PDF is a presentation format, not a translation format — which makes it the hardest common file type to translate cleanly. This post walks through why, the three viable approaches, and the quality checks worth running on every PDF translation. Aimed at teams handling contracts, proposals, IR docs, and product manuals.
Why PDF translation is hard
1. Absolute layout and font dependencies
PDFs position elements on absolute page coordinates. Japanese and English render at different lengths (English is typically 1.3–1.5× longer for the same content), so direct substitution overflows boxes, breaks line wrapping, and shifts pagination.
2. Text PDFs vs scanned PDFs
Two very different beasts under the same extension:
- Text PDF — characters embedded directly. Selectable. Tools can read it cleanly.
- Scanned PDF (image PDF) — paper scanned to image. Requires OCR before any translation.
Open the file: if you can select text, it's the first; if not, the second. OCR accuracy caps the achievable translation quality on scanned files, so handle them separately.
3. Tables, diagrams, and text inside images
Chart labels, flowchart boxes, org charts, captions inside images — most PDF translation tools either skip them or destroy the layout while translating them. Critical for IR decks, training materials, and product specs.
Three approaches to PDF translation
A. Extract to Word, translate, re-export
The traditional path. Convert PDF → Word with Adobe Acrobat, translate, re-export. Layout breaks badly and you spend hours reformatting.
Use when: content matters more than layout (memos, meeting minutes). Avoid when: contracts, proposals, IR — anywhere look matters.
B. AI translation that preserves layout
Modern AI translation tools (BizHonyaku included) translate while preserving the original PDF layout — text blocks, tables, lists, headings stay structurally intact.
- Pros: almost zero reformat work. Output looks like the original.
- Cons: very complex layouts (multi-column with dense figures) may still need touch-up.
This is now the default approach for business document translation.
C. OCR first, then translate (for scanned PDFs)
Scanned PDFs need OCR (optical character recognition)before any translation tool can work on them. Accuracy depends on:
- Clean printed text → 95%+ OCR accuracy
- Low-resolution scans, mixed handwriting → 60–80%
- Old documents, copy-of-copy → can drop below 50%
Always spot-check OCR output before translation — misrecognition propagates straight into bad translation.
Five quality checks to run on every PDF translation
1. Numbers, dates, proper nouns
Layout shifts can split numbers or change digit counts. Reconcile financials, contract amounts, and dates against the source before signing off.
2. Page numbers and TOC consistency
Different language length = different pagination. Verify table-of-contents page numbers and any "see p. 5" references still resolve.
3. Table cells and column widths
Cell widths are fixed; longer English text overflows. Visually inspect every table after translation.
4. Headers and footers
Company name, classification stamps, page numbers — decide whether these get translated, replaced, or left as-is. Many tools silently skip them.
5. Attachments and exhibits
When a parent PDF references separate exhibit PDFs, exhibits often go out untranslated. Maintain a checklist of every attached file and confirm each is translated before delivery.
Recommended approach by document type
- Contracts (PDF): Approach B + mandatory legal review
- Proposals / sales decks: Approach B. Touch up text inside images by hand.
- IR / annual reports: Approach B with native review on the final draft
- Manuals: Approach B with strict glossary enforcement
- Old scanned documents: Approach C, then Approach B
How BizHonyaku translates PDFs
- Text blocks, tables, lists, and heading hierarchies preserved
- Custom glossary applied to lock proper-noun and term translations
- Source ↔ target parallel view for clause-by-clause review
- One-page preview free (watermarked) before you commit
Scanned PDFs are also supported; for very low OCR-confidence files, contact us for handling.
Summary
PDF translation quality comes down to three variables: (a) PDF type (text vs scanned), (b) layout complexity, (c) post-translation QA process. Layout-preserving AI tools have made the old "convert to Word, translate, re-export" workflow obsolete for most use cases.
Start with a one-page preview to validate quality on your specific documents, then move to production volume.