OCR PDF With Multiple Languages (Free, No Upload)

A bilingual contract, a German invoice with English line items, a Spanish form with an Arabic name on it: real documents mix languages all the time. If you OCR a PDF with multiple languages but only tell the engine about one, it tries to force every word into that language’s alphabet — accents vanish, foreign words turn into look-alike nonsense and whole lines in another script come out as garbage. The fix is simple, but there are a few traps.

How do you OCR a PDF with multiple languages?

Select every language the document contains before you start, not just the main one. The OCR engine then loads a model for each and can recognise characters from all of them on the same page. In Slate PDF’s OCR tool you can pick up to three of 44 languages, and the scan stays on your device.

1. 1Open OCR PDF and choose your scanned PDF.
2. 2English is selected by default. Keep it if the document contains English; otherwise remove it. Then search for and add the other languages that appear — up to three in total.
3. 3Click Start OCR. Each language model downloads once; recognition then runs in your browser.
4. 4Download the searchable PDF and test it by searching for a word from each language.

Which language combinations work well?

Results depend far more on the scripts involved than on the languages themselves. Two languages that share an alphabet are easy; mixing directions or writing systems takes more care. This is what to expect from the engine behind Slate PDF’s OCR (the open-source Tesseract engine, running in the browser):

Why selecting more languages isn’t always better

It is tempting to tick every language that might appear. Don’t. Each extra model is another download and more work per page, so OCR gets slower. More importantly, every added language gives the engine more ways to misread a word: with English, Dutch and German all selected, a short smudged word has three dictionaries’ worth of plausible answers instead of one. That is why Slate PDF caps the selection at three.

• Choose the languages that actually appear in the text, not ones that only appear in a logo or a stamp.
• Add the main language first (remove the default English and re-add it later if English is only the secondary language); the first language is the engine’s primary model.
• If a document has one page in each of two languages, OCR each part separately with one language each — split it with Extract Pages first — and merge the results.
• A single foreign name in an otherwise English letter rarely needs a second language. A full paragraph does.

Edge cases worth knowing

Right-to-left scripts mixed with English

Arabic, Persian, Urdu and Hebrew run right to left, while numbers and English words embedded in them run left to right. The words themselves are recognised, but when a single line mixes directions, the order in which copied text comes out can differ from how you read it on screen. Searching for a word works well; copying a whole mixed paragraph into another program may need a quick tidy-up.

Serbian, Croatian and Bosnian

Serbian is written in both Cyrillic and Latin script. The Serbian model in the list is the Cyrillic one, so for Serbian (or Bosnian) printed in Latin letters choose Croatian, which covers the same č, ć, đ, š and ž letters.

Chinese, Japanese and Korean

CJK characters are dense, so resolution matters more than for Latin text: a character that is 10 pixels tall in a low-resolution scan has lost the strokes that tell similar characters apart. Rescan at 300 DPI or higher if you can. Japanese documents often mix kanji with some English; select Japanese first and English second.

A language that isn’t in the list

The tool offers 44 languages. If yours is missing but shares an alphabet with one that is listed — for example a Latin-script language with few special letters — the closest listed language will often recognise most words, but expect errors on letters it does not know. Check the output carefully before relying on it.

Check the result before you rely on it

1. 1Press Ctrl+F (Cmd+F on a Mac) and search for a distinctive word from each language. Every one should be found.
2. 2Copy a paragraph from each language into a text editor and read it. Look for missing accents and swapped letters such as rn for m, or 0 for O.
3. 3Check numbers, dates and names by eye — they are the costliest mistakes and no dictionary helps the engine with them.
4. 4If one language comes out badly, rerun OCR with that language listed first, or OCR those pages separately.

What the OCR tool does and doesn’t do

Being precise matters here. The output is a searchable PDF: each page is kept as an image and the recognised text is laid invisibly over it, so it looks exactly like your scan but can be searched, selected and copied. Every page is re-drawn this way, including any that already had text, so run OCR on scanned documents only. It is not a translator — if you need the content in another language, see how to translate a scanned PDF, which starts from this same OCR step.

Privacy-wise, the only thing that is downloaded is each language model, once; your PDF is recognised on your own device and never uploaded. That makes it a reasonable choice for passports, medical letters and contracts that you would rather not send to an online OCR service.