# OCR for Vertical Japanese and Chinese Books: Check the Reading Order

Identify vertical and mixed layouts, plan conversion or translation, and review representative pages for column order, small annotations, and page transitions.

A vertical book can contain correctly recognized characters in the wrong sequence. The OCR system may recover the text but join columns, captions, or small annotations incorrectly. For reading and translation, order matters as much as individual characters.

Start with representative pages and the original PDF. Do not assume that one clean vertical page establishes reliable processing for every layout in a long book.

## Identify vertical and mixed arrangements

Look at an ordinary body page, a chapter opening, and a page with illustrations or notes. A book may combine vertical prose with horizontal headings, tables, Latin-letter abbreviations, or small pronunciation annotations.

Note the intended sequence before processing. Which column comes next? Where does a caption belong? Does a small line annotate nearby characters, or form part of the main sentence? Record those relationships on a few selected pages so you have a concrete acceptance target.

If the scan includes two facing pages in one image, also check their order and the binding area. Cropping or shadow can hide the last characters of a line. Translation cannot reliably recover text absent from the image.

## Check OCR order before judging the translation

After conversion, read a continuous source passage from beginning to end while following the PDF image. Look for columns joined in the wrong direction, annotations inserted into prose, and running headers repeated inside sentences.

A useful sample includes:

| Page type                          | Main risk to inspect                        |
| ---------------------------------- | ------------------------------------------- |
| Ordinary vertical prose            | Column sequence and paragraph joins         |
| Mixed horizontal and vertical text | Headings or labels inserted into the body   |
| Small annotations                  | Their relationship to the main text         |
| Illustration or table              | Caption placement and interruption of prose |
| Page or chapter boundary           | Continuation versus a genuine break         |

If a translated passage sounds strange, the source order may be wrong. Compare the page rather than trying to repair meaning from the target-language sentence alone. The [OCR and translation explanation](https://pdfcraft.ai/en/blog/ocr-and-translate-workflow/) separates those stages.

## Choose conversion or translation explicitly

In the [PDF workspace](https://pdfcraft.ai/en/pdf-craft/), upload the PDF and choose the output around your goal. EPUB suits continued reading with adjustable text. Markdown suits editing and research notes. Translation is optional for those outputs and must be enabled when wanted; PDF output is a translation task.

The resulting reading layout need not reproduce the original vertical page. Evaluate whether the chosen output retains the content and useful sequence rather than assuming a facsimile. PDF Craft does not promise lossless reproduction of every vertical or mixed arrangement.

Add concise background when translation is needed: the subject, preferred names, and important terminology. Do not assume instructions can compensate for missing characters or scrambled source order. Review current file limits and [Pricing](https://pdfcraft.ai/en/pricing/) before submitting. New users receive trial credits; processing is charged by usage.

## Use a difficult sample to decide what needs review

Pick a page where the layout changes and a page crossing a boundary. Compare the end of the preceding page, the current page, and the start of the next. Check whether a complete sentence survives in order without a header or page number inserted.

For notes, identify the marker and complete note text rather than accepting small-type recognition alone. For a table, check row and column associations. For characters that are faint or ambiguous, consult a clearer source if possible instead of letting a fluent translation hide uncertainty.

Do not claim a percentage from a few sampled pages. The sample helps identify the book's failure modes and review burden; it is not a document-wide accuracy measurement.

## Keep the source close during reading

Save the downloaded output separately and record problem locations by original page and a recognizable phrase. EPUB pages change with font and screen size, so output page numbers alone are weak references.

If many pages share the same ordering problem, stop extending the workflow until you understand it. If only a few are difficult, mark them for closer source comparison. The [scanned PDF translation guide](https://pdfcraft.ai/en/blog/translate-scanned-pdf/) covers downstream content and layout review without treating fluency as proof that recognition was correct.