How to translate a textbook into your native language (PDF + EPUB guide)

Translating a textbook is not translating one paragraph eight thousand times. Layout matters. Footnotes matter. Equations, figures, glossary cross-references, the way the author refers back to a definition from chapter two — all of it carries meaning. Get the wrong tool and you end up with a translation that's technically complete and practically useless. I've watched friends abandon their study plans because of this. Here's how to avoid it.

Start with a clean source file

Translation quality starts at the source. Garbage in, garbage out applies harder here than in most places because errors at the OCR layer propagate into the translation, and then into your reading, and you'll never know whether the weird phrasing was the author or the pipeline.

EPUB is better than PDF when you have the choice. EPUB stores text as actual text, with structural metadata for chapters, sections, and footnotes that translation tools can work with directly. Modern textbook PDFs (anything published in the last ten years or so) embed their text and extract cleanly too. The trouble starts with scanned PDFs — books that are essentially photographs of pages rather than searchable documents. Those need an OCR pass before translation, and the OCR quality determines everything downstream.

Quick test: open your PDF and try to select a sentence with your cursor. If you can highlight it as text, you're fine. If your cursor draws a selection box but no text gets highlighted, it's a scan — OCR it first. Most modern tools (Translify included) do this automatically; if you're doing it yourself, Adobe Acrobat, ABBYY FineReader, and Tesseract all work, and 300 DPI is the minimum scan resolution worth bothering with.

Pick a tool that actually preserves layout

For a textbook specifically — not a novel, not an essay collection, a textbook with figures and equations and footnotes — layout preservation isn't a nice-to-have. It's the only thing that matters between “I can study from this” and “I give up after three chapters.”

The shortlist is small. Translify if you want the translation plus the reading experience around it — chat with the book, highlight-to-ask, quiz mode. DeepL Pro if you want the highest raw translation quality for European languages and you'll do your reading and studying elsewhere — its Document Translator preserves formatting cleanly and the European pairs are best-in-class. BookTranslate.ai is worth a look for books over 500 pages where terminology drift between chapters becomes a real problem; their multi-pass mode is slower but more consistent. Immersive Translateif you also translate web pages and videos and you want one tool that does all of it.

What to avoid: Google Translate's free document tier (reflows everything, loses your layout), older OCR-and-translate combos (they routinely mangle equations into nonsense by trying to OCR them as text), and anything that doesn't explicitly promise to preserve footnotes and figures. You'll waste an evening finding out the hard way.

Take a minute on the settings

Before you click translate, two things to think about.

First, your target language. Pick the one you read most fluently, which isn't necessarily your native language. If you grew up in Brazil but studied at MIT and now read English faster than Portuguese, translate to English. The point is comprehension speed, not nostalgia. I've seen people insist on translating to their native language out of principle and then read slower than if they'd stuck with the original — which defeats the purpose.

Second, the translation engine matters more than people assume, and it varies a lot by language pair. For European pairs (EN to or from DE, FR, ES, IT, PT, NL), DeepL is the best in class and Claude is close. For Chinese, Japanese, Korean to English, Claude or GPT-4o win; DeepL handles Japanese decently but is weaker on Chinese and Korean. Arabic to or from English is Claude or GPT-4o territory — DeepL doesn't support Arabic at all. For less-common pairs like Indonesian, Malay, Vietnamese, or Turkish, Claude and Google are both reliable. Translify picks automatically based on the pair; if your tool gives you the choice, choose deliberately.

One more setting worth knowing about if your tool supports it: glossaries. Textbooks are full of terms-of-art that the translator must render consistently across hundreds of pages. DeepL Pro lets you upload a glossary. Translify lets you pin term preferences. For a linear algebra textbook, telling the tool that Eigenwert should always be “eigenvalue” (and not “characteristic value” or “proper value,” both of which are technically correct and entirely confusing if used inconsistently) saves you 200 small papercuts.

Translate, then verify before you commit

Upload. Wait. Five to thirty minutes depending on length and tool. Get the translated file. Now — before you spend the next month studying from it — do three spot checks.

Pick a random page near the beginning, one in the middle, and one toward the end. Read them side-by-side with the source. If the translations read naturally and the technical terms are correct, you're done. Then pick an equation-dense page; equations themselves should be identical to the source (they're images or LaTeX, not translated), but the surrounding explanation text should read like a real sentence. Then check a footnote page — footnote markers should line up with the same paragraphs as the original, and the footnote text itself should actually be in your target language. (Sometimes tools translate the body and forget the footnotes. Worth catching now.)

If any of those checks fail, don't try to power through. Re-translate with a different tool or engine. The two hours it costs to redo the translation is much cheaper than the forty hours of confusion you'd get reading a bad one.

Studying from it

Translation gets you a readable book. Studying from it requires more, and this is where the right tool actually earns its place.

Keep the source language version accessible. Translify lets you toggle on any page; with other tools you'll need a second window open. You won't reach for it often — but when you hit a concept that feels wrong, you want the original right there, not three clicks away.

Use a highlight-and-ask flow on dense passages. Textbook authors compress on purpose — they're trying to fit decades of thought into one paragraph. Selecting a paragraph and asking the AI to explain it in plain language using the surrounding chapters is the single move that makes hard textbooks readable. It feels like cheating the first few times. It isn't.

Quiz yourself after each chapter. Five to ten multiple-choice questions with citations back to the page each answer came from. This is more useful than re-reading. Re-reading feels like learning; quizzing reveals what you actually retained. (If your tool doesn't do this, Anki works — it's just slower to set up.)

And keep a glossary of the field's vocabulary as you go. By chapter ten you'll have a bilingual term sheet that's more valuable than any single chapter of the textbook. Eventually you'll stop needing the translation altogether for that field. That's the win condition.

Things that will go wrong, and what they mean

Your pages got reflowed. You used the wrong tool. Re-translate with one that preserves layout. There's no fix downstream — this is a source-quality problem.

Footnotes ended up in the wrong place. Almost always an OCR-quality issue from a scanned PDF. Re-OCR at higher resolution, or find an EPUB version of the book if one exists.

Technical terms come out inconsistently from chapter to chapter. Your translation tool doesn't have a large enough context window to keep terminology stable across the whole book. Use a multi-pass tool (BookTranslate.ai) or one that supports a glossary (DeepL Pro, Translify).

Equations got translated into broken text. Your tool tried to OCR an equation image and butchered it. Either the source PDF stored equations as raster images (find a better source) or the tool is misdetecting equations as regular text (find a better tool).

The translation is technically correct but reads stiffly. Machine translation has a long-standing reputation for grammatical but unnatural prose. For dense technical content this barely matters — you're reading for information, not style. For narrative passages it matters more. Claude and GPT-4o give you the most natural reading; DeepL is the most accurate but sometimes wooden. Trade-off.

The whole thing in five lines

Get a clean source file (or OCR it). Pick a layout-preserving tool. Take a minute on settings — target language, engine, glossary. Spot-check the result before committing. Then study actively: source alongside, highlight-and-ask, quiz yourself, glossary as you go. That's it.

Try this with Translify — upload your first textbook free for 14 days, see if the workflow clicks for you. 30-day money-back on every paid plan.