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 | CITATION.cff Software Authors, Version and Citation Metadata

To translate CITATION.cff research-software citation files correctly, begin by separating human-readable language from the structured values that identify, route or validate the record. This guide answers the practical question: how to translate CITATION.cff YAML metadata without changing software citation identity. It is written for open-source maintainers, research librarians, translators and software documentation editors, with original examples, a repeatable audit process and clear boundaries around what a translator may change.

The visible mistake is usually simple: A translator sees an English citation message and an abstract in a repository, changes every adjacent string into another language, then discovers that the release date, person names and identifier have also been rewritten. The YAML may still look reasonable to a reader, yet the research object has become difficult to locate or cite correctly. A correct translation must let an ordinary reader understand the record and let the receiving system behave as intended. The standard is not stylistic similarity alone; citation identity remains stable while human-readable guidance becomes understandable in the reader’s language. Start by classifying each piece of information before typing a target-language equivalent.

Quick reference: Translate explanatory prose where permitted; protect identifiers, references, technical tokens and factual conditions; work from the specification rather than intuition; maintain a field-level change log; check original and translated renderings together; run structural checks; use real-world reader tasks as final proof. The following sections explain each action through CITATION.cff research-software citation files, rather than offering generic translation advice.

The precise translation problem and the limits of this guide

A CITATION.cff file is human-readable YAML which repository tools may parse to help people cite research software. The familiar English message is guidance, not the DOI; the title identifies an object, while `cff-version`, field names and structured identifiers carry machine-readable meaning. Translation therefore requires classifying every value before choosing words.

This article teaches human translation decisions inside a Citation File Format record. It does not teach the general practice of choosing bibliography styles, and it does not replace the canonical guide for BibTeX, CSL JSON and citation keys.

A file can be syntactically correct while carrying the wrong meaning, and a fluent translation can be syntactically invalid. For CITATION.cff research-software citation files, both kinds of failure must be found before publication. In practical work, record which values are display language, which are authoritative identity, which are program instructions and which are evidence or measurements. A reviewer should be able to defend every change in those terms.

First principles: the three layers that must remain distinct

Layer 1: the record that machines interpret

A structured record is not a continuous paragraph. It is a collection of named fields with defined relationships. Keys, identifiers and field types often use English-looking words, but their appearance does not make them phrases for translation. A program does not understand a localized synonym for a protocol key unless the format explicitly defines such a mechanism. Keep the original machine layer reproducible and testable.

The first diagnostic is to point to every changed character and ask whether the change is expected by the schema. If it is not, a human-readable explanation may belong in a label or documentation field instead. This distinction prevents attractive but destructive edits from passing a casual proofreading review. A useful exercise is to highlight every token that must remain byte-for-byte identical before translating the first sentence.

Layer 2: the statement a person will read

Human-facing titles, instructions, summaries, notices and descriptions are evaluated by what a reader can accurately recover. Translation should keep who did what, when, why and under what conditions. Where a sentence signals possibility, limitation or responsibility, preserve those limits. A fluent rewrite which implies a stronger permission, a completed action or a certified capability does not satisfy the source, even if it sounds elegant.

Read a candidate translation aloud as a decision instruction. Would a user make the same choice after seeing the target-language version? Could they identify the same author, object, service, field or publication? If not, the translation is not ready. In structured information, tiny words such as may, expected, only, optional, duplicate and experimental often carry disproportionate practical significance.

Layer 3: the evidence that lets another person verify it

The most reliable review combines a source snapshot, a field-by-field translation decision and a rendered output. That is stronger than comparing prose alone because the source may contain facts that the interface hides. Evidence includes actual links, identifiers, dates, numeric values and the documented meaning of status or type fields. A reviewer must distinguish what the record explicitly says from an editorial inference.

A defensible translation has an audit trail. Keep one entry for every disputed expression: original value, proposed target form, reason, authoritative reference and reviewer decision. When someone later asks why a field stayed in English or why a particular status was qualified, that record gives a precise answer. Such notes are especially important when different teams maintain the source record and its translated display.

Field-by-field guide: what to translate, what to protect

`cff-version`

`cff-version` represents the specification version which tells a parser how to interpret the record. The safe translation decision is to keep the key and version literal. The corresponding boundary is that changing a semantic version because another locale displays numbers differently. For example, `cff-version: 1.2.0` stays unchanged. A good reviewer verifies this field both in isolation and in the whole record, because a locally plausible edit may change a link, a data relationship or a reader’s inference elsewhere.

Ask two separate questions here: does the value obey its structural contract, and does any visible text convey the correct human meaning? If the first answer is no, restore the source identifier or rule before editing prose. If the second answer is no, revise the explanatory language without silently rewriting the authoritative data. This discipline makes corrections traceable rather than a succession of guesses.

`message`

`message` represents reader-facing instruction about how to cite the work. The safe translation decision is to translate its instruction and polite tone. The corresponding boundary is that never turn the sentence into a promise that the repository endorses a citation not supplied. For example, `Please cite this software` becomes a natural request, not `the citation is mandatory`. A good reviewer verifies this field both in isolation and in the whole record, because a locally plausible edit may change a link, a data relationship or a reader’s inference elsewhere.

Ask two separate questions here: does the value obey its structural contract, and does any visible text convey the correct human meaning? If the first answer is no, restore the source identifier or rule before editing prose. If the second answer is no, revise the explanatory language without silently rewriting the authoritative data. This discipline makes corrections traceable rather than a succession of guesses.

`title`

`title` represents the named research software or research object. The safe translation decision is to normally preserve the official title and introduce a separately labelled explanatory translation only if policy permits. The corresponding boundary is that a translation must not invent a new official product or project name. For example, `Harbour Data Toolkit` remains its title in a multilingual explanation. A good reviewer verifies this field both in isolation and in the whole record, because a locally plausible edit may change a link, a data relationship or a reader’s inference elsewhere.

Ask two separate questions here: does the value obey its structural contract, and does any visible text convey the correct human meaning? If the first answer is no, restore the source identifier or rule before editing prose. If the second answer is no, revise the explanatory language without silently rewriting the authoritative data. This discipline makes corrections traceable rather than a succession of guesses.

`abstract`

`abstract` represents the prose summary describing what the software does. The safe translation decision is to translate meaning, qualifications, capabilities and limitations. The corresponding boundary is that do not inflate an experimental toolkit into an approved clinical system. For example, `exploratory analysis` must not become `validated diagnosis`. A good reviewer verifies this field both in isolation and in the whole record, because a locally plausible edit may change a link, a data relationship or a reader’s inference elsewhere.

Ask two separate questions here: does the value obey its structural contract, and does any visible text convey the correct human meaning? If the first answer is no, restore the source identifier or rule before editing prose. If the second answer is no, revise the explanatory language without silently rewriting the authoritative data. This discipline makes corrections traceable rather than a succession of guesses.

`authors`

`authors` represents the ordered people or organisations credited for the work. The safe translation decision is to preserve the intended contributors and the role of each listed person. The corresponding boundary is that never add, omit, merge or reorder people as an incidental language edit. For example, two author entries remain two credited authors. A good reviewer verifies this field both in isolation and in the whole record, because a locally plausible edit may change a link, a data relationship or a reader’s inference elsewhere.

Ask two separate questions here: does the value obey its structural contract, and does any visible text convey the correct human meaning? If the first answer is no, restore the source identifier or rule before editing prose. If the second answer is no, revise the explanatory language without silently rewriting the authoritative data. This discipline makes corrections traceable rather than a succession of guesses.

`given-names` and `family-names`

`given-names` and `family-names` represents structured personal-name components. The safe translation decision is to retain the entered names unless the maintainers approve a specific name representation. The corresponding boundary is that do not split or reorder names according to a guess about another culture’s naming pattern. For example, `given-names: Ana` and `family-names: Silva` are not translation instructions. A good reviewer verifies this field both in isolation and in the whole record, because a locally plausible edit may change a link, a data relationship or a reader’s inference elsewhere.

Ask two separate questions here: does the value obey its structural contract, and does any visible text convey the correct human meaning? If the first answer is no, restore the source identifier or rule before editing prose. If the second answer is no, revise the explanatory language without silently rewriting the authoritative data. This discipline makes corrections traceable rather than a succession of guesses.

`name`

`name` represents a supported name representation for an author entity, often an organisation. The safe translation decision is to respect the identity string and any official translated name. The corresponding boundary is that do not automatically break an organisation name into person-name fields. For example, `The Harbour Data Team` stays a named contributor. A good reviewer verifies this field both in isolation and in the whole record, because a locally plausible edit may change a link, a data relationship or a reader’s inference elsewhere.

Ask two separate questions here: does the value obey its structural contract, and does any visible text convey the correct human meaning? If the first answer is no, restore the source identifier or rule before editing prose. If the second answer is no, revise the explanatory language without silently rewriting the authoritative data. This discipline makes corrections traceable rather than a succession of guesses.

`orcid`

`orcid` represents the author identifier which disambiguates a research contributor. The safe translation decision is to preserve the identifier and exact link. The corresponding boundary is that never translate the digits, hostname or path. For example, `https://orcid.org/0000-0002-1825-0097` stays literal when used as a format example. A good reviewer verifies this field both in isolation and in the whole record, because a locally plausible edit may change a link, a data relationship or a reader’s inference elsewhere.

Ask two separate questions here: does the value obey its structural contract, and does any visible text convey the correct human meaning? If the first answer is no, restore the source identifier or rule before editing prose. If the second answer is no, revise the explanatory language without silently rewriting the authoritative data. This discipline makes corrections traceable rather than a succession of guesses.

`doi`

`doi` represents the digital object identifier of the cited object. The safe translation decision is to keep its characters unchanged and separately verify what object it resolves to. The corresponding boundary is that avoid substituting a general project DOI for the release-specific reference without authority. For example, `10.0000/illustrative` is only a fictional training placeholder. A good reviewer verifies this field both in isolation and in the whole record, because a locally plausible edit may change a link, a data relationship or a reader’s inference elsewhere.

Ask two separate questions here: does the value obey its structural contract, and does any visible text convey the correct human meaning? If the first answer is no, restore the source identifier or rule before editing prose. If the second answer is no, revise the explanatory language without silently rewriting the authoritative data. This discipline makes corrections traceable rather than a succession of guesses.

`identifiers`

`identifiers` represents additional typed identifiers for a software artifact or collection. The safe translation decision is to translate only qualifying human descriptions when appropriate. The corresponding boundary is that keep `type` categories and machine identities consistent with the schema. For example, an archived snapshot DOI is not interchangeable with a collection DOI. A good reviewer verifies this field both in isolation and in the whole record, because a locally plausible edit may change a link, a data relationship or a reader’s inference elsewhere.

Ask two separate questions here: does the value obey its structural contract, and does any visible text convey the correct human meaning? If the first answer is no, restore the source identifier or rule before editing prose. If the second answer is no, revise the explanatory language without silently rewriting the authoritative data. This discipline makes corrections traceable rather than a succession of guesses.

`version`

`version` represents the named release version of the software. The safe translation decision is to retain the release token exactly. The corresponding boundary is that do not localize a semantic version or add a release which does not exist. For example, `2.0.4` remains `2.0.4`. A good reviewer verifies this field both in isolation and in the whole record, because a locally plausible edit may change a link, a data relationship or a reader’s inference elsewhere.

Ask two separate questions here: does the value obey its structural contract, and does any visible text convey the correct human meaning? If the first answer is no, restore the source identifier or rule before editing prose. If the second answer is no, revise the explanatory language without silently rewriting the authoritative data. This discipline makes corrections traceable rather than a succession of guesses.

`date-released`

`date-released` represents the declared release date. The safe translation decision is to keep ISO-style date value and verify it against the release record. The corresponding boundary is that do not move the date one day because of an unrequested timezone interpretation. For example, `2026-07-18` is a date value, not prose. A good reviewer verifies this field both in isolation and in the whole record, because a locally plausible edit may change a link, a data relationship or a reader’s inference elsewhere.

Ask two separate questions here: does the value obey its structural contract, and does any visible text convey the correct human meaning? If the first answer is no, restore the source identifier or rule before editing prose. If the second answer is no, revise the explanatory language without silently rewriting the authoritative data. This discipline makes corrections traceable rather than a succession of guesses.

`url` and `repository-code`

`url` and `repository-code` represents locations used to find the software and its source. The safe translation decision is to leave the URLs untouched and test that the endpoints still identify the right object. The corresponding boundary is that never translate a hostname, URL slug or repository path. For example, an https repository address remains one intact address. A good reviewer verifies this field both in isolation and in the whole record, because a locally plausible edit may change a link, a data relationship or a reader’s inference elsewhere.

Ask two separate questions here: does the value obey its structural contract, and does any visible text convey the correct human meaning? If the first answer is no, restore the source identifier or rule before editing prose. If the second answer is no, revise the explanatory language without silently rewriting the authoritative data. This discipline makes corrections traceable rather than a succession of guesses.

`preferred-citation`

`preferred-citation` represents the citation object preferred by the maintainers when distinct from the software itself. The safe translation decision is to preserve the difference between citing a software release and citing a related paper. The corresponding boundary is that do not invent a preferred paper merely because a paper is common in the discipline. For example, a paper title in a nested citation is not automatically a software title. A good reviewer verifies this field both in isolation and in the whole record, because a locally plausible edit may change a link, a data relationship or a reader’s inference elsewhere.

Ask two separate questions here: does the value obey its structural contract, and does any visible text convey the correct human meaning? If the first answer is no, restore the source identifier or rule before editing prose. If the second answer is no, revise the explanatory language without silently rewriting the authoritative data. This discipline makes corrections traceable rather than a succession of guesses.

`keywords`

`keywords` represents discovery words accompanying the object. The safe translation decision is to translate descriptors if this record’s policy and target consumers accept localized keywords. The corresponding boundary is that avoid silently removing useful retrieval forms or keyword variants. For example, a technical phrase may need a preserved original plus an approved local equivalent. A good reviewer verifies this field both in isolation and in the whole record, because a locally plausible edit may change a link, a data relationship or a reader’s inference elsewhere.

Ask two separate questions here: does the value obey its structural contract, and does any visible text convey the correct human meaning? If the first answer is no, restore the source identifier or rule before editing prose. If the second answer is no, revise the explanatory language without silently rewriting the authoritative data. This discipline makes corrections traceable rather than a succession of guesses.

`license`

`license` represents a license identifier or designation. The safe translation decision is to preserve the exact identifier as data. The corresponding boundary is that a legal identifier is not permission to rewrite the license terms. For example, `MIT` stays `MIT`; changing it does not translate the license. A good reviewer verifies this field both in isolation and in the whole record, because a locally plausible edit may change a link, a data relationship or a reader’s inference elsewhere.

Ask two separate questions here: does the value obey its structural contract, and does any visible text convey the correct human meaning? If the first answer is no, restore the source identifier or rule before editing prose. If the second answer is no, revise the explanatory language without silently rewriting the authoritative data. This discipline makes corrections traceable rather than a succession of guesses.

A repeatable twelve-stage translation and publication method

1. Identify the reader and the use case

Write down the audience, the task they want to complete and the language in which they will act. A support dashboard, academic citation and feed reader ask different things of the same phrase. In CITATION.cff research-software citation files, the practical focus at this stage is the specification version which tells a parser how to interpret the record. It is tempting to move quickly because the field appears familiar, yet the safer decision is to keep the key and version literal. Keep that rule in the working checklist so another translator can repeat it rather than relying on the first editor’s memory.

Test the decision against a concrete situation: message: Please cite this software if you use results in a publication. The review should not accidentally produce this alternative: Render the instruction as a required legal condition without checking the project. Instead, Translate this as the maintainers’ request for attribution, preserving the conditional reference to use in publications. Record the source value, the reason for the edit and the structural check that proves the intended behavior survived. If the reasoning cannot be explained in one or two clear sentences, escalate the ambiguity before publication.

2. Fix the canonical source version

Record the version, source URL and responsible owner before editing. Translation of an unstable or unapproved snapshot creates review problems when the upstream record changes. In CITATION.cff research-software citation files, the practical focus at this stage is the prose summary describing what the software does. It is tempting to move quickly because the field appears familiar, yet the safer decision is to translate meaning, qualifications, capabilities and limitations. Keep that rule in the working checklist so another translator can repeat it rather than relying on the first editor’s memory.

Test the decision against a concrete situation: abstract: Experimental package for comparing river-sensor observations; not for regulatory reporting. The review should not accidentally produce this alternative: Translate it as a certified official reporting product. Instead, Preserve experimental status, comparative function and explicit limitation. Record the source value, the reason for the edit and the structural check that proves the intended behavior survived. If the reasoning cannot be explained in one or two clear sentences, escalate the ambiguity before publication.

3. Inventory all fields before translating

Extract keys and values into a field inventory. Mark reader-facing sentences, schema tokens, codes, names, measurements, references and absent fields separately. In CITATION.cff research-software citation files, the practical focus at this stage is a supported name representation for an author entity, often an organisation. It is tempting to move quickly because the field appears familiar, yet the safer decision is to respect the identity string and any official translated name. Keep that rule in the working checklist so another translator can repeat it rather than relying on the first editor’s memory.

Test the decision against a concrete situation: given-names: Mei Lin; family-names: Tan The review should not accidentally produce this alternative: Swap the fields because target-language name order is different. Instead, Keep structured name data as recorded; handle display ordering at the presentation layer. Record the source value, the reason for the edit and the structural check that proves the intended behavior survived. If the reasoning cannot be explained in one or two clear sentences, escalate the ambiguity before publication.

4. Fence the protected data

Mark anything that must not change without a separately approved data migration. Identifiers, links, timestamps and validation tokens commonly belong here. In CITATION.cff research-software citation files, the practical focus at this stage is additional typed identifiers for a software artifact or collection. It is tempting to move quickly because the field appears familiar, yet the safer decision is to translate only qualifying human descriptions when appropriate. Keep that rule in the working checklist so another translator can repeat it rather than relying on the first editor’s memory.

Test the decision against a concrete situation: authors: – name: Regional Coastal Data Group The review should not accidentally produce this alternative: Invent a surname and given name to satisfy a target-language expectation. Instead, Retain organisation attribution as an organisation. Record the source value, the reason for the edit and the structural check that proves the intended behavior survived. If the reasoning cannot be explained in one or two clear sentences, escalate the ambiguity before publication.

5. Define language and terminology policy

Select an appropriate target variety, preferred equivalents and a rule for official names. A glossary supports consistency without turning every borrowed term into literal prose. In CITATION.cff research-software citation files, the practical focus at this stage is locations used to find the software and its source. It is tempting to move quickly because the field appears familiar, yet the safer decision is to leave the URLs untouched and test that the endpoints still identify the right object. Keep that rule in the working checklist so another translator can repeat it rather than relying on the first editor’s memory.

Test the decision against a concrete situation: identifiers: two DOIs, one for all archived versions and one for release 2.0.4 The review should not accidentally produce this alternative: Pick whichever DOI looks shorter in the translated message. Instead, Keep both with accurate descriptions of their respective scopes. Record the source value, the reason for the edit and the structural check that proves the intended behavior survived. If the reasoning cannot be explained in one or two clear sentences, escalate the ambiguity before publication.

6. Translate meaning with its conditions intact

Render each human-readable statement while preserving uncertainty, agency, negation, scope, time and technical qualifications. Test what the reader will infer. In CITATION.cff research-software citation files, the practical focus at this stage is a license identifier or designation. It is tempting to move quickly because the field appears familiar, yet the safer decision is to preserve the exact identifier as data. Keep that rule in the working checklist so another translator can repeat it rather than relying on the first editor’s memory.

Test the decision against a concrete situation: title: RiverWatch Analysis Kit The review should not accidentally produce this alternative: Rename the project into an unapproved target-language product name. Instead, Preserve the canonical title and explain its meaning in separate prose if needed. Record the source value, the reason for the edit and the structural check that proves the intended behavior survived. If the reasoning cannot be explained in one or two clear sentences, escalate the ambiguity before publication.

7. Resolve ambiguous terms against the schema

When a phrase seems ordinary but its field has a specialized meaning, return to the specification and example records instead of trusting intuition. In CITATION.cff research-software citation files, the practical focus at this stage is the named research software or research object. It is tempting to move quickly because the field appears familiar, yet the safer decision is to normally preserve the official title and introduce a separately labelled explanatory translation only if policy permits. Keep that rule in the working checklist so another translator can repeat it rather than relying on the first editor’s memory.

Test the decision against a concrete situation: message: "Cite this release: check the version and date." The review should not accidentally produce this alternative: Delete quotation marks and punctuation until the YAML fails or changes shape. Instead, Maintain valid YAML quoting while translating the sentence inside it. Record the source value, the reason for the edit and the structural check that proves the intended behavior survived. If the reasoning cannot be explained in one or two clear sentences, escalate the ambiguity before publication.

8. Preserve the document structure

Maintain required keys, nesting, data types, array shapes and references. Formatting and quoting changes are reviewed like code changes, not typography. In CITATION.cff research-software citation files, the practical focus at this stage is structured personal-name components. It is tempting to move quickly because the field appears familiar, yet the safer decision is to retain the entered names unless the maintainers approve a specific name representation. Keep that rule in the working checklist so another translator can repeat it rather than relying on the first editor’s memory.

Test the decision against a concrete situation: preferred-citation: type: article with its own title and authors The review should not accidentally produce this alternative: Replace the software’s main authors with paper authors. Instead, Keep nested paper attribution distinct from the main software author list. Record the source value, the reason for the edit and the structural check that proves the intended behavior survived. If the reasoning cannot be explained in one or two clear sentences, escalate the ambiguity before publication.

9. Compare field-level before and after states

Use a structured diff. A reviewer should see exactly which human-readable values changed and which protected values stayed identical. In CITATION.cff research-software citation files, the practical focus at this stage is the digital object identifier of the cited object. It is tempting to move quickly because the field appears familiar, yet the safer decision is to keep its characters unchanged and separately verify what object it resolves to. Keep that rule in the working checklist so another translator can repeat it rather than relying on the first editor’s memory.

Test the decision against a concrete situation: repository-code: https://example.org/research/river-tools The review should not accidentally produce this alternative: Translate the path segments into a localized URL without changing the destination server. Instead, Keep the literal URL unless a maintainer supplies a tested localized endpoint. Record the source value, the reason for the edit and the structural check that proves the intended behavior survived. If the reasoning cannot be explained in one or two clear sentences, escalate the ambiguity before publication.

10. Validate machine readability

Parse or validate the target record using the tools appropriate to the format. A file that cannot be consumed is not a successful translation artifact. In CITATION.cff research-software citation files, the practical focus at this stage is the declared release date. It is tempting to move quickly because the field appears familiar, yet the safer decision is to keep ISO-style date value and verify it against the release record. Keep that rule in the working checklist so another translator can repeat it rather than relying on the first editor’s memory.

Test the decision against a concrete situation: message: Please cite this software if you use results in a publication. The review should not accidentally produce this alternative: Render the instruction as a required legal condition without checking the project. Instead, Translate this as the maintainers’ request for attribution, preserving the conditional reference to use in publications. Record the source value, the reason for the edit and the structural check that proves the intended behavior survived. If the reasoning cannot be explained in one or two clear sentences, escalate the ambiguity before publication.

11. Test a real user journey

Open the translated object in a relevant reader or application. Verify not just that text appears but that its actions, links, identity and interpretation remain correct. In CITATION.cff research-software citation files, the practical focus at this stage is discovery words accompanying the object. It is tempting to move quickly because the field appears familiar, yet the safer decision is to translate descriptors if this record’s policy and target consumers accept localized keywords. Keep that rule in the working checklist so another translator can repeat it rather than relying on the first editor’s memory.

Test the decision against a concrete situation: abstract: Experimental package for comparing river-sensor observations; not for regulatory reporting. The review should not accidentally produce this alternative: Translate it as a certified official reporting product. Instead, Preserve experimental status, comparative function and explicit limitation. Record the source value, the reason for the edit and the structural check that proves the intended behavior survived. If the reasoning cannot be explained in one or two clear sentences, escalate the ambiguity before publication.

12. Approve, publish and preserve a rollback

Obtain an editorial sign-off on nontrivial language decisions. Preserve the source and a reversible revision history for future corrections. In CITATION.cff research-software citation files, the practical focus at this stage is reader-facing instruction about how to cite the work. It is tempting to move quickly because the field appears familiar, yet the safer decision is to translate its instruction and polite tone. Keep that rule in the working checklist so another translator can repeat it rather than relying on the first editor’s memory.

Test the decision against a concrete situation: given-names: Mei Lin; family-names: Tan The review should not accidentally produce this alternative: Swap the fields because target-language name order is different. Instead, Keep structured name data as recorded; handle display ordering at the presentation layer. Record the source value, the reason for the edit and the structural check that proves the intended behavior survived. If the reasoning cannot be explained in one or two clear sentences, escalate the ambiguity before publication.

Worked cases: how accurate translation changes the result

Worked case 1: A citation message for a global research team

Source situation. message: Please cite this software if you use results in a publication. This is a deliberately simplified teaching example, not a live institutional record. The first job is to identify whether the statement is evidence, an instruction, a display label or machine data. That classification determines how far the translator can legitimately rewrite it.

Common wrong move. Render the instruction as a required legal condition without checking the project. Such a rewrite may appear helpful to a reader who sees only the target text, but it crosses a boundary in the source. If a later reviewer compares real outcomes against the translated record, that unsupported change becomes a factual error rather than a matter of stylistic preference.

Correct move and reason. Translate this as the maintainers’ request for attribution, preserving the conditional reference to use in publications. Translate the human request; leave the YAML key unchanged. Verify the target conveys when and why the citation is sought, without assigning a penalty or mandatory legal rule. The practical proof is to place the original and localized versions side by side, check the protected fields, then ask whether both versions support the same defensible decision. That last test catches subtle semantic drift that an automated parser cannot detect.

Worked case 2: An experimental analysis abstract

Source situation. abstract: Experimental package for comparing river-sensor observations; not for regulatory reporting. This is a deliberately simplified teaching example, not a live institutional record. The first job is to identify whether the statement is evidence, an instruction, a display label or machine data. That classification determines how far the translator can legitimately rewrite it.

Common wrong move. Translate it as a certified official reporting product. Such a rewrite may appear helpful to a reader who sees only the target text, but it crosses a boundary in the source. If a later reviewer compares real outcomes against the translated record, that unsupported change becomes a factual error rather than a matter of stylistic preference.

Correct move and reason. Preserve experimental status, comparative function and explicit limitation. The word `experimental` affects how a reader should use the software. When removing it, the translator creates a false capability claim rather than merely changing style. The practical proof is to place the original and localized versions side by side, check the protected fields, then ask whether both versions support the same defensible decision. That last test catches subtle semantic drift that an automated parser cannot detect.

Worked case 3: A researcher with two name fields

Source situation. given-names: Mei Lin; family-names: Tan This is a deliberately simplified teaching example, not a live institutional record. The first job is to identify whether the statement is evidence, an instruction, a display label or machine data. That classification determines how far the translator can legitimately rewrite it.

Common wrong move. Swap the fields because target-language name order is different. Such a rewrite may appear helpful to a reader who sees only the target text, but it crosses a boundary in the source. If a later reviewer compares real outcomes against the translated record, that unsupported change becomes a factual error rather than a matter of stylistic preference.

Correct move and reason. Keep structured name data as recorded; handle display ordering at the presentation layer. Human reading order and machine field semantics are separate decisions. An output UI can display names according to locale without corrupting the source field contents. The practical proof is to place the original and localized versions side by side, check the protected fields, then ask whether both versions support the same defensible decision. That last test catches subtle semantic drift that an automated parser cannot detect.

Worked case 4: An organisation among individual authors

Source situation. authors: – name: Regional Coastal Data Group This is a deliberately simplified teaching example, not a live institutional record. The first job is to identify whether the statement is evidence, an instruction, a display label or machine data. That classification determines how far the translator can legitimately rewrite it.

Common wrong move. Invent a surname and given name to satisfy a target-language expectation. Such a rewrite may appear helpful to a reader who sees only the target text, but it crosses a boundary in the source. If a later reviewer compares real outcomes against the translated record, that unsupported change becomes a factual error rather than a matter of stylistic preference.

Correct move and reason. Retain organisation attribution as an organisation. An institutional contributor does not become a person when a language lacks a convenient equivalent. Translators should escalate unusual author structures to maintainers. The practical proof is to place the original and localized versions side by side, check the protected fields, then ask whether both versions support the same defensible decision. That last test catches subtle semantic drift that an automated parser cannot detect.

Worked case 5: A release DOI versus a collection DOI

Source situation. identifiers: two DOIs, one for all archived versions and one for release 2.0.4 This is a deliberately simplified teaching example, not a live institutional record. The first job is to identify whether the statement is evidence, an instruction, a display label or machine data. That classification determines how far the translator can legitimately rewrite it.

Common wrong move. Pick whichever DOI looks shorter in the translated message. Such a rewrite may appear helpful to a reader who sees only the target text, but it crosses a boundary in the source. If a later reviewer compares real outcomes against the translated record, that unsupported change becomes a factual error rather than a matter of stylistic preference.

Correct move and reason. Keep both with accurate descriptions of their respective scopes. The translator must understand which object each identifier points at. Different identifiers can be valid without being interchangeable. The practical proof is to place the original and localized versions side by side, check the protected fields, then ask whether both versions support the same defensible decision. That last test catches subtle semantic drift that an automated parser cannot detect.

Worked case 6: An apparently translatable software title

Source situation. title: RiverWatch Analysis Kit This is a deliberately simplified teaching example, not a live institutional record. The first job is to identify whether the statement is evidence, an instruction, a display label or machine data. That classification determines how far the translator can legitimately rewrite it.

Common wrong move. Rename the project into an unapproved target-language product name. Such a rewrite may appear helpful to a reader who sees only the target text, but it crosses a boundary in the source. If a later reviewer compares real outcomes against the translated record, that unsupported change becomes a factual error rather than a matter of stylistic preference.

Correct move and reason. Preserve the canonical title and explain its meaning in separate prose if needed. A name can be meaningful English yet function as an identity label. The governing question is what a person would search for to find the same repository. The practical proof is to place the original and localized versions side by side, check the protected fields, then ask whether both versions support the same defensible decision. That last test catches subtle semantic drift that an automated parser cannot detect.

Worked case 7: A YAML string containing punctuation

Source situation. message: "Cite this release: check the version and date." This is a deliberately simplified teaching example, not a live institutional record. The first job is to identify whether the statement is evidence, an instruction, a display label or machine data. That classification determines how far the translator can legitimately rewrite it.

Common wrong move. Delete quotation marks and punctuation until the YAML fails or changes shape. Such a rewrite may appear helpful to a reader who sees only the target text, but it crosses a boundary in the source. If a later reviewer compares real outcomes against the translated record, that unsupported change becomes a factual error rather than a matter of stylistic preference.

Correct move and reason. Maintain valid YAML quoting while translating the sentence inside it. The syntax layer and language layer must be reviewed independently. A fluent sentence cannot compensate for an invalid source file. The practical proof is to place the original and localized versions side by side, check the protected fields, then ask whether both versions support the same defensible decision. That last test catches subtle semantic drift that an automated parser cannot detect.

Worked case 8: A paper supplied as preferred citation

Source situation. preferred-citation: type: article with its own title and authors This is a deliberately simplified teaching example, not a live institutional record. The first job is to identify whether the statement is evidence, an instruction, a display label or machine data. That classification determines how far the translator can legitimately rewrite it.

Common wrong move. Replace the software’s main authors with paper authors. Such a rewrite may appear helpful to a reader who sees only the target text, but it crosses a boundary in the source. If a later reviewer compares real outcomes against the translated record, that unsupported change becomes a factual error rather than a matter of stylistic preference.

Correct move and reason. Keep nested paper attribution distinct from the main software author list. Two citations may describe related but different objects. Scope controls attribution, and translation must not erase that boundary. The practical proof is to place the original and localized versions side by side, check the protected fields, then ask whether both versions support the same defensible decision. That last test catches subtle semantic drift that an automated parser cannot detect.

Worked case 9: A repository address in translator notes

Source situation. repository-code: https://example.org/research/river-tools This is a deliberately simplified teaching example, not a live institutional record. The first job is to identify whether the statement is evidence, an instruction, a display label or machine data. That classification determines how far the translator can legitimately rewrite it.

Common wrong move. Translate the path segments into a localized URL without changing the destination server. Such a rewrite may appear helpful to a reader who sees only the target text, but it crosses a boundary in the source. If a later reviewer compares real outcomes against the translated record, that unsupported change becomes a factual error rather than a matter of stylistic preference.

Correct move and reason. Keep the literal URL unless a maintainer supplies a tested localized endpoint. A URL is actionable address data. Guessing its localized form creates a new and potentially broken location. The practical proof is to place the original and localized versions side by side, check the protected fields, then ask whether both versions support the same defensible decision. That last test catches subtle semantic drift that an automated parser cannot detect.

Failure modes and how to diagnose them

Failure mode 1: A citation message for a global research team

This failure arises when an editor treats `message` as though it were unrestricted prose or assumes that render the instruction as a required legal condition without checking the project. The warning sign is a translated record that sounds clearer but can no longer support the original task. To diagnose it, locate the precise source claim, inspect whether the target added certainty or removed an essential condition, and identify the affected field or downstream link.

Repair it by returning to the recorded meaning: Translate this as the maintainers’ request for attribution, preserving the conditional reference to use in publications. Then re-run both checks: does the structured object still parse and map to the same entities, and can a person using the target language make the same practical interpretation? Translate the human request; leave the YAML key unchanged. Verify the target conveys when and why the citation is sought, without assigning a penalty or mandatory legal rule. A resolved issue needs both forms of evidence, not an editor’s impression that the passage now reads more smoothly.

Failure mode 2: An experimental analysis abstract

This failure arises when an editor treats `title` as though it were unrestricted prose or assumes that translate it as a certified official reporting product. The warning sign is a translated record that sounds clearer but can no longer support the original task. To diagnose it, locate the precise source claim, inspect whether the target added certainty or removed an essential condition, and identify the affected field or downstream link.

Repair it by returning to the recorded meaning: Preserve experimental status, comparative function and explicit limitation. Then re-run both checks: does the structured object still parse and map to the same entities, and can a person using the target language make the same practical interpretation? The word `experimental` affects how a reader should use the software. When removing it, the translator creates a false capability claim rather than merely changing style. A resolved issue needs both forms of evidence, not an editor’s impression that the passage now reads more smoothly.

Failure mode 3: A researcher with two name fields

This failure arises when an editor treats `abstract` as though it were unrestricted prose or assumes that swap the fields because target-language name order is different. The warning sign is a translated record that sounds clearer but can no longer support the original task. To diagnose it, locate the precise source claim, inspect whether the target added certainty or removed an essential condition, and identify the affected field or downstream link.

Repair it by returning to the recorded meaning: Keep structured name data as recorded; handle display ordering at the presentation layer. Then re-run both checks: does the structured object still parse and map to the same entities, and can a person using the target language make the same practical interpretation? Human reading order and machine field semantics are separate decisions. An output UI can display names according to locale without corrupting the source field contents. A resolved issue needs both forms of evidence, not an editor’s impression that the passage now reads more smoothly.

Failure mode 4: An organisation among individual authors

This failure arises when an editor treats `authors` as though it were unrestricted prose or assumes that invent a surname and given name to satisfy a target-language expectation. The warning sign is a translated record that sounds clearer but can no longer support the original task. To diagnose it, locate the precise source claim, inspect whether the target added certainty or removed an essential condition, and identify the affected field or downstream link.

Repair it by returning to the recorded meaning: Retain organisation attribution as an organisation. Then re-run both checks: does the structured object still parse and map to the same entities, and can a person using the target language make the same practical interpretation? An institutional contributor does not become a person when a language lacks a convenient equivalent. Translators should escalate unusual author structures to maintainers. A resolved issue needs both forms of evidence, not an editor’s impression that the passage now reads more smoothly.

Failure mode 5: A release DOI versus a collection DOI

This failure arises when an editor treats `given-names` and `family-names` as though it were unrestricted prose or assumes that pick whichever doi looks shorter in the translated message. The warning sign is a translated record that sounds clearer but can no longer support the original task. To diagnose it, locate the precise source claim, inspect whether the target added certainty or removed an essential condition, and identify the affected field or downstream link.

Repair it by returning to the recorded meaning: Keep both with accurate descriptions of their respective scopes. Then re-run both checks: does the structured object still parse and map to the same entities, and can a person using the target language make the same practical interpretation? The translator must understand which object each identifier points at. Different identifiers can be valid without being interchangeable. A resolved issue needs both forms of evidence, not an editor’s impression that the passage now reads more smoothly.

Failure mode 6: An apparently translatable software title

This failure arises when an editor treats `name` as though it were unrestricted prose or assumes that rename the project into an unapproved target-language product name. The warning sign is a translated record that sounds clearer but can no longer support the original task. To diagnose it, locate the precise source claim, inspect whether the target added certainty or removed an essential condition, and identify the affected field or downstream link.

Repair it by returning to the recorded meaning: Preserve the canonical title and explain its meaning in separate prose if needed. Then re-run both checks: does the structured object still parse and map to the same entities, and can a person using the target language make the same practical interpretation? A name can be meaningful English yet function as an identity label. The governing question is what a person would search for to find the same repository. A resolved issue needs both forms of evidence, not an editor’s impression that the passage now reads more smoothly.

Failure mode 7: A YAML string containing punctuation

This failure arises when an editor treats `orcid` as though it were unrestricted prose or assumes that delete quotation marks and punctuation until the yaml fails or changes shape. The warning sign is a translated record that sounds clearer but can no longer support the original task. To diagnose it, locate the precise source claim, inspect whether the target added certainty or removed an essential condition, and identify the affected field or downstream link.

Repair it by returning to the recorded meaning: Maintain valid YAML quoting while translating the sentence inside it. Then re-run both checks: does the structured object still parse and map to the same entities, and can a person using the target language make the same practical interpretation? The syntax layer and language layer must be reviewed independently. A fluent sentence cannot compensate for an invalid source file. A resolved issue needs both forms of evidence, not an editor’s impression that the passage now reads more smoothly.

Failure mode 8: A paper supplied as preferred citation

This failure arises when an editor treats `doi` as though it were unrestricted prose or assumes that replace the software’s main authors with paper authors. The warning sign is a translated record that sounds clearer but can no longer support the original task. To diagnose it, locate the precise source claim, inspect whether the target added certainty or removed an essential condition, and identify the affected field or downstream link.

Repair it by returning to the recorded meaning: Keep nested paper attribution distinct from the main software author list. Then re-run both checks: does the structured object still parse and map to the same entities, and can a person using the target language make the same practical interpretation? Two citations may describe related but different objects. Scope controls attribution, and translation must not erase that boundary. A resolved issue needs both forms of evidence, not an editor’s impression that the passage now reads more smoothly.

Failure mode 9: A repository address in translator notes

This failure arises when an editor treats `identifiers` as though it were unrestricted prose or assumes that translate the path segments into a localized url without changing the destination server. The warning sign is a translated record that sounds clearer but can no longer support the original task. To diagnose it, locate the precise source claim, inspect whether the target added certainty or removed an essential condition, and identify the affected field or downstream link.

Repair it by returning to the recorded meaning: Keep the literal URL unless a maintainer supplies a tested localized endpoint. Then re-run both checks: does the structured object still parse and map to the same entities, and can a person using the target language make the same practical interpretation? A URL is actionable address data. Guessing its localized form creates a new and potentially broken location. A resolved issue needs both forms of evidence, not an editor’s impression that the passage now reads more smoothly.

Practice and transfer: exercises with model reasoning

Exercise 1: audit a citation message for a global research team

Task. You receive this source cue: message: Please cite this software if you use results in a publication. A teammate proposes the following shortcut: Render the instruction as a required legal condition without checking the project. Write a correction note that identifies the protected information, the meaning that must remain visible and one test that would catch an error. Use `title` as an additional cross-check in your answer.

Model reasoning. Translate this as the maintainers’ request for attribution, preserving the conditional reference to use in publications. The reasoning is not merely that the original wording is preferable: Translate the human request; leave the YAML key unchanged. Verify the target conveys when and why the citation is sought, without assigning a penalty or mandatory legal rule. The additional check should ask whether the named research software or research object remains valid after localization. Where the field is absent or evidence is incomplete, the correct answer is to say what is unknown rather than supplying an attractive guess.

Exercise 2: audit an experimental analysis abstract

Task. You receive this source cue: abstract: Experimental package for comparing river-sensor observations; not for regulatory reporting. A teammate proposes the following shortcut: Translate it as a certified official reporting product. Write a correction note that identifies the protected information, the meaning that must remain visible and one test that would catch an error. Use `abstract` as an additional cross-check in your answer.

Model reasoning. Preserve experimental status, comparative function and explicit limitation. The reasoning is not merely that the original wording is preferable: The word `experimental` affects how a reader should use the software. When removing it, the translator creates a false capability claim rather than merely changing style. The additional check should ask whether the prose summary describing what the software does remains valid after localization. Where the field is absent or evidence is incomplete, the correct answer is to say what is unknown rather than supplying an attractive guess.

Exercise 3: audit a researcher with two name fields

Task. You receive this source cue: given-names: Mei Lin; family-names: Tan A teammate proposes the following shortcut: Swap the fields because target-language name order is different. Write a correction note that identifies the protected information, the meaning that must remain visible and one test that would catch an error. Use `authors` as an additional cross-check in your answer.

Model reasoning. Keep structured name data as recorded; handle display ordering at the presentation layer. The reasoning is not merely that the original wording is preferable: Human reading order and machine field semantics are separate decisions. An output UI can display names according to locale without corrupting the source field contents. The additional check should ask whether the ordered people or organisations credited for the work remains valid after localization. Where the field is absent or evidence is incomplete, the correct answer is to say what is unknown rather than supplying an attractive guess.

Exercise 4: audit an organisation among individual authors

Task. You receive this source cue: authors: – name: Regional Coastal Data Group A teammate proposes the following shortcut: Invent a surname and given name to satisfy a target-language expectation. Write a correction note that identifies the protected information, the meaning that must remain visible and one test that would catch an error. Use `given-names` and `family-names` as an additional cross-check in your answer.

Model reasoning. Retain organisation attribution as an organisation. The reasoning is not merely that the original wording is preferable: An institutional contributor does not become a person when a language lacks a convenient equivalent. Translators should escalate unusual author structures to maintainers. The additional check should ask whether structured personal-name components remains valid after localization. Where the field is absent or evidence is incomplete, the correct answer is to say what is unknown rather than supplying an attractive guess.

Exercise 5: audit a release doi versus a collection doi

Task. You receive this source cue: identifiers: two DOIs, one for all archived versions and one for release 2.0.4 A teammate proposes the following shortcut: Pick whichever DOI looks shorter in the translated message. Write a correction note that identifies the protected information, the meaning that must remain visible and one test that would catch an error. Use `name` as an additional cross-check in your answer.

Model reasoning. Keep both with accurate descriptions of their respective scopes. The reasoning is not merely that the original wording is preferable: The translator must understand which object each identifier points at. Different identifiers can be valid without being interchangeable. The additional check should ask whether a supported name representation for an author entity, often an organisation remains valid after localization. Where the field is absent or evidence is incomplete, the correct answer is to say what is unknown rather than supplying an attractive guess.

Exercise 6: audit an apparently translatable software title

Task. You receive this source cue: title: RiverWatch Analysis Kit A teammate proposes the following shortcut: Rename the project into an unapproved target-language product name. Write a correction note that identifies the protected information, the meaning that must remain visible and one test that would catch an error. Use `orcid` as an additional cross-check in your answer.

Model reasoning. Preserve the canonical title and explain its meaning in separate prose if needed. The reasoning is not merely that the original wording is preferable: A name can be meaningful English yet function as an identity label. The governing question is what a person would search for to find the same repository. The additional check should ask whether the author identifier which disambiguates a research contributor remains valid after localization. Where the field is absent or evidence is incomplete, the correct answer is to say what is unknown rather than supplying an attractive guess.

Exercise 7: audit a yaml string containing punctuation

Task. You receive this source cue: message: "Cite this release: check the version and date." A teammate proposes the following shortcut: Delete quotation marks and punctuation until the YAML fails or changes shape. Write a correction note that identifies the protected information, the meaning that must remain visible and one test that would catch an error. Use `doi` as an additional cross-check in your answer.

Model reasoning. Maintain valid YAML quoting while translating the sentence inside it. The reasoning is not merely that the original wording is preferable: The syntax layer and language layer must be reviewed independently. A fluent sentence cannot compensate for an invalid source file. The additional check should ask whether the digital object identifier of the cited object remains valid after localization. Where the field is absent or evidence is incomplete, the correct answer is to say what is unknown rather than supplying an attractive guess.

Exercise 8: audit a paper supplied as preferred citation

Task. You receive this source cue: preferred-citation: type: article with its own title and authors A teammate proposes the following shortcut: Replace the software’s main authors with paper authors. Write a correction note that identifies the protected information, the meaning that must remain visible and one test that would catch an error. Use `identifiers` as an additional cross-check in your answer.

Model reasoning. Keep nested paper attribution distinct from the main software author list. The reasoning is not merely that the original wording is preferable: Two citations may describe related but different objects. Scope controls attribution, and translation must not erase that boundary. The additional check should ask whether additional typed identifiers for a software artifact or collection remains valid after localization. Where the field is absent or evidence is incomplete, the correct answer is to say what is unknown rather than supplying an attractive guess.

Exercise 9: audit a repository address in translator notes

Task. You receive this source cue: repository-code: https://example.org/research/river-tools A teammate proposes the following shortcut: Translate the path segments into a localized URL without changing the destination server. Write a correction note that identifies the protected information, the meaning that must remain visible and one test that would catch an error. Use `version` as an additional cross-check in your answer.

Model reasoning. Keep the literal URL unless a maintainer supplies a tested localized endpoint. The reasoning is not merely that the original wording is preferable: A URL is actionable address data. Guessing its localized form creates a new and potentially broken location. The additional check should ask whether the named release version of the software remains valid after localization. Where the field is absent or evidence is incomplete, the correct answer is to say what is unknown rather than supplying an attractive guess.

A compact quality-assurance checklist to use before releasing translations

  • Can a reader identify the same object, action, condition and limit in both languages?
  • Are all field names, keys and machine tokens still valid under the governing specification?
  • Have IDs, URLs, numeric units, author or agency names, timestamps and reference relations been preserved or explicitly verified?
  • Does the target-language UI clearly distinguish a documented fact from a possibility, example or estimate?
  • Does the structured file parse and does a real reader/application behave correctly?
  • Have examples and tests covered the edge cases rather than only the easy path?
  • Is there a versioned source, diff, reviewer decision and rollback point?

Do not mark an item complete just because it passed a spell checker. The highest-risk translation mistakes frequently preserve perfect grammar while changing obligations, provenance, units or identity. Conversely, a machine validator may report no errors while a translated summary asserts something unsupported. Good publication control therefore assigns separate responsibility for linguistic accuracy and structural integrity.

Frequently asked questions

Can a CITATION.cff title be translated?

Only if the project’s editorial policy explicitly allows a localized display title and the citation identity remains clear. Do not silently replace the official title in the source record. For an actual release, confirm the specific record against its governing schema and authoritative publisher. When the same expression has both an ordinary-language and a technical meaning, make the difference explicit in the reviewer note. Correct translation is demonstrated by reproducible decisions, not confidence in a phrase alone.

Are field names in CITATION.cff translatable?

No. Machine-readable YAML property names identify schema fields. Translators usually work with selected human-readable values such as explanatory messages and abstracts. For an actual release, confirm the specific record against its governing schema and authoritative publisher. When the same expression has both an ordinary-language and a technical meaning, make the difference explicit in the reviewer note. Correct translation is demonstrated by reproducible decisions, not confidence in a phrase alone.

What should happen to a DOI?

Preserve its exact value and verify its intended referent. Do not localize punctuation, digits or path segments and do not swap a release DOI for a collection DOI. For an actual release, confirm the specific record against its governing schema and authoritative publisher. When the same expression has both an ordinary-language and a technical meaning, make the difference explicit in the reviewer note. Correct translation is demonstrated by reproducible decisions, not confidence in a phrase alone.

How should an organisation author be treated?

Keep it as the named organisational contributor. Do not fabricate a given name and family name to fit a person template. For an actual release, confirm the specific record against its governing schema and authoritative publisher. When the same expression has both an ordinary-language and a technical meaning, make the difference explicit in the reviewer note. Correct translation is demonstrated by reproducible decisions, not confidence in a phrase alone.

Is CITATION.cff the same as BibTeX?

No. Citation File Format is a source metadata file; tools can render citations or export formats such as BibTeX. The translation decision begins with the CFF source. For an actual release, confirm the specific record against its governing schema and authoritative publisher. When the same expression has both an ordinary-language and a technical meaning, make the difference explicit in the reviewer note. Correct translation is demonstrated by reproducible decisions, not confidence in a phrase alone.

What happens if two authors have different name scripts?

Preserve authoritative identity fields, use approved script variants when provided, and document any display transliteration without modifying the underlying author identity silently. For an actual release, confirm the specific record against its governing schema and authoritative publisher. When the same expression has both an ordinary-language and a technical meaning, make the difference explicit in the reviewer note. Correct translation is demonstrated by reproducible decisions, not confidence in a phrase alone.

Does translating a software abstract change the license?

No. An abstract is a description of purpose and capabilities. License identifiers and license terms need separate legal and editorial treatment. For an actual release, confirm the specific record against its governing schema and authoritative publisher. When the same expression has both an ordinary-language and a technical meaning, make the difference explicit in the reviewer note. Correct translation is demonstrated by reproducible decisions, not confidence in a phrase alone.

Why validate YAML after editing?

Because a broken quote, indentation change or invalid scalar can make a plausible-looking record unreadable by tools even when every translated sentence sounds natural. For an actual release, confirm the specific record against its governing schema and authoritative publisher. When the same expression has both an ordinary-language and a technical meaning, make the difference explicit in the reviewer note. Correct translation is demonstrated by reproducible decisions, not confidence in a phrase alone.

Who approves a localized release title?

The repository maintainers or responsible publication team should govern changes to canonical names and citations. A translator can recommend alternatives but should not invent them. For an actual release, confirm the specific record against its governing schema and authoritative publisher. When the same expression has both an ordinary-language and a technical meaning, make the difference explicit in the reviewer note. Correct translation is demonstrated by reproducible decisions, not confidence in a phrase alone.

What is the quickest safe review?

Inspect a text diff, run a YAML/CFF validator, compare authors and identifiers byte-for-byte, then test the displayed citation or repository citation prompt. For an actual release, confirm the specific record against its governing schema and authoritative publisher. When the same expression has both an ordinary-language and a technical meaning, make the difference explicit in the reviewer note. Correct translation is demonstrated by reproducible decisions, not confidence in a phrase alone.

What to read next and how to transfer this method

Use this guide when the problem is specifically how to translate CITATION.cff YAML metadata without changing software citation identity. For broader decisions about multilingual vocabulary, English meaning and schema-aware translation, choose the relevant adjacent owner rather than duplicating its coverage. Useful eduKateSG reading routes are listed below.

Authoritative technical references for this field include Citation File Format project and examples and GitHub: About CITATION files. These references define the data contract; the original examples and teaching exercises above show how to apply that contract without turning translation into unreviewed data editing.

The transferable skill is straightforward: classify the source, protect what remains factual or machine-readable, translate what a person must understand, and verify the resulting action. A translator who can explain the boundary between those layers can apply the same reasoning to new records without assuming that all files or industries obey identical rules.