Build a subtitle glossary for a video series

Give repeated concepts one shared reference

For a series of subtitled videos, keep a small glossary beside the projects. Record each important name or concept, its meaning in this series, the preferred wording in each target language and the evidence behind that choice. Then check the actual cues against it. One entry should represent one meaning, rather than every possible use of the same word.

This is a manual review workflow for recordings and subtitle text you own or have permission to process and distribute. The glossary can be an ordinary Markdown document. Polyglot creates, translates and edits timed subtitles; this guide does not describe a glossary-import feature or automatic enforcement of your terminology decisions.

Choose terms that could confuse the viewer

Start with repeated names, product labels, specialist concepts and abbreviations whose meaning matters to the recording. Avoid copying every noun into the glossary. A short reference with clear decisions is easier to use than a dictionary full of obvious translations. Include a rare term when getting it wrong would change what the viewer understands.

Microsoft’s writing guidance recommends consistent terminology for a concept and using sentence context to remove ambiguity. For subtitles, apply that principle without rewriting the speaker into a different person. Different spoken words may refer to the same concept, while the same spoken word may refer to different things. Record that distinction before choosing a preferred target term.

Gather references from the actual project: the approved script, an owner-confirmed name list, a product screen shown in the video or the specialist who supplied the material. Listen to the recording as well. A script can establish spelling, but it does not prove the performer spoke that exact sentence. Keep genuinely unclear speech unresolved until someone can verify it.

Write an entry that survives outside its first scene

For each entry, record the source term, a brief definition, an example cue or timestamp, the target language and locale, and the preferred target wording. Add an explicit instruction when a name should stay unchanged. Keep capitalization, diacritics and allowed variants visible. A reviewer opening the document next week should understand the decision without remembering the original conversation.

Include a reference, a decision status and the glossary revision. Use statuses such as proposed, needs subject review and approved, and only mark approval when the relevant person has actually checked it. A content owner can confirm what an object is called; a competent language reviewer still needs to judge how to express it naturally for the target audience.

Google Cloud’s glossary documentation identifies product names and ambiguous words as glossary use cases. That supports the distinction between a name and a meaning-specific term. Its service-specific matching rules are not Polyglot behavior: the reference here is for people, and no subtitle text needs to be sent to Google to follow this workflow.

Example: a printing plate is not a dinner plate

Imagine an original printmaking series that mentions a printing plate in one lesson and a dinner plate used as a circular template in another. Translating every English occurrence of “plate” into the same Dutch word would make the series less accurate. Create two entries with different contexts, even though their source spelling matches.

The fictional draft below illustrates the structure. The suggested Dutch terms are editorial examples, not a record of a real reviewer’s approval. Mira Studio is an invented studio name retained unchanged. The glossary revision and filenames are suggestions for your own notes, not app-generated metadata.

Series: studio-lessons-v2
Glossary: r1 / EN → nl-NL
Status: proposed for review

Term: plate [printmaking]
Meaning: inked printing matrix
NL proposal: drukplaat
Context: lesson 01, cue 8
EN: Ink the plate evenly.
NL: Inkt de drukplaat
gelijkmatig in.

Term: plate [tableware]
Meaning: dish used as a
template
NL proposal: bord
Context: lesson 03, cue 12
EN: Trace around the plate.
NL: Trek de omtrek van
het bord over.

Name: Mira Studio
Rule: keep spelling unchanged
Reference: fictional title
card

Before approval, add:
- verified project reference
- language reviewer decision
- date and glossary revision

Allow grammar without losing the concept

A preferred term is not necessarily an exact string that must appear unchanged in every sentence. The target language may require a plural, a different word order or an inflected form. Record useful variants so reviewers can distinguish a grammatical change from an inconsistent translation. Do not reject a correct sentence just because a literal search finds a different ending.

Keep separate rules for proper names, visible interface labels and ordinary descriptive language. When a video shows an untranslated button, the subtitle may need to preserve the visible label so the viewer can find it. If a localized screen is used in another episode, check that version separately. The rule should follow the actual screen and audience, not an assumed universal translation.

If the preferred wording makes a cue too long, revise the sentence around it while retaining the necessary meaning. Do not invent an unexplained abbreviation solely to fit. Review line breaks and reading time with the final video; a terminology decision does not provide extra screen time.

Apply the glossary across the series deliberately

Before reviewing every episode, try the proposed decisions against several representative passages: an introduction, a dense explanation, a plural form and an unrelated use of an ambiguous word. In the plate example, the dinner-plate cue is a useful counterexample: it should not inherit the printing term. Resolve conflicts while the glossary is still small.

Then check every relevant occurrence in every intended language track. A text search can help locate repeated spellings, but it can miss pronouns, variants and recognition mistakes. Read the surrounding cues and replay the passage. Correct source transcription errors before relying on the translation, and edit target wording only with enough language knowledge to judge the result.

Keep a separate coverage note listing the video revision, language, glossary revision and review state. “Glossary approved” and “all episodes checked” are different claims. A new episode may contain an unseen sense of an old word; add a new contextual entry rather than silently expanding the earlier rule. For changes to already delivered subtitles, use the source-correction workflow linked below.

Verify the exports and retain the decisions

Export the reviewed tracks and inspect the actual files with their matching videos. Check names with accents, agreed labels, grammatical variants and the ambiguous counterexamples. Confirm that unrelated text and timing remain intact. For a series, verify that every delivered file corresponds to the glossary revision recorded in the coverage note; do not assume checking the first episode checks the others.

Retain the glossary with the source references and subtitle exports. Keep an earlier revision when decisions change so a collaborator can understand why one delivery differs from another. Unresolved entries should stay clearly pending, with the affected videos identified. The note is an external record you maintain, not an automatic audit trail from Polyglot.

Terminology consistency is only one part of subtitle quality. W3C’s caption guidance calls for accurate speech and the relevant non-speech information needed to understand the content. A perfectly consistent name does not catch missing negatives, wrong quantities, omitted sound cues or unreadable timing. Finish with a full meaning and playback review before publishing.

Keep exploring.