README.md

metadata

language:
  - da
license: cc-by-4.0
size_categories:
  - 1K<n<10K
task_categories:
  - image-to-text
pretty_name: Danish Handwritten Text 1841-1939
tags:
  - OCR
  - HTR
  - handwritten
  - historical
dataset_info:
  features:
    - name: image
      dtype: image
    - name: docId
      dtype: int64
    - name: docTitle
      dtype: string
    - name: sequence
      dtype: int16
    - name: alto
      dtype: string
    - name: page
      dtype: string
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Dataset Description

Dataset Summary

The Danish handwriting dataset is a Danish-language dataset containing around 3200 pages of proofread handwritten text.

Languages

All the text is in Danish. The BCP-47 code for Danish is da.

Dataset Structure

Data Instances

Each data instance describes and represents a single scanned, segmented and transcribed physical page with handwritten text. The handwriting originates in

Data Fields

List and describe the fields present in the dataset. Mention their data type, and whether they are used as input or output in any of the tasks the dataset currently supports. If the data has span indices, describe their attributes, such as whether they are at the character level or word level, whether they are contiguous or not, etc. If the datasets contains example IDs, state whether they have an inherent meaning, such as a mapping to other datasets or pointing to relationships between data points.

• id: an integer representing the unique id of each page
• image: a jpeg image containing a scan of the original physical page
• title: a string containing both the title of the book/protocol to which the page belongs, plus an optional page-range, indicating which document-chunk the page belongs to
• sequence: an integer representing the specific pagenumber of the page within the current page chunk
• alto: an xml-encoded string containing layout and content information of the physical page, stored in ALTO XML, version 4.2
• page: an xml-encoded string containing layout and content information of the physical page, stored in PAGE XML, version 2013-07-15

Data Splits

[N/A]

Dataset Creation

Curation Rationale

The dataset originates from a need to make physical public protocols discoverable via fulltext search and filtering.

Source Data

The source data consists of written minutes from different City and Parish Council meetings, dating from 1841 to 1939.

Initial Data Collection and Normalization

The original physical minutes where collected and curated as part of the judiciary demands of the city archives in Denmark.

Who are the source language producers?

The digitized and transcribed minutes were originally written by the secretaries of the individual City and Parish Councils.

The layout segmentation and transcription of the digitized minutes is primarily done by volenteers and employees at the participating Danish city archives.

All layout segmentation and transcription is done using the Transkribus platform, either through the desktop client or via the web-based interface.

Annotations

Annotation process

When using the Transkribus clients to segment and transcribe the scanned pages, it is possible to add TEI-tags to single words or whole paragraphs.

These annotations have not been applied in any consistent manner, and should be ignored.

Who are the annotators?

The annotations are produced inside Transkribus by the volunteers and employees transcribing and proofreading the documents.

Personal and Sensitive Information

No efforts were made to anonymize the data.

The dataset might contain data that can be considered sensitive (e.g., data that reveals racial or ethnic origins, sexual orientations, religious beliefs, political opinions or union memberships, or locations; financial or health data; biometric or genetic data; forms of government identification, such as social security numbers; criminal history).

Additional Information

Dataset Curators

The Danish handwriting dataset is curated by Aarhus City Archives, Denmark.

Licensing Information

The Danish handwriting dataset version 1.0.0 is released under the Creative Commons Attribution 4.0 International

Contributions

Many thanks to the many volunteers that have participated in the segmentation, trancription and proof reading of the archives!

• Point of Contact: Claus Juhl Knudsen