Translating a PDF without breaking the layout

A PDF is not a document, it is a printed page

In a Word file, text flows in paragraphs: change a word and everything adjusts. In a PDF, each fragment of text sits at fixed coordinates, with its typeface and size. There are no paragraphs, columns or tables — only pieces of text placed side by side, and a human reader who reconstructs them by eye.

Translating a PDF therefore means removing each fragment, replacing it with another of a different length, and putting it back in a box that hasn't changed size. Hence the three classic accidents: text overflows, text overlaps, or the tool gives up and spits out a shapeless text file.

Pitfall no. 1: the scanned PDF

A PDF from a scanner or a photograph contains no text: it's an image. No translation tool can access it without first running character recognition, which introduces its own errors — and on a table of figures, a recognition error is worse than a translation error.

The test takes two seconds: open the file and try to select a word with the mouse. If the selection catches a blue rectangle instead of words, your PDF is an image. Find the source file — the Word, the InDesign, the accounting export — before anything else. It almost always exists.

Pitfall no. 2: text expansion

The same sentence doesn't come out the same length from one language to another. Orders of magnitude, from French:

• into English: around 10 to 15% shorter ;
• into German: 20 to 30% longer, with compound words that don't break;
• into Spanish or Italian: 10 to 20% longer;
• into Chinese or Japanese: much shorter in characters, but each character is wider;
• into Arabic: reading direction reversed, which flips the entire layout.

An interface button, a table cell, a plan label: these are the first to break. If your document contains many of them, plan for a visual proofread, whatever the tool.

What Manuce does, and what it doesn't do

We read the PDF block by block: position, size, colour, weight, and table rows line by line. The source text is removed from its box, the translation is reinserted, the size reduced just enough if it overflows. Images, backgrounds, logos and structure are never touched. The original typeface is reused when it's embedded in the file and covers the characters of the target language; otherwise we take an equivalent with the same metrics, preserving bold and italic.

Two limits, stated frankly: scanned PDFs without a text layer are not supported; and into Arabic, Indian and Thai scripts, the original layout is not available — these languages are delivered in Word and in reformatted PDF.

If you have the source Word, PowerPoint or Excel file, send it rather than the PDF: the result will be better, and you'll be able to adjust it afterwards.

The two families of PDF, and why everything depends on them

A PDF is not a format, it's a container. Two apparently identical files can be radically different inside, and that determines what you can do with them.

PDF with text layer. Created by export from Word, InDesign, a word processor or accounting software. The text is present as text: you can select it with the mouse, copy it, search it. This is the favourable case — the text is extracted with its position, font, size and colour, and put back exactly in place after translation.

Image PDF. Created by a scanner, photograph or fax. It contains no text, only an image of text. No tool can translate it directly: optical character recognition is first required, which introduces its own errors.

The test takes three seconds: open the PDF and try to select a word. If the cursor doesn't catch it, you have an image PDF.

What breaks, technically, and how to avoid it

Four things break a translated layout, in this order of frequency.

Text expansion. A French text translated into German lengthens by 20 to 30%. In a fixed-size block, it overflows. The remedy is to reduce the font size slightly rather than let the text spill — this is what Manuce does, and this is why an eye must go over a brochure before printing.

Fonts. If the PDF embeds a font that does not cover the characters of the target language — Cyrillic, Greek or Chinese — the translated text appears as empty squares. You then need to substitute a font with equivalent metrics, preserving weight and italics.

Hyphenation. A word hyphenated at the end of a line in the original PDF is extracted in two pieces. Without rejoining them, the translation operates on two fragments that make no sense.

Tables. A table is, in a PDF, a series of text blocks with no connection between them. Translating them one by one loses the logic of the rows. They must be reconstituted into rows before translation.

Scripts where the original layout cannot be maintained

Certain scripts cannot be put back in place in a PDF designed for the Latin alphabet: Arabic and Hebrew are written right to left, Indian scripts form ligatures that change the width of words, Thai is written without spaces.

In these cases, Manuce does not promise what it cannot deliver: the document is supplied in Word and in reformatted PDF, and the unavailability of the option is announced at the point of upload — before payment, not after. This is a deliberate choice: better an informed client than a disappointed one.

The right reflex: start from the source file

If you still have the Word, PowerPoint or Excel file that produced the PDF, send it rather than the PDF. The result will be better, and you will get back an editable file instead of a frozen PDF.

PDF is a distribution format, not a working format. It is only translated well by default — when the source has been lost, or belongs to someone else.