Glossaries & consistency
Lock names and terms across a series — automatic detection, context groups, QA.
8 min read
The glossary
A glossary is a list of terms: character names, attacks, places, honorifics. Every entry can carry a gender and a voice note, and translations are stored per target language. During translation, matching terms are injected into the prompt as mandatory terminology — the model must use them.
Matching is smart: NFKC normalization, kana folding, Latin plural forms. You do not need to enumerate every declension.
Automatic detection
When a chapter’s OCR settles and the project uses an auto-detect glossary, Jokai scans the text and proposes recurring names and terms — one detection per chapter, per target language. Suggestions land as auto entries you can review, rename or dismiss.
- 1Enable auto-detection on the project glossary
- 2Upload and OCR a chapter — detection runs when OCR settles
- 3Review suggested entries in the glossary panel
- 4Re-apply glossary terms to already-translated blocks with one click
Shared context across pages
Glossaries lock names; shared context keeps the story coherent. In the Import Studio, the Translation context setting offers three modes: None (each page on its own), All pages (one rolling context for the chapter), or Manual groups — assign pages to groups from the toolbar.
Grouped pages translate sequentially with a rolling story summary and the previous page’s source→target pairs injected into each prompt, so references and tone carry across pages. A group waits for all its pages to finish OCR — it never translates half-cocked.
- 1In the Import Studio, set Translation context to All pages or Manual groups
- 2In manual mode, select pages and assign them a group from the toolbar
- 3Grouped pages show a G badge on the chapter page
- 4Translate — the group shares one rolling context
QA checks
Glossary terms are highlighted in the editor and the script panel, and QA checks flag blocks where a locked term is missing or mangled. Entry impact analysis shows where a term is used before you change its translation — rename once, fix everywhere.
Video coming soon
Watch: auto-detection to locked terminology (3 min)
Context groups (set at import) feed rolling story context — glossaries handle names, context handles tone and references.
Whole-chapter context
On top of context groups, whole-chapter mode translates every page as one continuous story: a rolling summary is checkpointed as the chapter progresses, and the previous chapter’s recap seeds the very first page. Names, tone and plot points stay consistent from the first panel to the last.
Identical source lines inside the run reuse the first translation — a repeated catchphrase reads the same on page 3 and page 30.
The consistency audit
When a chapter finishes translating, an audit compares every line against the glossary and the rest of the chapter: repeated source lines translated differently, drifted terminology, broken tone, empty translations. Deterministic checks run on every chapter for free.
Open the Consistency report drawer for the full findings with source and target side by side — retranslate a finding in one click, or dismiss it if the variation was intentional. A deeper LLM audit can be launched from the same drawer whenever you want a second pass.
