You translate a PDF and the output looks nothing like the original. The two-column layout collapsed into a single stream. The table became a wall of numbers with no column headers. The section numbering went missing.
This is the most common complaint about PDF translation tools — and it has a specific technical cause.
Why Formatting Gets Destroyed
PDF files do not store text as a flowing document. They store each character as an object with a position on the page. When a translation tool extracts text from a PDF, it reassembles those characters into a reading order — but spatial relationships like "these characters form a table row" or "this text is in column 2" are often lost in the process.
For scanned PDFs, the problem starts even earlier. The text is not stored as characters at all. It is an image of a page. OCR has to figure out what the text says, in what order it should be read, and which parts belong to which structure.
Bad OCR does not just produce wrong characters — it produces wrong structure. A two-column academic paper read in the wrong order produces alternating sentences from column A and column B, which is unreadable.
What Format Preservation Actually Requires
The translate PDF keep format workflow handles formatting in two layers:
OCR structure recognition: Before translation starts, OCR identifies not just the text but the structural units — paragraphs, table cells, numbered list items, section headers. These are kept as separate units through the translation step.
Bilingual side-by-side output: After translation, the original and translated versions appear next to each other on each page. This is the only way to verify that formatting survived. If the OCR output shows "Table: Revenue | Q1 | Q2 | Q3" and the translation shows the same structure, you know the table made it through. If it shows a collapsed list of numbers, you can catch that before the document is used.
The Specific Problem of Tables
Tables are the formatting element most likely to break in PDF translation. Here is why:
A table in a scanned PDF is a grid of lines and text. OCR sees the lines and text separately. If the OCR engine does not recognize that the text is inside a table, it reads the cells in order from top-left to bottom-right — which means all the content from row 1, then all from row 2, and so on — but without the structural markers that tell you which cell contains what.
When that gets translated, the column-row relationship is gone. What was a quarterly revenue breakdown becomes a list of numbers with no axis labels.
Format-preserving translation keeps table cells as discrete units. The translation fills each cell rather than treating the table as a block of text.
Multi-Column Layouts
Academic papers and news-style documents often use two-column layouts. OCR reads these in one of two ways:
Correct: Column 1 from top to bottom, then Column 2 from top to bottom.
Wrong: Alternating lines from Column 1 and Column 2, because the OCR read left-to-right across both columns simultaneously.
The second output is unusable. A sentence from the introduction of an article spliced with a sentence from the experimental results section is not a paragraph — it is noise.
The AI scanned PDF translator uses column detection to read multi-column documents in the right order before translation starts.
Checking Formatting in the Preview
Before processing a full document, use the free OCR preview to check three things:
- Reading order: Is the text extracted in the sequence you would read it on the page?
- Table structure: Are cells in the right rows and columns, or has the table been flattened?
- Section boundaries: Do chapter breaks and section headers appear in the right places?
If any of these look wrong in the preview, the full document will have the same problem. The scanned PDF translation workflow explains what a useful preview should show and what warning signs to watch for.
Common Document Types Where Formatting Matters Most
Financial reports: Balance sheets, income statements, and cash flow tables. A transposed column in a financial translation makes the document actively misleading. See the translate PDF documents guide for how financial PDFs are handled.
Legal contracts: Numbered clauses with sub-clauses. If the numbering gets shuffled or the indentation collapses, cross-references within the document become incorrect.
Technical specifications: Engineering data sheets with mixed text, diagrams, and measurement tables. The text and the figures need to stay associated with each other.
Research papers: Citations, footnotes, and figure references. A footnote that gets moved to the middle of a paragraph becomes a different document.
Bilingual Output as a Verification Tool
The bilingual side-by-side view is not just a convenience feature. It is a quality control tool. When you can compare original and translated text on the same screen, you can:
- Verify that translated tables have the same number of rows and columns as the original
- Check that numbered lists end at the same item in both versions
- Confirm that translated section headers correspond to the right sections
For long documents, this is especially important. The large PDF translator guide covers why page-by-page processing makes this review manageable even for 100+ page files.
What to Do When Formatting Still Breaks
Some documents are genuinely difficult: poor scan quality, overlapping text from stamps and handwriting, rotated pages, or tables with merged cells. For these, the preview will show the problem before you commit to the full file.
The alternatives overview covers when a different approach is appropriate and what to look for in a tool when your document has layout challenges that go beyond standard formatting.
Summary
Formatting breaks in PDF translation because tools extract text without preserving structure. The fix is not a better translation engine — it is better OCR that identifies structural units before translation starts, and bilingual output that lets you verify the result.
Start with a free preview. Check reading order and table structure. If those look right, the full document will too.