Translated PDF Accessibility: Language, Tags, Reading Order, and Alt Text

A translated PDF needs two separate reviews: does it communicate the correct meaning on the page, and does its underlying structure communicate that meaning to assistive technology? A screenshot can help answer the first question. It cannot answer the second.

Start with the final translated file, not the source's accessibility report. Check its document language, language changes within passages, semantic tags, logical reading order, image alternatives, and keyboard navigation. Then test it in the PDF reader and assistive-technology setup your audience uses.

The most useful rule is repair the document's meaning and structure, not just its appearance. Our downloadable example makes that distinction visible: two PDFs render identically, but one contains deliberately incorrect accessibility information.

This guide owns the accessibility checks after translation. For missing text, clipped tables, substituted fonts, and page-by-page visual inspection, use the separate PDF translation QA checklist. For the broader source-to-output process, see how to translate a PDF while keeping its format.

Two identical pages, three hidden differences

Download these one-page teaching files:

• Before: Spanish text with incorrect language, order, and alt text.
• After: the same page with those three properties corrected.
• Complete structural inspection results and the reproducible Python generator.

The page contains two columns of invented Spanish instructions about preparing and reviewing a document. The left column contains steps A1 and A2; the right contains B1 and B2. A folder-to-document diagram sits below them. An English interface phrase, “Read only,” remains intentionally untranslated.

We authored this synthetic example for this article on September 7, 2026. It is not a customer document, a BookTranslator output, or a translation-quality benchmark. The generator uses ReportLab 4.4.9 and pypdf 6.10.0. It creates both PDFs, reopens them, and reads the actual document catalog, tag tree, marked-content text, and figure alternative.

The inspection produced these results:

We also rendered both files with Poppler at 100 dpi and compared their 850 × 1100 pixel images. The rendered images were identical. Correcting the structure did not move a single visible element.

Here is the abbreviated tag-order trace, read from each PDF rather than inferred from its screenshot:

Before:
1. Revisar el archivo
2. Revisar la entrega
A1. Confirme que el archivo contiene todas las páginas.
B1. Compare los títulos con el archivo original.
A2. Guarde una copia del original antes de editarlo.
B2. Revise las notas antes de compartir el resultado.

After:
1. Revisar el archivo
A1. Confirme que el archivo contiene todas las páginas.
A2. Guarde una copia del original antes de editarlo.
2. Revisar la entrega
B1. Compare los títulos con el archivo original.
B2. Revise las notas antes de compartir el resultado.

This is a structural trace, not a recording or transcript of a screen reader. We did not run an assistive-technology test on these files. Actual speech, navigation, and language switching depend on the reader, screen reader, installed voices, and settings. The “after” file corrects the illustrated properties; it is not a certified accessible template or a PDF/UA conformance example.

That limitation matters: the fixture proves that appearance and these underlying properties can differ independently. It does not prove how every reading application will interpret them.

Check language metadata at two levels

Set the language of the translated document

Translating visible words does not establish that the PDF's language metadata changed with them. Inspect the final file's default language. In the example, the Spanish page starts with a document-level /Lang value of en-US; the corrected value is es-ES.

W3C's PDF16 technique describes the document catalog's /Lang entry and ways to inspect it. Correct language information helps supporting software choose appropriate pronunciation rules. It is an instruction available to the software, not a guarantee that a particular voice is installed or selected.

Use the language variety that accurately describes the document. Do not select a region just because it is the first option in a menu. If the translated file is predominantly Spanish but its cover retains an English book title, the title does not make English the document's default language.

Mark passages that genuinely remain in another language

A bilingual PDF needs more than one language setting. The default describes the main language; an enclosing tag or passage can override it for text in another language. W3C's PDF19 technique explains this narrower language assignment.

In our corrected example, “Read only” has an explicit en-US value while the surrounding text inherits es-ES. That is more precise than changing the whole paragraph to English to accommodate two words.

Apply the same reasoning to a facing-page bilingual report: inspect the English and translated text separately. A single file-wide language choice cannot describe both sides accurately. For a quotation retained in its original language, mark that passage; do not relabel nearby translated commentary with it.

During a listening test, record a pronunciation issue before deciding its cause. A correct tag with an unavailable language voice calls for a different response than a Spanish paragraph incorrectly marked as English.

Inspect the tag tree, not only the numbered order overlay

A tagged PDF exposes structure such as headings, paragraphs, lists, and figures. Tags should describe what content is, not merely where it sits. A large bold line is not necessarily a navigable heading unless the structure identifies it appropriately. W3C's PDF9 technique covers heading tags.

For a translated document, work through these questions:

1. Does a tag tree exist, and does it include the meaningful target-language content?
2. Do headings identify the actual section hierarchy?
3. Are list items, table cells, captions, and figures attached to the right structures?
4. Does the order express the intended argument or procedure?
5. Did translation or export introduce duplicated, empty, or misplaced structures?

The answer is not always “left to right.” Our Spanish fixture intentionally finishes the left column before starting the right. A bilingual comparison might need paired source and target blocks; a sidebar might belong after the paragraph it qualifies. Decide the intended sequence from meaning, then inspect whether the final structure expresses it.

W3C's PDF3 technique explains that tagged-PDF reading order is determined primarily by tag order, while content within a tag also has its own order. It specifically identifies multi-column layouts as a potential failure point.

That is why a page can have the right number of tags and still be wrong. In our “before” PDF, every text block is present, but the tree interleaves two separate procedures. Reordering those elements fixes the illustrated sequence without altering the page layout.

Copying a page into a text editor is a useful warning test, not final acceptance. Extraction tools can follow different paths through a PDF. Likewise, an order overlay is a useful editing view, but check the resulting tag tree and the actual reading experience after saving. Adobe's Reading Order tool documentation distinguishes reading-order work from more detailed tag repairs.

If the editable source exists, correct its heading styles and structural relationships before generating the next PDF. Otherwise, the next export may discard a careful one-off repair. Reserve direct PDF remediation for defects that the source workflow cannot reliably carry through.

Translate the information outside the visible body text

Accessibility information can carry language even when it does not appear on the page. Include figure alternatives, descriptive link text, form-field tooltips, document titles, and bookmarks in the localization review where those elements exist.

For a meaningful image, the target-language alternative should convey its purpose in context. Translating the visible caption alone does not update the figure's alternative. W3C's PDF1 technique describes the /Alt entry used for image alternatives and the need to convey meaningful words embedded in images.

Our diagram's English alternative remains hidden in the “before” file even though every visible instruction is Spanish. The corrected Spanish alternative describes the same folder-to-document relationship. It is not a claim that the diagram itself was machine-translated.

Two additional checks make this review more useful:

• A translated chart with unchanged alt text: compare the alternative against the final chart, not just the source caption. If labels or units changed, a fluent translation of an outdated description may still mislead.
• A translated form with an English tooltip: the visible label may be localized while the field's accessible explanation remains in English. Check the actual control's properties and its keyboard interaction, not only the printed label.

For tables, confirm that translated row and column headers retain their relationships to the data. For links, check both the translated description and the destination. These are structural and functional checks, not a request to translate identifiers, URLs, or every proper name.

Section508.gov's PDF remediation module treats table headers, form tooltips, alternatives, language properties, reading order, and tab order as separate repair tasks. Keep them separate in the defect log too: “translation looks good” does not close any of them by itself.

Use a failure matrix to decide who fixes what

Download the translated PDF accessibility review sheet. It separates the check, evidence, responsible role, and acceptance result so an untested item cannot disappear into a general “approved” label.

The following matrix is our suggested handoff model, not a conformance checklist:

The translator or editor owns linguistic meaning. The document author or remediator owns structural repairs. A reviewer using the intended reading setup checks whether the result is usable. One person may perform several roles, but the evidence should still distinguish them.

For scans, start earlier: OCR must recover accurate text before the accessibility review can judge the target document. The scanned PDF translation guide covers that preparation. Selectable text is a prerequisite for many workflows, not proof that headings, relationships, and language have been represented correctly.

Run the final check in the delivered file

Freeze the exact PDF you intend to distribute. Record its filename, revision or checksum, target language, and intended reading environment. Run the automated checks your authoring or remediation tool provides, then resolve its manual-check items. Adobe's accessibility-check documentation explicitly distinguishes automatic results from items requiring manual verification.

For the manual pass, use a short task sequence:

1. Start reading from the document title. Confirm that the opening content and main language make sense.
2. Navigate by headings. Reach a specific section without visually hunting for it.
3. Read a difficult page continuously. Include columns, a sidebar, or footnotes; compare the observed sequence with the intended one.
4. Encounter an image in context. Check whether its alternative supplies the information needed at that point.
5. Cross a language change. Listen with the required language support available and record what actually happens.
6. Navigate interactive elements with the keyboard. Where links or forms exist, verify focus order, descriptions, and completion of the intended action.
7. Repeat the affected tasks after the last repair or export. Retain evidence for that final file, not an earlier revision.

Write down the operating system, PDF reader, assistive technology, versions, relevant voice settings, tested pages, observations, and unresolved defects. Do not replace an observed result with an expected transcript. If no screen-reader test was possible, say that the structural review is complete but that part of acceptance remains untested.

Our downloadable fixture deliberately contains no interactive controls or tables. It cannot validate those tasks for your publication, and its inspection script is not a general-purpose accessibility checker. A public report with forms, complex tables, or contractual accessibility requirements needs an appropriately scoped review beyond this example.

Where BookTranslator fits

Use BookTranslator's PDF translation workflow for the language-conversion stage, then review the downloaded file as a new deliverable. When the input is scanned or image-heavy, OCR mode reconstructs translated content rather than promising an exact copy of the original layout.

Neither translated text nor preserved appearance establishes accessible PDF structure. BookTranslator does not guarantee PDF/UA or Section 508 conformance. Plan for dedicated authoring or accessibility-remediation tools and the relevant human checks when the document must meet those requirements.

If an editable source is available, keep it with the translation and repair notes so future revisions do not restart the process from a flattened PDF. The handoff is complete when meaning, structure, and the intended reading experience have each been checked on the file you will actually distribute.