Multilingual recognition and text layout

Recognition

“Recognition language” defaults to “Auto”. Thai keywords use the Thai model, Korean keywords use the Korean model, and keywords with Vietnamese tone letters use the Vietnamese model. Digits and Latin letters follow whichever model is needed. You can also choose one language manually.

Script	Recognizer	Source
Simplified Chinese, Traditional Chinese, Japanese, English, and Latin-script languages such as French, German, Spanish, Portuguese, Italian and Indonesian	PaddleOCR PP-OCRv6	Built in
Thai	PaddleOCR	Built in
Korean; Cyrillic; Greek; Arabic (including Persian and Urdu); Devanagari; Tamil; Telugu; Kannada	PaddleOCR	Language pack (8–14 MB)
Vietnamese; Hebrew; Bengali; Punjabi; Gujarati; Odia; Malayalam; Sinhala; Lao; Khmer; Burmese; Tibetan; Georgian; Armenian; Amharic	Tesseract 5	Language pack (1–6 MB)

The table lists the scripts that have a recognition model; it does not mean every one has been tested. Character error rates, replacement results and usability levels on real scans for each script are in the OCR language support matrix for scanned PDFs: Simplified Chinese and English are primary, 8 scripts are usable with review, 8 are experimental, and Lao scans are not usable. The capability matrix keeps the four kinds of language information apart.

Text layout

• Thai and Indic scripts: HarfBuzz handles glyph composition, and vowels and tone marks are stacked according to the font’s layout rules.
• Arabic (including Persian and Urdu), Hebrew and other right-to-left scripts: the current version does not yet support searching, replacing or redacting by keyword in these scripts. If the text to find or the new text contains such characters, the job stops before anything is changed.
• Vertical Japanese and Chinese: vertical columns are detected automatically and searched top to bottom, right to left. New characters stay upright. Brackets and the long vowel mark “ー” are rotated for vertical text, and ”、” and ”。” are placed in the top-right corner of their cell. Tested scope: vertical Japanese was tested on 3 real scans (8 of 9 accepted); vertical Simplified Chinese is supported in code but untested on real scans; vertical Traditional Chinese pages with zhuyin are refused. Other scripts are horizontal only.
• Japanese furigana: with the “Japanese reading dictionary” language pack installed, replacing a word that has furigana generates the new reading from the dictionary. Without it, the old furigana stays unchanged and the result shows a notice.

Fonts

“Replacement font: Auto” picks a local font based on the script of the new text: for example SimSun for Simplified Chinese, PMingLiU for Traditional Chinese, a Mincho font for Japanese, Malgun Gothic for Korean, Tahoma for Thai, Nirmala UI for Indic scripts and Arial for Latin. The font menu lists these fonts if they are installed, and you can choose one manually.

If one replacement contains both Chinese and Thai, it needs a font that covers both scripts. If no such font is installed, you are asked to replace them separately.

FAQ

Which interface languages are available?

Simplified Chinese, Traditional Chinese, English, Thai, Vietnamese, Indonesian, Japanese and Korean. On first launch, the app follows the Windows display language.

Can I install a language pack without internet access?

Yes. Download it on a computer with internet access, then use "Import from file" in "Recognition language packs…". The SHA-256 is checked on import as well.

Can Thai text be searched after replacement?

Yes. Thai is written as vector outlines with an invisible layer of the correct text on top, so it displays correctly and can be searched and copied.