Scanned PDF translation: OCR and layout checks

A scanned PDF needs text recognition before a text-based translator can work with its sentences. The useful sequence is source inspection, OCR, a small translation sample, then a checked output copy. Translation and page reconstruction deserve separate checks: readable wording does not establish that every label, number and footnote survived.

This guide focuses on page images, including mixed files that contain both selectable text and scans. It describes a practical review workflow rather than a translation-quality benchmark. The right outcome depends on whether the result is for private reading, an editable note or a document that must resemble the source.

The difference between an image and a text layer

A page that looks like a paragraph can still contain only image data. Adobe's OCR documentation explains that recognition creates selectable, searchable text from a scan. That text layer is an interpretation of the image. It needs an accuracy check before translation.

A useful inspection starts with a body paragraph, a caption and a table cell. If selection captures no characters, OCR is a likely prerequisite. If selection works but pasted text is nonsense, the existing text layer also needs attention. A successful copy from the cover does not establish that the remaining pages are text-based.

The original scan should remain the reference throughout. An OCR error and a translation error require different corrections. The first changes the source supplied to the translator. The second changes how a correctly recognised source passage is expressed in the target language.

The output choice before processing

Three output goals call for different acceptance checks.

A reading aid can tolerate a simpler layout if the source location remains clear. A translated handout may need its captions and table headings close to the material they describe. Choosing the output first prevents time being spent restoring page furniture that has no role in the actual task.

For an archival workflow, retain a short record of the source filename, source language, selected pages and processing date. That record can live in an ordinary note. Its purpose is to distinguish the original from later corrected or translated copies.

OCR before translation

A sensible sample contains difficult material: a faint paragraph, small print and a page with multiple columns. The review should compare recognised text directly with those regions. Names, decimal separators, negatives and subscripts deserve explicit attention because a plausible sentence can conceal a changed character.

Adobe documents page-range and recognition-language controls in its OCR workflow. These are separate from choosing the translation language. Correct recognition is about reading the source script; translation is about expressing that source in another language. Adobe also recommends reviewing the recognition result.

In Oyren, scanned material can become editable notes through OCR. The scanned PDF guide describes page selection and the credit estimate. Preparing text for PDF editing and converting a PDF into a Markdown note are separate operations. A note conversion should not be described as a guaranteed replacement text layer in every exported PDF.

Tool support differs for scans

A file extension is not a capability check. Google Translate's desktop document help states that text in images and scanned PDF pages appears in the output but is not translated. An unchanged scan inside an output file therefore does not show that its words were processed.

DeepL's PDF help discusses scanned documents and warns that scan quality can affect the translation. It also suggests using the original editable document when a PDF translation is unsatisfactory. These are product-specific statements, not a universal ranking between translation engines.

The practical selection question is whether the chosen workflow supports this source and this output. A trial on the actual difficult page is more useful than assuming that general document support includes every kind of scan. Current account requirements and processing limits also belong in that check.

A small translation sample

The first translation should be narrow enough to inspect completely. A representative passage with a heading, a technical term and a reference provides a useful starting point. The review can then separate three questions: whether the source was recognised correctly, whether the meaning survived and whether the output fits.

A terminology note helps make decisions explicit. It can record the original term, a preferred target term and any abbreviation that should remain unchanged. This is a review aid; it does not imply that every translation tool accepts an enforceable glossary.

For a technical paper, formulas should be compared with the original rather than interpreted from a translated paraphrase. Variable names, equation references and the surrounding explanation each need attention. An unchanged equation image does not establish that its caption or explanation was translated correctly.

Layout review and saved copies

A layout check works block by block. Headings, paragraphs, footnotes and table labels should remain present, legible and associated with the right content. Longer target text needs particular attention around narrow columns and small boxes. The layout translation article covers that separate problem.

Oyren's available PDF translation route works on an editable text block. Its AI box offers Translate to…, which fills a prompt for a target language. The generated result can be reviewed before insertion or replacement. Save as new PDF exports saved block edits into a separate workspace file. AI processing uses credits, and the result still needs source and layout checks.

The exported copy deserves its own inspection after saving. The checklist should include the first page, the difficult sample and a page near the end. A filename indicating the target language and revision helps distinguish a reviewed copy from an earlier trial.

A finished reading copy

Completion means that the intended pages are translated, the important claims match the source and unresolved problems remain visible in a note. A doubtful word can stay marked for later review. Silent certainty is less useful than a specific question tied to its page.

The Oyren PDF editor describes block editing and saved copies for keeping reviewed text with the original document.