II.6 Language and localisation
Three different languages in play
When you use a Documentize app, up to three languages are in the room, and they are independent:
Interface language -> the words on the buttons and messages
Document language -> the language of the text in your file
Output language -> the language the app writes in (AI apps)
Mixing them up is a classic source of confusing results. A Spanish contract can be uploaded through an English page and recognised as Spanish and summarised in German. Each language is chosen in a different place.
1. Interface language
The shared texts of the app pages are translated into many languages. The code has English plus 42 translations of the shared interface texts, among them Arabic, Chinese (simplified and traditional), Czech, Dutch, French, German, Hindi, Japanese, Korean, Polish, Portuguese, Russian, Spanish, Turkish, Ukrainian and Vietnamese. The list on the live site may differ; check the language picker.
The interface language changes labels, messages, FAQ and help. It does not change how your document is processed.
Some AI apps follow the interface language for what they write (Ask PDF answers in the language of the page you use; the Resume Analyzer report is written in the site's language). Others have their own picker.
2. Document language: what the recogniser must know
For recognition (OCR), the app has to know which alphabet and words to expect. Here the choice is yours and it matters more than any other setting:
| App | Option | Default | What happens if wrong |
|---|---|---|---|
| OCR Image | Document language, from a list of about 35 | English | Gibberish. The chapter calls it the commonest cause of garbage |
| Searchable | Document language | English | Wrong words that look right in the searchable file |
| Converter with OCR | Languages (tick all that are present) | English preselected | Scans lose or mangle text on the way to Word |
| Search on scans | Language for the recognition step | -- | "No results" does not prove absence |
| Merger with OCR | Language for recognition | -- | As above |
Practical rules:
- Always set the language to match the text in the picture, not the language of your interface.
- If a document mixes languages, tick or pick all that are present where the app lets you.
- Sideways or crooked pages are recognised badly whatever the language. Fix rotation and skew first.
- Right-to-left languages (Arabic, Persian, Hebrew) are laid out right-to-left in the AI apps that produce PDFs (Abstracts, Cheat Sheet, Resume report).
3. Output language of AI apps
Where an app writes new text, it usually offers a language picker:
| App | How the language is chosen |
|---|---|
| Abstracts | A Language option; the abstract can be in any language of the list, whatever the document's language. Default follows the page language |
| Cheat Sheet | A Language option; English by default in the engine |
| Ask PDF | No output-language selector; answers come in the interface language, and you may ask in a language different from the documents' (verify on the live app) |
| Checklist, Chat | See the chapters; check the output |
| SVG Maker | If your interface language is not English, the app asks that labels in the drawing be in that language |
An AI app can write fluent text in a language it handles less well. Have a speaker check translations when it matters. Trust and verification puts it plainly: fluent but wrong is a real failure mode.
Language of the file name and content
Documents with non-Latin names or text are not special as far as the flow is concerned. The practical risk is fonts: converting to other formats, or rebuilding pages (Deskew, Recovery), can substitute fonts, so check non-Latin text in the result.
Quick table: which language do I set?
| Situation | Set |
|---|---|
| Scan of a French invoice, I use the English site | OCR language = French |
| Spanish report; I want an English abstract | Abstracts language = English |
| Arabic book scan with some English words | Pick both where allowed |
| Ask PDF with documents in German; I write in Polish | Just ask; the answer comes in the page language |
Try it
Take a printed page in a language other than your interface language. Run OCR Image once with the wrong language and once with the right one. Count the errors. You will not forget the setting afterwards.