Which file to send for translation: DOCX, PDF or EPUB

The question sounds technical, but it determines a large part of the outcome. The format you send decides what survives: the text always, the structure usually, the page design sometimes. Here is how that plays out.

DOCX (a Word file) — the right default

In a DOCX file, text is text, and that is the whole difference. Headings are marked as styles, paragraphs are preserved as paragraphs, notes and lists stay in place. A translated book can come back looking like the original, and it remains easy to edit afterwards. If you have the source file of your book, this is the one to send.

Two things are worth confirming before you send it. That headings are marked with a heading style rather than merely set in large bold type, and that page breaks are actual page breaks rather than twenty presses of the return key. Both affect directly how the translated book will look.

EPUB (the standard ebook format) — good, with a caveat

An EPUB is essentially packaged HTML, so its text is accessible and its structure explicit. It suits books that have already been published digitally. The caveat is that its appearance is governed by a stylesheet, so it renders differently in every reading app — less suitable when pagination is part of the point.

PDF — possible, but an extra stage

PDF was designed for display, not processing. Text sits in it as positioned glyphs rather than a dependable sequence. Correct extraction is work in itself, and on some files it loses reading order, breaks apart hyphenated words, or drags running heads and page numbers into the body text.

A book scanned from paper is a separate case: it contains no text at all, only images, so character recognition is required, and that adds its own layer of errors — especially in languages with diacritics or a non-Latin alphabet.

What to check before sending

Four quick checks. Whether any text exists only inside an image, such as a caption on an illustration or a photographed table. Whether the file is the final version rather than a draft with tracked changes. Whether some parts are not meant for translation, such as a dedication or bibliography, and if so, marking them. And whether the names and terms you care about are written down somewhere, even as a short list.

Books with heavy design

Cookbooks, children's books, textbooks and anything with text sitting on artwork are a category of their own. The text is short, the layout carries the meaning, and translated text rarely occupies the same space as the original — some languages run longer, and a right-to-left language reverses the direction the page was built around. Decide early whether you want the design rebuilt around the new text, or the new text fitted into the existing design, because the two lead to different files and different work.

And the format you get back

That is a separate question from the one you sent. A book for screen reading, a book for print and a file for further editing are not the same thing, so say up front what you need: an editable file for more work, a file for digital readers, or a file ready to upload to a store.

We cover what happens to the text along the way in translating a book with AI, and checking the result in how to check a translation you cannot read.