welsh-texts

The National Library of Wales and The Welsh Government have authorized the hosting, distribution, and use of this dataset for public use, including research, scholarship, and machine learning.

License: CC-BY-SA

This dataset contains a variety of printed / handwritten material from Welsh sources, mostly in the Welsh language:
- Drych y Prif Oesoedd by Theophilus Evans - a book on the early history of Wales (published 1716)
- Enwogion Cymreig by Thomas Morgan - a book cataloging prominent figures in Welsh history (published 1907)
- Cronicl Elis Gruffudd by Elis Gruffudd - a handwritten manuscript on history (published 1552).
- Bibliographic index cards (bitonal digitization from microfiche). Contains a "blue" set (lower legibility, mix of print and handwritten, 1-up to a page) and a "green" set (better legibility, variable typeface but all printed, 4-up to a page). These are largely in English
- Emigrant Letters (Mostly welsh language, handwritten) from Henry Jones who migrated to Holland Patent, New York in 1850
- American Civil War letters - Welsh letters from members of the Winsconsin Volunteers (Union Army) 23rd & 24th regiment.

The dataset is packaged with Parquet. Each source is formatted as a collection of image files (either JPEG2000 or PNG).

For a subset of the material (Drych y Prif Oesoedd, Enowogion Cymreig, Bibliographic catalog index cards), we also provide OCR transcriptions from GPT 4-o (for this cases, the Parquet includes content in a `text` column). These transcriptions are known to contain some errors, although we provide them as they may be helpful in searching through the text or use as a baseline when developing other OCR systems.

威尔士国家图书馆与威尔士政府已授权本数据集的托管、分发与公共使用，允许其应用于科研、学术研究及机器学习等领域。
许可证：CC-BY-SA

本数据集包含源自威尔士本土的各类印刷与手写资料，其中绝大多数以威尔士语撰写：
- 《Drych y Prif Oesoedd》（作者Theophilus Evans）：威尔士早期历史专著，首版于1716年
- 《Enwogion Cymreig》（作者Thomas Morgan）：威尔士历史名人名录，首版于1907年
- 《Cronicl Elis Gruffudd》（作者Elis Gruffudd）：手写历史手稿，成书于1552年
- 书目索引卡片：基于微缩胶片完成双色调数字化处理，包含「蓝色组」与「绿色组」两个子集。蓝色组可读性较弱，混合印刷与手写文本，每页仅收录1份内容；绿色组可读性更佳，字体多样但均为印刷体，每页可容纳4份内容。该批资料以英文为主。
- 移民信件：主要为威尔士语手写信件，作者为1850年移民至美国纽约州荷兰帕滕特的Henry Jones
- 美国内战信件：隶属于联邦军威斯康星志愿兵第23、24团的威尔士籍士兵所撰写的信件。

本数据集以Parquet格式封装，每个数据源均以图像文件集合形式存储，图像格式涵盖JPEG2000与PNG。

针对部分资料（《Drych y Prif Oesoedd》、《Enwogion Cymreig》、书目索引卡片），我们还提供了由GPT-4-o生成的OCR转录文本：此类内容对应的Parquet文件包含名为`text`的字段。已知该转录文本存在一定错误，但仍予以提供，因其可用于文本检索，或作为开发其他OCR系统的基准参照。