to a live production system. Use copies or invented sample records, not private accounts or operational files. For each task, write the source fact, the protected fields, the proposed target-language change and the test. A response earns credit when another person can reproduce the reasoning and identify what evidence would make a currently uncertain claim safe to publish. Translation quality is demonstrated by deliberate choices, not merely confident wording.

Practice 1: Build a bilingual contentMap

Draft two reviewed strings keyed by valid language tags and show a separate, responsible fallback plan.

Model reasoning: begin by marking what readers must understand, then isolate the values that must remain stable. Reject any version that silently introduces a new actor, level of certainty, permission, number, deadline, destination or technical identity. A sound answer may offer two natural target-language phrasings if both satisfy the same constraints, but it must explain why one is preferable in the specified interface. End with a positive test, such as a rendered preview or structured comparison, and a negative test deliberately designed to expose the tempting mistake.

Transfer question: what would change if the same idea appeared in a longer paragraph, a short button, a screen-reader announcement or an automated client? The surface wording may change. The source commitment and the underlying record must not. Give yourself a mark only after explaining the consequence of a false pass; this helps distinguish understanding from memorization.

Practice 2: Separate actor id and name

Mark structural identifiers and human-readable labels in a sample actor object; explain why only the latter is localized.

Model reasoning: begin by marking what readers must understand, then isolate the values that must remain stable. Reject any version that silently introduces a new actor, level of certainty, permission, number, deadline, destination or technical identity. A sound answer may offer two natural target-language phrasings if both satisfy the same constraints, but it must explain why one is preferable in the specified interface. End with a positive test, such as a rendered preview or structured comparison, and a negative test deliberately designed to expose the tempting mistake.

Transfer question: what would change if the same idea appeared in a longer paragraph, a short button, a screen-reader announcement or an automated client? The surface wording may change. The source commitment and the underlying record must not. Give yourself a mark only after explaining the consequence of a false pass; this helps distinguish understanding from memorization.

Practice 3: Audit a followers-only announcement

Prove that no distribution fields changed when a translated version is prepared.

Model reasoning: begin by marking what readers must understand, then isolate the values that must remain stable. Reject any version that silently introduces a new actor, level of certainty, permission, number, deadline, destination or technical identity. A sound answer may offer two natural target-language phrasings if both satisfy the same constraints, but it must explain why one is preferable in the specified interface. End with a positive test, such as a rendered preview or structured comparison, and a negative test deliberately designed to expose the tempting mistake.

Transfer question: what would change if the same idea appeared in a longer paragraph, a short button, a screen-reader announcement or an automated client? The surface wording may change. The source commitment and the underlying record must not. Give yourself a mark only after explaining the consequence of a false pass; this helps distinguish understanding from memorization.

Practice 4: Review a corrected post

A source author updates a numerical claim. Explain how to find and amend the target-language copy without creating a false new original.

Model reasoning: begin by marking what readers must understand, then isolate the values that must remain stable. Reject any version that silently introduces a new actor, level of certainty, permission, number, deadline, destination or technical identity. A sound answer may offer two natural target-language phrasings if both satisfy the same constraints, but it must explain why one is preferable in the specified interface. End with a positive test, such as a rendered preview or structured comparison, and a negative test deliberately designed to expose the tempting mistake.

Transfer question: what would change if the same idea appeared in a longer paragraph, a short button, a screen-reader announcement or an automated client? The surface wording may change. The source commitment and the underlying record must not. Give yourself a mark only after explaining the consequence of a false pass; this helps distinguish understanding from memorization.

Practice 5: Test a quote and a reply

Keep inReplyTo and attribution stable while changing the reader-facing sentence.

Model reasoning: begin by marking what readers must understand, then isolate the values that must remain stable. Reject any version that silently introduces a new actor, level of certainty, permission, number, deadline, destination or technical identity. A sound answer may offer two natural target-language phrasings if both satisfy the same constraints, but it must explain why one is preferable in the specified interface. End with a positive test, such as a rendered preview or structured comparison, and a negative test deliberately designed to expose the tempting mistake.

Transfer question: what would change if the same idea appeared in a longer paragraph, a short button, a screen-reader announcement or an automated client? The surface wording may change. The source commitment and the underlying record must not. Give yourself a mark only after explaining the consequence of a false pass; this helps distinguish understanding from memorization.

Practice 6: Compare two receiving clients

Design a checklist for one client that supports language maps and one that displays only a default string.

Model reasoning: begin by marking what readers must understand, then isolate the values that must remain stable. Reject any version that silently introduces a new actor, level of certainty, permission, number, deadline, destination or technical identity. A sound answer may offer two natural target-language phrasings if both satisfy the same constraints, but it must explain why one is preferable in the specified interface. End with a positive test, such as a rendered preview or structured comparison, and a negative test deliberately designed to expose the tempting mistake.

Transfer question: what would change if the same idea appeared in a longer paragraph, a short button, a screen-reader announcement or an automated client? The surface wording may change. The source commitment and the underlying record must not. Give yourself a mark only after explaining the consequence of a false pass; this helps distinguish understanding from memorization.

A usable quality-assurance rubric

Score each translated unit on seven separate dimensions rather than merging everything into a vague claim of quality: source meaning, field selection, structural integrity, audience interpretation, linguistic naturalness, accessibility and reproducible verification. On a four-point scale, zero means the dimension is untested, one means an observed defect, two means partly supported with unresolved risk, and three means adequately supported by evidence. A technically correct string with the wrong scope cannot be rescued by excellent style. Equally, a perfectly preserved file with unreadable public text is not a completed localization.

Before approval, compare every sensitive word carrying negation, condition, number, status, rights, obligation or consequence. Then compare identifiers and field structure mechanically where possible. Run the sample in a consuming interface, inspect how it displays at realistic width, and ask whether a reader would choose the correct action. This separation of linguistic and technical review is not bureaucracy: it is a practical way to catch errors that one form of expertise cannot see alone.

Build a compact editorial record containing the source reference, the target locale, the approver, the protected-field list, a link or screenshot of the tested result, any unresolved ambiguity and the next review trigger. That record allows a future editor to understand why two apparently similar strings were handled differently. It also makes it easier to revise a translation when the source changes without starting from zero.

Frequently asked questions

Is ActivityPub the same as ActivityStreams?

ActivityPub specifies federated social operations using ActivityStreams 2.0 objects. The data model’s language-map semantics are a separate layer.

For this question, the deciding evidence is the field’s defined role plus what a user or software client actually does with it. Check the official specification before editing technical fields, and seek a fluent target-language review before stating that a phrase is fully equivalent. Where the implementation has product-specific rules, document those as local policy rather than implying the standard guarantees them universally.

Can I translate an actor id?

No. Localize names and summaries, not the identity URI or delivery endpoints.

For this question, the deciding evidence is the field’s defined role plus what a user or software client actually does with it. Check the official specification before editing technical fields, and seek a fluent target-language review before stating that a phrase is fully equivalent. Where the implementation has product-specific rules, document those as local policy rather than implying the standard guarantees them universally.

What is contentMap?

An ActivityStreams natural-language map giving equivalent content strings under language tags.

For this question, the deciding evidence is the field’s defined role plus what a user or software client actually does with it. Check the official specification before editing technical fields, and seek a fluent target-language review before stating that a phrase is fully equivalent. Where the implementation has product-specific rules, document those as local policy rather than implying the standard guarantees them universally.

Does a receiving client always display maps?

No. Implementation support varies; test fallback behaviour and avoid promising automatic per-reader translation.

For this question, the deciding evidence is the field’s defined role plus what a user or software client actually does with it. Check the official specification before editing technical fields, and seek a fluent target-language review before stating that a phrase is fully equivalent. Where the implementation has product-specific rules, document those as local policy rather than implying the standard guarantees them universally.

Can a translation change who receives the post?

It should not. Audience and recipient fields are not ordinary translatable copy.

For this question, the deciding evidence is the field’s defined role plus what a user or software client actually does with it. Check the official specification before editing technical fields, and seek a fluent target-language review before stating that a phrase is fully equivalent. Where the implementation has product-specific rules, document those as local policy rather than implying the standard guarantees them universally.

How should an unknown language be tagged?

The map mechanism provides und for undetermined language; use it deliberately rather than inventing certainty.

For this question, the deciding evidence is the field’s defined role plus what a user or software client actually does with it. Check the official specification before editing technical fields, and seek a fluent target-language review before stating that a phrase is fully equivalent. Where the implementation has product-specific rules, document those as local policy rather than implying the standard guarantees them universally.

Does translation permit republishing a private post?

No. Consent, audience and origin attribution remain relevant; a new language is not new authorization.

For this question, the deciding evidence is the field’s defined role plus what a user or software client actually does with it. Check the official specification before editing technical fields, and seek a fluent target-language review before stating that a phrase is fully equivalent. Where the implementation has product-specific rules, document those as local policy rather than implying the standard guarantees them universally.

How do I handle an edited or deleted source?

Maintain a traceable source reference and update or withdraw localized copies according to the source and moderation policy.

For this question, the deciding evidence is the field’s defined role plus what a user or software client actually does with it. Check the official specification before editing technical fields, and seek a fluent target-language review before stating that a phrase is fully equivalent. Where the implementation has product-specific rules, document those as local policy rather than implying the standard guarantees them universally.

References, boundaries and contextual next routes

Technical references are listed here as primary documentation. They describe the applicable data model and operations; they do not verify every invented teaching string in the examples above. The editorial rule is to compare a precise implementation against the version of the standard and the product it actually uses, and to treat external search snippets as discovery rather than as permission to alter software behaviour.

For the general meaning-transfer decision, begin with the existing eduKateSG guide on translating meaning rather than words. For audience relationships and degree of formality, use the context, tone and register owner. For jokes and cultural references, use the idioms and culture owner rather than trying to force cultural analysis into this technical field guide. For lexical depth, follow the Vocabulary Learning Hub; for grammar and meaning relationships, use How English Works. These are neighboring routes with separate jobs, not competing pages for the same search intent.

The next move is practical: select one small, representative sample, make an inventory of what can be translated, apply the twelve-stage workflow and test the result with someone who has not seen the source commentary. Then move to a second sample that changes one crucial condition. Repeat until the method works when the visual details differ. This is how translation becomes transferable skill rather than a collection of memorized corrections.

The central lesson is that correct translation respects both a human message and the environment carrying that message. For ActivityPub and ActivityStreams 2.0 human-language values in federated social objects, the standard of success is a natural and faithful target text that remains attached to the right data, source and action. When evidence is incomplete, a documented question is better than an invented answer. Precision is something readers and maintainers can test, not simply a tone of confidence.