Can it tell the Persian from the Arabic when both are on the page?

Yes. The two are distinguished by script, size and position rather than by ink colour, and the transcription can label each stretch so that a bilingual folio comes back as an Arabic text with its Persian rendering attached, not as one continuous line.

Does it add the dots for پ, چ, ژ and گ that the scribe left out?

Only if you ask. The default keeps the letters as written, because for a text-critical edition the undotted form is the evidence. Ask for a normalised version and the modernised letters are supplied, with the manuscript form retained alongside if you want both.

How are word-by-word interlinear glosses handled?

They are read as a separate layer and can be returned aligned to the Arabic word each one translates. Because they are written on a slant and often overlap the line below, uncertain alignments are flagged rather than guessed at.

Will archaic spellings be modernised without warning?

No. Old forms are treated as readings rather than mistakes and are reproduced. A normalised text for searching is a second output; the first stays close to the page so the two can be compared.

Is eastern Kufic legible to the model?

The angular New Style is recognised as a distinct hand, including its plaited form in headings where interlace fills the space between the shafts. Gold-on-red headings and faded ink are the harder cases, and letters that cannot be read are marked instead of invented.

Persian · c. 900–1200

Read Persian as it was first written down

The first Persian books were copied in an alphabet built for Arabic, and the scribes did not always add the marks that keep the two languages apart: p, ch, zh and g wear the coats of b, j, z and k, and no short vowel is written at all. Evidano reads these folios line by line, keeps an Arabic base text separate from the Persian glossing it, and puts every reading beside the page so you can check it.

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.

Gold plaited Kufic sura heading above a line of large angular Kufic with minute Persian words written under each Arabic word — Sample 1. Verso of a detached Qurʾan leaf, Washington, Library of Congress, 1-90-154.150, made in Iran between the eleventh and thirteenth centuries. The heading of sūrat Maryam is in plaited eastern Kufic, gold on red, its shafts knotted with interlace; the basmala below is in the same hand, and under each of its words a minute Persian naskh gloss — bi-nām-i khudāy, rūzī rasānanda, va-āmurzanda — renders it word for word.

Nine lines of tall angular brown Kufic on cream paper with small red and blue vowel marks and two gold rosettes — Sample 2. A leaf from a dispersed Qurʾan, Washington, Arthur M. Sackler Gallery, S1997.95, Iran, eleventh or twelfth century. This is the eastern Kufic book hand at its most disciplined: uprights of equal height, sharply cut heads, and long tails that curl under the following word. The red and blue specks are vowels and hamzas; the two gold rosettes mark the ends of verses.

Seven lines of large sweeping Arabic script with tiny Persian words written diagonally beneath each Arabic word — Sample 3. A leaf from an interlinear Qurʾan, Washington, Library of Congress, 1-89-154.172, Iran, fourteenth century. The Arabic of sūrat Āl ʿImrān is in tawqīʿ, a script that joins letters normally kept apart; under every word a Persian equivalent is written on a slant in a much smaller naskh, with no change of ink colour to separate the two. Gold rosettes and a lamp-shaped medallion in the margin mark verse ends and a section.

Persian prose in small brown naskh below a large Arabic chapter heading, with one line in red and a gold and blue illuminated panel at the foot — Sample 4. The left-hand page of the opening bifolium of Balʿamī’s Persian version of al-Ṭabarī’s chronicle, Washington, Library of Congress, 1-85-154.69, copied in Iran or Central Asia under the Ilkhanids. The Arabic section heading al-qawl fī mā khalaqa Allāh al-khalq li-ajlih runs large across the middle; the Persian prose beneath it opens ammā baʿd, and the Qurʾanic tag closing the block is picked out in red.

What an early Persian page looks like

New Persian appears as a written language in Khurasan and Transoxiana in the ninth century, set down in Arabic letters on rag paper. Almost nothing survives in a copy made before c. 1200 — the oldest dated example is a pharmacological handbook written out in 1055 by the poet Asadī Ṭūsī and now in Vienna — so the period is studied from Qurʾans copied in Iran, from translations glossed between the lines of an Arabic text, and from thirteenth- and fourteenth-century copies of tenth- and eleventh-century works. Two book scripts carry them: eastern Kufic, the angular New Style, with tall leaning shafts and bowls swept far below the line, knotted with interlace in its plaited form and kept for gold headings long after the body text had moved on; and naskh, rounded and evenly spaced, which becomes the ordinary hand of Persian prose in the eleventh century and never leaves it.

The alphabet was a poor fit and the early scribes patched it unevenly. Persian needed four letters Arabic did not have, and پ, چ, ژ and گ are routinely written as plain ب, ج, ز and ک, so that گفت and کفت, پیش and بیش are the same shape on the page. Early orthography also keeps distinctions modern Persian has dropped: a ذ where later copies write د, and prefixes and suffixes such as بی، می، ها and تر set apart from their word or joined to it by no fixed rule. Short vowels are not written and neither is the eżāfe that links a noun to its adjective; the vowel signs that do appear were often added by a later reader in red, with hamzas, shaddas and corrections that belong to a different century from the text.

Layout is doing work as well. In an interlinear Qurʾan the Arabic runs large across the page and the Persian is squeezed under each word on a slant, so the eye moves horizontally through one text and diagonally through the other; in other copies the Persian is continuous and picked out in red. Gold rosettes divide verses, lamp-shaped medallions in the margin mark the thirtieth parts, marginal notes tell a reciter how to lighten a consonant, and a catchword at the foot of the verso gives the first word of the next leaf.

Why ordinary OCR struggles here

One skeleton, five or six letters

Strip the dots and ب، ت، ث، پ، ن and ی collapse into one shape, as do ج، چ، ح and خ. Early scribes place dots loosely, group them into a slanting dash, or leave them off where the word is obvious to them. An engine trained on printed Persian, which is always fully dotted, settles these by guessing the commonest word rather than by weighing the sentence.

The letters Persian added and the scribes omitted

Because پ، چ، ژ and گ are usually written as ب، ج، ز and ک, deciding between گور and کور, or پرده and برده, is a question about the sentence rather than about the ink. The transcription has to record what stands on the page and, if you want it, the modernised form beside it, so that a lexicographer can see which is which.

Two languages interleaved on the same lines

An interlinear Qurʾan is not one text but two, in different scripts, at different sizes and on different angles. Read straight through and the Arabic and the Persian fuse into nonsense; the reading order has to be rebuilt word by word, with base text and gloss kept in separate streams.

Spellings that no modern dictionary lists

Archaic forms, the ذ for د, detached prefixes and vocabulary that fell out of use by the fifteenth century all look like errors to a system tuned on modern Persian, which repairs them silently. Vowel points and corrections added centuries later sit in the same space and are easily taken for the scribe’s own.

Who works with this material

Editors of early Persian prose

Anyone establishing the text of a tenth- or eleventh-century work from later copies needs a reading that preserves each copy’s own spelling rather than a tidied modern one. Variants become visible only when the ذ, the detached prefix and the missing gāf stroke stand as the scribe left them.

Scholars of Qurʾan translation

The interlinear versions are the earliest continuous evidence for Persian prose, and their value lies in the pairing of each Arabic word with its Persian equivalent. What such a project needs is an aligned output — Arabic token, Persian token — not a flat block of text in which the two have been merged.

Cataloguers of detached folios

Museums and libraries hold thousands of single leaves cut from dispersed manuscripts. Identifying one means reading enough of it to locate the passage. A transcription of the visible lines turns an unidentified leaf into a searchable record.

Historical lexicography and corpus building

Dictionaries of early Persian are built from dated attestations. A corpus assembled from manuscript pages is only usable if archaic forms survive the transcription and if uncertain readings are flagged instead of quietly regularised.

Getting the best transcription

Step 1

Say which language you want back

On a bilingual page, ask for the Arabic and the Persian as two labelled streams, and say whether the glosses should be aligned to the word above them or listed line by line. Settling this before the run saves untangling the output afterwards.

Step 2

Fix a rule for the four Persian letters

Choose whether ب، ج، ز and ک are transcribed as written or normalised to پ، چ، ژ and گ where the word requires it. If you want both, ask for the manuscript form first and the modern form in brackets; set once, the rule holds across a whole codex.

Step 3

Protect the old orthography

State that the ذ, separated prefixes and unfamiliar spellings must be kept, and that anything the model would ordinarily correct is to be marked rather than changed. Ask for later additions in red — vowels, hamzas, corrections — to be reported apart from the black text.

Step 4

Test on a page with a heading

Start with a folio that mixes gold Kufic in a heading, black naskh in the body and a marginal note. If the three are kept apart there, the plain pages of the same codex will present nothing new.

Frequently asked questions

Can it tell the Persian from the Arabic when both are on the page?: Yes. The two are distinguished by script, size and position rather than by ink colour, and the transcription can label each stretch so that a bilingual folio comes back as an Arabic text with its Persian rendering attached, not as one continuous line.
Does it add the dots for پ, چ, ژ and گ that the scribe left out?: Only if you ask. The default keeps the letters as written, because for a text-critical edition the undotted form is the evidence. Ask for a normalised version and the modernised letters are supplied, with the manuscript form retained alongside if you want both.
How are word-by-word interlinear glosses handled?: They are read as a separate layer and can be returned aligned to the Arabic word each one translates. Because they are written on a slant and often overlap the line below, uncertain alignments are flagged rather than guessed at.
Will archaic spellings be modernised without warning?: No. Old forms are treated as readings rather than mistakes and are reproduced. A normalised text for searching is a second output; the first stays close to the page so the two can be compared.
Is eastern Kufic legible to the model?: The angular New Style is recognised as a distinct hand, including its plaited form in headings where interlace fills the space between the shafts. Gold-on-red headings and faded ink are the harder cases, and letters that cannot be read are marked instead of invented.

Related scripts and pages

Persian Chancery Taʿliq, c. 1200–1500The hanging documentary hand that grew out of naskh in the Ilkhanid and Timurid chanceries.Mashriqi Naskh and Related Book Hands, c. 1000–1500The same naskh book hand in the Arabic manuscripts copied alongside these Persian ones.All Persian hands and document typesEvery Persian domain from early Kufic and naskh to shikasta letters and Indo-Persian records.Image to TextSearch all supported languages, scripts and document types.

Transcribe your early Persian folios

Upload a leaf, choose Early New Persian Kufic and Naskh Manuscripts as the domain, and settle the glossing rules on one page before you run the codex.

Start transcribing Create a free account