PDF Translation QA Checklist: Layout, Tables, Fonts, and Missing Text

A translated PDF passes quality assurance only when its text is complete and accurate, its logical reading order still works, and every page remains usable at the intended screen or print size. A file that opens without an error has passed almost none of those checks.

Use this order:

1. Freeze the exact source and target files.
2. Compare page and content-block completeness.
3. Inspect reading order and source-to-target meaning.
4. Stress-test tables, fonts, figures, links, headers, and page boundaries.
5. Review difficult pages at normal size and high zoom.
6. Log defects by severity and search for the same pattern throughout the PDF.
7. Recheck the final exported file after fixes.

Download the PDF translation QA checklist to record page-level evidence and acceptance status.

Establish the acceptance target first

PDFs serve different jobs. Define the intended use before judging the output.

A visual facsimile may be the priority for a form or designed brochure. Logical reading order may dominate for an accessible report. Both matter for many ebooks and public documents, but the acceptance weights are not identical.

Phase 1: prove the source and target belong together

Record the source filename, revision, checksum or immutable link, page count, language, and whether the file contains selectable text. Record the same details for the translated PDF.

Then compare:

• first and last page;
• page count and page labels;
• section and heading sequence;
• table, figure, footnote, appendix, and attachment counts;
• headers, footers, watermarks, form fields, and annotations;
• visible links and destinations; and
• intentionally excluded content.

Different page counts are not automatically wrong: translated text can reflow or a production workflow can insert pages. But every difference must be explainable. If the target has fewer pages, do not assume it is more efficient; prove that content was not dropped.

For scanned sources, verify the OCR layer before evaluating translation. The scanned PDF translation workflow covers skew, image quality, language selection, suspect characters, handwriting, and mixed printed/handwritten pages.

Phase 2: detect missing and duplicated text

Missing text is often less visible than awkward text because there is nothing to attract the reviewer's attention. Use structural comparisons before line editing.

Check:

• title, subtitle, author, revision, and legal pages;
• every heading and list item;
• the first and last paragraph on every page;
• text inside shapes, diagrams, callouts, captions, and sidebars;
• table headers, cells, notes, and continuation labels;
• footnote markers and footnote bodies;
• cross-references, citations, URLs, and page references;
• headers, footers, stamps, and marginal content; and
• appendices and pages that contain little text.

Search for repeated opening phrases to find duplicated segments. Search for source-language residues, but maintain a do-not-translate list for names, formulas, code, trademarks, and quoted material that should remain unchanged.

When the PDF permits text extraction, compare normalized source and target block counts as an alarm, not a quality score. Language expansion, ligatures, OCR, and segmentation can change counts. Any large or localized discrepancy should send a reviewer to the corresponding page.

Phase 3: verify logical reading order

A two-column PDF can look correct while its extracted text alternates between columns. Screen readers, copy and paste, search indexing, and downstream translation can then receive the wrong sentence order.

Run these checks:

1. Select and copy a multi-column section into a plain-text editor.
2. Move through headings, paragraphs, lists, figures, and notes with the keyboard in a PDF reader that exposes navigation.
3. Inspect the tag or reading-order panel when accessibility is required.
4. Compare the order around floating images, sidebars, and footnotes.
5. Repeat the test on the translated PDF; reconstruction can create a new order defect even when the source was correct.

For every failure, identify whether the defect originated in source extraction, translation segmentation, or target reconstruction. Fixing visual coordinates will not necessarily fix the logical order.

Phase 4: inspect layout at the stress points

Do not review only the attractive pages. Force coverage of the structures most likely to break.

Text expansion and clipping

Target text can be longer or shorter than the source. Inspect narrow columns, buttons, labels, table cells, figure callouts, footers, and headings near page boundaries.

Reject:

• text cut off at a frame edge;
• overlapping lines or objects;
• headings stranded from their first paragraph;
• a single line pushed onto an otherwise empty page without a deliberate design choice;
• text reduced below the agreed readable size; and
• concealed overflow that becomes visible only when selected or copied.

Review once at normal reading size and again at 200–400% zoom. Normal size exposes hierarchy and usability; high zoom exposes overlap, broken glyphs, and rasterized text.

Tables

Tables combine meaning and geometry. Compare row and column headers, cell order, units, decimal separators, footnotes, repeated headers, and page breaks.

Check that:

• every source row and column has a target counterpart;
• values remain attached to the correct labels;
• minus signs, ranges, percentages, and units are preserved;
• cells do not overflow, disappear, or cover borders;
• split tables repeat the correct header; and
• the logical extraction order matches the visible table.

A beautifully aligned table with one value under the wrong heading is a major content defect.

Figures, diagrams, and captions

Verify that every figure remains present, legible, and paired with the correct caption. Text baked into an image may be omitted from a text-only translation workflow. Decide whether it must be localized, rebuilt, or explicitly left in the source language.

Check cross-references such as "see Figure 8" after pagination changes. Confirm alt text separately when the delivery requires accessibility; a visible caption is not automatically a useful alternative description.

Headers, footers, and page furniture

Inspect running heads, page numbers, dates, version marks, confidentiality labels, logos, and footers. These elements are frequently processed separately from body text and can disappear, duplicate, or remain untranslated.

Phase 5: test fonts and scripts

Font failures include more than obvious empty boxes. Inspect:

• accented characters and combining marks;
• Arabic and Indic shaping;
• Hebrew and Arabic bidirectional behavior;
• CJK punctuation and line breaking;
• mathematical symbols, currency, superscripts, and subscripts;
• bold, italic, small caps, and fallback styles;
• copied text compared with the visible glyphs; and
• embedding status when the PDF will be printed or distributed.

A fallback font may preserve readable characters but change line length enough to clip or reflow the page. Check every style, not only body text. Search the whole document after finding one missing or substituted glyph because the same font is likely reused.

For Arabic and Hebrew outputs, continue with the RTL book translation checks; direction, numbers, URLs, tables, cover order, and page progression require a dedicated pass.

Phase 6: verify links, forms, and interactive elements

Click every unique link type and sample repeated links across the file:

• table-of-contents entries;
• internal page and section references;
• external URLs;
• email links;
• footnote and return links; and
• bookmarks.

Confirm that translated visible text still describes the destination. A working hyperlink with a mistranslated label is not correct.

If the source contains forms, decide whether the translated deliverable is expected to remain fillable. Test field order, labels, validation, checkboxes, signatures, calculations, and saved values. Flattening can be intentional, but it must be an explicit delivery decision.

Classify defects by release consequence

The ISO 5060:2024 overview covers analytic translation evaluation using error types and configurable penalties and discusses sampling. PDF release QA also includes production, structure, accessibility, and interaction checks beyond translation-output evaluation.

Sample deliberately, then escalate

Even when a complete visual review is impossible, do not select only random pages. Include:

1. First page, last page, table of contents, and every section opening.
2. The densest page and the page with the least text.
3. Every distinct table, figure, note, form, column, and list structure.
4. Pages with uncommon scripts, symbols, names, dates, and numbers.
5. A page before and after every major layout transition.
6. Any passage where an error would cause material harm.

Escalate the scope when defects cluster:

• one cosmetic defect: log it and continue;
• one major defect: search the complete PDF for the same pattern and inspect adjacent pages;
• two major defects of the same type: treat the issue as systematic until disproved; and
• any blocker: stop acceptance and rerun all checks affected by the cause.

Final translated-PDF release gate

Accept the PDF only when:

• the source and target versions are unambiguous;
• all expected pages and content structures are present;
• reading order works for the intended use;
• tables preserve labels, values, units, and structure;
• no required text is clipped, overlapping, or unreadably small;
• fonts cover every required glyph and style;
• figures, captions, notes, links, and bookmarks work as specified;
• every blocker and major issue is fixed and rechecked;
• the final export—not an intermediate file—received the last inspection; and
• the issue log and acceptance decision are stored with the deliverable.

Use the broader book and long-document QA checklist for meaning, terminology, and sampling rules. Use translate a PDF while keeping its format to choose and test the production workflow. BookTranslator PDF Translator can produce a translated PDF for supported files; the page-level checks above determine whether that result is ready for its intended use.