Translating a PDF without losing its layout

Why copy-and-paste breaks the layout

A PDF does not store paragraphs the way a word processor does. It stores positioned text: “draw these characters at this point, in this font and size”. Columns, table cells, headings and captions are only a matter of where things are drawn. When you select everything and copy it, that position is thrown away — two columns become one jumbled stream, table cells run together, and line breaks land in the middle of sentences.

A translator given that stream does its best with broken sentences, and you get back plain text that has to be laid out again by hand. For a one-page letter that is an annoyance; for a brochure, a form or a report with tables it is most of the work.

How a layout-keeping translation works

1. Read the page as blocks. Each block of text is found with its position, width, font size, colour and weight, and its lines are joined back into whole sentences so the translator sees real sentences.
2. Translate the sentences. With Translate PDF this happens on our own server with open translation models — the document is not sent to a translation company.
3. Take the old words off, keep everything else. Only the text objects are removed; drawings, colours, logos and photos stay exactly as they were, still sharp.
4. Write the translation into the same blocks. A translation that runs longer is allowed to use the free space beside it and is set slightly smaller if it still does not fit, so it never spills over the next block.

Translating into Arabic, the text is set right-to-left and aligned to the right; translating out of Arabic, it is set left-to-right.

What to check before you share it

• Names, numbers and dates. Machine translation sometimes “translates” a surname, reorders a date or drops a digit. Compare them with the original — Compare PDFs puts the two side by side.
• Legal, medical and official wording. A machine can get the general meaning right and a precise term wrong. For contracts, prescriptions, certificates or anything you will sign or submit, have a qualified translator check it.
• Small fixes. If one phrase comes out badly, you do not need to translate again: open the translated PDF in Edit PDF Text, click the line and type the better wording.
• Tone. Translations can sound flatter or more formal than the original. For a customer letter, read it aloud once.

Scans, long documents and other languages

A scan is a photograph of a page, so its words are first read with OCR, then painted out with the colour around them, and the translation is written on top. A sharp, straight scan gives a much better result than a photo taken at an angle; up to ten scanned pages are translated at a time.

Long documents are best translated a section at a time: split the PDF into parts, translate each, and merge them again. It is faster, and a problem in one part does not mean starting again.

Languages: English, Arabic, French, German and Spanish, in any direction (between two languages other than English, the translation goes through English).