From the editorial desk · English
How to Translate an EPUB While Preserving Formatting
Preserving an EPUB means preserving a reader's ability to use the book: move between chapters, follow the text, recognise emphasis and see images where they belong. An output file with an .epub extension is only the start of that check.
The archive's Would You Rather book tag asks about paper and ereaders. The review of De win-winsituatie describes an ebook reading experience. When the text changes language, that same reading comfort still matters.
Record what the source edition contains
Make a working copy and list its chapters, images, footnotes, tables and special text. Mark a few passages with italics or unusual punctuation. Save the chapter list as readers see it. You need something to compare against after translation.
An EPUB packages content documents, resources and navigation. The W3C EPUB specification describes those relationships. A translation workflow may rebuild the package instead of editing the original documents in place. In that case, you must check what the rebuild kept.
What happened in our excerpt test
Method and ownership disclosure. About my bookshelf and Skillsfor.ai share an owner. On 10 September 2026 we used the real application upload controller, text extraction, translation backend and output assembly in an isolated QA environment. Sign-in was simulated; no payment was made. This tests file processing, not a customer's live checkout or queue time. No independent reviewer or ereader device panel was involved.
We constructed a small Alice EPUB from two English excerpt scenes. It had two chapter documents, two named navigation links, two heading elements, six emphasis elements, twelve prose paragraphs and two occurrences of one original diagnostic image. We also supplied an Austen excerpt as a minimal DOCX diagnostic file, plus a valid one-page PDF to test rejection. These are constructed test files, not untouched commercial editions or complete novels.
| Input and language | Upload / output | Sample / excerpt processing |
|---|---|---|
| Alice EPUB, English → Spanish | Accepted; EPUB returned | 9.19 s / 41.91 s |
| Alice EPUB, English → Dutch | Accepted; EPUB returned | 8.88 s / 44.52 s |
| Austen DOCX, English → Spanish | Accepted; TXT by default; EPUB and HTML also generated | 8.07 s / 32.92 s |
| Rejected at upload; no translation created | Not processed |
These are measured processing durations for very short files in this environment. They exclude a real customer queue and payment and should not be extrapolated to a novel. The application quoted $9.99 for each accepted file; the quoted amount was not charged.
| Feature | Source | Spanish and Dutch output |
|---|---|---|
| Chapter documents | 2 | 1 |
| Table of contents | “Opening”, “Next scene” | One link labelled “1” |
| Heading elements | 2 | 0; heading words became ordinary text |
| Emphasis elements | 6 | 0 |
| Images shown | 2 occurrences | 2 occurrences, with identical image bytes |
| Paragraph elements | 12 prose paragraphs | 16 text paragraphs, including unwanted titles and CSS |
| CSS visible in the text | No | Yes |
The source image was duplicated into two output image files; both hashes matched the source. Its two markers remained in order between the corresponding paragraphs in the translated text. That establishes what happened to these two occurrences, not every possible illustration layout. Image alternative text was not preserved in the generated EPUB.
The most visible failure was a stylesheet fragment entering the prose: em{color:#123f47} p{margin:1em 0}. It came from the EPUB document's head. The original titles also became body text. Counting sixteen extracted and sixteen translated blocks would therefore have given false reassurance: the input had only twelve prose paragraphs to begin with.
Conclusion for these files: images survived, but the original reading structure and emphasis did not. The returned EPUBs required cleanup, chapter reconstruction and proofing before publication. DOCX acceptance did not produce a DOCX result, and the default TXT result could not carry the illustrations.
Download the evidence package (ZIP): test inputs, extracted text, unedited translations, generated files, observations, hashes and method. Text sources are Alice's Adventures in Wonderland and Pride and Prejudice; the diagnostic image is our own. Our novel study looks at selected language problems in the same results.
Translate the text while keeping a structural reference
- Keep the source package. Preserve a clean master and the original chapter list. Never make the only copy your working translation file.
- Inspect extraction before trusting a sample. Look at the beginning and end of each section. Remove stylesheet text, duplicate titles and other material that is not meant to be read. Confirm that captions and notes are intentional.
- Keep a chapter map. Match each original section to its target section. A heading that merely looks large is different from a real heading linked from the contents.
- Review language and rebuild the reading file. Restore emphasis where its meaning calls for it; it may not attach to precisely the same words in the other language. Place images with their relevant passages and provide appropriate alternative text.
- Regenerate navigation and metadata. Confirm the translated title, language, chapter labels and reading order. Do not retain a plausible-looking original title list that points to the wrong locations.
- Proof the actual output. Reopen the file you intend to distribute, not an earlier preview. A later conversion can introduce a new problem.
If you cannot inspect and rebuild the package, include that work explicitly in your formatting brief. Ask for an EPUB proof with a working table of contents and a record of any unsupported features. A generic promise to “preserve formatting” does not tell you which of your book's features were checked.
A reader's final check
Open every chapter from the contents. Change type size. Move through an image and its caption. Follow each note link and return. Search for two repeated names and one unusual term. Check the last sentence of each chapter against the approved translation. Inspect the file in the reading applications you intend to support.
For technical validation, use an EPUB conformance checker as an additional check. A technically valid package can still contain the wrong words, flat dialogue or a misleading chapter list. It complements editorial proofing rather than deciding whether the edition is good to read.
Download the EPUB reading checklist (CSV). It separates completeness, navigation, visual features and language metadata so a successful image check cannot hide a failed chapter check.
Return to Book Translation for Authors to connect file preparation with language review and publication, or explore International Books from the reader's perspective.