Translate PDF: every page read and translated on your device

What this tool does, exactly

Every page is read and translated in this browser tab; nothing of the document is sent anywhere. The page's words come out in reading order — straight from the file when it carries real text, through the OCR reader when it is a scan — and your browser's own translator, in Chrome 138 or Edge 148 and newer on a laptop or desktop, turns each block into the language you choose. The original and the translation sit side by side, page by page, with no page cap.

What it does

Reads each page into blocks — paragraphs, headings, table rows as lines — in reading order, and passes each block to your browser's translator. The translation is shown beside the original and can be edited before you copy or download it. Page 1 is read first; Translate all pages carries on from the page you are on, and a page you move to that is not translated yet says so. The downloads are the translation as plain text with the page breaks marked, and as Markdown with a heading for each page.

Where the translation happens

In the translator built into your browser, on this device. The first time you translate a pair of languages, the browser downloads that pair's language pack from its own vendor, and the status line shows the download. That request is the browser's, not this page's, and it carries nothing of your document. There is no translation service of ours to send a page to.

Which browsers

Translating needs Chrome 138 or Edge 148 and newer, on a laptop or desktop. Anywhere else — Safari, a phone, an older browser — the page says so before you choose a file, still reads the text out of the PDF, and leaves it ready to copy into any translator you trust.

What it cannot do yet

It does not put the translation back into the PDF with the layout kept, and it does not make a bilingual PDF: both are behind the buttons under the result, in preview. A table comes back as rows of text rather than a grid. A scanned page is read in the scripts the reader knows — the Latin alphabet, Chinese, Japanese, and Russian, Ukrainian or Belarusian Cyrillic. The reading of a scan is a guess and the translation is another, so check anything that matters against the original.

What it costs to load

A PDF with real text is read without downloading the reader; a scan is read by the same reader the OCR tools use. That reader is about sixteen megabytes the first time, cached by your browser afterwards, and reads an A4 page in about 0.6 seconds on a graphics chip, measured on an Apple M4 Mac. A Cyrillic scan needs a second recogniser, eight megabytes more, fetched when you choose Russian, Ukrainian or Belarusian in From. The language pack is the browser's own download.