Receiving a Scanned PDF written in a language you do not speak is more common than ever in global business, academic research, and legal document review. You need to extract the text, translate it, and understand what the document says. The challenge is not just running OCR on the file. It is running OCR with the correct language settings so the extracted text is accurate enough to translate. Selecting the wrong language during OCR PDF processing produces garbled output that no translation engine can salvage.
OCR engines do not see letters. They see shapes and match those shapes against character patterns for a specific language. If you tell the OCR engine the document is in English but it is actually in Russian, the engine maps Cyrillic shapes to the closest Latin characters it knows. The result is a string of random-looking letters that bears no relationship to the original text. Getting the language setting right, or using auto-detection where available, is the single most important decision in the entire OCR-to-translation pipeline.
WukongPDF's PDF to Text and OCR tools extract text from scanned documents for translation and analysis. Pairing the correct OCR language selection with a reliable translation service turns an unreadable foreign-language scan into an understandable document in under five minutes.
Step 1: Identify the Document Language
Before you touch any OCR software, identify the document's language with certainty. If the document metadata or filename hints at the language, start there. If not, open the PDF and look at a few pages. Even without reading the text, visual cues narrow the possibilities significantly. Script families are distinctive: Latin script serves most European languages. Cyrillic is used for Russian, Ukrainian, Bulgarian, and Serbian. Arabic script covers Arabic, Persian, and Urdu. CJK covers Chinese, Japanese, and Korean. Devanagari covers Hindi, Marathi, and Nepali.
If you cannot visually identify even the script family, take a screenshot of a representative paragraph and upload it to Google Translate's image translation feature or a reverse image search. These tools identify both the script and the specific language in most cases. Knowing that the document is Thai rather than Lao, or Korean rather than Japanese, makes the difference between accurate OCR output and useless character noise.
Step 2: Select an OCR Tool That Supports the Target Language
Not all OCR tools support all languages. Adobe Acrobat Pro ships with OCR engines for about 40 languages. Its Recognize Text tool includes a language selection dropdown in the settings dialog. Tesseract, the free open-source OCR engine maintained by Google, supports over 100 languages through downloadable language data files. Google Cloud Vision API has the widest language coverage at over 200 languages and includes automatic language detection as an optional feature.
For a document in a single language you can identify, any of these tools works as long as you specify the correct language before running OCR. For a document that contains multiple languages, such as a bilingual contract with clauses in both English and French, you need a tool that either supports multiple simultaneous languages or can auto-detect on a per-paragraph basis. Google Cloud Vision and Tesseract with the appropriate language data packs both handle multi-language documents, though the accuracy on language-switch boundaries is never perfect.
Step 3: Configure and Run the OCR With the Correct Language
In Adobe Acrobat Pro, open the scanned PDF, go to Tools, then Scan and OCR, and select Recognize Text, then In This File. Click the Edit button next to the language selector. In the Recognize Text dialog, choose the correct primary language from the dropdown. If the document contains a second language, click the Add Language button and select it as well. Acrobat processes both languages simultaneously. Click OK, then Recognize Text. Acrobat runs the OCR pass and embeds the recognized text as an invisible layer behind the scanned image. The document is now searchable.
In Tesseract, the command-line invocation specifies the language with the -l flag: tesseract input.pdf output -l rus for Russian, tesseract input.pdf output -l jpn for Japanese, or tesseract input.pdf output -l fra+eng for a bilingual French-English document. Install the required language data files first; Tesseract ships with English only by default. For Google Cloud Vision, the API call includes a languageHints parameter that accepts an array of BCP-47 language codes. Cloud Vision also offers an automatic language detection mode that requires no language hint at all, making it the simplest option when you genuinely do not know the document's language.
Step 4: Copy the Extracted Text and Translate
After OCR completes, verify the text quality before sending it to translation. In Acrobat, open the Find tool and search for a common word you recognize in the document. If the search finds matches, the OCR worked. Select a paragraph of the extracted text, copy it, and paste it into a plain text editor. Scan for obvious OCR errors: repeated symbols, words that break in the middle, or large blocks of unrecognized characters. If more than 5 percent of the text appears corrupted, re-run the OCR with a different language setting or a higher resolution setting.
Once the extracted text passes a visual sanity check, copy the entire text and paste it into a translation tool. Google Translate, DeepL, or Microsoft Translator all accept plain text input. For longer documents, DeepL and Google Translate support direct file upload of searchable PDFs, which skips the manual copy step. The translation output is good enough for comprehension, research, and internal business use. For legal, medical, or publication-grade accuracy, send the searchable PDF to a professional human translator, who will use the OCR layer as a starting point and correct the machine translation against the original scanned pages.
Handling Documents With Mixed Scripts or Non-Standard Fonts
Documents that mix scripts, such as an Arabic research paper that includes English technical terms in Latin script, confuse OCR engines that expect a single script. Configure the OCR for the dominant script first, then run a second pass targeting the minority script on the same document. Merge the results by exporting both OCR layers and combining them in a text editor. The accuracy on the minority script passages will be lower because the engine is optimized for the primary script.
Non-standard fonts, including decorative typefaces, old typewriter fonts, and handwritten-like cursive fonts, reduce OCR accuracy even with the correct language selected. Pre-process these documents by converting the PDF pages to high-resolution images at 400 DPI or higher, applying a sharpening filter, and increasing contrast before feeding them into the OCR engine. Tesseract's built-in preprocessing, enabled with --psm 3 for automatic page segmentation, handles moderate font variation. For severely degraded or stylized text, the Google Cloud Vision API generally outperforms local OCR engines because its models have been trained on a larger corpus of real-world font samples.
| OCR Tool | Multi-Lang Support | Auto-Detection | Best For |
|---|---|---|---|
| Adobe Acrobat Pro | 40+ languages | Manual selection | Professional documents, high accuracy |
| Tesseract (open source) | 100+ languages | Manual selection required | Batch processing, scripting, free |
| Google Cloud Vision | 200+ languages | Auto-detect available | API integration, mixed scripts |
| Online OCR services | 10-50 languages | Manual selection | Quick single-document OCR |
Verifying Accuracy Before Acting on Translated Content
OCR followed by machine translation introduces two layers of potential error: OCR misrecognition and translation ambiguity. For critical documents, spot-check the output by asking a bilingual colleague or a professional translator to compare a randomly selected paragraph of the translation against the original PDF. A five-minute verification catches catastrophic failures like a wrong language setting that produced plausible-looking but entirely incorrect text.
If you regularly handle foreign-language scanned PDFs, build a checklist: identify the language, select the matching OCR engine, run OCR with the correct language parameter, spot-check the extracted text for character accuracy, translate, and verify a sample paragraph. After two or three documents, the workflow becomes routine and takes less than five minutes from opening the file to reading the translated result.
When it comes to document processing, for security-sensitive documents, offline OCR avoids cloud exposure entirely. Tesseract runs locally with no network requests after you download the language data files once. Adobe Acrobat Pro also performs OCR locally via the Recognize Text feature without signing in to Document Cloud. For confidential foreign-language materials such as legal discovery or sensitive business correspondence, local OCR paired with offline text review keeps original document content within your controlled environment.
When OCR Alone Is Not Enough: Assisted Transcription Services
For documents where OCR accuracy is unusably low due to extreme font degradation, handwritten text, or heavily mixed scripts, assisted transcription services provide human-verified output. These services combine an initial machine OCR pass with human review and correction, typically charging per page. Turnaround ranges from a few hours to a few days. For a single critical document where accuracy is non-negotiable, such as a foreign-language birth certificate or a patent filing, the cost of human transcription is justified by the risk of acting on incorrectly recognized text. Run machine OCR first to assess the quality. If more than 15 percent of words are unrecognized or obviously wrong, escalate to assisted transcription rather than spending hours manually correcting machine output.