CatScribe Docs

#Maintaining Consistency

Consistency is the difference between a rough draft and a translation that feels intentional. Long projects drift when names, tone, and repeated phrases are decided differently in each chapter or batch. CatScribe gives you several enforcement layers — this playbook shows how to combine them.

#The Consistency Toolset

Lever Where What it does
Glossary groups Glossaries screen, attached per job Enforce fixed translations for names and terms
"Verify Glossary Terms" File Translation options Post-translation double-check that glossary terms landed correctly
QA Flags CAT Editor Flags chunks where a glossary term was not translated as expected
"Find and Replace" Translations job detail Bulk-correct a wording decision across all chunks
Chunk Override CAT Editor Re-run individual chunks with different settings — see Chunk Overrides
Translation memory Automatic Chunks you approve are stored and reused as consistency context by AI refinement stages

#Practical Consistency Workflow

  1. Create one glossary group per book or series on the exact source → target language pair you translate with. A group created for en → pt will not attach to an en → pt-BR job.
  2. Add the obvious terms before the first translation: character names, places, organizations, titles, honorifics, invented vocabulary.
  3. Translate a small sample and review it. Every naming or terminology decision you make during review goes into the glossary immediately.
  4. Translate chapter by chapter (or file by file), attaching the group each time.
  5. After each batch, update glossary entries before continuing. A term fixed after chapter 3 costs one entry; the same term fixed after chapter 30 costs a re-run pass.
  6. Reserve chunk overrides for segments that must match exact full-sentence wording (mottos, recurring epigraphs, legal boilerplate).
  7. Run a final consistency pass across the whole project (see below).

#What To Track

Track decisions that readers will notice:

  • Character names and aliases.
  • Place names and organizations.
  • Honorifics, titles, ranks, and forms of address.
  • Invented terms, spells, technologies, or product names.
  • Recurring slogans, chapter titles, and signature phrases.
  • Tone decisions, such as formal vs informal voice (keep these in a notes file — a glossary stores terms, not register).

#Fixing An Inconsistent Translation

When you find a term rendered two different ways:

  1. Decide the preferred wording.
  2. Add or update the glossary entry so future runs are correct.
  3. Open the job in the Translations screen and use the chunk search ("Search source or target text...") to find every affected chunk.
  4. Use "Find and Replace" with "Replace in target text" for a mechanical swap — enable "Case sensitive" when the term could collide with common words. The modal previews how many chunks match before you commit with "Replace All".
  5. If the surrounding sentence needs rewording (not just the term), re-run those chunks via the "Chunk Override" panel or edit them by hand.
  6. Confirm the exported file uses the preferred form everywhere.

#Maintaining The Glossary Itself

The Glossaries screen has group-level maintenance tools:

  • "Replace" — find and replace inside a group's entries, with "Match case" and "Whole word only" options.
  • "Merge" — combine groups; matching terms are combined and the target group keeps its own values on conflict. Useful when consolidating per-chapter experiments into one canonical group.
  • "Stats" — entry counts and category breakdown per group.
  • "Export CSV" — back up your glossary. Note the CSV export carries the core columns only; part-of-speech, inflection flags, and variants are not included, so treat the app's database as the source of truth and CSV as an interchange format.

#Translation Memory

CatScribe stores a translation memory entry each time you approve a chunk in the CAT Editor — machine output never enters the memory on its own. Stored segments are then supplied as consistency context to the AI refinement stages of later translations. There is no memory panel to manage; the practical implication is simple: approving chunks is not just bookkeeping — it teaches the system your decisions. Approve as you review, especially for recurring phrasing.

#Outcome

Glossary-first setup, per-batch glossary updates, targeted re-runs instead of full retranslations, and consistent approval discipline keep chapters coherent across long projects — without redoing work every time a decision changes.