README.md

metadata

language:
  - la
dataset_info:
  features:
    - name: image
      dtype: image
    - name: img_url
      dtype: string
    - name: bounding_boxes
      sequence:
        sequence: int32
    - name: transcriptions
      sequence: string
    - name: text
      dtype: string
  splits:
    - name: train
      num_bytes: 4526339380.8
      num_examples: 4014
  download_size: 4189436834
  dataset_size: 4526339380.8
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Guiding Principles of this Dataset

Segmentation

• The goal is here to segment the "main text" while ignoring paratext. This means page numbers, catch words, marginal annotations, are purposefully not segmented

Transcription

• The guiding transcription principles are to
  • expand abbreviations
  • preserve orthography