Telugu
Telugu pages, signs and manuscripts turned into text
Telugu letters are built from loops rather than straight strokes, a habit that goes back to writing on palm leaf, where a straight line risked splitting the leaf along its grain. Evidano reads printed Telugu books, government signage and civic notices, and older palm-leaf manuscripts, and sets the transcription beside the image so every subscript conjunct and reshaped vowel can be checked.
Sample pages
Real pages from public collections, shown beside their transcriptions. Pages whose transcription is still being checked are marked.
Source · CC BY-SA 4.0 · విశ్వనాధ్.బి.కె.
Transcription being verified
Printed editions of the Basava Purana exist, but no transcription has been matched to this specific manuscript bundle.
What the model is told to watch for: Produce a diplomatic transcription in Telugu.
Jangamwadi Math collection, digitised by eGangotri, via Internet Archive, 1926 · Source · CC0 1.0 · Sheshadri Ramana Kavi
Transcription
The first paragraphs of the preface shown on this page, as proofread on Telugu Wikisource against this same scan (proofread once, quality level 1 of 4 — not yet checked by a second proofreader). Wiki markup and the page-running-head template are removed; paragraph breaks follow the source.
పీఠిక రసికులారా! నిజాము రాష్ట్రీయుల నుద్బోధించుచు మేము నాలుగేండ్లక్రింద నీలగిరిపత్రికకు ఐదుపద్యములను వాసితిమి. ఒక సభాసందర్భమున హైదరాబాదులోని యాంధ్రవిద్యాధికులాపద్దెములఁ జదివించి మమ్ముఁ బ్రోత్సహించిరి. వారి ప్రోత్సాహము నాధారపఱచికొని యింక నిరువది పద్యముల వ్రాసి గ్రంథాలయ 'నిజాము రాష్ట్ర ప్రశంస' యను పేరునఁ జిన్ని పొత్తమును బ్రచురించితిమి. పత్రికాసంపాదకు లాచిఱుపొత్తము పై సదభిప్రాయములిచ్చి మాయందలి వాత్సల్యము చేఁ బరిశోధనోద్యమమున నిజాము రాష్ట్రమునకు మేము చేయు సేవను బ్రశంసించిరి. ముద్రితమైన 500 ప్రతులు రెండు మాసములలో గర్చుపడెను. మిత్రులు మఱికొందఱును నీలగిరిసంపాదకులు ఇటులె మఱికొన్ని పద్యముల వ్రాయఁ బ్రోత్సహించిరి, మిత్రులప్రోత్సాహమున ౧౦౦ పద్యములను వ్రాసితిమి.
Transcription from పుట:Nizam Rashtra Prashamsa ( Telugu) By Sheshadri Ramana Kavi, 1926.djvu/7, Telugu Wikisource (CC0 1.0).
Source · CC BY-SA 4.0 · Mouryan
Transcription being verified
A standard legal-aid notice; no separately published transcription of this specific board exists.
What the model is told to watch for: Produce a diplomatic transcription in Telugu.
Source · CC BY-SA 4.0 · Rajasekhar1961
Transcription being verified
No published transcription exists for this specific signboard.
What the model is told to watch for: Produce a diplomatic transcription in Telugu.
What Telugu text looks like on the page
Telugu and Kannada share a common ancestor script that split into two recognisably separate traditions by around the thirteenth century; Telugu literary writing is usually dated from Nannaya’s eleventh-century translation of the Mahabharatam. Both scripts owe their rounded, looping letterforms to the palm leaf: a stylus dragged in a straight line along the leaf’s fibres tends to split it, so scribes favoured curves, and print later inherited the same shapes. Devotional and narrative literature such as the Basava Purana and the Andhra Mahabhagavatam survive today mostly in palm-leaf bundles copied by generations of temple and monastery scribes.
The script is an abugida with a large resulting glyph set. Consonant clusters are written with a vattu, a reduced form of the second consonant attached below the main letter rather than beside it, so க்ஷ-type combinations such as క్ష, స్త్ర and ద్ర are built downward rather than across the line. Several vowel signs do not simply add a stroke to a consonant but reshape its lower curve entirely, so the same base consonant can look markedly different depending on which vowel it carries. Telugu also has its own numerals, now rare outside older printing, and classical verse is marked with abbreviations for its metre — సీ for sīsa, తే.గీ. for tēṭagīti and others — set into the line rather than kept as a separate heading.
Everyday material spans this whole range: a temple’s palm-leaf holdings sit alongside a small-press book from the early twentieth century, whose type can already look slightly different from a modern font, and government notices and signage that mix formal Telugu with English department names, addresses and phone numbers on the same board.
Why ordinary OCR struggles here
Conjuncts that hang below the letter
A vattu consonant cluster is built downward from the main letter rather than written beside it, and several vattu shapes bear little resemblance to the full-size consonant they represent. Reading the cluster right means recognising the pair as a unit, not two unrelated marks.
A vowel sign that changes the consonant’s shape
Some Telugu vowel signs do not simply attach a stroke to a consonant; they reshape its lower curve outright, so the same consonant looks different from one vowel to the next. A model has to learn a family of related shapes as one letter rather than treat each as a separate glyph.
A very large glyph inventory
Combining roughly sixty base letters with vattu conjuncts and vowel reshaping produces thousands of possible visual forms, most of them rare. Metal-type Telugu printing needed hundreds of separate sorts for exactly this reason, and recognition faces the same long tail of uncommon combinations.
Verse metre markers mistaken for stray text
Classical poetry is printed with short abbreviations such as సీ or తే.గీ. marking the metre of the lines that follow, set into the body of the page rather than as a heading. Generic OCR can read these as garbled words or drop them, losing information about the poem’s structure.
Who works with this material
Temple and matha manuscript projects
Purana and devotional literature such as the Basava Purana survives largely in palm-leaf bundles held by temples and mathas, and cataloguing them starts with a reliable reading of each leaf’s title, extent and condition.
Historians of Hyderabad State and Telugu small-press printing
Early twentieth-century books from presses like Osmania Press carry ownership stamps, dedication verses and metre markers that a researcher needs read accurately to place a copy within a text’s publishing history.
Civic and welfare administration
Government notices, cooperative-society signage and welfare-scheme boards mix formal Telugu with English department names and contact details, and an index needs the numbered eligibility points and addresses kept intact.
Genealogists and land-record researchers
Family papers and revenue records combine printed forms with handwritten Telugu entries, and getting names, village names and dates right matters more than the boilerplate wording around them.
Getting the best transcription
Say what kind of material the page is
Tell the model whether it is looking at a palm-leaf manuscript, an early-twentieth-century press edition, or modern print or signage, since letterforms and layout conventions differ across the three.
Ask for vattu conjuncts to be kept as one cluster
Request that a subscript vattu consonant be transcribed as the cluster it forms with the letter above it, rather than split into two unrelated characters.
Preserve verse structure and metre markers
For classical poetry, ask for metre abbreviations such as సీ or తే.గీ. and the poem’s own line breaks to be kept, rather than reflowed into an ordinary prose paragraph.
Start with the hardest image in a batch
Run a manuscript or a multi-panel sign before a run of uniform print. If the model correctly separates the panels or leaves and reads the conjuncts, the rest of the batch usually needs only a light check.
Frequently asked questions
- Can it read vattu conjunct consonants correctly?
- Yes. A vattu form attached below a consonant is read as the cluster it represents, including the reduced shapes that no longer closely resemble their full-size letter.
- Does it know that some vowel signs reshape the consonant rather than just adding a mark?
- Yes. The related family of shapes a single consonant can take with different vowel signs is recognised as that one consonant, not as separate, unrelated letters.
- Can it read an old palm-leaf Telugu manuscript?
- Incised, blackened text on palm leaf is read as such, including several leaves photographed together, though a badly cracked or very worn leaf may still need a specialist’s check.
- Will classical verse metre markers and line breaks be kept?
- Yes, when you ask for them. Metre abbreviations and a poem’s own line divisions can be preserved rather than reflowed into prose paragraphs.
- Can it read Telugu numerals?
- Yes, along with the standard Arabic numerals used in most modern Telugu printing; both are transcribed as written.
- Can it handle a bilingual Telugu-English sign or government notice?
- Yes. Telugu and English panels or lines are transcribed in place and kept distinguishable from each other rather than merged into one block of text.
Related scripts and pages
Transcribe your Telugu documents
Upload a scan or a photograph — a printed page, a government notice, a sign or a manuscript leaf — and check the Telugu transcription line by line before you export it.