Master Art of Translation | The Language Documentation, Interlinear Glossing & Linguistic Archive Translation System — How Morphemes, Speaker Consent, Recordings and Language IDs Keep Their Meaning

How to translate interlinear glosses, field recordings, linguistic corpora and language-archive metadata accurately

How do you translate an interlinear glossed text, a linguistic fieldwork transcript, an endangered-language recording, or a lexical database without changing what the speaker said or what the grammar shows? Language documentation translation is more demanding than replacing a story with fluent English. Every segment has a voice, an original form, an interpretation, an annotation history, a recording time, and sometimes a restriction on who may hear or reuse it. This complete guide explains a practical system for translating linguistic fieldwork and interlinear glosses while keeping original evidence distinct from the analyst’s interpretation.

The crucial questions are practical: What is the difference between transcription, morphological segmentation, morpheme-by-morpheme gloss and free translation? How do Leipzig Glossing Rules, ELAN annotations, FieldWorks Language Explorer, CLDF tables, ISO 639 identifiers and Glottolog references fit together? How should researchers and community editors handle uncertainty, speaker consent, time-aligned audio, orthography choices, multiple language varieties and sensitive oral histories? These are related tasks, but solving one does not automatically solve the others.

A reliable multilingual language-documentation workflow protects the recording and speaker’s intent, preserves the alignment between observed forms and analysis, makes terminology explicit, and checks that a published translation remains traceable to a versioned source. It also treats access permission as a substantive part of the result. A perfect gloss is not ethical to publish if the material was not cleared for that audience. This article provides searchable definitions, a reusable review procedure, worked diagnostics, language-data quality checks, transfer exercises and links to authoritative references. All invented sample utterances and record IDs here are teaching examples, not claims about any real language or community.

Series: EDKSG-TRANS-MASTER-WORLD-860. Audience: translators, linguistic fieldworkers, documentation teams, archive curators, community language workers, editors, data stewards and students of linguistic analysis. Research review: 1 November 2026. This is a world-facing methods article, not a licence to disclose restricted recordings or an assertion that one analytic theory governs every language.

Start with the translation root: Master Art of Translation | The Complete System for Moving Meaning Between Languages. This child owns the special problems of documentation and interlinear data; it does not take over the root’s general theory, the protected vocabulary-learning owner or the protected English-grammar owner.

Quick navigation through the complete learning route

• Search intent and scope
• The hidden problem
• The evidence-and-language mechanism
• The repeatable workflow
• Concept and field atlas
• Worked diagnostic laboratory
• Transfer workshops
• Failure analysis
• Working glossary
• Questions and answers
• Four-week practice plan
• Research and standards
• Next routes

1. Search intent: which language-documentation translation problem is the reader trying to solve?

A translator may receive an audio recording and be asked for subtitles. A researcher may receive a four-line example consisting of an original utterance, a segmented line, morpheme glosses and a free translation. A community editor may be asked to correct an ancestral-language story whose public archive caption has turned a tentative identification into an asserted historical fact. A data engineer may be converting a collection of lexical entries into CLDF tables. Each task uses the word translation, but each has different evidence, permissions, output structure and failure conditions.

Before working, identify the requested deliverable: a faithful rendering of what was said; a readable summary; a morpheme-level analytical gloss; a bilingual lexicon entry; a subtitle linked to audio; a language-learning adaptation; or an archival metadata statement. A free translation is not an analysis of morphology. A glossary is not a speaker-approved dictionary. An ISO language code is not a community’s preferred name. An accessible website is not proof that a recording may be republished. The task specification must say what the new text will allow its reader to infer or do.

The page’s search language is intentionally specific: interlinear gloss translation, language documentation best practices, Leipzig Glossing Rules examples, ELAN tier alignment, FLEx lexical sense translation, endangered-language archive permissions, CLDF dataset validation, Glottolog language identifiers, orthography consistency, oral-history subtitles and speaker-consent records. Readers searching for these terms need more than definitions: they need an operating method for deciding what cannot be silently changed.

2. The hidden problem: a recording, its analysis and its publication are not the same object

Consider a completely fictional clip marked SESSION-Q07. Two people speak at once. One repeats a noun, changes a verb ending, laughs, and gestures toward an object off camera. An editor produces a smooth target-language sentence, deletes the repetition, interprets the laughter as agreement, assigns a single speaker and adds a place name from memory. The sentence might read naturally while misrepresenting the conversation in five separate ways. There has been no gross dictionary error; the loss occurred because layers of evidence were merged before anyone checked their authority.

The original waveform is primary evidence of an acoustic event, not a complete representation of what it means. A transcription is an interpretation of heard speech under an orthography policy. Segmentation proposes word and morpheme boundaries; morphological glossing states analyses that may be disputed. A free translation produces intelligible discourse in another language. Metadata describes circumstances, people, rights, access and data versions. Each layer can be corrected independently, and each should link back to its source. When these layers collapse into one prose paragraph, later readers cannot identify whether a phrase came from a speaker, a translator, an annotator or a catalogue editor.

The ethical problem is equally hidden. The subject of a recording may not have agreed that raw audio, personal names, ritual knowledge, exact locations or recognizable voices would be publicly searchable. Deposit permission, public display permission, educational reuse, commercial reuse, training-data reuse and translation distribution are not one universal yes. Restrictions may differ by item, participant, time, platform and audience. Accordingly, an apparently excellent transcription can still be an unacceptable publication. The safe workflow treats access and attribution as content requirements rather than administrative footnotes.

A third error involves linguistic prestige. A published reference name or standardized code is useful for interoperable cataloguing but does not determine how people identify their language. Some sources describe a cluster of varieties while a speaker identifies a more specific one. A historical code can change its scope. The translator should preserve identifiers for machine matching, retain original labels, record preferred community expressions and explain any unresolved distinction. Replacing all variants with the most famous label is not normalization; it is the loss of identity metadata.

The solution is a chain of typed assertions. Keep the audio segment, transcript, grammatical analysis, free translation, community revision, identifier mapping and access decision as separate records. Record who asserted each one, why, from which source version and under what licence or permission. Then construct the reader-facing output from that verified chain. Translating meaning becomes evidence management plus linguistically skilled writing, rather than a detached rewrite.

3. Mechanism: the nine layers that make a linguistic record interpretable

1. Acoustic and visual observation

A waveform, video frame and time range fix what was recorded, but not automatically who spoke or why. Silence, overlapping speech, background music and gesture all affect interpretation. Before translating a sentence, the reviewer can replay its interval and distinguish audible detail from a note added after the recording.

A practical audit begins by asking who has authority over this layer and what evidence would overturn the current interpretation. For acoustic and visual observation, store an explicit owner, version, review state and record link. If the layer is revised, enumerate every dependent caption, gloss, subtitle, dictionary sense, public search result and translation extract. Releasing only the visible sentence while leaving the surrounding metadata unchanged can make a later correction impossible to interpret. A mature documentation project therefore treats the translated output as a maintained view of evidence, not as an independent source of historical truth.

2. Participant and occasion context

A narrative, elicitation session, argument, demonstration or conversation changes how an utterance functions. Participant roles and permissions can differ even inside one session. Treat the setting as contextual evidence; do not use it to put words in a speaker’s mouth or disclose sensitive identities.

A practical audit begins by asking who has authority over this layer and what evidence would overturn the current interpretation. For participant and occasion context, store an explicit owner, version, review state and record link. If the layer is revised, enumerate every dependent caption, gloss, subtitle, dictionary sense, public search result and translation extract. Releasing only the visible sentence while leaving the surrounding metadata unchanged can make a later correction impossible to interpret. A mature documentation project therefore treats the translated output as a maintained view of evidence, not as an independent source of historical truth.

3. Orthographic representation

Orthography represents speech under a writing convention. A phonemic transcription, broad practical spelling and narrow phonetic rendering are different editorial choices. Preserve a reversible link to the original string and document normalization so later editors can revisit disputed forms.

A practical audit begins by asking who has authority over this layer and what evidence would overturn the current interpretation. For orthographic representation, store an explicit owner, version, review state and record link. If the layer is revised, enumerate every dependent caption, gloss, subtitle, dictionary sense, public search result and translation extract. Releasing only the visible sentence while leaving the surrounding metadata unchanged can make a later correction impossible to interpret. A mature documentation project therefore treats the translated output as a maintained view of evidence, not as an independent source of historical truth.

4. Word and morpheme segmentation

Morpheme segmentation hypothesizes meaningful components and their boundaries. Leipzig conventions help align a segmented line with glosses but do not themselves prove an analysis. If morphology changes during review, recheck all dependent glosses and translations rather than editing only the line visible to readers.

A practical audit begins by asking who has authority over this layer and what evidence would overturn the current interpretation. For word and morpheme segmentation, store an explicit owner, version, review state and record link. If the layer is revised, enumerate every dependent caption, gloss, subtitle, dictionary sense, public search result and translation extract. Releasing only the visible sentence while leaving the surrounding metadata unchanged can make a later correction impossible to interpret. A mature documentation project therefore treats the translated output as a maintained view of evidence, not as an independent source of historical truth.

5. Glossing and grammatical categories

Short labels such as PST, NEG, 1SG, ERG or EVID summarize analyses using project-defined conventions. A grammatical label is not a lexical translation. Maintain a key to abbreviations and distinguish an uncertain parse from an observed acoustic fact.

A practical audit begins by asking who has authority over this layer and what evidence would overturn the current interpretation. For glossing and grammatical categories, store an explicit owner, version, review state and record link. If the layer is revised, enumerate every dependent caption, gloss, subtitle, dictionary sense, public search result and translation extract. Releasing only the visible sentence while leaving the surrounding metadata unchanged can make a later correction impossible to interpret. A mature documentation project therefore treats the translated output as a maintained view of evidence, not as an independent source of historical truth.

6. Discourse and pragmatic interpretation

Meaning includes turn-taking, stance, presupposition, reference, politeness, correction and genre. Some content is intentionally indirect or belongs to a shared history. Translate discourse force without adding unsupported details, especially when writing for people who lack the original social context.

A practical audit begins by asking who has authority over this layer and what evidence would overturn the current interpretation. For discourse and pragmatic interpretation, store an explicit owner, version, review state and record link. If the layer is revised, enumerate every dependent caption, gloss, subtitle, dictionary sense, public search result and translation extract. Releasing only the visible sentence while leaving the surrounding metadata unchanged can make a later correction impossible to interpret. A mature documentation project therefore treats the translated output as a maintained view of evidence, not as an independent source of historical truth.

7. Lexical structure and sense relations

A dictionary form may have different senses, constructions, cultural restrictions and etymologies. One source form does not necessarily map to one target-language headword. Link each translated definition to a specified sense and example, rather than borrowing the first dictionary equivalent.

A practical audit begins by asking who has authority over this layer and what evidence would overturn the current interpretation. For lexical structure and sense relations, store an explicit owner, version, review state and record link. If the layer is revised, enumerate every dependent caption, gloss, subtitle, dictionary sense, public search result and translation extract. Releasing only the visible sentence while leaving the surrounding metadata unchanged can make a later correction impossible to interpret. A mature documentation project therefore treats the translated output as a maintained view of evidence, not as an independent source of historical truth.

8. Identifiers, versions and interoperability

Stable internal IDs, language-code snapshots, dataset references and file versions allow a record to travel between applications and archives. Names can be displayed differently, but identifier reassignment requires a documented mapping and must not break citations or joins.

A practical audit begins by asking who has authority over this layer and what evidence would overturn the current interpretation. For identifiers, versions and interoperability, store an explicit owner, version, review state and record link. If the layer is revised, enumerate every dependent caption, gloss, subtitle, dictionary sense, public search result and translation extract. Releasing only the visible sentence while leaving the surrounding metadata unchanged can make a later correction impossible to interpret. A mature documentation project therefore treats the translated output as a maintained view of evidence, not as an independent source of historical truth.

9. Consent, attribution and circulation

Permissions, access conditions, speaker credit, culturally sensitive material and correction channels define what may circulate, to whom and for what purpose. A technically valid file can remain unpublished. Release requires authorized decisions and a way to revise or restrict access.

A practical audit begins by asking who has authority over this layer and what evidence would overturn the current interpretation. For consent, attribution and circulation, store an explicit owner, version, review state and record link. If the layer is revised, enumerate every dependent caption, gloss, subtitle, dictionary sense, public search result and translation extract. Releasing only the visible sentence while leaving the surrounding metadata unchanged can make a later correction impossible to interpret. A mature documentation project therefore treats the translated output as a maintained view of evidence, not as an independent source of historical truth.

4. Repeatable method: twenty-four gates from recording request to approved reuse

01. Identify the use case

The purpose of this gate is to specify whether the deliverable is transcription, translation, gloss, dictionary, subtitle or public summary. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: The commission mixes distinct outputs and reviewers approve different objects. The required deliverable at this stage is a one-page delivery brief naming each output and reader. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

02. Establish custodianship

The purpose of this gate is to identify who may authorize access to the recording and its derivatives. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: A requester possesses a file but does not control its circulation. The required deliverable at this stage is a signed or otherwise documented authorization path. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

03. Freeze the source

The purpose of this gate is to record file IDs, checksums, revision and the exact session or page. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: Reviewers compare translations to different media edits. The required deliverable at this stage is a stable source manifest with a named version. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

04. Map participants

The purpose of this gate is to record speaker, interviewer, transcriber and reviewer roles with privacy settings. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: A translator mistakes an interviewer prompt for narrator speech. The required deliverable at this stage is a role table linked to time ranges. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

05. Capture consent scope

The purpose of this gate is to distinguish recording, archiving, discovery, display, teaching, redistribution and computational reuse. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: Publication is assumed because research recording was permitted. The required deliverable at this stage is an item-level permissions matrix. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

06. Check restricted knowledge

The purpose of this gate is to ask authorized community reviewers about sensitive stories, names, rituals and locations. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: Material enters public search before consent issues are considered. The required deliverable at this stage is a release decision and an access-aware redaction plan. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

07. Identify language and variety

The purpose of this gate is to record preferred names, documented identifiers, variety and uncertainty. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: A broad macrolanguage replaces an individual variety without evidence. The required deliverable at this stage is a multilingual identity and code-mapping note. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

08. Specify writing convention

The purpose of this gate is to choose practical orthography, phonetic notation, phonemic transcription or parallel representations. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: Editors silently standardize away meaningful distinctions. The required deliverable at this stage is a versioned orthography guide with examples. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

09. Synchronize media

The purpose of this gate is to review utterance start and end, overlapping turns and unmatched intervals. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: Subtitles imply the wrong speaker or event order. The required deliverable at this stage is a media-alignment ledger. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

10. Preserve raw transcription

The purpose of this gate is to retain a source-near text layer separate from normalized and translated views. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: Corrections erase evidence of what the annotator originally heard. The required deliverable at this stage is an immutable or versioned transcription snapshot. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

11. Review uncertainty

The purpose of this gate is to mark unintelligible stretches, alternatives and low confidence without fabricating words. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: Plausible guesses are promoted to asserted quotations. The required deliverable at this stage is an uncertainty log tied to timestamps. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

12. Segment cautiously

The purpose of this gate is to use documented word, morpheme and clitic boundaries subject to community and linguistic review. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: A boundary marker is mistaken for an ordinary punctuation mark. The required deliverable at this stage is a segmentation table with unresolved parses flagged. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

13. Define gloss labels

The purpose of this gate is to document abbreviations, categories and analytic choices consistently. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: Another reviewer interprets PST, PFV or INCL differently. The required deliverable at this stage is a project-specific gloss-key with examples. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

14. Separate literal and free translation

The purpose of this gate is to produce meaning-oriented text without treating it as proof of morphology. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: The free translation becomes the sole surviving witness of an utterance. The required deliverable at this stage is paired source, gloss, translation and annotation links. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

15. Verify referents

The purpose of this gate is to check pronouns, demonstratives, kin terms and omitted arguments against context. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: A target sentence inserts an unverified actor or place. The required deliverable at this stage is a reference-resolution note with source anchors. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

16. Check speech act and stance

The purpose of this gate is to distinguish request, order, report, quotation, irony, joke and uncertainty. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: Polite advice is translated as a command or vice versa. The required deliverable at this stage is a pragmatic-equivalence review. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

17. Confirm numerical content

The purpose of this gate is to preserve dates, counts, duration, sequence and uncertainty in the source. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: A cultural unit or approximate count becomes false precision. The required deliverable at this stage is a structured quantity and provenance register. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

18. Review names and genres

The purpose of this gate is to preserve self-identifications, title conventions, sacred registers and optional pseudonyms. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: A generic translated label overrides speaker-preferred naming. The required deliverable at this stage is an authority-controlled name and genre sheet. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

19. Carry citations and IDs

The purpose of this gate is to keep session, annotation, sense, language and source reference identifiers stable. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: Dataset conversion drops the link to the recording. The required deliverable at this stage is a machine-checkable referential-integrity report. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

20. Run a bilingual meaning review

The purpose of this gate is to have a capable reviewer test the target against audio and approved annotation. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: Grammar is fluent but a negation, participant or evidential is wrong. The required deliverable at this stage is a signed discrepancy-and-resolution record. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

21. Perform community review

The purpose of this gate is to invite relevant authorized speakers and stakeholders to approve public wording and conditions. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: Outside analysis is mistaken for community consent. The required deliverable at this stage is a review log specifying accepted and unresolved comments. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

22. Validate target formats

The purpose of this gate is to test ELAN tiers, exported lexicon, CLDF relationships, rendered captions and search indexes. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: File validates as XML but translated labels point to wrong speakers. The required deliverable at this stage is format validation plus visual inspection. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

23. Approve versioned release

The purpose of this gate is to publish only authorized derivatives and label their source and correction histories. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: A withdrawn clip remains accessible through mirrors or cached previews. The required deliverable at this stage is a release manifest and access-control audit. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

24. Maintain corrections

The purpose of this gate is to define how speakers and custodians can challenge, refine, remove or reclassify records. In a language-documentation project, this decision must be made against actual evidence and with the people who have the relevant authority. It is not enough to insert a translated phrase into a spreadsheet: identify the artifact, its version, the responsible reviewer and the people whose rights may be implicated. When the decision is not yet known, mark it pending rather than choosing a conventional default.

The recognizable failure is: A misleading gloss persists because nobody owns post-publication maintenance. The required deliverable at this stage is a durable change log with correction contacts. A useful handoff states what was checked, which source version it refers to, which exceptions were accepted, and which downstream items must be retested if the decision changes. This gate is complete only when another editor can reconstruct the reasoning without relying on the first editor’s memory. Archive constraints and community permissions take precedence over a publishing deadline.

5. Field atlas: distinctions a translation must preserve rather than flatten

The following atlas is a practical concept reference, not a substitute for a grammar of a particular language. It focuses on the collisions that cause inaccurate subtitles, misleading searchable dictionaries and corrupted interlinear data. Each entry identifies a semantic boundary and a test; the fact that a technical field exists in a tool does not by itself license the proposed value.

Field 01: Language name

Meaning boundary. Keep endonyms, exonyms and source-specific reference labels distinct. The misleading shortcut is: An external catalogue label is assumed to be the community’s preferred name. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Ask the named community reviewer which public form is appropriate for each audience. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Language name contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 02: Language identifier

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Keep code, standard, version date and claimed variety in separate fields. The frequent mistake is: The file’s code is treated as permanent proof of dialect identity. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Reconstruct the original identifier mapping after a registry revision. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Language identifier contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 03: Macrolanguage

Analytic distinction. Distinguish group scope from the actual variety and speaker metadata. Teams lose the distinction when a broad umbrella code is read as an exact description of the recording. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Compare the cited code’s scope with the project’s documented speakers. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Macrolanguage contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 04: Glottocode

Editorial rule. Retain database version and avoid converting a classification into cultural identity. Avoid the assumption that a Glottolog ID is presented as a guarantee of linguistic relationship. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Check the cited Glottolog release and local description independently. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Glottocode contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 05: Variety label

Meaning boundary. Make the claimed level of classification and claimant explicit. The misleading shortcut is: A dialect name is silently rewritten as a language name or the reverse. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Ask whether the speaker’s self-description was preserved. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Variety label contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 06: Endonym

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Keep the original name and relevant transliteration when permission allows. The frequent mistake is: The community’s self-name is replaced by a better-known outsider term. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Verify spelling with an authorized speaker. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Endonym contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 07: Exonym

Analytic distinction. Separate bibliographic discovery aliases from display names. Teams lose the distinction when a historical outsider name is treated as neutral or preferred. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Review the public label for unwanted implications. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Exonym contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 08: Orthography policy

Editorial rule. Preserve original transcript alongside a separately labelled normalized view. Avoid the assumption that spellings are silently regularized to a modern standard. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Compare a disputed line with the waveform and editorial rules. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Orthography policy contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 09: Script conversion

Meaning boundary. Record source script, output script and reversible mapping if applicable. The misleading shortcut is: Transliteration is confused with translation or transcription. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Reconstruct characters from mapping and log ambiguous cases. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Script conversion contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 10: Phonetic transcription

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Keep IPA or other phonetic notation labelled with the analytic purpose. The frequent mistake is: A narrow sound notation is mistaken for standard spelling. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Ask whether readers will mistake a bracketed form for an endorsed orthography. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Phonetic transcription contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 11: Phonemic representation

Analytic distinction. Separate hypothesized contrasts from acoustic observation. Teams lose the distinction when a phoneme-based analysis is presented as raw sound. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Test contrasting words against the source and stated analysis. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Phonemic representation contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 12: Vowel length

Editorial rule. Preserve duration contrast or explicitly mark an editorial limitation. Avoid the assumption that a long vowel is shortened because target readers omit diacritics. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Replay contrasting tokens and test the writing convention. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Vowel length contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 13: Tone

Meaning boundary. Document tone notation separately from punctuation and intonation. The misleading shortcut is: Pitch differences marking lexical or grammatical contrast disappear. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Check minimal or near-minimal contrasts when available. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Tone contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 14: Stress

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Separate lexical stress, focus and orthographic accents. The frequent mistake is: Emphasis is inferred from a written accent mark with another function. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Compare heard prominence to the annotation convention. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Stress contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 15: Pause

Analytic distinction. Distinguish hesitation, turn boundary and editorial segmentation. Teams lose the distinction when silence is removed as if it were irrelevant formatting. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Check the exact audio interval and discourse sequence. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Pause contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 16: Overlap

Editorial rule. Keep parallel time spans and speaker-attribution uncertainty visible. Avoid the assumption that two simultaneous speakers are merged into a single sentence. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Review multi-tier playback around the overlap. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Overlap contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 17: Gesture

Meaning boundary. Keep the demonstrative and annotate gestures as separate evidence. The misleading shortcut is: A deictic utterance is expanded with an invented object or direction. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: View synchronized video before resolving reference. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Gesture contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 18: Speaker identity

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Keep the participant ID and attribution confidence attached to each segment. The frequent mistake is: A turn is assigned to the principal narrator by default. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Listen to nearby turns and consult the role map. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Speaker identity contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 19: Interview prompt

Analytic distinction. Preserve contextual turns that make the answer interpretable. Teams lose the distinction when a question from the interviewer is removed from the record. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Read the exchange rather than isolated lines. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Interview prompt contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 20: Elicitation frame

Editorial rule. Label elicited, naturalistic, reconstructed and illustrative materials. Avoid the assumption that a prompted paradigm example is presented as spontaneous speech. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Check recording notes for the collection method. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Elicitation frame contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 21: Time-alignment

Meaning boundary. Keep segment timing synchronized to the relevant utterance. The misleading shortcut is: A subtitle starts where it looks attractive rather than where speech occurs. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Replay captions with the original media. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Time-alignment contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 22: Transcription uncertainty

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Represent unreadable, inaudible and competing analyses honestly. The frequent mistake is: An unclear word receives a plausible but unmarked guess. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Review the annotated time span and uncertainty status. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Transcription uncertainty contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 23: Normalization

Analytic distinction. Separate normalized, editorial and original layers. Teams lose the distinction when a conservative transcript is replaced by a polished literary version. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Run a diff that identifies every changed form. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Normalization contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 24: Word boundary

Editorial rule. Segment from the source language evidence, not the English syntax. Avoid the assumption that a fluent target phrase is used to infer source word divisions. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Compare boundary hypotheses with distributions across tokens. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Word boundary contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 25: Morpheme boundary

Meaning boundary. Treat the segmentation as a proposed analysis. The misleading shortcut is: Hyphens are inserted to match the gloss a researcher wants. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Test each boundary against paradigms and alternate parses. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Morpheme boundary contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 26: Clitic boundary

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Distinguish clitic notation from affix notation under the stated convention. The frequent mistake is: Equals signs are converted to hyphens during reformatting. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Recheck each gloss-line delimiter against its source line. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Clitic boundary contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 27: Root gloss

Analytic distinction. Assign a context-specific lexical meaning compatible with the parse. Teams lose the distinction when a lexical root is glossed by a target noun despite verbal use. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Check related forms and attested examples. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Root gloss contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 28: Grammatical abbreviation

Editorial rule. Distinguish labels such as NEG, PST or ERG from lexical glosses. Avoid the assumption that a compact label is translated as an ordinary word. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Inspect the project’s abbreviation inventory. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Grammatical abbreviation contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 29: Portmanteau exponent

Meaning boundary. Represent fused features and disclose the analytic convention. The misleading shortcut is: One source form encodes multiple features but the gloss forces one. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Compare the proposed gloss to the actual paradigmatic contrasts. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Portmanteau exponent contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 30: Zero exponent

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Mark zero morphology as analysis rather than heard material. The frequent mistake is: No audible form is invented to complete a symmetrical diagram. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Show where evidence supports the zero category. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Zero exponent contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 31: Allomorphy

Analytic distinction. Distinguish lexeme identity and surface realization. Teams lose the distinction when conditioned forms are treated as unrelated dictionary entries. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Test the distribution across environments. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Allomorphy contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 32: Syncretism

Editorial rule. Keep form identity separate from the feature value inferred in context. Avoid the assumption that identical surface endings are declared to have identical meaning. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Compare paradigms rather than one token. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Syncretism contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 33: Person indexing

Meaning boundary. Distinguish argument realization from agreement marking. The misleading shortcut is: A prefix is rendered as an independent pronoun when it is agreement. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Check whether independent pronouns can co-occur. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Person indexing contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 34: Inclusive and exclusive

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Preserve whether the addressee is included when the grammar encodes it. The frequent mistake is: First-person plural is always translated as an undifferentiated we. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Annotate the relevant speech participants. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Inclusive and exclusive contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 35: Dual number

Analytic distinction. Keep the grammatical value even if free translation is less explicit. Teams lose the distinction when a two-person category is flattened into general plural in a teaching gloss. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Check the participant count and paradigm. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Dual number contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 36: Case marking

Editorial rule. Interpret its construction and grammatical relation. Avoid the assumption that a case suffix is translated as a fixed English preposition. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Test alternative contexts and syntax. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Case marking contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 37: Ergative alignment

Meaning boundary. Keep the case system analysis and alignment distinction visible. The misleading shortcut is: The ergative marker is equated with an English subject marker. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Compare transitive and intransitive clauses. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Ergative alignment contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 38: Inverse marking

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Distinguish direct-inverse systems from voice changes. The frequent mistake is: A marker describing participant hierarchy is glossed as passive voice. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Check whether grammatical roles actually reverse. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Inverse marking contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 39: Obviation

Analytic distinction. Preserve discourse prominence and reference alternatives. Teams lose the distinction when third-person reference-tracking is lost in a pronoun translation. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Track proximate and obviative participants across clauses. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Obviation contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 40: Evidentiality

Editorial rule. Keep hearsay, inference and witnessed evidence distinct when grammatically relevant. Avoid the assumption that a marker for information source is translated as generic certainty. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Ask whose source is encoded and whether a target paraphrase misstates it. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Evidentiality contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 41: Aspect

Meaning boundary. Distinguish aspectual structure from tense. The misleading shortcut is: Completion and viewpoint are replaced by a calendar date. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Test the event boundaries separately from temporal anchoring. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Aspect contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 42: Tense

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Preserve the time relation that the construction actually asserts. The frequent mistake is: A grammatical temporal value is mistaken for a speaker’s prediction. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Check reference time and surrounding sequence. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Tense contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 43: Mood

Analytic distinction. Maintain speaker commitment and possible-world status. Teams lose the distinction when a conditional or counterfactual construction is rewritten as factual. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Test the inferred truth conditions. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Mood contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 44: Negation scope

Editorial rule. Record exactly what the negation takes scope over. Avoid the assumption that a negative marker is moved from a constituent to the whole clause. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Ask which propositions remain affirmed. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Negation scope contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 45: Focus

Meaning boundary. Distinguish information structure from emotional stance. The misleading shortcut is: An emphatic particle is rendered as anger. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Compare alternative focus placements. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Focus contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 46: Topic

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Keep topic, subject and actor analytically separate. The frequent mistake is: A topic marker is equated with the agent of an event. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Test cases with non-agent topics. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Topic contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 47: Classifier

Analytic distinction. Keep agreement or classification separate from lexical identity. Teams lose the distinction when a noun classifier is treated as the name of an object. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Check classifier alternations and counted referents. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Classifier contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 48: Noun class

Editorial rule. Do not infer human gender or sex from a grammatical class. Avoid the assumption that a class prefix is translated as biological gender. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Compare class membership and agreement. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Noun class contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 49: Reduplication

Meaning boundary. Check whether repetition marks intensity, plurality, aspect or another function. The misleading shortcut is: Repeated morphology is deleted as a typo. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Compare repeated and unrepeated tokens. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Reduplication contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 50: Direction and motion

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Preserve path or motion associated with the event. The frequent mistake is: An associated-motion affix is omitted because the main verb is fluent. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Ask who moved and when relative to the main action. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Direction and motion contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 51: Applicative

Analytic distinction. Keep beneficiary, instrument or location relations where evidenced. Teams lose the distinction when an added participant is lost when the target is simplified. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Map argument structure before translating. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Applicative contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 52: Causative

Editorial rule. Distinguish causing from doing. Avoid the assumption that a caused action becomes an action performed by the causer. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Identify both the instigator and immediate actor. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Causative contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 53: Nominalization

Meaning boundary. Preserve whether the action is named, asserted or embedded. The misleading shortcut is: A clause treated as an event noun becomes an asserted finite sentence. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Check the distribution and syntactic role. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Nominalization contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 54: Quotative

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Keep the quoted source and reporting predicate. The frequent mistake is: Reported speech is turned into a narrator’s direct assertion. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Mark speaker boundaries and citation scope. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Quotative contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 55: Kinship term

Analytic distinction. Document the relationship dimensions rather than forcing a familiar kin label. Teams lose the distinction when a culturally specific relationship becomes a generic mother or uncle. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Consult validated kinship diagrams and context. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Kinship term contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 56: Honorific

Editorial rule. Preserve relationship and register without inventing titles. Avoid the assumption that a respectful form is translated as social rank without evidence. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Ask whether status is encoded or merely suggested. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Honorific contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 57: Taboo register

Meaning boundary. Explain use restrictions and publication permission. The misleading shortcut is: An avoidance form is presented as an ordinary synonym. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Consult authorized community conventions. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Taboo register contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 58: Ceremonial speech

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Keep genre, formulaic repetition and authorised editorial choices visible. The frequent mistake is: A ritual performance is rewritten in ordinary conversational register. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Review with performers or delegated custodians. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Ceremonial speech contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 59: Metaphor

Analytic distinction. Preserve imagery where possible and label explanatory adaptation. Teams lose the distinction when a culturally specific image is replaced by a factual paraphrase. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Ask whether the image functions as evidence or rhetoric. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Metaphor contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 60: Speech act

Editorial rule. Keep request, advice, warning and order distinct. Avoid the assumption that an invitation is promoted to a command in the target. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Check who has authority and whether refusal remains possible. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Speech act contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 61: Epistemic stance

Meaning boundary. Preserve degree and source of commitment. The misleading shortcut is: A suggestion or supposition is stated as a witnessed fact. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Highlight every inference added in translation. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Epistemic stance contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 62: Place name

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Preserve permitted names and withheld precision. The frequent mistake is: A hidden or ambiguous location is filled from the researcher’s recollection. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Apply the source’s access restrictions to metadata and captions. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Place name contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 63: Personal name

Analytic distinction. Keep identity policy consistent across transcription, tags and filename. Teams lose the distinction when a pseudonym is accidentally replaced with the person’s real name. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Search released files and thumbnails for leaks. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Personal name contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 64: Narrative genre

Editorial rule. Distinguish memory, tradition, report, fiction and analytic inference. Avoid the assumption that a historical memory is labeled a factual official chronology. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Ask a relevant storyteller or reviewer about genre. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Narrative genre contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 65: Lexical entry

Meaning boundary. Separate form, lexeme, sense, grammatical behavior and usage restrictions. The misleading shortcut is: One surface form is assumed to equal one sense. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Check at least two authentic usage contexts. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Lexical entry contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 66: Borrowing

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Separate contact history from inheritance hypotheses. The frequent mistake is: A borrowed item is classified as a cognate because it looks similar. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Check sound correspondences and external evidence. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Borrowing contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 67: Cognate set

Analytic distinction. Keep confidence and scholarly source attached to cognacy judgments. Teams lose the distinction when a tentative etymological grouping is reported as proven genealogy. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Recheck forms across comparative data. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Cognate set contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 68: CLDF language row

Editorial rule. Use stable row identifiers and versioned references. Avoid the assumption that a display name is used as a relational primary key. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Run integrity checks on every foreign key. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for CLDF language row contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 69: CLDF parameter row

Meaning boundary. Keep the feature scope, coding rule and source references. The misleading shortcut is: A feature definition is replaced with a shorter misleading label. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Audit sample values against parameter definition. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for CLDF parameter row contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 70: CLDF form row

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Retain original orthography and normalization provenance. The frequent mistake is: A cleaned spelling replaces attested forms without marking edits. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Trace one exported form back to its original record. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for CLDF form row contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 71: CLDF example row

Analytic distinction. Maintain example IDs and segmentation/gloss alignments. Teams lose the distinction when a translation is separated from the example and gloss it explains. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Select random rows and reconstruct their full context. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for CLDF example row contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 72: Archive identifier

Editorial rule. Keep stable item identifiers distinct from display strings. Avoid the assumption that a title or filename is substituted for a persistent item reference. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Follow references after renaming and file migration. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Archive identifier contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 73: Provenance

Meaning boundary. Record source speaker, transcriber, gloss editor and translator separately. The misleading shortcut is: A translator’s wording is attributed to the speaker. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Review credit for each derivative layer. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Provenance contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 74: Consent status

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Record the requested uses and current authorization. The frequent mistake is: A general research consent is treated as permission for public video reuse. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Compare permissions to publication channels. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Consent status contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 75: Access category

Analytic distinction. Check item-level access and derived copies independently. Teams lose the distinction when the catalogue is open so the attached audio is assumed to be open. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Verify file permissions by a least-privileged test. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Access category contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 76: Licensing

Editorial rule. Separate licence grants from personal and community conditions. Avoid the assumption that a database-level licence is treated as overriding a sensitive performance. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Review the actual recording and associated documents. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Licensing contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 77: Revision

Meaning boundary. Version the annotation and explain what was changed. The misleading shortcut is: A later correction replaces earlier forms without history. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Reconstruct published excerpts against archived versions. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Revision contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 78: Takedown

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Maintain a derivative inventory and withdrawal path. The frequent mistake is: A withdrawn record survives in transcripts, captions or search thumbnails. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Test every known distribution location. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Takedown contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 79: Search snippet

Analytic distinction. Apply access controls before indexing or preview generation. Teams lose the distinction when an auto-generated excerpt reveals a restricted personal name. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Search with an unauthenticated account. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Search snippet contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 80: Teaching extract

Editorial rule. Treat new audience and format as a permission question. Avoid the assumption that a field note is reused in a lesson beyond the original agreement. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Check proposed teaching use against consent. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Teaching extract contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 81: Machine translation

Meaning boundary. Mark machine-generated suggestions and require human/community review. The misleading shortcut is: A model output is passed off as a speaker-approved translation. A multilingual record must allow later users to reconstruct whether the entry is observed speech, a proposed analysis, a community-approved display term, or an archival assertion. Record the authority of this particular field; a technical term is not an excuse to erase the status of the underlying claim.

Diagnostic question: Verify source alignment and attribution. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Machine translation contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 82: Speech recognition

What can change? User-facing language can be made clearer, but the evidence cannot be silently revised. Keep ASR output provisional and compare to audio. The frequent mistake is: Automatic transcription substitutes standard-language words for unfamiliar speech. Label the chosen output as transcript, gloss, translation or metadata before an editor works on it. Otherwise a natural-looking revision may alter the basis on which the project and later readers rely.

Diagnostic question: Audit out-of-vocabulary tokens and language switches. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Speech recognition contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 83: Public summary

Analytic distinction. Label translation, summary and paraphrase as different products. Teams lose the distinction when a descriptive public abstract is mistaken for a verbatim quote. This can contaminate subsequent wordlists, subtitles, comparative tables or teaching materials. Keep a pointer from the displayed wording to the original segment or source field, and preserve uncertainty rather than smoothing it away.

Diagnostic question: Try reconstructing the underlying quote and fail safely. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Public summary contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

Field 84: Correction request

Editorial rule. Record who reviewed, why and how the disagreement was resolved. Avoid the assumption that a speaker’s correction is dismissed because it conflicts with an older grammar. A reviewer needs the source context and any approved project conventions, not merely a bilingual dictionary. Where the source analysis remains disputed, show that status instead of choosing a confident target equivalent.

Diagnostic question: Expose review status without publishing private correspondence. The verification is more specific than proofreading: identify a source witness, inspect the typed relationship and decide whether a different reader could make a different real-world inference from the proposed rendering. If so, hold that line and record the repair before updating the public output. In particular, test whether another editor could reproduce the decision without access to private information that must not be released.

A useful record for Correction request contains the original value, an approved display form where needed, a source citation or timestamp, responsible reviewer, status, version and permitted audience. When these pieces are separated, one language can be corrected without pretending that every other language, interface or downstream dataset has automatically been updated. This is how a field definition becomes a sustainable translation practice rather than an isolated wording preference.

6. Worked diagnostic laboratory: realistic fictional failures and their repairs

All diagnostics in this section are invented teaching cases. No example is evidence about any named community or language, and no permission for publishing a recording should be inferred from its appearance here. Each case gives a concrete source claim, a plausible but damaging mistake, a meaning-preserving repair and a test before release. The exercises can be performed with a harmless simulated corpus.

Recording, transcription and alignment

The cases in this laboratory group are not interchangeable. Examine the original witness and then ask whether the target changes who spoke, what was asserted, how the analysis was obtained, what other data it connects to, or which uses are authorized. The goal is not to make every text literal; it is to make every editorial change accountable and the resulting communication fit for its intended reader.

Diagnostic 001: A session includes a parent and child

Initial conditions. A session includes a parent and child. Assume the researcher must provide an accessible translation while preserving the evidence trail. Start by naming the kind of source involved: recording, transcript, segmentation, gloss, contextual note or access record. A source type determines which edits are even candidates for approval.

Misleading outcome. The subtitle attributes both voices to the named storyteller. Its apparent fluency hides a shifted claim. An editor should make that shift explicit in a discrepancy log so that the reviewer is assessing a proposition rather than merely preferring one wording over another.

Repair. Keep a separate speaker tier, protecting identities that require restriction. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Replay the exchange with speaker turns visible and public names redacted. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 002: A narrator says a phrase twice before correcting it

Case setting. A narrator says a phrase twice before correcting it. The source contains a specific observation or condition, not a blank invitation to normalize. Record the exact relevant token, time range, data field or rights entry before drafting the final sentence. If that reference cannot be recovered, the work requires source repair before publication.

Tempting shortcut. The translation keeps only the corrected phrase and presents it as an uninterrupted quote. It makes the text superficially simpler while removing a distinction on which attribution, grammatical analysis, participant reference or permitted reuse may depend. A high-quality translation must not be evaluated solely by how effortless it feels to read.

Repair. Retain repetition in transcript and describe a cleaned paraphrase as editorial. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Compare transcript and audio at the corrected interval. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 003: Two speakers overlap during a joking exchange

What the evidence says. Two speakers overlap during a joking exchange. Treat this as an example of the gap between documentary evidence and editorial interpretation. A reader may reasonably expect the translation to reflect the same event, uncertainty and source relationship. The form of expression can change; the speaker’s factual and social commitments must not quietly be replaced.

What goes wrong. A merged subtitle assigns one speaker the other person’s denial. This error propagates because future systems often reuse published captions, lexical databases and excerpts as if they were primary evidence. The immediate fix is important, but the team must also record which derivatives may already contain the misleading value.

Repair. Keep simultaneous tiers with careful segment timing and separate turns. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Replay both channels and map each word to a voice. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 004: A long pause precedes an emotionally difficult answer

Evidence to preserve. A long pause precedes an emotionally difficult answer. Begin with the document or recording, not with the most plausible target phrase. Determine whether the proposed output is a direct translation, analytical annotation, explanatory note or summary, and make that status visible. A claim with no source pointer remains a claim without a usable audit trail.

Failure mode. The target removes the pause and calls the reply immediate. The reader might then infer a different event, grammatical construction, identity or permission status. Smooth wording cannot compensate for that misrepresentation; retaining a deliberate ambiguity can be the most accurate editorial choice.

Repair. Represent timing or annotate the pause rather than adding a reaction. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Inspect waveform and surrounding interaction. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 005: A recording includes a gesture toward an object off camera

Source witness. A recording includes a gesture toward an object off camera. This fact is the case’s entire authorized premise; other details are not available. Before choosing target-language words, make a note of what is observed and what an interpreter merely suspects. A different representation can be readable without claiming to establish the omitted background.

Unsafe transformation. The target names a specific tool never spoken aloud. This is a semantic or evidential change, not an improvement in style. When a translator crosses the boundary between recorded content and an unverified explanation, later archive users may unknowingly cite the explanation as speaker testimony.

Repair. Translate demonstrative reference conservatively and annotate visible gesture. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Have a reviewer inspect the relevant frames. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 006: The transcriber marks one phrase as uncertain

Initial conditions. The transcriber marks one phrase as uncertain. Assume the researcher must provide an accessible translation while preserving the evidence trail. Start by naming the kind of source involved: recording, transcript, segmentation, gloss, contextual note or access record. A source type determines which edits are even candidates for approval.

Misleading outcome. The public transcript silently replaces the marker with a plausible word. Its apparent fluency hides a shifted claim. An editor should make that shift explicit in a discrepancy log so that the reviewer is assessing a proposition rather than merely preferring one wording over another.

Repair. Keep unresolved readings marked and only publish an approved alternative. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Locate the uncertain interval and review proposed readings. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 007: The audio begins after the first syllable of a word

Case setting. The audio begins after the first syllable of a word. The source contains a specific observation or condition, not a blank invitation to normalize. Record the exact relevant token, time range, data field or rights entry before drafting the final sentence. If that reference cannot be recovered, the work requires source repair before publication.

Tempting shortcut. The target restores an entire sentence as if recorded. It makes the text superficially simpler while removing a distinction on which attribution, grammatical analysis, participant reference or permitted reuse may depend. A high-quality translation must not be evaluated solely by how effortless it feels to read.

Repair. Identify missing audio and distinguish reconstruction from observed content. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Compare the first waveform samples to the annotated words. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 008: A speaker switches between two languages in one line

What the evidence says. A speaker switches between two languages in one line. Treat this as an example of the gap between documentary evidence and editorial interpretation. A reader may reasonably expect the translation to reflect the same event, uncertainty and source relationship. The form of expression can change; the speaker’s factual and social commitments must not quietly be replaced.

What goes wrong. The translation tags the entire session as monolingual. This error propagates because future systems often reuse published captions, lexical databases and excerpts as if they were primary evidence. The immediate fix is important, but the team must also record which derivatives may already contain the misleading value.

Repair. Mark spans and avoid assuming each word belongs to a single grammar. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Examine language tags at segment rather than session level. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 009: A participant laughs while disagreeing with a suggestion

Evidence to preserve. A participant laughs while disagreeing with a suggestion. Begin with the document or recording, not with the most plausible target phrase. Determine whether the proposed output is a direct translation, analytical annotation, explanatory note or summary, and make that status visible. A claim with no source pointer remains a claim without a usable audit trail.

Failure mode. The subtitle turns the laugh into explicit agreement. The reader might then infer a different event, grammatical construction, identity or permission status. Smooth wording cannot compensate for that misrepresentation; retaining a deliberate ambiguity can be the most accurate editorial choice.

Repair. Translate actual words and annotate laughter separately when relevant. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Ask whether the transcript supports the claimed agreement. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 010: A recording contains interviewer prompts in the background

Source witness. A recording contains interviewer prompts in the background. This fact is the case’s entire authorized premise; other details are not available. Before choosing target-language words, make a note of what is observed and what an interpreter merely suspects. A different representation can be readable without claiming to establish the omitted background.

Unsafe transformation. The target omits the prompt and makes the reply sound spontaneous. This is a semantic or evidential change, not an improvement in style. When a translator crosses the boundary between recorded content and an unverified explanation, later archive users may unknowingly cite the explanation as speaker testimony.

Repair. Preserve elicitation context or explicitly label a decontextualized excerpt. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Review prompt-answer adjacency before release. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 011: An older cassette is digitized with uneven speed

Initial conditions. An older cassette is digitized with uneven speed. Assume the researcher must provide an accessible translation while preserving the evidence trail. Start by naming the kind of source involved: recording, transcript, segmentation, gloss, contextual note or access record. A source type determines which edits are even candidates for approval.

Misleading outcome. Editors interpret a pitch artifact as lexical tone. Its apparent fluency hides a shifted claim. An editor should make that shift explicit in a discrepancy log so that the reviewer is assessing a proposition rather than merely preferring one wording over another.

Repair. Review recording transfer quality before making tonal conclusions. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Compare repeated tokens and document the digitization condition. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 012: A storyteller begins to quote another participant

Case setting. A storyteller begins to quote another participant. The source contains a specific observation or condition, not a blank invitation to normalize. Record the exact relevant token, time range, data field or rights entry before drafting the final sentence. If that reference cannot be recovered, the work requires source repair before publication.

Tempting shortcut. The translation turns quoted speech into narrator opinion. It makes the text superficially simpler while removing a distinction on which attribution, grammatical analysis, participant reference or permitted reuse may depend. A high-quality translation must not be evaluated solely by how effortless it feels to read.

Repair. Mark quotation spans and assign speech roles with context. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Ask who is represented as holding the proposition. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 013: A mobile recorder captures multiple distant voices

What the evidence says. A mobile recorder captures multiple distant voices. Treat this as an example of the gap between documentary evidence and editorial interpretation. A reader may reasonably expect the translation to reflect the same event, uncertainty and source relationship. The form of expression can change; the speaker’s factual and social commitments must not quietly be replaced.

What goes wrong. Automatic speech recognition inserts familiar words over unintelligible talk. This error propagates because future systems often reuse published captions, lexical databases and excerpts as if they were primary evidence. The immediate fix is important, but the team must also record which derivatives may already contain the misleading value.

Repair. Treat ASR as a draft and confirm by listening and speaker checking. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Audit low-confidence spans without inventing a transcript. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 014: The video and audio files differ by two seconds after export

Evidence to preserve. The video and audio files differ by two seconds after export. Begin with the document or recording, not with the most plausible target phrase. Determine whether the proposed output is a direct translation, analytical annotation, explanatory note or summary, and make that status visible. A claim with no source pointer remains a claim without a usable audit trail.

Failure mode. Subtitles display a warning before the speaker utters it. The reader might then infer a different event, grammatical construction, identity or permission status. Smooth wording cannot compensate for that misrepresentation; retaining a deliberate ambiguity can be the most accurate editorial choice.

Repair. Apply verified synchronization offsets and preserve the original timing. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Watch the exported media rather than only the annotation editor. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 015: A ceremonial speech includes a repeated formula

Source witness. A ceremonial speech includes a repeated formula. This fact is the case’s entire authorized premise; other details are not available. Before choosing target-language words, make a note of what is observed and what an interpreter merely suspects. A different representation can be readable without claiming to establish the omitted background.

Unsafe transformation. The subtitle editor treats repetition as accidental duplication. This is a semantic or evidential change, not an improvement in style. When a translator crosses the boundary between recorded content and an unverified explanation, later archive users may unknowingly cite the explanation as speaker testimony.

Repair. Preserve performance structure or label an abridged version clearly. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Ask authorized reviewers whether repetition carries genre function. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 016: An old translation was created from a different recording take

Initial conditions. An old translation was created from a different recording take. Assume the researcher must provide an accessible translation while preserving the evidence trail. Start by naming the kind of source involved: recording, transcript, segmentation, gloss, contextual note or access record. A source type determines which edits are even candidates for approval.

Misleading outcome. An editor anchors it to the new clip because wording is similar. Its apparent fluency hides a shifted claim. An editor should make that shift explicit in a discrepancy log so that the reviewer is assessing a proposition rather than merely preferring one wording over another.

Repair. Bind the translation to the exact source ID and timestamp. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Check file hashes and excerpt boundaries before publishing. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Orthography, grammar and interlinear analysis

The cases in this laboratory group are not interchangeable. Examine the original witness and then ask whether the target changes who spoke, what was asserted, how the analysis was obtained, what other data it connects to, or which uses are authorized. The goal is not to make every text literal; it is to make every editorial change accountable and the resulting communication fit for its intended reader.

Diagnostic 017: An invented form naa has a long vowel marked by double letters

Case setting. An invented form naa has a long vowel marked by double letters. The source contains a specific observation or condition, not a blank invitation to normalize. Record the exact relevant token, time range, data field or rights entry before drafting the final sentence. If that reference cannot be recovered, the work requires source repair before publication.

Tempting shortcut. The target normalizes it to na and merges two proposed lexemes. It makes the text superficially simpler while removing a distinction on which attribution, grammatical analysis, participant reference or permitted reuse may depend. A high-quality translation must not be evaluated solely by how effortless it feels to read.

Repair. Keep vowel-length distinctions under the chosen orthography. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Compare attested contrast and project spelling policy. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 018: A practical alphabet writes one sound with two characters

What the evidence says. A practical alphabet writes one sound with two characters. Treat this as an example of the gap between documentary evidence and editorial interpretation. A reader may reasonably expect the translation to reflect the same event, uncertainty and source relationship. The form of expression can change; the speaker’s factual and social commitments must not quietly be replaced.

What goes wrong. The analyst segments the digraph as two morphemes. This error propagates because future systems often reuse published captions, lexical databases and excerpts as if they were primary evidence. The immediate fix is important, but the team must also record which derivatives may already contain the misleading value.

Repair. Separate grapheme boundaries from morphological evidence. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Test independent distribution and known spellings. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 019: A fieldworker uses hyphens between morphemes

Evidence to preserve. A fieldworker uses hyphens between morphemes. Begin with the document or recording, not with the most plausible target phrase. Determine whether the proposed output is a direct translation, analytical annotation, explanatory note or summary, and make that status visible. A claim with no source pointer remains a claim without a usable audit trail.

Failure mode. The publishing script removes hyphens during line wrapping. The reader might then infer a different event, grammatical construction, identity or permission status. Smooth wording cannot compensate for that misrepresentation; retaining a deliberate ambiguity can be the most accurate editorial choice.

Repair. Maintain correspondence between segmentation and gloss lines. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Count aligned segments in both lines. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 020: A source uses equals signs for clitics

Source witness. A source uses equals signs for clitics. This fact is the case’s entire authorized premise; other details are not available. Before choosing target-language words, make a note of what is observed and what an interpreter merely suspects. A different representation can be readable without claiming to establish the omitted background.

Unsafe transformation. The export converts clitics into ordinary suffixes. This is a semantic or evidential change, not an improvement in style. When a translator crosses the boundary between recorded content and an unverified explanation, later archive users may unknowingly cite the explanation as speaker testimony.

Repair. Preserve notation and document the analytic difference. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Validate delimiters after conversion. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 021: One affix carries person and number simultaneously

Initial conditions. One affix carries person and number simultaneously. Assume the researcher must provide an accessible translation while preserving the evidence trail. Start by naming the kind of source involved: recording, transcript, segmentation, gloss, contextual note or access record. A source type determines which edits are even candidates for approval.

Misleading outcome. The target creates two audible affixes to match two gloss labels. Its apparent fluency hides a shifted claim. An editor should make that shift explicit in a discrepancy log so that the reviewer is assessing a proposition rather than merely preferring one wording over another.

Repair. Represent fused grammatical values without inventing forms. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Compare source form count to annotation and paradigm. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 022: A negative suffix scopes only over an embedded verb

Case setting. A negative suffix scopes only over an embedded verb. The source contains a specific observation or condition, not a blank invitation to normalize. Record the exact relevant token, time range, data field or rights entry before drafting the final sentence. If that reference cannot be recovered, the work requires source repair before publication.

Tempting shortcut. The target denies the entire event. It makes the text superficially simpler while removing a distinction on which attribution, grammatical analysis, participant reference or permitted reuse may depend. A high-quality translation must not be evaluated solely by how effortless it feels to read.

Repair. Map negation’s syntactic scope before writing free translation. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Identify the exact proposition under negation. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 023: A suffix indicates hearsay rather than speaker doubt

What the evidence says. A suffix indicates hearsay rather than speaker doubt. Treat this as an example of the gap between documentary evidence and editorial interpretation. A reader may reasonably expect the translation to reflect the same event, uncertainty and source relationship. The form of expression can change; the speaker’s factual and social commitments must not quietly be replaced.

What goes wrong. The target states the speaker personally witnessed the event. This error propagates because future systems often reuse published captions, lexical databases and excerpts as if they were primary evidence. The immediate fix is important, but the team must also record which derivatives may already contain the misleading value.

Repair. Convey the information-source distinction in an appropriate rendering. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Ask whether witnesshood is falsely implied. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 024: A verb marker indicates completed viewpoint rather than past time

Evidence to preserve. A verb marker indicates completed viewpoint rather than past time. Begin with the document or recording, not with the most plausible target phrase. Determine whether the proposed output is a direct translation, analytical annotation, explanatory note or summary, and make that status visible. A claim with no source pointer remains a claim without a usable audit trail.

Failure mode. The target inserts yesterday despite no time phrase. The reader might then infer a different event, grammatical construction, identity or permission status. Smooth wording cannot compensate for that misrepresentation; retaining a deliberate ambiguity can be the most accurate editorial choice.

Repair. Maintain aspect and avoid invented temporal anchors. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Check the surrounding timeline. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 025: An inclusive first-person plural addresses speaker and listener

Source witness. An inclusive first-person plural addresses speaker and listener. This fact is the case’s entire authorized premise; other details are not available. Before choosing target-language words, make a note of what is observed and what an interpreter merely suspects. A different representation can be readable without claiming to establish the omitted background.

Unsafe transformation. The target treats we as the speaker’s family excluding the listener. This is a semantic or evidential change, not an improvement in style. When a translator crosses the boundary between recorded content and an unverified explanation, later archive users may unknowingly cite the explanation as speaker testimony.

Repair. Add explanatory context where necessary without changing the participant set. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Write who is included and excluded. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 026: A dual pronoun refers to precisely two people

Initial conditions. A dual pronoun refers to precisely two people. Assume the researcher must provide an accessible translation while preserving the evidence trail. Start by naming the kind of source involved: recording, transcript, segmentation, gloss, contextual note or access record. A source type determines which edits are even candidates for approval.

Misleading outcome. The target treats it as an unlimited plural in a grammatical commentary. Its apparent fluency hides a shifted claim. An editor should make that shift explicit in a discrepancy log so that the reviewer is assessing a proposition rather than merely preferring one wording over another.

Repair. Retain dual in gloss and clarify in free translation if consequential. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Check number in the event and grammar. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 027: A noun-class prefix occurs on inanimate objects

Case setting. A noun-class prefix occurs on inanimate objects. The source contains a specific observation or condition, not a blank invitation to normalize. Record the exact relevant token, time range, data field or rights entry before drafting the final sentence. If that reference cannot be recovered, the work requires source repair before publication.

Tempting shortcut. The translation interprets the class as biological gender. It makes the text superficially simpler while removing a distinction on which attribution, grammatical analysis, participant reference or permitted reuse may depend. A high-quality translation must not be evaluated solely by how effortless it feels to read.

Repair. Label grammatical class rather than inferring identity. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Review lexical membership and agreement. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 028: A source clause is marked hypothetical

What the evidence says. A source clause is marked hypothetical. Treat this as an example of the gap between documentary evidence and editorial interpretation. A reader may reasonably expect the translation to reflect the same event, uncertainty and source relationship. The form of expression can change; the speaker’s factual and social commitments must not quietly be replaced.

What goes wrong. The target rewrites it as a confirmed historical occurrence. This error propagates because future systems often reuse published captions, lexical databases and excerpts as if they were primary evidence. The immediate fix is important, but the team must also record which derivatives may already contain the misleading value.

Repair. Keep mood and nonassertion intact. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Ask whether the source entails the event happened. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 029: An agreement prefix repeats subject features

Evidence to preserve. An agreement prefix repeats subject features. Begin with the document or recording, not with the most plausible target phrase. Determine whether the proposed output is a direct translation, analytical annotation, explanatory note or summary, and make that status visible. A claim with no source pointer remains a claim without a usable audit trail.

Failure mode. The gloss calls it a separate person acting in the clause. The reader might then infer a different event, grammatical construction, identity or permission status. Smooth wording cannot compensate for that misrepresentation; retaining a deliberate ambiguity can be the most accurate editorial choice.

Repair. Distinguish indexing from independent argumenthood. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Check co-occurrence with explicit noun phrases. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 030: A focus particle highlights one object

Source witness. A focus particle highlights one object. This fact is the case’s entire authorized premise; other details are not available. Before choosing target-language words, make a note of what is observed and what an interpreter merely suspects. A different representation can be readable without claiming to establish the omitted background.

Unsafe transformation. The target calls the speaker angry or sarcastic. This is a semantic or evidential change, not an improvement in style. When a translator crosses the boundary between recorded content and an unverified explanation, later archive users may unknowingly cite the explanation as speaker testimony.

Repair. Render focus without unlicensed emotional attribution. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Compare answers to alternative information questions. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 031: A causative construction has a causer and a separate actor

Initial conditions. A causative construction has a causer and a separate actor. Assume the researcher must provide an accessible translation while preserving the evidence trail. Start by naming the kind of source involved: recording, transcript, segmentation, gloss, contextual note or access record. A source type determines which edits are even candidates for approval.

Misleading outcome. The target makes the causer perform the action directly. Its apparent fluency hides a shifted claim. An editor should make that shift explicit in a discrepancy log so that the reviewer is assessing a proposition rather than merely preferring one wording over another.

Repair. Track both argument roles through translation. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Annotate who initiates and who executes. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 032: An applicative marker introduces a beneficiary

Case setting. An applicative marker introduces a beneficiary. The source contains a specific observation or condition, not a blank invitation to normalize. Record the exact relevant token, time range, data field or rights entry before drafting the final sentence. If that reference cannot be recovered, the work requires source repair before publication.

Tempting shortcut. The free translation omits who benefits. It makes the text superficially simpler while removing a distinction on which attribution, grammatical analysis, participant reference or permitted reuse may depend. A high-quality translation must not be evaluated solely by how effortless it feels to read.

Repair. Preserve the added participant where context establishes one. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Create an argument inventory before and after. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 033: A reduplicated form signals repeated action

What the evidence says. A reduplicated form signals repeated action. Treat this as an example of the gap between documentary evidence and editorial interpretation. A reader may reasonably expect the translation to reflect the same event, uncertainty and source relationship. The form of expression can change; the speaker’s factual and social commitments must not quietly be replaced.

What goes wrong. The target treats repetition as transcription noise. This error propagates because future systems often reuse published captions, lexical databases and excerpts as if they were primary evidence. The immediate fix is important, but the team must also record which derivatives may already contain the misleading value.

Repair. Gloss grammatical reduplication and translate iteration. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Compare unreduplicated examples. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 034: A single surface ending has two possible morphological analyses

Evidence to preserve. A single surface ending has two possible morphological analyses. Begin with the document or recording, not with the most plausible target phrase. Determine whether the proposed output is a direct translation, analytical annotation, explanatory note or summary, and make that status visible. A claim with no source pointer remains a claim without a usable audit trail.

Failure mode. The published gloss asserts one without noting ambiguity. The reader might then infer a different event, grammatical construction, identity or permission status. Smooth wording cannot compensate for that misrepresentation; retaining a deliberate ambiguity can be the most accurate editorial choice.

Repair. Preserve competing parses until evidence settles them. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Record what example would distinguish the analyses. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 035: A grammatical tone differs from discourse intonation

Source witness. A grammatical tone differs from discourse intonation. This fact is the case’s entire authorized premise; other details are not available. Before choosing target-language words, make a note of what is observed and what an interpreter merely suspects. A different representation can be readable without claiming to establish the omitted background.

Unsafe transformation. An editor uses a question mark to encode the lexical tone. This is a semantic or evidential change, not an improvement in style. When a translator crosses the boundary between recorded content and an unverified explanation, later archive users may unknowingly cite the explanation as speaker testimony.

Repair. Keep tone notation independent of punctuation. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Compare intonation and lexical patterns. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 036: A root acquires different vowels in related forms

Initial conditions. A root acquires different vowels in related forms. Assume the researcher must provide an accessible translation while preserving the evidence trail. Start by naming the kind of source involved: recording, transcript, segmentation, gloss, contextual note or access record. A source type determines which edits are even candidates for approval.

Misleading outcome. The dictionary imports all allomorphs as unrelated entries. Its apparent fluency hides a shifted claim. An editor should make that shift explicit in a discrepancy log so that the reviewer is assessing a proposition rather than merely preferring one wording over another.

Repair. Connect forms to a lexeme when the analysis supports it. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Check paradigmatic and lexical evidence. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 037: An inverse marker tracks person hierarchy

Case setting. An inverse marker tracks person hierarchy. The source contains a specific observation or condition, not a blank invitation to normalize. Record the exact relevant token, time range, data field or rights entry before drafting the final sentence. If that reference cannot be recovered, the work requires source repair before publication.

Tempting shortcut. The target labels the clause passive and changes agent roles. It makes the text superficially simpler while removing a distinction on which attribution, grammatical analysis, participant reference or permitted reuse may depend. A high-quality translation must not be evaluated solely by how effortless it feels to read.

Repair. Translate event roles and explain the actual grammatical construction. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Test a direct counterpart if available. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 038: A topic-marked noun is not the event’s agent

What the evidence says. A topic-marked noun is not the event’s agent. Treat this as an example of the gap between documentary evidence and editorial interpretation. A reader may reasonably expect the translation to reflect the same event, uncertainty and source relationship. The form of expression can change; the speaker’s factual and social commitments must not quietly be replaced.

What goes wrong. A free translation makes it the person who acted. This error propagates because future systems often reuse published captions, lexical databases and excerpts as if they were primary evidence. The immediate fix is important, but the team must also record which derivatives may already contain the misleading value.

Repair. Separate discourse topic from grammatical participant roles. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Identify event participants independently. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 039: A word has a zero-marked singular under one analysis

Evidence to preserve. A word has a zero-marked singular under one analysis. Begin with the document or recording, not with the most plausible target phrase. Determine whether the proposed output is a direct translation, analytical annotation, explanatory note or summary, and make that status visible. A claim with no source pointer remains a claim without a usable audit trail.

Failure mode. An editor inserts an invented singular suffix to make a table symmetrical. The reader might then infer a different event, grammatical construction, identity or permission status. Smooth wording cannot compensate for that misrepresentation; retaining a deliberate ambiguity can be the most accurate editorial choice.

Repair. Keep zero analysis visible while preserving the surface form. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Compare utterance against the recorded morphological inventory. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 040: An honorific marks respect to an addressee

Source witness. An honorific marks respect to an addressee. This fact is the case’s entire authorized premise; other details are not available. Before choosing target-language words, make a note of what is observed and what an interpreter merely suspects. A different representation can be readable without claiming to establish the omitted background.

Unsafe transformation. The target invents an official noble title. This is a semantic or evidential change, not an improvement in style. When a translator crosses the boundary between recorded content and an unverified explanation, later archive users may unknowingly cite the explanation as speaker testimony.

Repair. Preserve register relationship without adding rank. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Ask whether the source specifies a title. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 041: A nominalized clause names an activity

Initial conditions. A nominalized clause names an activity. Assume the researcher must provide an accessible translation while preserving the evidence trail. Start by naming the kind of source involved: recording, transcript, segmentation, gloss, contextual note or access record. A source type determines which edits are even candidates for approval.

Misleading outcome. The translation turns it into a factual complete sentence. Its apparent fluency hides a shifted claim. An editor should make that shift explicit in a discrepancy log so that the reviewer is assessing a proposition rather than merely preferring one wording over another.

Repair. Keep embedded status and avoid new truth claims. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Check clause distribution in the source. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Meaning, culture and discourse

The cases in this laboratory group are not interchangeable. Examine the original witness and then ask whether the target changes who spoke, what was asserted, how the analysis was obtained, what other data it connects to, or which uses are authorized. The goal is not to make every text literal; it is to make every editorial change accountable and the resulting communication fit for its intended reader.

Diagnostic 042: A kinship term covers relatives distinguished in English

Case setting. A kinship term covers relatives distinguished in English. The source contains a specific observation or condition, not a blank invitation to normalize. Record the exact relevant token, time range, data field or rights entry before drafting the final sentence. If that reference cannot be recovered, the work requires source repair before publication.

Tempting shortcut. The target selects uncle without contextual evidence. It makes the text superficially simpler while removing a distinction on which attribution, grammatical analysis, participant reference or permitted reuse may depend. A high-quality translation must not be evaluated solely by how effortless it feels to read.

Repair. Use a documented relationship explanation or preserve local term. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Confirm genealogy and situation with an authorized expert. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 043: A storyteller makes a humorous understatement

What the evidence says. A storyteller makes a humorous understatement. Treat this as an example of the gap between documentary evidence and editorial interpretation. A reader may reasonably expect the translation to reflect the same event, uncertainty and source relationship. The form of expression can change; the speaker’s factual and social commitments must not quietly be replaced.

What goes wrong. The target reports a literal impossibility as historical fact. This error propagates because future systems often reuse published captions, lexical databases and excerpts as if they were primary evidence. The immediate fix is important, but the team must also record which derivatives may already contain the misleading value.

Repair. Keep the joke’s communicative role while indicating any explanatory adaptation. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Have a reviewer identify the intended rhetorical force. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 044: A narrative invokes a culturally restricted personal name

Evidence to preserve. A narrative invokes a culturally restricted personal name. Begin with the document or recording, not with the most plausible target phrase. Determine whether the proposed output is a direct translation, analytical annotation, explanatory note or summary, and make that status visible. A claim with no source pointer remains a claim without a usable audit trail.

Failure mode. A searchable translation expands the name everywhere. The reader might then infer a different event, grammatical construction, identity or permission status. Smooth wording cannot compensate for that misrepresentation; retaining a deliberate ambiguity can be the most accurate editorial choice.

Repair. Follow name-use permissions and approved public wording. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Test full-text search, captions and metadata. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 045: A proverb uses an animal image familiar to the audience

Source witness. A proverb uses an animal image familiar to the audience. This fact is the case’s entire authorized premise; other details are not available. Before choosing target-language words, make a note of what is observed and what an interpreter merely suspects. A different representation can be readable without claiming to establish the omitted background.

Unsafe transformation. A translation replaces it with an unrelated business metaphor. This is a semantic or evidential change, not an improvement in style. When a translator crosses the boundary between recorded content and an unverified explanation, later archive users may unknowingly cite the explanation as speaker testimony.

Repair. Preserve semantic function and imagery or label an adaptation. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Compare metaphorical implications with a speaker’s explanation. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 046: An origin story includes more than one accepted retelling

Initial conditions. An origin story includes more than one accepted retelling. Assume the researcher must provide an accessible translation while preserving the evidence trail. Start by naming the kind of source involved: recording, transcript, segmentation, gloss, contextual note or access record. A source type determines which edits are even candidates for approval.

Misleading outcome. The target declares one version the sole authentic history. Its apparent fluency hides a shifted claim. An editor should make that shift explicit in a discrepancy log so that the reviewer is assessing a proposition rather than merely preferring one wording over another.

Repair. Attribute the narrated version to its source and context. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Check how version and narrator are identified. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 047: A community speaker gives a tentative identification

Case setting. A community speaker gives a tentative identification. The source contains a specific observation or condition, not a blank invitation to normalize. Record the exact relevant token, time range, data field or rights entry before drafting the final sentence. If that reference cannot be recovered, the work requires source repair before publication.

Tempting shortcut. The museum-style caption asserts certainty without qualification. It makes the text superficially simpler while removing a distinction on which attribution, grammatical analysis, participant reference or permitted reuse may depend. A high-quality translation must not be evaluated solely by how effortless it feels to read.

Repair. Keep tentative language and reviewer notes. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Highlight every added certainty word in target. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 048: A reported saying is introduced by an evidential phrase

What the evidence says. A reported saying is introduced by an evidential phrase. Treat this as an example of the gap between documentary evidence and editorial interpretation. A reader may reasonably expect the translation to reflect the same event, uncertainty and source relationship. The form of expression can change; the speaker’s factual and social commitments must not quietly be replaced.

What goes wrong. The target presents it as the recorder’s direct observation. This error propagates because future systems often reuse published captions, lexical databases and excerpts as if they were primary evidence. The immediate fix is important, but the team must also record which derivatives may already contain the misleading value.

Repair. Keep chain of speakers and evidence. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Map who said what and who witnessed it. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 049: A speaker uses a polite indirect request

Evidence to preserve. A speaker uses a polite indirect request. Begin with the document or recording, not with the most plausible target phrase. Determine whether the proposed output is a direct translation, analytical annotation, explanatory note or summary, and make that status visible. A claim with no source pointer remains a claim without a usable audit trail.

Failure mode. A target-language imperative suggests coercive authority. The reader might then infer a different event, grammatical construction, identity or permission status. Smooth wording cannot compensate for that misrepresentation; retaining a deliberate ambiguity can be the most accurate editorial choice.

Repair. Translate intended social force with context-sensitive wording. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Have a second reviewer compare politeness and refusability. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 050: A conversation uses teasing rather than insult

Source witness. A conversation uses teasing rather than insult. This fact is the case’s entire authorized premise; other details are not available. Before choosing target-language words, make a note of what is observed and what an interpreter merely suspects. A different representation can be readable without claiming to establish the omitted background.

Unsafe transformation. The translation labels it an abusive statement. This is a semantic or evidential change, not an improvement in style. When a translator crosses the boundary between recorded content and an unverified explanation, later archive users may unknowingly cite the explanation as speaker testimony.

Repair. Preserve interactional context rather than one dictionary meaning. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Read preceding responses and participant relationship. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 051: A culturally specific plant term covers several botanical taxa

Initial conditions. A culturally specific plant term covers several botanical taxa. Assume the researcher must provide an accessible translation while preserving the evidence trail. Start by naming the kind of source involved: recording, transcript, segmentation, gloss, contextual note or access record. A source type determines which edits are even candidates for approval.

Misleading outcome. A translator inserts one scientific species name without evidence. Its apparent fluency hides a shifted claim. An editor should make that shift explicit in a discrepancy log so that the reviewer is assessing a proposition rather than merely preferring one wording over another.

Repair. Preserve folk category and distinguish botanical identification. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Check specimen or authorized expert evidence. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 052: A traditional measurement is approximate and context-specific

Case setting. A traditional measurement is approximate and context-specific. The source contains a specific observation or condition, not a blank invitation to normalize. Record the exact relevant token, time range, data field or rights entry before drafting the final sentence. If that reference cannot be recovered, the work requires source repair before publication.

Tempting shortcut. The target turns it into an exact decimal quantity. It makes the text superficially simpler while removing a distinction on which attribution, grammatical analysis, participant reference or permitted reuse may depend. A high-quality translation must not be evaluated solely by how effortless it feels to read.

Repair. Keep approximation and explain any conversion as estimate. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Record method and uncertainty of proposed conversion. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 053: The story employs intentional ambiguity about the actor

What the evidence says. The story employs intentional ambiguity about the actor. Treat this as an example of the gap between documentary evidence and editorial interpretation. A reader may reasonably expect the translation to reflect the same event, uncertainty and source relationship. The form of expression can change; the speaker’s factual and social commitments must not quietly be replaced.

What goes wrong. The translator assigns the action to a named person. This error propagates because future systems often reuse published captions, lexical databases and excerpts as if they were primary evidence. The immediate fix is important, but the team must also record which derivatives may already contain the misleading value.

Repair. Keep unresolved reference unless evidence disambiguates it. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Ask whether other participants remain possible. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 054: A speaker corrects their own earlier account

Evidence to preserve. A speaker corrects their own earlier account. Begin with the document or recording, not with the most plausible target phrase. Determine whether the proposed output is a direct translation, analytical annotation, explanatory note or summary, and make that status visible. A claim with no source pointer remains a claim without a usable audit trail.

Failure mode. The public summary retains only the original assertion. The reader might then infer a different event, grammatical construction, identity or permission status. Smooth wording cannot compensate for that misrepresentation; retaining a deliberate ambiguity can be the most accurate editorial choice.

Repair. Represent self-correction as part of the evidence. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Compare event order across the entire passage. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 055: An elder says a practice used to occur in the past

Source witness. An elder says a practice used to occur in the past. This fact is the case’s entire authorized premise; other details are not available. Before choosing target-language words, make a note of what is observed and what an interpreter merely suspects. A different representation can be readable without claiming to establish the omitted background.

Unsafe transformation. The translation says it must be performed today. This is a semantic or evidential change, not an improvement in style. When a translator crosses the boundary between recorded content and an unverified explanation, later archive users may unknowingly cite the explanation as speaker testimony.

Repair. Distinguish description from current instruction. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Identify tense, modality and intended audience. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 056: A lullaby includes sound-symbolic syllables

Initial conditions. A lullaby includes sound-symbolic syllables. Assume the researcher must provide an accessible translation while preserving the evidence trail. Start by naming the kind of source involved: recording, transcript, segmentation, gloss, contextual note or access record. A source type determines which edits are even candidates for approval.

Misleading outcome. The target replaces them with invented semantic words. Its apparent fluency hides a shifted claim. An editor should make that shift explicit in a discrepancy log so that the reviewer is assessing a proposition rather than merely preferring one wording over another.

Repair. Keep performance features and label nonlexical syllables appropriately. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Review the song with performer permission. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 057: A gesture changes the reference of here

Case setting. A gesture changes the reference of here. The source contains a specific observation or condition, not a blank invitation to normalize. Record the exact relevant token, time range, data field or rights entry before drafting the final sentence. If that reference cannot be recovered, the work requires source repair before publication.

Tempting shortcut. The translation inserts an exact geographic coordinate. It makes the text superficially simpler while removing a distinction on which attribution, grammatical analysis, participant reference or permitted reuse may depend. A high-quality translation must not be evaluated solely by how effortless it feels to read.

Repair. Preserve deictic meaning without publishing a guessed location. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Check consent and evidence for place identification. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Archive metadata, consent and interoperability

The cases in this laboratory group are not interchangeable. Examine the original witness and then ask whether the target changes who spoke, what was asserted, how the analysis was obtained, what other data it connects to, or which uses are authorized. The goal is not to make every text literal; it is to make every editorial change accountable and the resulting communication fit for its intended reader.

Diagnostic 058: A file is deposited under restricted-access conditions

What the evidence says. A file is deposited under restricted-access conditions. Treat this as an example of the gap between documentary evidence and editorial interpretation. A reader may reasonably expect the translation to reflect the same event, uncertainty and source relationship. The form of expression can change; the speaker’s factual and social commitments must not quietly be replaced.

What goes wrong. A translator copies the transcript into an unrestricted blog post. This error propagates because future systems often reuse published captions, lexical databases and excerpts as if they were primary evidence. The immediate fix is important, but the team must also record which derivatives may already contain the misleading value.

Repair. Request authorized public reuse or provide only permitted cataloguing metadata. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Check the release against item-level access rules. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 059: A speaker consented to listening in a classroom

Evidence to preserve. A speaker consented to listening in a classroom. Begin with the document or recording, not with the most plausible target phrase. Determine whether the proposed output is a direct translation, analytical annotation, explanatory note or summary, and make that status visible. A claim with no source pointer remains a claim without a usable audit trail.

Failure mode. A publisher assumes permission for open commercial redistribution. The reader might then infer a different event, grammatical construction, identity or permission status. Smooth wording cannot compensate for that misrepresentation; retaining a deliberate ambiguity can be the most accurate editorial choice.

Repair. Separate educational access from wider distribution rights. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Read the consent record for the proposed audience. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 060: A collection’s catalogue is openly searchable

Source witness. A collection’s catalogue is openly searchable. This fact is the case’s entire authorized premise; other details are not available. Before choosing target-language words, make a note of what is observed and what an interpreter merely suspects. A different representation can be readable without claiming to establish the omitted background.

Unsafe transformation. An editor infers every attached audio file is public. This is a semantic or evidential change, not an improvement in style. When a translator crosses the boundary between recorded content and an unverified explanation, later archive users may unknowingly cite the explanation as speaker testimony.

Repair. Respect item-specific file permissions and derivative access rules. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Open links as an anonymous viewer and inspect restrictions. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 061: A researcher uses a pseudonym in public materials

Initial conditions. A researcher uses a pseudonym in public materials. Assume the researcher must provide an accessible translation while preserving the evidence trail. Start by naming the kind of source involved: recording, transcript, segmentation, gloss, contextual note or access record. A source type determines which edits are even candidates for approval.

Misleading outcome. A subtitle track exposes the legal name. Its apparent fluency hides a shifted claim. An editor should make that shift explicit in a discrepancy log so that the reviewer is assessing a proposition rather than merely preferring one wording over another.

Repair. Carry identity policy through all outputs including filenames. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Scan captions, metadata and thumbnails. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 062: Two recordings share an informal session name

Case setting. Two recordings share an informal session name. The source contains a specific observation or condition, not a blank invitation to normalize. Record the exact relevant token, time range, data field or rights entry before drafting the final sentence. If that reference cannot be recovered, the work requires source repair before publication.

Tempting shortcut. The translator attaches a caption to the wrong digital object. It makes the text superficially simpler while removing a distinction on which attribution, grammatical analysis, participant reference or permitted reuse may depend. A high-quality translation must not be evaluated solely by how effortless it feels to read.

Repair. Use persistent identifiers plus versioned time references. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Verify identifiers across export and archive. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 063: An ISO language identifier has been revised in a newer registry

What the evidence says. An ISO language identifier has been revised in a newer registry. Treat this as an example of the gap between documentary evidence and editorial interpretation. A reader may reasonably expect the translation to reflect the same event, uncertainty and source relationship. The form of expression can change; the speaker’s factual and social commitments must not quietly be replaced.

What goes wrong. The export overwrites the historical code without mapping. This error propagates because future systems often reuse published captions, lexical databases and excerpts as if they were primary evidence. The immediate fix is important, but the team must also record which derivatives may already contain the misleading value.

Repair. Store code-set version and explicit old-new relationships. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Reconstruct the original catalogue join. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 064: A Glottolog classification changes in a newer release

Evidence to preserve. A Glottolog classification changes in a newer release. Begin with the document or recording, not with the most plausible target phrase. Determine whether the proposed output is a direct translation, analytical annotation, explanatory note or summary, and make that status visible. A claim with no source pointer remains a claim without a usable audit trail.

Failure mode. The project deletes the earlier lineage annotation. The reader might then infer a different event, grammatical construction, identity or permission status. Smooth wording cannot compensate for that misrepresentation; retaining a deliberate ambiguity can be the most accurate editorial choice.

Repair. Keep cited release and note changed classification. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Trace all previous published references. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 065: A CLDF Wordlist has LanguageTable and FormTable references

Source witness. A CLDF Wordlist has LanguageTable and FormTable references. This fact is the case’s entire authorized premise; other details are not available. Before choosing target-language words, make a note of what is observed and what an interpreter merely suspects. A different representation can be readable without claiming to establish the omitted background.

Unsafe transformation. A translated language display name replaces the key used by FormTable. This is a semantic or evidential change, not an improvement in style. When a translator crosses the boundary between recorded content and an unverified explanation, later archive users may unknowingly cite the explanation as speaker testimony.

Repair. Keep the stable ID and localize only suitable display fields. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Run referential integrity on imported files. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 066: A cognate link is a provisional scholarly hypothesis

Initial conditions. A cognate link is a provisional scholarly hypothesis. Assume the researcher must provide an accessible translation while preserving the evidence trail. Start by naming the kind of source involved: recording, transcript, segmentation, gloss, contextual note or access record. A source type determines which edits are even candidates for approval.

Misleading outcome. The public dictionary labels the words identical in origin. Its apparent fluency hides a shifted claim. An editor should make that shift explicit in a discrepancy log so that the reviewer is assessing a proposition rather than merely preferring one wording over another.

Repair. Keep cognacy status and attribution distinct from demonstrated meaning. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Review supporting phonological comparisons. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 067: A lexical database stores multiple senses for one headword

Case setting. A lexical database stores multiple senses for one headword. The source contains a specific observation or condition, not a blank invitation to normalize. Record the exact relevant token, time range, data field or rights entry before drafting the final sentence. If that reference cannot be recovered, the work requires source repair before publication.

Tempting shortcut. The output collapses senses into one target word. It makes the text superficially simpler while removing a distinction on which attribution, grammatical analysis, participant reference or permitted reuse may depend. A high-quality translation must not be evaluated solely by how effortless it feels to read.

Repair. Retain sense IDs and usage contexts. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Trace example sentences to the correct sense. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 068: An annotation exports from ELAN as a separate file

What the evidence says. An annotation exports from ELAN as a separate file. Treat this as an example of the gap between documentary evidence and editorial interpretation. A reader may reasonably expect the translation to reflect the same event, uncertainty and source relationship. The form of expression can change; the speaker’s factual and social commitments must not quietly be replaced.

What goes wrong. A renamed media file breaks the reference to the recording. This error propagates because future systems often reuse published captions, lexical databases and excerpts as if they were primary evidence. The immediate fix is important, but the team must also record which derivatives may already contain the misleading value.

Repair. Preserve and test the relation between annotation and media. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Open the exported project and replay several intervals. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 069: An archive includes video of a child speaking

Evidence to preserve. An archive includes video of a child speaking. Begin with the document or recording, not with the most plausible target phrase. Determine whether the proposed output is a direct translation, analytical annotation, explanatory note or summary, and make that status visible. A claim with no source pointer remains a claim without a usable audit trail.

Failure mode. An editor assumes adult participant consent covers the child. The reader might then infer a different event, grammatical construction, identity or permission status. Smooth wording cannot compensate for that misrepresentation; retaining a deliberate ambiguity can be the most accurate editorial choice.

Repair. Check the specific approvals, privacy safeguards and governing rules. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Request suitable authorization or withhold the extract. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 070: A recorded location is considered sensitive

Source witness. A recorded location is considered sensitive. This fact is the case’s entire authorized premise; other details are not available. Before choosing target-language words, make a note of what is observed and what an interpreter merely suspects. A different representation can be readable without claiming to establish the omitted background.

Unsafe transformation. The map caption converts descriptive context into precise coordinates. This is a semantic or evidential change, not an improvement in style. When a translator crosses the boundary between recorded content and an unverified explanation, later archive users may unknowingly cite the explanation as speaker testimony.

Repair. Respect location suppression across structured and prose metadata. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Audit search and geospatial fields for leaks. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 071: An authorial credit identifies only the outside researcher

Initial conditions. An authorial credit identifies only the outside researcher. Assume the researcher must provide an accessible translation while preserving the evidence trail. Start by naming the kind of source involved: recording, transcript, segmentation, gloss, contextual note or access record. A source type determines which edits are even candidates for approval.

Misleading outcome. The translation erases speakers and local annotators’ contribution. Its apparent fluency hides a shifted claim. An editor should make that shift explicit in a discrepancy log so that the reviewer is assessing a proposition rather than merely preferring one wording over another.

Repair. Use permitted contributor roles and community-preferred credit. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Check provenance against the actual collaborators. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 072: A community editor asks to revise a public translation

Case setting. A community editor asks to revise a public translation. The source contains a specific observation or condition, not a blank invitation to normalize. Record the exact relevant token, time range, data field or rights entry before drafting the final sentence. If that reference cannot be recovered, the work requires source repair before publication.

Tempting shortcut. The archival system keeps an old cached extract uncorrected. It makes the text superficially simpler while removing a distinction on which attribution, grammatical analysis, participant reference or permitted reuse may depend. A high-quality translation must not be evaluated solely by how effortless it feels to read.

Repair. Record correction, notify derivative holders and update outputs. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Follow revision identifiers through distribution copies. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 073: A depositor withdraws permission for one public clip

What the evidence says. A depositor withdraws permission for one public clip. Treat this as an example of the gap between documentary evidence and editorial interpretation. A reader may reasonably expect the translation to reflect the same event, uncertainty and source relationship. The form of expression can change; the speaker’s factual and social commitments must not quietly be replaced.

What goes wrong. A new teaching compilation still serves an older download. This error propagates because future systems often reuse published captions, lexical databases and excerpts as if they were primary evidence. The immediate fix is important, but the team must also record which derivatives may already contain the misleading value.

Repair. Inventory derived files and apply appropriate restriction and takedown procedures. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Test public endpoints after remediation. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 074: A translation is generated using machine-learning tools

Evidence to preserve. A translation is generated using machine-learning tools. Begin with the document or recording, not with the most plausible target phrase. Determine whether the proposed output is a direct translation, analytical annotation, explanatory note or summary, and make that status visible. A claim with no source pointer remains a claim without a usable audit trail.

Failure mode. The article implies a native speaker endorsed it. The reader might then infer a different event, grammatical construction, identity or permission status. Smooth wording cannot compensate for that misrepresentation; retaining a deliberate ambiguity can be the most accurate editorial choice.

Repair. Declare workflow and require qualified human review. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Compare claimed approval with recorded reviewer action. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 075: An outsider asks for a restricted ceremony transcript

Source witness. An outsider asks for a restricted ceremony transcript. This fact is the case’s entire authorized premise; other details are not available. Before choosing target-language words, make a note of what is observed and what an interpreter merely suspects. A different representation can be readable without claiming to establish the omitted background.

Unsafe transformation. An automatic public translation service processes the upload. This is a semantic or evidential change, not an improvement in style. When a translator crosses the boundary between recorded content and an unverified explanation, later archive users may unknowingly cite the explanation as speaker testimony.

Repair. Stop disclosure and route the request to the authorized custodian. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Check data handling and permission before processing. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 076: A collaborative dictionary exports publicly but a term is marked private

Initial conditions. A collaborative dictionary exports publicly but a term is marked private. Assume the researcher must provide an accessible translation while preserving the evidence trail. Start by naming the kind of source involved: recording, transcript, segmentation, gloss, contextual note or access record. A source type determines which edits are even candidates for approval.

Misleading outcome. The translated search index includes the restricted entry. Its apparent fluency hides a shifted claim. An editor should make that shift explicit in a discrepancy log so that the reviewer is assessing a proposition rather than merely preferring one wording over another.

Repair. Apply record-level disclosure policy to every language and index. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Search for forbidden terms from an unauthenticated account. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

Diagnostic 077: A later archive migration changes all file paths

Case setting. A later archive migration changes all file paths. The source contains a specific observation or condition, not a blank invitation to normalize. Record the exact relevant token, time range, data field or rights entry before drafting the final sentence. If that reference cannot be recovered, the work requires source repair before publication.

Tempting shortcut. The article links to filenames rather than persistent item identifiers. It makes the text superficially simpler while removing a distinction on which attribution, grammatical analysis, participant reference or permitted reuse may depend. A high-quality translation must not be evaluated solely by how effortless it feels to read.

Repair. Use durable record IDs and maintain redirect or resolver mappings. The revised text or metadata must preserve the documented relationship and make any additional explanatory material distinguishable from the source. Do not silently insert certainty, authority, words or actors that the witness does not support. The appropriate reviewer may be a speaker, linguistic analyst, archivist, permissions custodian or more than one of these people; record who actually approved the correction.

Release test. Revalidate every outbound reference after migration. Compare the candidate output with the relevant primary evidence, then ask a second reader to paraphrase its implication without access to the translator’s private assumptions. If that reader reaches a claim the source would not warrant, fail the test and revise. Preserve a brief issue record showing the source version, rejected rendering, corrected rendering, rationale and affected downstream files. This same test can be repeated whenever language tags, formats, access settings or textual interpretations change.

7. Transfer workshops: use the method beyond one file format or one language pair

Workshop 01: Audio narrative to subtitles

Setup. Start with a three-minute fictional story containing a correction and a repeated place name. Work product. Make two subtitle tracks, one source-near and one reader-facing, and retain the same speaker/time identity. The learner should deliver both a reader-facing artifact and a compact evidence ledger: original identifier, revised text, type of transformation, reviewer, access condition and version. The point is to practice communication and accountability together, not to create a decorative bilingual copy.

Critical guardrail. Do not mistake normalization for evidence or reveal restricted locations. Ask two separate questions during review. First, can the reader recover the intended linguistic or documentary meaning? Second, is the reader entitled to receive this particular material in this particular form? A correct translation passes the first question; an ethical, usable release must pass both. Store a sample of rejected decisions to make the method teachable and repeatable.

Workshop 02: ELAN annotation to prose exhibit label

Setup. Use a simulated EAF file with speech, gesture and translation tiers. Work product. Write an exhibition label citing the licensed excerpt while preserving the original roles. The learner should deliver both a reader-facing artifact and a compact evidence ledger: original identifier, revised text, type of transformation, reviewer, access condition and version. The point is to practice communication and accountability together, not to create a decorative bilingual copy.

Critical guardrail. Keep visual description separate from direct quotation. Ask two separate questions during review. First, can the reader recover the intended linguistic or documentary meaning? Second, is the reader entitled to receive this particular material in this particular form? A correct translation passes the first question; an ethical, usable release must pass both. Store a sample of rejected decisions to make the method teachable and repeatable.

Workshop 03: FLEx lexicon to learners’ dictionary

Setup. Take a synthetic lexeme with two senses and one restricted example. Work product. Produce sense-level entries and replace only the restricted example under authorization. The learner should deliver both a reader-facing artifact and a compact evidence ledger: original identifier, revised text, type of transformation, reviewer, access condition and version. The point is to practice communication and accountability together, not to create a decorative bilingual copy.

Critical guardrail. Never infer that one spelling means one sense. Ask two separate questions during review. First, can the reader recover the intended linguistic or documentary meaning? Second, is the reader entitled to receive this particular material in this particular form? A correct translation passes the first question; an ethical, usable release must pass both. Store a sample of rejected decisions to make the method teachable and repeatable.

Workshop 04: Morphological example to classroom worksheet

Setup. Use a made-up segmented form with a clitic and a fused person-number ending. Work product. Create an accessible exercise and an explicit analysis key. The learner should deliver both a reader-facing artifact and a compact evidence ledger: original identifier, revised text, type of transformation, reviewer, access condition and version. The point is to practice communication and accountability together, not to create a decorative bilingual copy.

Critical guardrail. Avoid inventing extra phonemes to match a tidy gloss. Ask two separate questions during review. First, can the reader recover the intended linguistic or documentary meaning? Second, is the reader entitled to receive this particular material in this particular form? A correct translation passes the first question; an ethical, usable release must pass both. Store a sample of rejected decisions to make the method teachable and repeatable.

Workshop 05: Historical orthography to modern searchable spellings

Setup. Use two approved forms of the same synthetic token recorded decades apart. Work product. Build a reversible form-index and mark modernization as derived. The learner should deliver both a reader-facing artifact and a compact evidence ledger: original identifier, revised text, type of transformation, reviewer, access condition and version. The point is to practice communication and accountability together, not to create a decorative bilingual copy.

Critical guardrail. Preserve original publication spelling. Ask two separate questions during review. First, can the reader recover the intended linguistic or documentary meaning? Second, is the reader entitled to receive this particular material in this particular form? A correct translation passes the first question; an ethical, usable release must pass both. Store a sample of rejected decisions to make the method teachable and repeatable.

Workshop 06: Interviewer-led elicitation to a public clip

Setup. Use an invented question-answer pair where the answer is just yes. Work product. Translate with enough prompt context to explain yes. The learner should deliver both a reader-facing artifact and a compact evidence ledger: original identifier, revised text, type of transformation, reviewer, access condition and version. The point is to practice communication and accountability together, not to create a decorative bilingual copy.

Critical guardrail. Label elicitation rather than staging it as spontaneous testimony. Ask two separate questions during review. First, can the reader recover the intended linguistic or documentary meaning? Second, is the reader entitled to receive this particular material in this particular form? A correct translation passes the first question; an ethical, usable release must pass both. Store a sample of rejected decisions to make the method teachable and repeatable.

Workshop 07: Mixed-language conversation to bilingual subtitles

Setup. Use a fabricated dialogue with three code switches. Work product. Tag spans, not merely the whole file, and preserve speaker roles. The learner should deliver both a reader-facing artifact and a compact evidence ledger: original identifier, revised text, type of transformation, reviewer, access condition and version. The point is to practice communication and accountability together, not to create a decorative bilingual copy.

Critical guardrail. Avoid forcing every token into the dominant language. Ask two separate questions during review. First, can the reader recover the intended linguistic or documentary meaning? Second, is the reader entitled to receive this particular material in this particular form? A correct translation passes the first question; an ethical, usable release must pass both. Store a sample of rejected decisions to make the method teachable and repeatable.

Workshop 08: Community correction to archive erratum

Setup. Assume a speaker rejects an early mistranslation of a place relationship. Work product. Record approved correction and locate all public derivatives. The learner should deliver both a reader-facing artifact and a compact evidence ledger: original identifier, revised text, type of transformation, reviewer, access condition and version. The point is to practice communication and accountability together, not to create a decorative bilingual copy.

Critical guardrail. Update search snippets and captions as well as article text. Ask two separate questions during review. First, can the reader recover the intended linguistic or documentary meaning? Second, is the reader entitled to receive this particular material in this particular form? A correct translation passes the first question; an ethical, usable release must pass both. Store a sample of rejected decisions to make the method teachable and repeatable.

Workshop 09: CLDF Wordlist import

Setup. Create five language rows, forms, examples and references with stable IDs. Work product. Translate readable descriptions while retaining relational keys. The learner should deliver both a reader-facing artifact and a compact evidence ledger: original identifier, revised text, type of transformation, reviewer, access condition and version. The point is to practice communication and accountability together, not to create a decorative bilingual copy.

Critical guardrail. Run foreign-key and Unicode checks. Ask two separate questions during review. First, can the reader recover the intended linguistic or documentary meaning? Second, is the reader entitled to receive this particular material in this particular form? A correct translation passes the first question; an ethical, usable release must pass both. Store a sample of rejected decisions to make the method teachable and repeatable.

Workshop 10: Glottolog reclassification note

Setup. Use a fictional project’s citation to an older classification release. Work product. Record the old citation and newer mapping without erasing the project history. The learner should deliver both a reader-facing artifact and a compact evidence ledger: original identifier, revised text, type of transformation, reviewer, access condition and version. The point is to practice communication and accountability together, not to create a decorative bilingual copy.

Critical guardrail. Do not rewrite the past dataset as if it used the new edition. Ask two separate questions during review. First, can the reader recover the intended linguistic or documentary meaning? Second, is the reader entitled to receive this particular material in this particular form? A correct translation passes the first question; an ethical, usable release must pass both. Store a sample of rejected decisions to make the method teachable and repeatable.

Workshop 11: Consent matrix for an oral tradition

Setup. Construct four uses: research listening, school study, public video and model training. Work product. Assign distinct authorization statuses and an approval owner. The learner should deliver both a reader-facing artifact and a compact evidence ledger: original identifier, revised text, type of transformation, reviewer, access condition and version. The point is to practice communication and accountability together, not to create a decorative bilingual copy.

Critical guardrail. Withhold unapproved derivatives. Ask two separate questions during review. First, can the reader recover the intended linguistic or documentary meaning? Second, is the reader entitled to receive this particular material in this particular form? A correct translation passes the first question; an ethical, usable release must pass both. Store a sample of rejected decisions to make the method teachable and repeatable.

Workshop 12: Pseudonym propagation audit

Setup. Assume one speaker wants a public pseudonym but internal IDs remain confidential. Work product. Review public text, filenames, embedded metadata and thumbnails. The learner should deliver both a reader-facing artifact and a compact evidence ledger: original identifier, revised text, type of transformation, reviewer, access condition and version. The point is to practice communication and accountability together, not to create a decorative bilingual copy.

Critical guardrail. A translated caption is not the only identity-leak path. Ask two separate questions during review. First, can the reader recover the intended linguistic or documentary meaning? Second, is the reader entitled to receive this particular material in this particular form? A correct translation passes the first question; an ethical, usable release must pass both. Store a sample of rejected decisions to make the method teachable and repeatable.

Workshop 13: Machine-aided transcript repair

Setup. Run a hypothetical ASR suggestion on a recording with unusual sounds. Work product. Mark suggestions provisional and audit low-confidence forms against audio. The learner should deliver both a reader-facing artifact and a compact evidence ledger: original identifier, revised text, type of transformation, reviewer, access condition and version. The point is to practice communication and accountability together, not to create a decorative bilingual copy.

Critical guardrail. Do not count hallucinated words as corpus occurrences. Ask two separate questions during review. First, can the reader recover the intended linguistic or documentary meaning? Second, is the reader entitled to receive this particular material in this particular form? A correct translation passes the first question; an ethical, usable release must pass both. Store a sample of rejected decisions to make the method teachable and repeatable.

Workshop 14: Cross-linguistic cognacy review

Setup. Make a toy table with sound similarities and a known borrowing. Work product. Separate inherited cognates from contact borrowing hypotheses. The learner should deliver both a reader-facing artifact and a compact evidence ledger: original identifier, revised text, type of transformation, reviewer, access condition and version. The point is to practice communication and accountability together, not to create a decorative bilingual copy.

Critical guardrail. Do not translate similarity into proven common ancestry. Ask two separate questions during review. First, can the reader recover the intended linguistic or documentary meaning? Second, is the reader entitled to receive this particular material in this particular form? A correct translation passes the first question; an ethical, usable release must pass both. Store a sample of rejected decisions to make the method teachable and repeatable.

Workshop 15: Publication embargo change

Setup. A clip becomes open only after a designated review date. Work product. Schedule authorized status review and recheck linked derivatives. The learner should deliver both a reader-facing artifact and a compact evidence ledger: original identifier, revised text, type of transformation, reviewer, access condition and version. The point is to practice communication and accountability together, not to create a decorative bilingual copy.

Critical guardrail. An embargo date is not proof that consent cannot be reconsidered. Ask two separate questions during review. First, can the reader recover the intended linguistic or documentary meaning? Second, is the reader entitled to receive this particular material in this particular form? A correct translation passes the first question; an ethical, usable release must pass both. Store a sample of rejected decisions to make the method teachable and repeatable.

Workshop 16: Multilingual archive search

Setup. Offer community-preferred names, historical aliases and stable identifiers. Work product. Design an index that supports discovery without forcing one identity label. The learner should deliver both a reader-facing artifact and a compact evidence ledger: original identifier, revised text, type of transformation, reviewer, access condition and version. The point is to practice communication and accountability together, not to create a decorative bilingual copy.

Critical guardrail. Searchability does not justify publishing sensitive participant information. Ask two separate questions during review. First, can the reader recover the intended linguistic or documentary meaning? Second, is the reader entitled to receive this particular material in this particular form? A correct translation passes the first question; an ethical, usable release must pass both. Store a sample of rejected decisions to make the method teachable and repeatable.

Workshop 17: Versioned oral-history anthology

Setup. Combine permitted excerpts from several recordings with different editors. Work product. Publish source citations and permission notes for each excerpt. The learner should deliver both a reader-facing artifact and a compact evidence ledger: original identifier, revised text, type of transformation, reviewer, access condition and version. The point is to practice communication and accountability together, not to create a decorative bilingual copy.

Critical guardrail. The anthology must not erase different source custodians. Ask two separate questions during review. First, can the reader recover the intended linguistic or documentary meaning? Second, is the reader entitled to receive this particular material in this particular form? A correct translation passes the first question; an ethical, usable release must pass both. Store a sample of rejected decisions to make the method teachable and repeatable.

Workshop 18: Interlinear gloss PDF to HTML

Setup. Make a synthetic four-line gloss with diacritics and clitic marks. Work product. Render it accessibly while retaining segment correspondence. The learner should deliver both a reader-facing artifact and a compact evidence ledger: original identifier, revised text, type of transformation, reviewer, access condition and version. The point is to practice communication and accountability together, not to create a decorative bilingual copy.

Critical guardrail. Line wrapping must not destroy analyzable alignment. Ask two separate questions during review. First, can the reader recover the intended linguistic or documentary meaning? Second, is the reader entitled to receive this particular material in this particular form? A correct translation passes the first question; an ethical, usable release must pass both. Store a sample of rejected decisions to make the method teachable and repeatable.

8. Failure-mode analysis and practical quality controls

01. Semantic overreach

Signal: An editor adds implied actors or locations. Diagnostic: Trace each added proposition to audio, transcript or approved contextual note. This is not just a proofreading error: it changes what another human or software pipeline may learn from the source. An audit should record severity, affected records, responsible reviewer and the point at which the problem first entered the workflow.

Corrective action: Withdraw the unsupported claim, not merely the word that sounds awkward. Recheck the original witness, determine whether the same defect appears in related languages and revalidate any structured export. Avoid treating a corrected public sentence as evidence that the underlying identifiers, source annotations, permissions and search copies have also been corrected. Closure requires a verified record of all affected outputs and authorized approval where rights or community interpretation are involved.

02. Alignment drift

Signal: Timing moves when media is edited. Diagnostic: Validate time offset and referenced media version. This is not just a proofreading error: it changes what another human or software pipeline may learn from the source. An audit should record severity, affected records, responsible reviewer and the point at which the problem first entered the workflow.

Corrective action: Re-export and inspect captions against final playback. Recheck the original witness, determine whether the same defect appears in related languages and revalidate any structured export. Avoid treating a corrected public sentence as evidence that the underlying identifiers, source annotations, permissions and search copies have also been corrected. Closure requires a verified record of all affected outputs and authorized approval where rights or community interpretation are involved.

03. Morphological mismatch

Signal: Segmented forms and gloss labels no longer correspond. Diagnostic: Compare delimiter count and morpheme mapping line by line. This is not just a proofreading error: it changes what another human or software pipeline may learn from the source. An audit should record severity, affected records, responsible reviewer and the point at which the problem first entered the workflow.

Corrective action: Correct the analysis and all dependent examples together. Recheck the original witness, determine whether the same defect appears in related languages and revalidate any structured export. Avoid treating a corrected public sentence as evidence that the underlying identifiers, source annotations, permissions and search copies have also been corrected. Closure requires a verified record of all affected outputs and authorized approval where rights or community interpretation are involved.

04. Code collapse

Signal: An individual language is replaced by a generic code. Diagnostic: Record registry, code version and actual variety. This is not just a proofreading error: it changes what another human or software pipeline may learn from the source. An audit should record severity, affected records, responsible reviewer and the point at which the problem first entered the workflow.

Corrective action: Restore original identification and maintain explicit mappings. Recheck the original witness, determine whether the same defect appears in related languages and revalidate any structured export. Avoid treating a corrected public sentence as evidence that the underlying identifiers, source annotations, permissions and search copies have also been corrected. Closure requires a verified record of all affected outputs and authorized approval where rights or community interpretation are involved.

05. Unauthorized disclosure

Signal: Restricted names appear in searchable metadata. Diagnostic: Audit all public views, filenames and derivatives. This is not just a proofreading error: it changes what another human or software pipeline may learn from the source. An audit should record severity, affected records, responsible reviewer and the point at which the problem first entered the workflow.

Corrective action: Restrict access and follow the custodian’s remediation procedure. Recheck the original witness, determine whether the same defect appears in related languages and revalidate any structured export. Avoid treating a corrected public sentence as evidence that the underlying identifiers, source annotations, permissions and search copies have also been corrected. Closure requires a verified record of all affected outputs and authorized approval where rights or community interpretation are involved.

06. Attribution loss

Signal: The translator is named as the original speaker. Diagnostic: Use layer-specific contributor roles. This is not just a proofreading error: it changes what another human or software pipeline may learn from the source. An audit should record severity, affected records, responsible reviewer and the point at which the problem first entered the workflow.

Corrective action: Correct citation and notify downstream repositories. Recheck the original witness, determine whether the same defect appears in related languages and revalidate any structured export. Avoid treating a corrected public sentence as evidence that the underlying identifiers, source annotations, permissions and search copies have also been corrected. Closure requires a verified record of all affected outputs and authorized approval where rights or community interpretation are involved.

07. False certainty

Signal: A disputed form becomes definitive prose. Diagnostic: Keep uncertainty and alternative readings visible. This is not just a proofreading error: it changes what another human or software pipeline may learn from the source. An audit should record severity, affected records, responsible reviewer and the point at which the problem first entered the workflow.

Corrective action: Relabel as tentative until resolved. Recheck the original witness, determine whether the same defect appears in related languages and revalidate any structured export. Avoid treating a corrected public sentence as evidence that the underlying identifiers, source annotations, permissions and search copies have also been corrected. Closure requires a verified record of all affected outputs and authorized approval where rights or community interpretation are involved.

08. Revision invisibility

Signal: A later correction has no history. Diagnostic: Store issue, change, reviewer and release version. This is not just a proofreading error: it changes what another human or software pipeline may learn from the source. An audit should record severity, affected records, responsible reviewer and the point at which the problem first entered the workflow.

Corrective action: Publish an honest correction record. Recheck the original witness, determine whether the same defect appears in related languages and revalidate any structured export. Avoid treating a corrected public sentence as evidence that the underlying identifiers, source annotations, permissions and search copies have also been corrected. Closure requires a verified record of all affected outputs and authorized approval where rights or community interpretation are involved.

09. Tool mismatch

Signal: Valid exported XML still misrepresents context. Diagnostic: Combine schema validation with linguistic playback and reading. This is not just a proofreading error: it changes what another human or software pipeline may learn from the source. An audit should record severity, affected records, responsible reviewer and the point at which the problem first entered the workflow.

Corrective action: Repair semantic mapping rather than only syntax. Recheck the original witness, determine whether the same defect appears in related languages and revalidate any structured export. Avoid treating a corrected public sentence as evidence that the underlying identifiers, source annotations, permissions and search copies have also been corrected. Closure requires a verified record of all affected outputs and authorized approval where rights or community interpretation are involved.

10. Cultural flattening

Signal: Registers and participant roles vanish. Diagnostic: Use review by authorized contextual experts. This is not just a proofreading error: it changes what another human or software pipeline may learn from the source. An audit should record severity, affected records, responsible reviewer and the point at which the problem first entered the workflow.

Corrective action: Restore function or add a clearly marked editorial note. Recheck the original witness, determine whether the same defect appears in related languages and revalidate any structured export. Avoid treating a corrected public sentence as evidence that the underlying identifiers, source annotations, permissions and search copies have also been corrected. Closure requires a verified record of all affected outputs and authorized approval where rights or community interpretation are involved.

11. Privacy-through-translation

Signal: A translated clue reveals a hidden individual. Diagnostic: Check inferences across joined public fields. This is not just a proofreading error: it changes what another human or software pipeline may learn from the source. An audit should record severity, affected records, responsible reviewer and the point at which the problem first entered the workflow.

Corrective action: Generalize or withhold only with authorized guidance. Recheck the original witness, determine whether the same defect appears in related languages and revalidate any structured export. Avoid treating a corrected public sentence as evidence that the underlying identifiers, source annotations, permissions and search copies have also been corrected. Closure requires a verified record of all affected outputs and authorized approval where rights or community interpretation are involved.

12. Unverifiable data reuse

Signal: A cited form has no source reference. Diagnostic: Require a resolvable item, page or interval. This is not just a proofreading error: it changes what another human or software pipeline may learn from the source. An audit should record severity, affected records, responsible reviewer and the point at which the problem first entered the workflow.

Corrective action: Treat it as unverified until a witness is found. Recheck the original witness, determine whether the same defect appears in related languages and revalidate any structured export. Avoid treating a corrected public sentence as evidence that the underlying identifiers, source annotations, permissions and search copies have also been corrected. Closure requires a verified record of all affected outputs and authorized approval where rights or community interpretation are involved.

9. Working glossary of language-documentation translation

Annotation. A layer of information linked to an utterance, media interval or linguistic object. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Archive deposit. A curated submission of files and metadata under repository and access conditions. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Alignment. The correspondence between source segments, tiers, glosses and translations. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Allomorph. A variant realization associated with a morpheme in a stated analysis. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Aspect. A grammatical category describing event viewpoint or internal temporal structure. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Attestation. A form supported by a recorded, written or otherwise documented occurrence. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Autonym. A self-designation used for a language, people or other entity. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Borrowing. A linguistic element adopted through language contact rather than necessarily inherited. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Clitic. A dependent element often represented differently from a morphological affix. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

CLDF. Cross-Linguistic Data Formats for structured exchange of comparative language datasets. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Code switching. The use of more than one language or variety within a conversation or passage. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Cognate. A form proposed to descend from a shared historical source. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Community review. Review by appropriately authorized speakers, custodians or representatives. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Consent scope. The uses, audiences, durations and conditions that permission actually covers. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Corpus. An organized collection of language evidence and associated metadata. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Data provenance. A trace of where a value originated and how it was transformed. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Derivation. A word-building operation that can change lexical category or meaning. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Dialect. A named language variety; classification and community identity can differ. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

EAF. ELAN Annotation Format linking annotated tiers and media. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

ELAN. Software for time-aligned multilingual and multimodal annotation. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Endonym. The name used by members of a group for themselves or their language. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Evidentiality. Grammatical marking of information source or evidential basis. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Exonym. An external name for a group, place or language. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Field recording. Audio or video source collected to document a speech event. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

FLEx. FieldWorks Language Explorer, software for lexicon and language analysis. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Free translation. A meaning-oriented sentence or passage not constrained to mirror source segmentation. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Gloss. An explanatory label corresponding to a source morpheme or lexical item. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Glottocode. A stable Glottolog identifier assigned under the database’s classification scheme. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Grammaticalization. Change through which expressions acquire grammatical functions. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Interlinear glossed text. Aligned lines representing linguistic forms, analyses and translation. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

IPA. International Phonetic Alphabet for recording phonetic distinctions. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

ISO 639. International language-code standard, with historically named identifier sets. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Language variety. A specified way of speaking that may differ by region, community or social setting. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Lexeme. An abstract lexical unit connecting forms and senses. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Lexical sense. A particular meaning and use of a lexeme. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Macrolanguage. A higher-scope language identifier grouping individual languages in some contexts. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Metadata. Descriptive, technical, administrative or rights information about a source. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Morpheme. A minimal meaningful or grammatical unit under a particular analysis. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Morphology. The analysis of word structure, forms and grammatical properties. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Normalization. A documented change from source-near representation to a conventional form. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Orthography. A writing convention for a language or variety. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Parallel text. Linked renditions of content in different languages or forms. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Parse. An analysis assigning structure and relations to an utterance. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Persistent identifier. A stable reference used to maintain identity across changes. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Phoneme. A sound category functioning contrastively under a phonological analysis. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Phonetic detail. Recorded or transcribed properties of actual speech realization. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Pragmatics. Meaning conveyed through context, speaker intent and social use. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Prosody. Pitch, stress, timing and rhythm used in speech. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Pseudonym. A public substitute name protecting or managing participant identity. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Referent. The person, entity, situation or event an expression identifies. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Restricted access. An archive condition limiting who may see or reuse an item. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Segmentation. Marking words and morphemes as separate analytical units. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Sense ID. A stable key for one distinct meaning in a lexical database. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Source witness. The evidence to which a transcription, gloss or translation is answerable. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Speaker tier. An annotation track associated with a participant. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Speech act. The social action performed by an utterance, such as asking or warning. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Stand-off annotation. Annotation maintained separately from the source media. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Syncretism. Sharing a surface form across distinct grammatical categories. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Time-aligned transcript. Text linked to specified media intervals. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Tone. Pitch patterns that may mark lexical or grammatical contrasts. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Transcription. A written representation of an observed speech event. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Transliteration. Systematic representation of writing in another script. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Translation memory. A store of source-target segments whose reuse requires context checks. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Uncertainty marker. A notation preserving doubt about source or analysis. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Version ledger. A record linking changes to their source and approved releases. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

Withdrawn access. A rights or access change requiring review of released derivatives. In this article, the term belongs to a particular evidential layer; its definition should not be used to silently override another layer’s observed form, analytic status, identity, permission or review history.

10. Questions readers ask about translating linguistic records

1. What are interlinear glosses?

They are aligned representations of linguistic forms and analyses, often including a source line, segmented line, morpheme gloss and free translation. The layers need not all have identical syntactic form, but the analytical correspondence must be explicit. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

2. Are Leipzig Glossing Rules compulsory?

No single proposed convention proves a language’s grammar. They offer widely used practices for presenting analyses and abbreviations; a project should document departures and added category labels. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

3. Is free translation the same as a literal gloss?

No. A free translation aims at discourse meaning for readers, while morpheme glossing marks lexical and grammatical components. One cannot serve as a substitute for the other’s evidence. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

4. Can an interlinear gloss have more than four lines?

Yes. Projects may include phonetic, orthographic, segmented, glossed, literal, free, translation and note lines. Name each line and keep its relations clear. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

5. Why keep an original spelling after normalization?

Normalization can aid reading and searching, but the original helps resolve historical, phonological and editorial questions. A public edition should reveal when a representation has been changed. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

6. Are grammatical gloss labels ordinary English words?

No. Labels such as PST or NEG denote analytical categories. A lexical gloss and a grammar label are not interchangeable and should have a documented key. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

7. Does a hyphen mean the same as an equals sign in glossing?

Not under the usual Leipzig-style conventions: the symbols can distinguish morpheme boundaries from clitic boundaries. Check the project’s exact stated policy. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

8. Can audio alignment itself tell us where morphemes begin?

Not reliably. Acoustic segmentation and morphological segmentation are different analyses. One morphological element may have no simple isolated time interval. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

9. How do I translate evidentiality accurately?

First determine the grammatical value of the source marker and its scope. Then preserve whether the speaker is reporting, inferring, witnessing or otherwise indicating information source where relevant. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

10. Why not use one English word for one source word?

A source form can have multiple senses, and a target word can cover different semantic territory. Verify the sense and construction in each context. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

11. What if speaker accounts disagree?

Keep competing versions attributed and reviewed. Do not merge them into a single invented consensus or mistake a researcher’s preferred analysis for unanimous community agreement. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

12. Can community speakers change an earlier transcription?

Yes, subject to the project’s agreed editorial process. Preserve earlier source versions and document the correction; a revision should not erase the history of the evidence. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

13. Is an ISO language code the true name of a language?

No. It is a standardized identifier. Community-preferred names, varieties and code-set versions require their own fields. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

14. What is a Glottocode for?

It identifies a classification entry in Glottolog and supports data linkage. It does not determine what a community should call itself or certify every historical subgrouping. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

15. What is CLDF used for?

It structures tabular cross-linguistic data and its metadata so tools can exchange language, form, parameter and related records. Correct foreign keys do not prove semantic correctness. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

16. Can I translate CLDF IDs?

Generally, preserve stable identifiers and translate appropriate human-readable labels or descriptive fields instead. Changes to IDs require a controlled migration. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

17. Can I publish every recording held by an archive?

No. Catalogue visibility, file access, public distribution and derivative reuse can have different conditions. Check current item-level permissions and community agreements. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

18. Is a CC licence enough if a person’s voice is identifiable?

Not automatically. Copyright permissions do not exhaust privacy, consent and community obligations. Use the archive’s actual conditions and responsible decision makers. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

19. Can AI transcribe an under-documented language without review?

Automated suggestions can help a project in controlled circumstances, but unfamiliar phonology and code-switching can produce persuasive false text. Compare against primary evidence and check processing permissions. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

20. How should uncertainty appear in translation?

Mark unresolved source, segmentation or interpretation as uncertain using agreed notation or carefully worded editorial explanation. Do not replace doubt with unearned certainty. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

21. What is a safe teaching extract?

One whose source, context, privacy and reuse conditions authorize the proposed audience and format. A freely found online clip is not automatically freely reusable. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

22. Do captions have to be literal?

Readable subtitles can compress some speech, but they should not change participants, speech acts, factual claims or uncertainty. Abridgment should be identified when material is documentary evidence. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

23. What is the best way to record translator credit?

Credit speakers, transcribers, analysts, translators and community reviewers according to their actual roles and agreed visibility. Avoid treating the outside publisher as the sole originator. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

24. What if a language identifier changes in 2026?

Record the versioned identifier and maintain a mapping to updated registries. Do not retroactively conceal which code a cited dataset actually used. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

25. How do I evaluate a translated glossary entry?

Test sense, grammatical behavior, example usage, register and authorization. A familiar word equivalent is only a starting hypothesis. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

26. Can a phonetic transcription substitute for a practical orthography?

Not without stating the change in purpose. Phonetic symbols document sound detail, while practical orthography follows a community or project writing convention. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

27. How should hidden locations be handled?

Keep withheld precision withheld across captions, databases, geocodes and search results. Do not infer exact coordinates from a contextual description. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

28. How does a correction reach every language?

Use identifiers and versioned derivative inventories. Recheck translations, glosses, transcripts, exported files, public snippets and archived copies affected by the corrected claim. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

29. Does a four-line example prove linguistic theory?

No. It illustrates an analysis backed by some evidence. A responsible publication exposes conventions and gives enough context or references to assess the claim. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

30. Where should I learn general translation strategy?

Begin with the translation root linked at the start and end of this article. This specialist branch is about linguistic evidence, interlinear data, archive permissions and maintainable derivatives. For an actual project, check the original evidence and the custodians’ current policies before converting this guidance into a publication decision. A source-language reviewer and an independent target reader can detect different errors, so both perspectives are valuable where the material allows review.

11. Four-week implementation plan: learn the method through completed artifacts

Week one: Build an evidence inventory

Goal. Select a harmless synthetic corpus of recordings, a transcription, a gloss and a free translation. Give every item a stable ID. Record participant roles, versions, permissions and an orthography policy. Audit whether the original and the derived representations can be separated without deleting information.. The deliverable is not a stack of attractive translations, but a compact and testable system of evidentiary links, role assignments, versions, permissions, analysis conventions and user-facing texts. Build a weekly reviewer meeting around disagreements: each case should identify what evidence would resolve it, who may make the decision and whether the resulting explanation may be published.

Passing standard. A colleague can reconstruct the evidence chain and find the source for any displayed sentence.. When the team cannot establish a safe answer, preserve uncertainty, withhold the disputed derivative or seek authorization. It is better to publish a smaller trustworthy corpus than a larger collection of polished but untraceable claims.

Week two: Repair the linguistic layers

Goal. Take fifteen toy examples and check alignment, segmentation, clitic notation, person values, evidence markers, pronoun reference, negation and speaker roles. Create an abbreviation key and a sense-level termbase. Review ambiguities as unresolved issues rather than guessing.. The deliverable is not a stack of attractive translations, but a compact and testable system of evidentiary links, role assignments, versions, permissions, analysis conventions and user-facing texts. Build a weekly reviewer meeting around disagreements: each case should identify what evidence would resolve it, who may make the decision and whether the resulting explanation may be published.

Passing standard. An independent reviewer can identify which distinctions each gloss encodes and which translations are only explanatory.. When the team cannot establish a safe answer, preserve uncertainty, withhold the disputed derivative or seek authorization. It is better to publish a smaller trustworthy corpus than a larger collection of polished but untraceable claims.

Week three: Validate formats and circulation

Goal. Export representative annotation tiers and a small CLDF-style table. Verify row keys, language references and the relation of translated text to media. Create an item-level reuse matrix and simulate a denied public request for restricted material.. The deliverable is not a stack of attractive translations, but a compact and testable system of evidentiary links, role assignments, versions, permissions, analysis conventions and user-facing texts. Build a weekly reviewer meeting around disagreements: each case should identify what evidence would resolve it, who may make the decision and whether the resulting explanation may be published.

Passing standard. No data join breaks, and no denied item appears in the public demonstration view or generated search extract.. When the team cannot establish a safe answer, preserve uncertainty, withhold the disputed derivative or seek authorization. It is better to publish a smaller trustworthy corpus than a larger collection of polished but untraceable claims.

Week four: Release and rehearse correction

Goal. Publish only the approved fictional exercises, source IDs, credit and review notes. Simulate a community correction to a kinship term and a withdrawn media permission. Track every affected derivative and generate a revised manifest.. The deliverable is not a stack of attractive translations, but a compact and testable system of evidentiary links, role assignments, versions, permissions, analysis conventions and user-facing texts. Build a weekly reviewer meeting around disagreements: each case should identify what evidence would resolve it, who may make the decision and whether the resulting explanation may be published.

Passing standard. A second editor can verify both the linguistic correction and the access-control repair from the change log.. When the team cannot establish a safe answer, preserve uncertainty, withhold the disputed derivative or seek authorization. It is better to publish a smaller trustworthy corpus than a larger collection of polished but untraceable claims.

12. Primary reference documents and terminology sources

This guide uses public technical documentation for format, terminology and method, not a claim that every repository, language community or legal system follows one uniform rule. Source versions and dates matter. Official terminology is a starting point for review, not permission to copy restricted source material or reproduce copyrighted standards text.

1. ISO 639:2023, Code for individual languages and language groups, official ISO overview — Published November 2023; integrated identifier-set framework and harmonized terminology.
2. Leipzig Glossing Rules, Max Planck Institute for Evolutionary Anthropology — Interlinear morpheme-by-morpheme conventions, ten rules and an abbreviation appendix.
3. CLDF, Cross-Linguistic Data Formats — Specification information for structured comparative datasets; CLDF 1.3 announcement.
4. Glottolog 5.3 as CLDF, Zenodo official release record — Release dated 2 March 2026; cited database version for language and classification references.
5. FieldWorks Language Explorer, SIL — Lexicon and interlinear analysis tools, publicly documented and maintained; 2026 release history.
6. ELAN User Guide, Max Planck Institute for Psycholinguistics — Tier structure and time-aligned multimedia annotation concepts.
7. PARADISEC deposit guidance — Practical source for archive access conditions, metadata and deposit procedure.
8. Endangered Languages Archive — Repository descriptions and access-condition context for multimedia language documentation.
9. DOBES Documentation of Endangered Languages archive — Language documentation archive context and collections.
10. ISO language-code agency, Library of Congress Set 2 — Official code-maintenance context and relation to the ISO 639:2023 framework.

Research verification date: 1 November 2026. The Leipzig conventions and referenced file formats govern representation, not the truth of a particular linguistic analysis. Glossing abbreviations used in this article are illustrative and should be checked against each project’s actual categories. Descriptions of archive access are general examples; the only authoritative access decision for a particular recording is its current, applicable record and the people empowered to approve use.

13. Next routes: carry meanings across languages without taking over specialist owners

For the general theory of equivalence, source and target text, pragmatic transfer, translator judgment, quality review and the broad architecture of language transfer, return to Master Art of Translation | The Complete System for Moving Meaning Between Languages. This specialist article adds a narrower competency: understanding what a documented language record actually asserts and what its custodians authorize.

If the reader wants to learn words, senses, collocations, retrieval practice and long-term vocabulary growth rather than translate an archived lexicon, the appropriate owner is the protected Vocabulary Learning Hub. If the reader needs the grammar of English itself, especially clause structure and reference in English explanations, use the protected How English Works route. Those pages remain the owners of their subjects; this child links outward only where a documented translation problem leads naturally to them.

The enduring method is to distinguish observation, analysis, translation, public explanation and permission. Preserve what was recorded; state what was inferred; label what was adapted; check with those who have linguistic and cultural authority; retain stable data references; and make correction possible. A translation is successful when a future reader can understand the speaker’s meaning without mistaking editorial convenience for evidence or unrestricted access for consent.