Build a topic index from an oral-history transcript

Give the listener useful places to begin

To build an oral-history topic index, review the interview, identify meaningful changes of subject and write a short entry for each useful starting point. Keep the entry’s title, summary and search terms separate from the narrator’s words. Then open each entry in the receiving archive’s player and check that it reaches the passage it describes.

A transcript helps someone search for words. An index can help them find a subject even when the narrator uses different vocabulary. The Nunn Center’s guide to indexing explains this distinction: descriptive topic metadata adds navigation within an interview. It does not replace the transcript, the recording or the need to listen in context.

This guide uses a fictional interview and a manual worksheet. Polyglot can help prepare reviewed text locally on an Apple Silicon Mac; it does not create topic indexes, identify speakers or export an OHMS archive record. Index authoring and publication happen separately in the archive’s chosen system.

Fix the access copy and the publication scope first

Choose the exact audio or video file that listeners will access. Record its identifier, revision, duration and any separate parts. An index built against an untrimmed master can point to the wrong words in a public copy with an introduction removed. Keep the original safely, but take navigation times from the approved access copy.

Confirm the agreed audience and review process for both media and metadata. The Oral History Association’s best practices address narrator review, access restrictions and contextualized use. A topic title or synopsis can reveal information even when the corresponding recording is restricted. Have the project owner decide what descriptions may be published; keep private review notes out of public fields.

Ask the receiving archive for its current authoring workflow, required fields, vocabulary conventions and review environment. OHMS is one example, not a required destination. Its official overview describes separate authoring and viewer components, with an XML record connecting the prepared metadata to the viewer. A TXT or SRT export alone is not that record.

Use reviewed text to find candidates, then listen

Review the original-language transcript against the recording before using it as evidence. Correct misheard names, places, dates and negatives. Retain uncertain wording under the project’s convention rather than inventing a confident reconstruction. If a topic title depends on an unresolved word, leave that entry pending.

Polyglot’s text transcript export can include cue-start timestamps. Keep that export as a reference copy, alongside the reviewed timed track. Those times help locate candidate passages; they do not establish the start of a topic or the timing of every word. A topic can begin halfway through a subtitle cue, or in the interviewer’s question before the answer.

Read for structure, then listen around each proposed transition. Look for the start of a story, a change of place or period, and a return to an earlier subject. Choose a point that gives the listener enough context to understand the response. Do not generate an entry for every subtitle or divide the interview into arbitrary equal blocks.

Example: one subject can need two entries

Imagine an authorized interview about working on a ferry. The narrator first discusses receiving wages, later describes a changed route, then returns to wages while recalling a payroll dispute. The times and content below are invented editorial examples. They are not a recording, a transcription test or a machine-import template.

The two wage passages deserve separate descriptions because they answer different questions. Giving both the title “Pay” would make browsing less useful. Combining them into one continuous segment would hide the intervening route discussion. Keep their separate positions and use a consistent search term to connect them.

The narrator is unsure whether the first payment arrived on a Friday. Preserve that uncertainty in the transcript and synopsis. An indexer should not convert a recollection into an established payroll fact. The proposed landing points also include the relevant question or lead-in, rather than beginning after the context has already been spoken.

Fictional manual index sheet
Media: OH-021-access-v2.wav
Clock: elapsed playback

02:10 · First wage payment
Term: wages
Synopsis: Recalls first pay;
uncertain which day it arrived.

05:42 · Change of ferry route
Term: ferry routes
Synopsis: Describes adapting
to a different morning route.

11:08 · Disputed wage payment
Term: wages
Synopsis: Recalls questioning
a later payment amount.

For each entry, record:
lead-in checked: ____
destination checked: ____
reviewer / status: ____

Write descriptions without rewriting the narrator

Give each segment a specific, neutral title. Write a short synopsis describing what the narrator discusses, with attribution and uncertainty where needed. Keep a partial transcript or quotation in its own field and check it word for word. A summary is editorial description; it should never be presented as something the narrator said verbatim.

Use the archive’s approved subjects and consistent keywords. The Nunn Center’s 2014 indexing guide distinguishes controlled subjects from more specific local terms and explains how a synopsis can clarify a title. Its conceptual guidance remains useful, but use current documentation for interface instructions. Avoid adding speculative diagnoses, motives or identities to make a segment easier to find.

Review the whole interview’s representation. An index containing only dramatic incidents can misrepresent a recording that mostly concerns ordinary work and family life. Include useful transitions across the full interview, and document any deliberate limits on indexing depth. A search term should lead to relevant content, not merely a passing word that happens to match.

Keep topic markers separate from transcript synchronization

OHMS’s September 2023 Aviary user guide describes index entries with a timestamp, required segment title and optional descriptive fields. Its transcript synchronization is a different process. The guide specifies supported document formats and interval-based timestamps formatted as [HH:MM:SS]. Consult the archive’s current system before preparing an import.

Polyglot’s timestamped TXT instead places [HH:MM:SS,mmm] cue-start references beside cue text. Irregular subtitle starts are not the documented OHMS synchronization intervals. Removing the milliseconds does not turn them into correctly placed interval markers. Likewise, renaming SRT to TXT or XML does not create a synchronized transcript or an index.

For this workflow, use the export to locate and check source passages, then author the topic entries in the receiving system. If the archive also wants a synchronized full transcript, prepare and verify that as a separate deliverable using its current instructions. No OHMS import or player integration was tested for this article.

Accept the index by following its entries

In the archive’s preview or test environment, select every entry and listen forward. Confirm that the correct recording opens, the lead-in is intelligible and the promised topic follows. Check entries near the beginning and end, repeated subjects and any boundaries between recording parts. A correctly formatted time can still point to the wrong passage.

Where the destination offers search, try an approved subject term and a word from the transcript. Check which type of result appears and whether selecting it reaches the expected content. Ask a second reader to find a passage from its description without your explanation. Record failures as concrete corrections to the title, description, time or media association.

Recheck public titles, snippets and metadata against the access decisions before release. Retain the approved transcript, index export, media revision and review record under the archive’s arrangements. The OHMS guide warns that changes to media timing disrupt index and synchronization markers: a replacement recording needs another entry-by-entry check, even if its filename stays the same.

The accepted result is a navigable interview whose descriptions can be checked against the source. Timed captions, a readable transcript and a topic index each serve a different purpose; keep their review status explicit so completing one is not mistaken for completing all three.

Keep exploring.