NewsEye/READ: French Newspaper Pages with Annotated Text

French newspaper pages from the 19th and early 20th centuries, provided by the French National Library. The dataset contains page images with annotated text in the PAGE XML format, produced via the Transkribus platform. It was created with support from the NewsEye and READ projects for training automated text recognition models.

Use Cases

• [Training OCR/HTR models] based on [annotated text regions in PAGE XML format]
• [Historical text analysis] based on [extracted content from 19th-20th century French newspapers]
• [Layout analysis of historical documents] based on [annotated page images]
• [Studying the evolution of French language and media] based on [text from a defined historical period]

Strengths

• Annotations follow the structured PAGE XML format, a standard for document layout analysis.
• Data provenance is clear, sourced from the French National Library and created with the Transkribus platform.
• Supports specific research projects (NewsEye, READ), indicating a defined use case.

Limitations

• The exact number of pages is unclear, with sources conflicting between 183 and 184 for the training set.
• Metadata is sparse; specific column names, file formats, and the last update date are unknown.
• The dataset's scale is limited, likely containing only a few hundred annotated pages.

Provenance

Source

French National Library (Bibliothèque nationale de France).

Collection Method

Page images were annotated using the Transkribus platform, following specific guidelines.

Time Range

19th and early 20th centuries.

Geography

France.