How to Translate a Scanned Chinese PDF to English

To translate a scanned Chinese PDF to English, the workflow that holds up is: OCR tuned for Chinese text, translation applied to whole passages rather than broken lines, and re-typesetting that puts the longer English back into the original page layout. Each of those steps has a Chinese-specific failure mode — word segmentation without spaces, red official stamps sitting on top of the text, dense tabular documents, and English translations that physically don't fit where the Chinese was. This guide goes through each one and what to do about it.

What makes Chinese scans harder than they look

No spaces between words

Chinese is written without spaces, so "words" exist only in the reader's segmentation of a continuous character stream. This has two consequences for scanned documents. First, a single misrecognized character doesn't produce a visible typo the way it does in English — it silently changes which words the sentence segments into, so the error propagates. Second, when OCR emits hard line breaks, the translator can't use spaces to tell whether a break falls inside a word, because there were never any spaces to begin with. A translation pipeline for Chinese has to reassemble full sentences before translating, or quality collapses at every line boundary.

Simplified versus traditional

Documents from mainland China use simplified characters; Taiwan, Hong Kong, and Macau use traditional; older documents and inscriptions on seals often use traditional forms regardless of origin. Many character pairs look nothing alike (广/廣, 与/與), so OCR configured for the wrong script family misreads systematically rather than occasionally. If a document mixes both — a mainland contract quoting a Hong Kong entity's registered name, say — the OCR needs to handle both on the same page. Before translating anything, identify which script family you're holding: a document from the mainland is simplified, one from Taiwan or Hong Kong is traditional, and a quick check of a distinctive character (广 vs 廣) settles it.

Red stamps over the text

Chinese official and business documents are validated by red circular company seals — 公章 — and they are stamped directly onto the text almost by design: over the company name, the amount, the signature line, the date. For OCR, red ink over black text is occlusion; the characters underneath come back wrong or not at all, and they're reliably the most important characters on the page. Related stamps (contract seals, finance seals, the fapiao seal on invoices) mean a single page can carry several overlapping red regions. Whatever workflow you use, treat every stamped region as unverified until you've checked it against the source image.

Dense tables everywhere

The Chinese documents people most often need in English — customs declarations, VAT invoices (发票), business licenses, bank statements, bills of lading, product spec sheets — are mostly table. Tight grids, merged header cells, two-line cells, and columns whose meaning depends entirely on position. Extract that as plain text and you've destroyed the information: a column of amounts means nothing once separated from its row labels. For these documents, preserving the table isn't cosmetic; it's the content.

English doesn't fit where the Chinese was

Chinese is dense. A four-character phrase can need a ten-word English clause; 有限公司 becomes "Company Limited." Translated English routinely runs substantially longer than the source, which means it physically overflows the table cell, form field, or column the Chinese occupied. (Occasionally the reverse: a long formal phrase collapses to one English word and leaves a hole.) Any tool that claims to preserve layout has to solve this — refitting and reflowing the translation into the original space — or the output has text colliding across cell borders.

The workflow, step by step

1. 1Get a clean scan. Straight pages, 300 DPI or equivalent, even lighting for phone photos. Chinese characters carry more strokes per glyph than Latin letters, so blur costs proportionally more accuracy. Clear printed pages OCR well; heavily handwritten material — common on older certificates — is unreliable for any OCR and should go straight to a human reader.
2. 2OCR with sentence reassembly. The OCR must handle simplified and traditional forms, keep table cells associated with their rows and columns, and rejoin line-broken text into complete sentences before anything gets translated.
3. 3Translate with context. Chinese omits plurals, tense, and often subjects; the English rendering of a clause depends on the sentences around it. Passage-level AI translation resolves this well; line-by-line translation of an unspaced language does not.
4. 4Re-typeset into the original layout. The English goes back where the Chinese was — resized and reflowed to absorb the expansion — so the invoice still reads as an invoice and every amount stays in its cell.
5. 5Verify the stamped regions and the numbers. Check every red-seal area against the source image, and spot-check amounts and dates. Chinese documents also use formal numerals (壹贰叁 instead of 一二三) on financial documents precisely to prevent alteration — confirm those converted correctly.

Doing it without the manual steps

Steps two through four are what Reglyph automates for scanned and photographed documents. OCR reads the page, the original Chinese is erased from the page image, and the English translation is typeset back in place — so the seals, the table grid, the figures, and the numbers stay exactly where they were, and what you download is a finished PDF rather than extracted text. That matters most on precisely the stamped, table-heavy documents this post is about: the 公章 stays visible on the output because the page image is preserved, and the translation is fitted back into the cells the Chinese occupied. For review, the bilingual side-by-side export puts the original and the English next to each other, which makes checking stamped regions and amounts fast. It runs in any browser including mobile, the first 5 pages are free with no credit card required, and paid use starts at $5.

Quick reference: document types and their traps

• Contracts (合同) — company seals on every signature page; check party names under the red ink, and expect traditional characters inside seal impressions.
• VAT invoices (发票) — dense fixed-format tables plus a dedicated invoice seal; the amounts appear in both Arabic and formal Chinese numerals and must match.
• Business licenses (营业执照) — single dense page, large seal, long registered names that expand dramatically in English.
• Customs and shipping documents — bilingual headers are common; make sure the tool translates the Chinese cells without mangling the English ones already there.
• Certificates (birth, marriage, degree) — often a mix of printed template and typed or handwritten entries; the handwritten fields need human verification.

None of these is exotic. They're the everyday paperwork of doing business with or emigrating from the Chinese-speaking world, and each one is tractable once the workflow respects what Chinese pages actually look like: unspaced text, red ink over the important parts, and tables that are the document.