The Gold Standard of Anthropology | Collaborative Captioning, Translation Choice and Family Voice in Digital Archives

1. Start with the audience and access purpose

Back to contents

Define who needs the captions and what they need to do with them. A Deaf relative following the video, a heritage-language learner, an archivist searching names and a researcher studying interaction may require different levels of timing, sound description and verbatim detail. A single caption track rarely serves every purpose. Record the intended audience before deciding what “accurate” means.

2. Separate transcription, captioning and translation

Back to contents

Transcription represents speech as text. Captioning also synchronises that text to audiovisual time and may identify speakers or meaningful sounds. Translation moves meaning across languages; transliteration moves between writing systems. These tasks can be combined, but their decisions should remain visible. A fluent English subtitle is not a verbatim transcript of a Cantonese, Malay or Tamil utterance simply because it appears beneath the same image.

3. Keep family voice in the decision process

Back to contents

Invite speakers and implicated relatives to review names, kin terms, dialect forms, code-switching, humour and sensitive passages. Collaboration does not mean majority vote. The person whose speech is represented should have meaningful authority, while others may hold rights or risks connected to the story. Record disagreements and the final governance decision rather than smoothing every difference into institutional house style.

4. Choose how much spoken texture to preserve

Back to contents

Pauses, repetition, overlap, laughter, particles and self-repair can carry stance and relationship. Keeping everything may make captions unreadable; removing everything can turn a hesitant recollection into confident testimony. Establish levels such as readable captions, fuller transcript and research annotation. Readers should know which layer they are seeing and which features were compressed for access.

5. Treat translation as a set of accountable choices

Back to contents

Kin terms, honorifics, idioms and culturally specific references may have no single equivalent. Retain the source wording where useful, add a concise translation and document alternatives in notes or metadata. Do not standardise a speaker’s register merely to make the target language sound polished. A translation can improve access while also shifting age, intimacy, humour or authority.

6. Worked example: one kin term changes the family map

Back to contents

In a bilingual video, an aunt uses a heritage-language kin term that distinguishes an older paternal uncle from other uncles. The first English caption says “uncle”, making two people appear to be the same relative. During review, a younger family member proposes the person’s name; the speaker prefers the kin term because it explains respect and lineage. The archive keeps the original term, adds “older paternal uncle” on first use and stores the family-approved name in metadata. The caption becomes clearer without erasing the relationship encoded in speech.

7. Design captions for legibility and multilingual use

Back to contents

Synchronise captions to speech, keep reading speed and line length manageable, identify speaker changes where necessary and describe meaningful non-speech audio. For multilingual recordings, test whether one track, selectable tracks or dual-language display works for real users. Scripts differ in density and line breaking. Accessibility is not achieved by generating text alone; timing, placement, contrast and language switching must be reviewed in the actual player.

8. Changed-context transfer check: an audio-only archive

Back to contents

Move from a family video to an audio interview with no visible speaker. The transferable questions remain: who is speaking, what sounds matter, how translation changes voice and who approves representation. The changed context removes visual cues, so speaker labels and descriptions of gestures heard indirectly may matter more. Caption placement disappears, but a time-aligned transcript and accessible player remain essential.

9. Know the limits and common misconceptions

Back to contents

A caption is not the speaker’s voice itself, and a polished translation is not neutral. Family review cannot guarantee one shared interpretation, especially when conflict or unequal authority exists. Automatic speech recognition may accelerate a draft but can fail unevenly across accents, dialects, code-switching and poor audio. Publishing two languages does not automatically produce language justice if one track is inaccurate, hard to select or treated as secondary.

10. Audit every representational layer

Back to contents

Retain the source recording, versioned transcript, time codes, speaker labels, translation alternatives, reviewer roles, permissions, correction history and player settings. Mark machine-generated drafts and human revisions. Preserve withdrawn or sensitive text only when consent and governance allow it. The audit should let a future curator see why a wording was chosen without exposing family information that contributors decided not to publish.

11. Frequently asked questions

Back to contents

Should captions be verbatim? It depends on audience and purpose; disclose any compression. Can family members translate their own recordings? Yes, but expertise, burden and power still need attention. Should dialect be normalised? Only through an explicit, reviewed policy. Can one bilingual caption track serve everyone? Often not. What makes the result credible? Visible source-to-caption decisions, speaker review where appropriate and tested accessibility.

12. The gold standard makes the caption an accountable edition

Back to contents

A trustworthy digital archive does not hide the work that turns speech into searchable text. It distinguishes transcription from translation, gives speakers a real place in review, tests multilingual accessibility and preserves uncertainty when no single wording is adequate. The gold standard treats captions as a carefully governed edition of family voice: useful, revisable and never mistaken for the whole encounter.