#Translating Books

Book translation is a long project. The main challenge is not only translating each sentence, but keeping names, voice, terminology, and formatting consistent from beginning to end. This playbook walks through the full pipeline in CatScribe: glossary preparation, upload configuration, the translation run, CAT Editor review, and export verification.

#Before You Start

• Prefer EPUB as the source format when you have a choice. CatScribe re-injects translated text into the original chapter structure, so styling, images, and inline formatting survive export. See Working With EPUB, DOCX, and PDF.
• File Translation accepts EPUB, DOCX, and PDF up to 50 MB.
• Decide the target language variant up front (for example pt vs pt-BR). Glossary groups are bound to an exact language pair, so switching variants mid-project disconnects your glossary.
• If you plan to use AI refinement or quality scoring, install the pieces first: local AI models on the Providers screen, and optionally COMET scoring (a ~1–2 GB download enabled in Settings with "Enable COMET quality scoring").

#Step 1: Build The Glossary First

Translate nothing until the obvious terms are captured. This is the single highest-leverage step for book quality.

1. Open Glossaries and use the "Local" tab.
2. Create a group with "Create Group" — one group per book or series, on the exact source → target pair you will translate.
3. Add entries with "Add New Entry Manually" or "Import CSV". The CSV columns are: Source Term, Target Term, Source Language, Target Language, Category, Group, Gender, Number.
4. Capture at minimum: character names, place names, organizations, titles and honorifics, and invented terms.

You do not need the glossary to be complete — you will add terms during review. But every term added before translation is one you never have to fix.

Full glossary reference: Glossaries.

#Step 2: Configure The Job

1. Open File Translation and upload the book ("Upload Document" or drag and drop).
2. CatScribe analyzes the file and shows a "Document Advisor" card with a star rating, workflow recommendation, and analysis chips (page count, whether OCR is needed, layout notes). Read it before proceeding — it flags scanned PDFs and layout-heavy documents early.
3. Choose your configuration style:
   • Auto Mode ON (recommended for most books): pick a "Quality" preset and set source language, target language, and output format. Everything else — engine, chunk size, AI refinement layers — is selected automatically from what is installed. See Auto Mode.
   • Auto Mode OFF: pick the provider, chunk size, glossary groups, and AI refinement stages manually.
4. Attach your glossary group in the "Glossary" selector. If the selector shows no groups, your language pair does not match the group's pair — fix one or the other.

#Choosing a quality preset

Higher presets use local AI models for validation and rewriting, which is slower and uses more RAM. The "Recommended setup" line under the preset reflects your hardware.

#Step 3: Translate In Batches

For a first book, translate a sample before committing to the whole file: upload one chapter (or a short excerpt saved as its own file), run it end to end, and inspect the export. This validates engine choice, glossary coverage, and formatting survival at minimal cost.

For the full run:

1. Start the job and watch the "Translation Progress" panel — it shows the current chunk, active layer, provider, and retry count.
2. Long books can run for hours. Enable "Prevent system sleep while translating" in Settings for overnight runs.
3. Failed chunks retry automatically ("Auto-Retry Failed Chunks" is on by default). Chunks that still fail are not lost — you handle them in review.
4. Manage the job from the Translations screen: "Pause" / "Resume", "Retry Failed", "Retry All", and "Generate Document" when translation completes.

#Step 4: Review In The CAT Editor

Open the finished job in the CAT Editor from the Translations screen ("Open CAT editor") or from Home recents.

Work in passes, not chunk by chunk from front to back:

Practical notes for book-length review:

• Chunk statuses are "Draft (Machine)", "Edited (Human)", and "Approved". Any edit promotes a chunk to Edited automatically; approval is your explicit sign-off.
• Use Alt+↓ to jump to the next unreviewed chunk. The sidebar shows "Review progress" as approved/total.
• Low-scoring chunks are your priority queue. Chunk scores blend quality estimation and glossary compliance ("Score = (2×Quality + Glossary) / 3"); with COMET enabled, scores of 85+ are rated Excellent and 70+ High. Sort your attention accordingly rather than re-reading everything.
• When a chunk's engine output is fundamentally wrong, use the "Chunk Override" panel to re-run just that chunk with a different engine — see Chunk Overrides.
• Add every terminology decision you make to the glossary as you go. Approved chunks also feed CatScribe's local translation memory, which the AI refinement stages reuse for consistency in later work.
• There is no autosave — click "Save" (or approve) before switching chunks.

#Suggested Passes Summary

#Production Tips

• Treat each pass as a separate quality gate. Fixing terminology before style review prevents repeated rework.
• Keep a plain-text notes file for style decisions a glossary cannot capture (register, dialect, how to render a narrator's tics). Consistency of decisions matters more than any single decision.
• For a series, reuse the same glossary group across books — groups are global per language pair, not per project.