How to Prepare a Book File for Translation
Prepare an EPUB, DOCX, or PDF for translation by freezing the source version, cleaning revisions, inventorying assets, defining terminology, and establishing a verifiable acceptance baseline.

Prepare a book for translation by creating one frozen, complete source package and an acceptance baseline. Resolve revisions, identify what should and should not be translated, inventory chapters and assets, define names and terminology, and choose the source format that preserves the structure you need in the target.
Do not begin with a file named final-final-2.docx and explain missing decisions later. Translation multiplies source ambiguity across every target language.
Use this preflight:
- Freeze the exact source edition.
- Remove or deliberately retain comments, revisions, and hidden content.
- Choose the best source format.
- Inventory front matter, chapters, back matter, notes, images, and tables.
- Mark translatable and do-not-translate content.
- Build the initial terminology and entity list.
- Package fonts, image sources, references, and instructions.
- Record counts and representative test passages.
- Define what makes the translated file acceptable.
Download the book translation source-preflight template and complete it before uploading the full file or handing it to a translator.
Freeze one authoritative source edition
Create a source record with:
- title and subtitle;
- author and edition;
- source language;
- filename and revision;
- date frozen;
- chapter and section count;
- word count or another stable content baseline;
- included front and back matter; and
- owner authorized to approve source changes.
Keep later corrections in a change log. Each entry should identify the source location, old content, new content, reason, date, and affected target editions. Do not overwrite the frozen source while a translation is in progress.
If you cannot explain which file is authoritative, stop. A translation team cannot reconcile versions you have not distinguished.
Resolve comments, tracked changes, and hidden content
For DOCX manuscripts, decide whether every tracked insertion or deletion is accepted or rejected and whether comments belong in the translation package. Hiding markup is not removal. Microsoft's current Word guidance states that tracked changes must be accepted or rejected to remove them; the “No Markup” view only hides them temporarily.
Create two files when editorial history is useful:
- a clean translation source with resolved text; and
- a read-only reference copy that preserves comments or revision history.
Also inspect hidden text, document properties, headers, footers, footnotes, endnotes, text boxes, alt text, and embedded objects. If the file contains private or unpublished material, run the document translation privacy checklist before sharing or uploading it.
Choose the source format by structure, not convenience
| Available source | Use it when | Prepare before translation | Main limitation |
|---|---|---|---|
| EPUB | The book is reflowable and chapter structure matters | Validate package, TOC, language metadata, links, notes, and images | Fixed-layout or unusual CSS may need separate testing |
| DOCX | The manuscript is editable and styles, tables, or comments matter | Resolve revisions, normalize styles, inspect text boxes and headers | It is not the final e-reader or print artifact |
| Text-based PDF | The PDF is the authoritative fixed-layout source | Verify selectable text, reading order, fonts, and page inventory | Editing and image text can be difficult |
| Scanned PDF | No better source exists | Run OCR on representative pages and record recognition risks | OCR output reconstructs content rather than exact layout |
Use the most upstream editable source you are authorized to use, and retain a reference export showing the intended appearance. Converting a clean EPUB or DOCX into PDF before translation often throws away useful structure.
For image-only material, use the scanned PDF translation preflight before treating OCR text as the source.
Inventory the complete book, not only the body chapters
List every content unit and its intended treatment:
- cover text;
- title and copyright pages;
- dedication, epigraph, and acknowledgments;
- table of contents;
- introductions and forewords;
- body chapters;
- tables, figures, captions, and image-internal text;
- footnotes and endnotes;
- bibliography and citations;
- appendices, glossary, and index;
- author biography and calls to action; and
- metadata and store description handled outside the interior file.
Assign a stable ID to each chapter, figure, table, and note. Record whether it is translatable, retained, replaced, recreated, or excluded. A missing appendix should be visible in the inventory before it becomes a silent omission in the target.
For EPUB, the W3C specification requires publication resources to be listed in the package manifest and uses package metadata for information such as title and author. The EPUB 3.3 specification is useful here because the package is a structured publication, not a folder of interchangeable text files.
Mark what should not be translated
Create a do-not-translate list for items such as:
- author and contributor names when their approved form is unchanged;
- trademarks and product names;
- URLs, email addresses, identifiers, and ISBNs;
- code, commands, file paths, and formulas;
- cited titles that must retain the source form;
- invented words awaiting an editorial decision; and
- quotations that should use an existing authorized translation.
Add a reason and context for each item. “DNT” without explanation can preserve the wrong string or suppress content that readers actually need translated.
The W3C Internationalization Tag Set includes data categories for whether content should be translated, terminology, and contextual information. Its ITS 2.0 requirements illustrate the underlying principle: translatability and context are explicit metadata, not guesses left to each translator.
Build the first terminology and entity inventory
Do not wait until chapter 12 to decide how a recurring name or concept should be handled. Start with:
- character, place, organization, and product names;
- titles and forms of address;
- defined technical or academic terms;
- invented vocabulary;
- abbreviations;
- repeated interface or procedural labels;
- units and measurement conventions; and
- terms that intentionally change by context.
For each entry, include source form, definition or context, approved target form when known, rejected variants, capitalization or inflection rules, first location, and decision owner.
The initial list does not need to predict every term. It needs to prevent obvious drift and provide a place to record later decisions. Use the book terminology consistency template as the working asset.
Package external assets and rights information
A book file may depend on materials that are not fully embedded:
- high-resolution image sources;
- editable diagrams;
- font files and license records;
- cover artwork and text layers;
- linked media;
- existing approved translations;
- permissions for quotations, lyrics, or images; and
- style guides or publisher specifications.
Do not assume that permission to publish the source automatically covers a translated edition in every market. Record the rights owner and approved scope for material that will be reproduced or adapted.
For fonts, identify the target scripts before translation. A Latin-only font can look fine in the source and fail completely for Arabic, Japanese, or Devanagari output. Provide the licensed font or an approved fallback instead of leaving the production team to improvise after translation.
Create a baseline that makes omissions detectable
Record practical counts before translation:
- chapters and major sections;
- headings by level;
- paragraphs or text blocks when the format exposes them;
- figures, tables, captions, notes, and links;
- TOC entries;
- front- and back-matter components; and
- source-language word count as a reference, not a target-language equality rule.
Target word count will change by language, so do not reject a translation merely because it is longer. Use counts to find missing, duplicated, or unexpectedly empty structures.
For EPUB, run a conformance check on the source. EPUBCheck is maintained by the DAISY Consortium on behalf of W3C and checks EPUB conformance. Fixing source-package errors before translation prevents confusion about whether the translation process created them.
Select representative passages before full translation
Choose a small test pack that exposes the book's real difficulty:
- opening prose;
- dialogue or voice-heavy passage;
- terminology-dense section;
- table, figure, footnote, or cross-reference;
- long paragraph and short fragmented text;
- culture-specific or ambiguous passage; and
- late chapter that tests continuity with earlier decisions.
Use the same pack to compare workflows and establish the style guide. Do not let a vendor or tool prove quality only on the easiest first page.
The test is not a claim that the complete book will behave identically. It is a cheap way to reject a bad source format, glossary, or review process before scaling it.
Define acceptance before translation starts
Write down:
- target language and locale;
- intended use: private reading, editorial draft, internal research, or publication;
- output format;
- required bilingual or target-only version;
- approved names and terminology;
- structures that must remain usable;
- expected human review;
- defect severity and escalation rules; and
- who approves language, subject matter, and final production.
If public release is the goal, include native-language editing and final-format proofing. If personal reading is the goal, completeness and usable meaning may be sufficient. The book translation QA checklist can become the acceptance log for the completed target.
Where BookTranslator fits
Once the source package is stable, upload the complete supported file through the BookTranslator workflow. EPUB, PDF, and DOCX require different checks, so translate the representative test pack or inspect a difficult section before committing to the full review plan. Use the automatic glossary for recurring names and terms, and use bilingual output where supported to inspect important passages.
BookTranslator does not resolve source ambiguity, approve rights, license fonts, or decide which revisions belong in the book. Those are source-preparation decisions. For scanned PDFs, OCR mode prioritizes recoverable translated content over exact original-page reproduction.
Final source-package gate
The book is ready for translation only when:
- one authoritative source revision is frozen;
- revisions, comments, hidden content, and exclusions are resolved;
- the best available EPUB, DOCX, PDF, or OCR path is selected;
- every chapter, asset, note, and back-matter item has a treatment;
- DNT and terminology decisions have owners;
- source assets, fonts, references, and rights information are packaged;
- structural counts and representative passages are recorded; and
- target locale, intended use, output, review depth, and acceptance criteria are explicit.
Good source preparation does not guarantee a good translation. It removes avoidable ambiguity and gives reviewers evidence when something goes missing.
Related Posts





