Japanese · c. 1600–1868

Read printed kuzushiji, furigana and all

An Edo book was not set in type: a carver copied a calligrapher’s brush drawing onto a block, so the printed page has the same joined kana and the same variant letterforms as a manuscript, only repeated identically on every copy. Evidano follows the main column, keeps the small reading gloss beside it as a separate layer, and works out the order in which blocks of text arranged around a picture are meant to be read.

Upload a page See the samples

Sample pages

Real pages from public collections, shown beside their transcriptions. Pages whose transcription is still being checked are marked.

Opening of an Edo printed novel, dense columns of carved cursive kana with tiny reading glosses down the right side of the kanji — Sample 1. The opening of Ihara Saikaku’s Kōshoku ichidai otoko, first printed at Osaka in 1682 and here in a first impression held by the National Institute of Japanese Literature. Carved cursive text within a ruled frame, with small kana glosses set beside the kanji and the book’s short title 男 printed in the margin outside the frame.

Printed book opening with a full-page portrait of a seated court poet facing a poem written in flowing carved kana inside a plain cartouche — Sample 2. The first poem of the Hyakunin isshu in Nishiki hyakunin isshu Azuma-ori, published in 1775, National Institute of Japanese Literature. The poet’s name heads the cartouche in bold characters and the poem follows in carved cursive kana, drawn from several source characters and joined across syllables, facing a portrait of Emperor Tenji behind a bamboo blind.

Two printed pages of a travel account in carved cursive script, columns running the full height of the ruled frame — Sample 3. An opening of a 1789 printing of Bashō’s Oku no hosomichi, National Institute of Japanese Literature: the end of the innkeeper Butsu Gozaemon episode on the right and the whole of the Nikkō entry on the left. The carved hand is a plain running style with few kanji, and the reproduction here is from a monochrome microfilm of the volume.

Two pages of a printed comic novel, every kanji carrying a small kana reading down its right side — Sample 4. An opening of Jippensha Ikku’s Tōkaidōchū hizakurige, the comic road novel first issued from 1802, in an undated Edo printing at the National Institute of Japanese Literature. Almost every kanji carries a furigana gloss at a third of the size of the main line, and the dialogue runs on without quotation marks or spaces between words.

Preface pages of a printed gazetteer, wide columns of flowing carved kana with long joined strokes and almost no kanji — Sample 5. The preface to the Edo meisho zue, the illustrated gazetteer of Edo published in 1834 and 1836, National Institute of Japanese Literature. The block cutter has reproduced a calligrapher’s brush draft: long joined descents, generously spaced columns, and kana drawn from several source characters within a few lines of each other.

How a woodblock page is put together

From the seventeenth century Japan printed millions of books from carved cherry blocks, one block to an opening. Because the block was cut from a calligrapher’s fair copy, the printed line keeps the writer’s ligatures, the sloping baseline of a column, and the whole repertoire of variant kana — the shapes now called hentaigana, one syllable written from several different source characters. A single page of a novel can use three shapes for ka and two for shi, chosen for the look of the column, and the printed page therefore poses much the same recognition problem as a manuscript with one difference: the same block prints the same shape thousands of times, so an error in reading it repeats through the whole edition.

Around the main column sit several other kinds of text. Furigana, a small kana gloss giving the reading, runs down the right side of a kanji and is set at half size or less; in popular fiction almost every kanji carries one, and the gloss may give a colloquial word rather than the character’s dictionary reading. Chinese prefaces are printed with return marks and small katakana endings so that they can be read aloud in Japanese. The centre fold of each sheet carries the hashira: the book’s short title, the volume and the leaf number, printed sideways relative to the text. Publisher marks, censor seals and the colophon at the end of the last volume carry the date and the shop.

Illustration is not separate from text. In a gazetteer a caption is fitted into the sky above a temple roof; in comic fiction the speech of each figure is written into the space beside them, so that reading order follows the picture rather than a column grid. Printing itself leaves traces to interpret: ink spread where a block was over-inked, broken strokes where it wore, thin paper letting the reverse show through, and later owners’ notes brushed into the margins of a printed page in an entirely different hand.

Why ordinary OCR struggles here

Several printed shapes for one syllable

A model trained on modern Japanese type has never seen the kana derived from 志, 者 or 曾. It will either refuse them or map them to whatever modern kana they most resemble, which produces text that scans as Japanese and is wrong in a way a spellchecker cannot catch.

Main text and gloss at two sizes

Furigana sits so close to the character it glosses that ordinary line-finding merges the two streams, giving a column in which reading and word alternate. Keeping them apart, and attaching each gloss to the right kanji, is a layout problem before it is a recognition problem.

Reading order set by the picture

When text is fitted around an illustration there is no column grid to follow. Blocks have to be ordered by where they sit in the scene and who is speaking, and a transcription that simply reads right to left across the sheet will interleave two speeches into nonsense.

Worn blocks, ink spread and show-through

Late impressions lose hairlines, heavy inking fills the counters of dense characters, and thin paper shows the reverse of the leaf. All three change the shape of a character without changing the text, so the reading has to be made from the whole word rather than from the outline of one glyph.

Who works with this material

Editors of early modern literature

Anyone preparing a text of a novel, a play or a haikai collection has to work from the block-printed editions and record where impressions differ. A first pass that keeps the variant kana as written lets those differences be found by search rather than by re-reading every copy.

Libraries digitising rare book collections

A scanned volume without text is unsearchable. Catalogue departments need the colophon, the hashira title and the leaf numbers captured reliably so that scattered volumes of a set can be matched, even where the body text is only a draft reading.

Historians using gazetteers and guidebooks

Meisho zue, road guides and directories record place names, shrine dedications and distances that appear nowhere else. Researchers need the captions fitted into the illustrations transcribed along with the running text, and each one tied to the place it labels.

Print and publishing historians

Publisher marks, censor seals, advertisement pages and the dated colophon are the evidence for how a title moved between shops and how often it was reissued. They are small, they sit at the edge of the sheet, and they are exactly what a general-purpose reader skips.

Getting the best transcription

Step 1

Ask for furigana as a parallel field

Request the main line and the gloss separately, with each gloss keyed to the kanji it sits beside. You then have a clean text for searching and a record of how a nineteenth-century reader was told to pronounce the word, which is often not the modern reading.

Step 2

Describe the page before you transcribe it

For an illustrated sheet, ask first for a list of the text blocks and where they sit, then for the transcription in the order you agree. Settling the reading order as its own step avoids a transcription that is right character by character and useless as a text.

Step 3

Say what to do with the hashira and the frame

Decide whether the centre-fold title and leaf number, the border rules and the marginal headings should be transcribed, listed separately or ignored. For a page-level index they are the most valuable thing on the sheet; inside a running text they are an interruption.

Step 4

Test a late impression as well as a clean one

Run one page from a worn or heavily inked copy before the rest of the volume. If filled-in strokes and show-through from the reverse are handled there, the crisp early impressions will need almost no correction.

Further reading on this hand

The palaeography guides the in-app selector points to for this domain, if you want to check a transcription against the standard references.

Kuzushiji (Japanese Cursive Characters)
This National Diet Library guide collects authoritative tools for deciphering kuzushiji in manuscripts and early Japanese printed books. It supplies hentaigana charts, links historical kana to their source characters, and identifies searchable dictionaries and character databases. The resource also points to fonts and learning materials that reproduce forms no longer used in ordinary modern writing. It is particularly useful for Edo woodblock books, whose carved lettering preserves brush-style cursive and numerous hentaigana.

Frequently asked questions

Is printed kuzushiji easier than handwriting?: A little, because a block repeats the same shape exactly and the columns are ruled. It is not easier in kind: the shapes being repeated are still cursive and still drawn from the old kana repertoire, so a reader trained only on modern type gains nothing from the regularity.
How are the variant kana handled in the output?: They are read as the syllable they stand for, and you can ask for the source character to be recorded alongside. Open resources such as the Kuzushiji-MNIST project, released under a Creative Commons Attribution-ShareAlike licence, catalogue these variants and are a good reference when checking a doubtful shape.
Can the furigana be dropped instead of transcribed?: Yes. Say so in the prompt and the gloss is skipped, which suits a plain reading text. Keep it if you are studying vocabulary or pronunciation, because the gloss often records a spoken word quite different from the characters it explains.
What about the seals and the colophon?: Publisher marks, censor seals and the dated colophon are read as their own items rather than as part of the text, with their position on the sheet noted. Where a seal is too worn to read it is marked as present and illegible rather than guessed from the usual formula.
Does it distinguish the printed text from a reader’s annotations?: It can, and it is worth asking for. Brushed notes, corrections and ownership inscriptions added to a printed page differ in ink, size and hand, and separating them keeps the edition’s text clean while preserving the evidence of who used the book.

Related scripts and pages

Edo-Period Kuzushiji Manuscripts, c. 1600–1868The handwritten side of the same centuries: diaries, petitions and account books in the hands the carvers copied.Meiji Transitional Cursive and Pre-Standard Kana, c. 1868–1900What replaced the block: movable type, standard kana and the last printed hentaigana.Medieval to Late Imperial Classical Chinese Manuscripts, c. 220–1900The Chinese prefaces and reading marks that Edo publishers printed alongside their Japanese text.All Japanese hands and periodsEvery Japanese domain, from early kana to Meiji type and transitional cursive.Image to TextSearch all supported languages, scripts and document types.

Transcribe your Edo printed books

Upload an opening from a woodblock volume, choose Edo-Period Woodblock Print and Hentaigana, and settle the furigana and reading-order rules on one page first.

Start transcribing Create a free account

© Evidano, All Rights Reserved.

Privacy Policy Terms of Service