Get the text out of a PDF for translation, clean enough to quote and translate

What a translator needs from you

Editable text with its paragraphs intact, in a file their software opens. Many translators work in a CAT tool (computer-assisted translation software) that splits the text into sentences, remembers earlier translations and puts the translation back into the same file format.

memoQ, one of the common CAT tools, states the trade plainly in its documentation (checked on 11 October 2026): it imports a PDF by converting it to a Word file first, or as plain text, which loses all formatting and is not recommended; it cannot export a PDF; it cannot import a password-protected PDF; and it does not read text from scanned pages. Its advice is to translate the original document where possible. So a Word file you made and checked yourself is a better thing to send than a PDF the translator’s software has to guess at.

The workflow, step by step

1. Check the PDF in 1 minute. Drop it on Inspect. “Pages with text” should match the page count; “Scanned pages (image, no text)” should be 0; “Copy allowed” should say yes. If any of these is off, the check before converting guide says what to do.
2. Convert with PDF to Word. In the tool above, leave Running lines at “left out” and decide on Pictures: “put in” keeps photos and logos where they sit, “left out” gives a lighter file of text only. Press Convert to Word.
3. Read what was done. The result names it, for example “Wrote 4 headings, 12 paragraphs, 0 list items, 0 tables and 0 pictures as editable Word content”.
4. Check the numbers, names and dates against the PDF, side by side. Amounts, article numbers, addresses and spellings like “Čermák” are where a wrong character costs the most once it is translated.
5. Count the words in the converted file (next section).
6. Send the .docx, with the PDF as a reference for how it looked.

The cleanup the converter does for you

Raw text out of a PDF is the printed page line by line: every line ends in a break, a word split at the margin stays split, and the header and page number of every page sit in the middle of the text. A translator’s software treats each of those breaks as the end of a segment, so 1 sentence becomes 3 half-sentences to translate.

On our invented 4-page lease, 60 printed lines of text became 12 paragraphs and 4 headings in Word. The header and “Page 1 of 4” lines were left out on every page, and “land-” at the end of one line and “lord” at the start of the next came back as “landlord”. Clause numbers such as 4.2.1 or (a) come through as printed, each followed by a tab, so the translation keeps the contract’s own numbering.

What is left for you: tables (check that cells landed in the right columns), footnotes (they come out as ordinary text at the end of their page), and any text inside pictures, which is not text at all and is not extracted. Read the line break guide for the one join no converter gets right every time: words that carry a real hyphen at the end of a line.

The word count for the quote

If your translator prices by the word, count before you ask for prices. Count the converted Word file, not text copied straight from the PDF. Word shows the count in the status bar at the bottom of the window, and Review > Word Count opens the full figures (Microsoft Support, “Show word count”, checked on 11 October 2026). If the status bar shows no count, right-click the status bar and tick Word Count.

Why not count the raw text? On the 4-page lease, raw text out of PDF to text counted 1,132 words; the Word file 1,092. The 40 extra words are the running header and page number, 10 words a page. On a 40-page contract that would be about 400 words you pay to have translated and then delete.

If the translator quotes per character instead, the PDF to text result line gives the characters directly, for example “Took 21,891 characters of text from 500 pages, in the order a PDF reader finds it, as UTF-8”. That count includes headers and page numbers, so treat it as an upper bound.

When plain text is what they asked for

Some translators, and every machine translation box, want plain text. PDF to text gives you all the text as UTF-8, but it keeps 1 line per printed line, and the headers and page numbers stay in. For text with whole paragraphs, use PDF to Markdown instead: it joins the lines into paragraphs, leaves out the running lines, and marks headings with # and lists with -, which you can delete or leave in. Paste the paragraphs, not the lines, and machine translation stops breaking sentences in half.

Layout is the translator’s last step, not yours

A copy of the page layout with the source text in it is not what you want to send. The translated text rarely fits the same boxes, so the layout has to be redone around the translation anyway, by the translator or a designer, in the original program or in Word. What you can do is keep the PDF as the reference for how it should look in the end.

What this cannot do

A scanned contract or certificate has no text to extract. Run it through OCR PDF first, then convert. OCR guesses and reports its confidence per page; names, stamps and numbers on a scan deserve a careful check, and handwriting is weaker still. The scan guide has the details.

The converter takes text in the order the file stores it. Most contracts and letters store it in reading order; a 2-column layout drawn line by line across the page comes out with the columns mixed, and no setting here reorders it. Check 1 page of a complicated layout before sending the whole file.

Finally, the privacy part. Contracts, medical letters, birth and marriage certificates are exactly the files that are sent for translation, and exactly the files not to put through a converter that uploads them. Here the conversion runs in your browser tab and nothing is uploaded; the privacy guide explains how to check that yourself. Whether a certified translation needs the original paper or a scan is a question for the translator or the office receiving it; this guide does not give legal advice.