ocr-projects-archive / ocr-tech /docs /ATTRIBUTIONS.md
prosperious's picture
Add 1000 files (OCR_INDRA/.gitignore ...)
f9bc0c1 verified
|
Raw History Blame Contribute Delete
13.2 kB
# Attributions
Required credits for data used in this project. Include these wherever
derived models, datasets, or numbers are published.
## Commons Bhujimol basic-glyph vector chart
“Bhujimol script characters (Unicode),” created and uploaded by Wikimedia
Commons user `Z3lvs` as own work on 2026-05-24 and dedicated to the public
domain under CC0 1.0. Source:
https://commons.wikimedia.org/wiki/File:Bhujimol_script_characters_(Unicode).svg.
The project uses the exact SVG only as an optional qualified-review reference.
It is a basic-glyph outline chart, not a font, mapped OCR dataset, conjunct or
shaping specification, or substitute for manuscript-specific script review.
## Nithya Ranjana 1.000 — Ranjana-style reference fonts
Nithya Ranjana DU and Nithya Ranjana NU, copyright 2024 the Nithya Ranjana
Project Authors. Designed, engineered, and maintained by Ek Type with support
from Callijatra. The official release names Tathagata Biswas, Noopur Datye,
Ananda K. Maharjan, Sarang Kulkarni, and project sponsor Murali K. Prahalad.
Licensed under the SIL Open Font License, Version 1.1. Official repository and
release: https://github.com/EkType/Nithya-Ranjana and
https://github.com/EkType/Nithya-Ranjana/releases/tag/1.000.
DU is based on Devanagari Unicode; NU is based on Newa Unicode. These are
Ranjana-style reference/render candidates, not evidence that Ranjana has an
assigned Unicode block, not Bhujimol fonts, and not real historical OCR data.
## Wellcome MS Indic epsilon 1 — Bhujimol candidate
Wellcome Collection, `MS Indic epsilon 1` / `b14901560`, *Aṣṭasāhasrikā
Prajñāpāramitā Sūtra*. Catalogue record:
https://wellcomecollection.org/works/jnsn9gjd. Complete IIIF manifest:
https://iiif.wellcomecollection.org/presentation/v2/b14901560.
The manifest applies Public Domain Mark 1.0 and requests attribution to
Wellcome Collection. Intended use is a sealed, manuscript-level Bhujimol
evaluation candidate only after remote per-canvas hashing, qualified script
review, human transcription, pixel-label audit, and leakage control. This
entry does not imply that any pixels or labels are currently admitted.
The GRETIL *Aṣṭasāhasrikā* e-text based on P. L. Vaidya's 1960 edition is
recorded only as a different-edition work-level reviewer reference. It is not
copied into the project, is not line-aligned to the Wellcome manuscript, and
is not admitted for training, labels, or redistribution.
## Cambridge MS Add.2137 Commons photograph — Bhujimol candidate
“1407 CE, Naradasmriti Hindu law manuscript, Sanskrit, Bhujimol script, Malla
kingdom,” photograph by Wikimedia Commons user Ms Sarah Welch, 29 October
2018, licensed under CC BY-SA 4.0. Source:
https://commons.wikimedia.org/wiki/File:1407_CE,_Naradasmriti_Hindu_law_manuscript,_Sanskrit,_Bhujimol_script,_Malla_kingdom.jpg.
The exact 2,332 x 1,800 JPEG has SHA-256
`938013af2c4c16ac84164a41f903cf34d131663b7a069e58df779bacfcd4258d`.
The image contains four folio photographs. The review packet displays bounded
regions corresponding to institutional canvases 48–50 (`28r`, `28v`, `29r`)
while retaining the complete Commons source identity. Those display crops are
transformations under the same attribution and ShareAlike conditions. The
Cambridge Digital Library scan and TEI are used only for institutional
provenance, folio crosswalk, and catalog facts; no Cambridge CC BY-NC source
pixels or complete transcription are admitted.
## BDRC MW1NPL99 — Bhujimol candidate
Buddhist Digital Resource Center, unidentified physical instance `MW1NPL99`,
scan `W1NPL99`, image group `I1NPL99`, Nepalese Sanskrit Manuscript Collection.
Catalogue: https://library.bdrc.io/show/bdr:MW1NPL99. IIIF manifest:
https://iiifpres.bdrc.io/vo:bdr:I1NPL99/manifest.
BDRC's access policy classifies scanned original manuscripts as public domain
and asks users to credit BDRC where possible:
https://www.bdrc.io/access-policies/. The exact scan is `AccessOpen`; the
physical record calls the work “Unidentified” and assigns
`ScriptNepaleseHooked`. The project maps that institutional script term to a
Bhujimol review profile but retains qualified page review as a gate. This is a
zero-label remote-review candidate, not an admitted OCR dataset or companion
text source.
## Wellcome MS Indic delta 33 — selected Ranjana references
Wellcome Collection, `MS Indic delta 33` / `b14901468`, selected digital-image
assets `L0021448`, `L0021450`, `L0035405`, and `L0035406`. Physical manuscript
record: https://wellcomecollection.org/works/bp979bf3. The four image records
are credited to Wellcome Collection under CC BY 4.0.
These are overlapping or partial reference views, not a complete digitized
manuscript and not an admitted OCR dataset. Any future reuse of the exact
selected assets must preserve Wellcome attribution and the CC BY 4.0 notice.
## BDRC MW1NPL132 — Ranjana candidate
Buddhist Digital Resource Center, *Ārya nāmasaṁgīti*, physical instance
`MW1NPL132`, scan `W1NPL132`, image group `I1NPL132`, Nepalese Sanskrit
Manuscript Collection. Catalogue:
https://library.bdrc.io/show/bdr:MW1NPL132. IIIF manifest:
https://iiifpres.bdrc.io/vo:bdr:I1NPL132/manifest.
BDRC's access policy classifies scanned original manuscripts as public domain
and asks users to credit BDRC where possible:
https://www.bdrc.io/access-policies/. The exact scan is `AccessOpen`; the
manifest supplies BDRC attribution but no license URI. Metadata CC0 is not
represented here as a grant over source pixels. This entry describes a
zero-label remote-review candidate, not an admitted OCR dataset.
## Zenodo 6967421 — Pracalit/Newa manuscript ground truth (CC-BY-4.0)
Alexander O'Neill (SOAS University of London). "OCR model for Pracalit for
Sanskrit and Newar MSS 16th to 19th C., Ground Truth." Zenodo, 2022. DOI:
10.5281/zenodo.6967421.
Underlying manuscript images/transcriptions credit: Felix Otter; Min
Bahadur Shakya & Naresh Man Bajracharya; Lokesh Chandra. Companion paper:
O'Neill & Hill, "Text Recognition for Nepalese Manuscripts in Pracalit
Script," Journal of Open Humanities Data 8:90 (2022), DOI: 10.5334/johd.90.
Used for: `data/newa-real-eval-pack` (3,180 line crops), the isolated Newa
CTC200 payload, its prediction-free Cambridge native-review bundle, and Newa
recognizer training/evaluation.
## Gorkhapatra daily — recovered Newar-Devanagari silver material
Gorkhapatra daily, Gorkhapatra Corporation/Sansthan, Nepal. Discovery pages
and matching newspaper PDFs came from the official Gorkhapatra publication and
epaper services recorded in `RESEARCH.md`. The recovered Newar-Devanagari pack
contains machine-converted newspaper line crops and is retained only as an
internal diagnostic while source rights remain `RISK` in
`docs/DATA_LICENSING.md`.
This general credit does not repair the recovered pack's missing per-row PDF
URL, PDF hash, page, bbox, date, or edition fields, and does not grant
permission to train, publish numbers, redistribute, or use commercially. Those
rights and provenance gates must be resolved before any row or metric is
admitted.
## Sikkim Herald — Government of Sikkim IPR editions
Sikkim Herald, Information & Public Relations Department, Government of
Sikkim. Official archive:
https://www.sikkim.gov.in/SikkimHerald/. The Information & Public Relations
Department identifies the Herald as its official government paper, published
in thirteen State languages. Source editions must be credited by language,
date, volume/issue when available, direct PDF URL, and SHA-256.
Use is under the Government of Sikkim State Portal's source-specific
[copyright permission](https://www.sikkim.gov.in/legal-link/copyright), not a
public-domain claim or standard open-content license. Reproduction must be
accurate, non-derogatory, and non-misleading, and the Government of Sikkim/IPR
source must be prominently acknowledged. Material explicitly identified as
third-party is excluded unless separately authorized. Intended use: rebuilding
edition-disjoint Lepcha, Kirat Rai, Sunuwar/Mukhia, and Gurung Khema real-print
OCR sources; no new frozen pack or model is implied by this attribution entry.
Current zero-label source-review set: Kirat Rai 2024-09-26 and 2025-01-13;
Lepcha 2024-07-26 and 2024-10-04; Sunuwar/Mukhia 2024-04-12 and 2025-01-26.
The Gurung Khema reopening adds the Gurung Editions of 2022-03-01 and
2022-04-22. Their direct State Portal PDFs and SHA-256 hashes are:
- https://www.sikkim.gov.in/uploads/SikkimHerald/gurung_1MARCH_0_20220301.pdf —
`e667bbd2eb4f6769209013fe569cdc37d3e1aff40a9f0121301b43a03afe5d14`
- https://www.sikkim.gov.in/uploads/SikkimHerald/Gurung_22April_0_20220504.pdf —
`2db16c7b7abe678e02103037397c7a35e21fb2fd34d4e4413d1979d1cf7f7c93`
The second PDF prints issue 5 on pages 1–3 and issue 4 on page 4; that source
anomaly is preserved rather than normalized. The semantic audit uses Unicode
proposal [L2/22-096](https://www.unicode.org/L2/L2022/22096-gurung-khema.pdf)
(SHA-256
`bd515093584bf2e5d2630f855d91340badecee6877f0a6ebf8fb896ebdb5cec0`)
as a legacy-font mapping and same-font attestation, and
[Unicode 16.0](https://www.unicode.org/versions/Unicode16.0.0/core-spec/chapter-13/#Gurung_Khema)
for the encoded repertoire. Those references do not grant a font
redistribution license or certify any OCR label.
The two complete press photographs on Lepcha 2024-10-04 page 1 are excluded
from any future derived pack. Each future row must additionally retain its
direct State Portal PDF URL, complete PDF hash, printed edition identity, page,
bbox, and the permission/source attribution above.
## Santali Wikipedia (CC BY-SA 4.0 / GFDL)
Text corpus from the sat.wikipedia.org pages-articles dump (Wikimedia
Foundation and contributors). Used for: Ol Chiki synthesis corpora and
charset dictionary. Share-alike applies to redistributed text.
## Videha e-zine Tirhuta editions (CC BY-SA 4.0)
Videha (विदेह) Maithili biweekly, editor Gajendra Thakur; Tirhuta editions
via Internet Archive items
[`videha_15_04_2008_tirhuta`](https://archive.org/details/videha_15_04_2008_tirhuta),
[`videha_15_12_2011_tirhuta`](https://archive.org/details/videha_15_12_2011_tirhuta),
and
[`videha_01_09_2012_tirhuta`](https://archive.org/details/videha_01_09_2012_tirhuta).
The item metadata identifies Editorial Staff Videha as creator/uploader and
marks the issues [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/).
The item description also says individual authors retain copyright and carries
an all-rights-reserved notice. This project relies on the uploader-applied
item-level CC BY-SA grant for the files while retaining that conflict as a
licensing caveat; redistribution must not imply that individual author rights
were waived beyond the recorded grant.
Used for Tirhuta converter derivation, the sample-scoped Janaki PDF glyph-ID
recovery audit, and real-line candidate construction. The issue-001 original
`videha_01_01_08_tirhuta.pdf` (SHA-256
`91ec43fdc5ccd22cf449457f94e159650b944fea5cf35c7baec89a695d146722`)
backs the 400-row unreviewed candidate. Separately downloaded originals used
for edition/font inventories and the edition-disjoint training candidate are
`videha_15_04_2008_tirhuta.pdf` (SHA-256
`740782ecf5bfa9466727029bcb7733d9c8b046c36d848b598ddc60efc1c51bd2`),
`videha_15_12_2011_tirhuta.pdf` (SHA-256
`acdd97d44d75f06331f728b2916b63cd59c5e5132f68c924baf842db2212d703`),
and `videha_01_09_2012_tirhuta.pdf` (SHA-256
`2ae96ac6e63bdb57fd75f312c8d989d0bb0cc8612b9bb139a34e4276cfc6133c`).
Recovered Unicode is a format conversion and diagnostic annotation of the
source text; attribution and ShareAlike terms apply to redistributed derived
text and crops. The April candidate archive contains 2,000 machine-converted
real-line crops plus a 1,999-row leakage-filtered training view; it remains
unreviewed and is not claim evidence.
## Aale Chhatka (Public Domain Mark 1.0 on 2023 item)
Ol Chiki literary monthly, editor Lakhan Murmu (archive.org uploads by the
editor's account). The 2023 compilation is Internet Archive item
`aale-chhatka-pdf-e-magazine-2023` and backs the public Ol Chiki real-print
eval candidate. Its item page carries the Creative Commons Public Domain Mark
1.0. The mark is not CC0 and is not itself a license; the project treats the
publisher/uploader-applied mark as public-evaluation evidence. The separate
September 2023 item has no license or rights statement on record and is not
public-claim eligible.
## OLCKOptimum fonts (SIL Open Font License 1.1)
The extracted embedded OLCKOptimum Medium and ExtraBlack font name tables carry
the full SIL Open Font License 1.1. Credit Sengellabs; the name tables identify
Birendra Nath Murmu as the font developer and N. C. Murmu as the original Chiki
glyph designer. The Medium file records copyright 2019 by Sengellabs and the
ExtraBlack file records copyright 2017 by Sengellabs.
These extracted font files came from the same Aale Chhatka 2023 compilation
used for real evaluation. Models synthesized with them, including the Ol Chiki
v2 diagnostic model, therefore do not provide source-asset-independent claim
evidence on that edition.