Nepali
Nepali letters, government papers and Kathmandu Valley manuscripts turned into text
Modern Nepali is set in Devanagari, where a headline stroke ties every syllable of a word together and consonants stack vertically into combined shapes a keyboard has no key for. Nepal’s older manuscript tradition used two further alphabets, Ranjana and Prachalit, that share almost none of Devanagari’s letterforms. Evidano reads Nepali across that whole range — Kathmandu Valley manuscripts, printed newspapers, and the government letters and notices that make up most everyday transcription work — and sets its output beside the image so every stacked conjunct and every Bikram Sambat date can be checked.
Sample pages
Real pages from public collections, shown beside their transcriptions. Pages whose transcription is still being checked are marked.
Asa Archives, Kathmandu, 1695 (Nepal Sambat 815) · Source · Public domain
Transcription being verified
No published edition of this specific leaf has been located; the Asa Archives catalogue describes the manuscript but does not print a transcription of it.
What the model is told to watch for: Produce a diplomatic transcription in Nepali.
Transcription being verified
No line-by-line transcription of this masthead has been published; the National Library of Nepal holds runs of the paper but has not issued an edited text of this issue.
What the model is told to watch for: Produce a diplomatic transcription in Nepali.
Office of the Rural Municipal Executive, Jagannath Rural Municipality, Bikram Sambat 2081.06.10 (2024) · Source · CC BY-SA 4.0 · Jagannathmun
Transcription being verified
An original 2024 municipal letter uploaded by the issuing office; there is no separate published transcription to check it against.
What the model is told to watch for: Produce a diplomatic transcription in Nepali.
Transcription being verified
A single information board; no independently published transcription exists outside the board itself.
What the model is told to watch for: Produce a diplomatic transcription in Nepali.
Endangered Archives Programme, uploaded by Archive Nepal for Wikisource Loves Manuscripts, 18th century · Source · Public domain
Transcription
All seven lines of this page, transcribed and marked proofread (quality level 3 of 4) on the multilingual Wikisource against this same scan, and checked line for line against the image above. The Wikisource text sets the Prachalit letters in Devanagari rather than reproducing the original letterforms, and keeps the manuscript’s line breaks, its double dandas and its raga headings.
लयाओ मन पूर ॥ ६ ॥ १ राग ॥ प्रथमं जरि ॥ प्र ॥ जय नृत्यश्वर देव महेश्वर निम्हन छम्ह थे विज्याक रे नंदि भिंदि ओ स भुत वेतालस मनया आनंद याक रे ॥ जोतया स्यासन थुसा क्षि आसन त्रिशुल दम्वरु थाक रे ।। जतस पो लस षु स्वान सिलस वालक चंद्रमा थिक रे ।। ।। लसिक वसिक नलि धु जुगुलि मोल मालक्षि कोषास रे ।। दया लक्ष्मीया नाथ अनाथया ल्हाक म्ह जय परकास रे ॥ ॥ ६ ॥ १ राग ॥ केदारा ।। ख ॥ भुवानी क्षि चरण शवनस दान ॥ थुगुलि संसारस कपत मय जुल धरम विवेक मयाक ॥
Transcription from Page:AN-PM-13.pdf/40, multilingual Wikisource (CC BY-SA 4.0).
What Nepali text looks like on the page
Nepali is written in Devanagari, an abugida in which every consonant carries an inherent vowel unless a different vowel sign is attached, and a horizontal headline, the shirorekha, runs along the top of a word and joins its syllables into one visual unit. Consonants that fall together without a vowel between them are not written side by side but combined into a single conjunct glyph, some barely resembling their parts — क्ष, ज्ञ and त्र are common examples — and Nepali keeps Devanagari’s retroflex-versus-dental distinctions, ट against त and ड against द, that Sanskrit needed and Nepali inherited. Official Nepali documents commonly date themselves by the Bikram Sambat calendar, written with Devanagari numerals and running some fifty-six or fifty-seven years ahead of the Gregorian year.
Before Devanagari became the standard printed alphabet, the Kathmandu Valley had its own writing traditions for Newar and for the Sanskrit texts its scribes copied. Ranjana is an ornate, angular script cut for monumental and decorative use — temple inscriptions, mantra banners, manuscript titles — and it is still used that way today, even though almost nobody reads it as a running text. Prachalit, sometimes simply called the Newar script, is its faster cursive relative and is what most working manuscripts from the valley are actually written in: centuries of Buddhist and Hindu texts in Newar and Sanskrit survive in Prachalit hands that look nothing like modern printed Devanagari, despite sharing the same distant Brahmic root.
Most Nepali transcription work today is neither Ranjana nor Prachalit, though: it is government correspondence, land and citizenship paperwork, and public notices. A municipal letter typically carries a printed letterhead with a crest, a letter number and a reference number, a body paragraph in ordinary printed Devanagari, and a handwritten signature pressed against a rubber-stamped seal, all on one page. Older Nepali newspapers add their own habit, a heavier, more angular metal type than any font in use today, often printed with a Bikram Sambat dateline running along the foot of the page.
Why ordinary OCR struggles here
Conjuncts built from more than one letter
A cluster of consonants with no vowel between them is written as a single combined glyph rather than as separate letters side by side, and some of the commonest conjuncts barely resemble the consonants that make them up. A system trained to recognise individual letters either misreads the conjunct as one unfamiliar character or splits it into the wrong pieces.
Two other alphabets with no shared letterforms
Ranjana and Prachalit do not look like modern Devanagari at all, so a model trained only on printed Nepali does not merely misread a Kathmandu Valley manuscript — it may not recognise that the marks on the page are letters in a related script rather than an unrelated decoration.
Bikram Sambat dates and Devanagari numerals
Official documents date themselves in the Bikram Sambat era using Devanagari numerals, in a year that runs roughly fifty-six to fifty-seven years ahead of the Gregorian calendar with no fixed offset across the whole year. Read carelessly, a date like २०८१।०६।१० can be transcribed as the wrong digits entirely or left unconverted where a reader expected the Gregorian equivalent.
Letterhead, stamp and signature on one page
A government letter mixes crisp printed letterhead type, a rubber-stamped seal, and a handwritten signature squeezed into the same block of the page. A model expecting one hand or one typeface throughout a document has to separate three different kinds of mark rather than transcribe them as if they were a single continuous text.
Who works with this material
Newar and Kathmandu Valley manuscript scholars
Reading Prachalit and Ranjana copies of Newar and Sanskrit texts means recognising each cursive or angular stroke on its own terms rather than mapping it onto modern Devanagari, and keeping colophons and verse numbers distinct from the running text.
Municipal and civil-registration staff
Renaming notices, recommendation letters and certificates like the ones rural and municipal offices issue every day need their letter numbers, reference numbers and body text read correctly before they can be filed, indexed or acted on.
Land and family historians
Land records and family papers reaching back generations date themselves in Bikram Sambat and use administrative vocabulary that has since changed; getting the era, the numerals and the old terms right is what lets a record be placed correctly in a family or a property history.
Press historians and archivists
Citing an old issue of Gorkhapatra or another historic Nepali paper accurately means transcribing dense period type, including the paper’s own Bikram Sambat dateline, rather than paraphrasing it from a modern summary.
Getting the best transcription
Say which alphabet the page is in
Tell the model whether it is looking at modern Devanagari, Ranjana or Prachalit, so a manuscript hand is read on its own terms rather than forced into the nearest Devanagari letter shapes.
Ask for conjuncts to be kept whole
Specify that a stacked consonant cluster should be transcribed as the single combined syllable it represents, not split into a guessed sequence of component letters.
Decide how Bikram Sambat dates should come out
Say whether a date should stay as written in Bikram Sambat, be converted to the Gregorian equivalent, or be given as both, since a silent conversion is where an off-by-one-year error is easiest to introduce.
Start with the busiest part of the batch
Run a letterhead with a stamp and signature, or the densest paragraph of a manuscript leaf, before the rest of a batch. If the model separates print, stamp and handwriting correctly there, the plainer pages need only a light check.
Frequently asked questions
- Can it read Devanagari’s stacked conjunct consonants correctly?
- Yes. Conjuncts such as क्ष, ज्ञ and त्र are recognised as the single combined syllables they represent rather than split into a guessed sequence of separate letters.
- Does it know Ranjana and Prachalit are different alphabets from printed Nepali?
- Yes, when you say which one the page uses. Ranjana’s angular monumental letterforms and Prachalit’s cursive manuscript hand are each read on their own terms rather than mapped onto modern Devanagari shapes.
- Can it read a Kathmandu Valley manuscript, not just modern print?
- Yes. Prachalit and Ranjana manuscripts are transcribed as the scripts they are, including verse numbers, colophons and the abbreviations a scribe used to save space at the end of a line.
- How does it handle Bikram Sambat dates and Devanagari numerals?
- Devanagari numerals are read as the digits they are, and a Bikram Sambat date is kept as written, converted to the Gregorian calendar, or given as both, according to what you ask for.
- Can it separate a letterhead, a stamp and a handwritten signature on one page?
- Yes. Printed letterhead text, a rubber-stamped seal and a handwritten signature are treated as separate elements rather than merged into one block, so each can be reviewed on its own.
Related scripts and pages
Transcribe your Nepali documents
Upload a government letter, a printed page or a Kathmandu Valley manuscript, and check the Nepali line by line before you export it.