VOC Data – IIIF Manifests with Transcriptions

This repository contains a Jupyter notebook for processing IIIF manifests from the NL-HaNA VOC Archives, augmented with page-level transcription text. The notebook downloads transcription TXT files and corresponding IIIF JSON manifests, merges the text into each manifest's canvases, and saves per-document outputs as well as a combined manifest.

Contents

• VOC_Data_Downloader.ipynb: Batch processor that downloads, merges, and exports manifests.
• upload_to_huggingface.py: Script to upload generated datasets to Hugging Face.
• globalise_transcriptions_v2_txt.tab: Source file with transcription URLs.

Accessing the Dataset

The processed dataset (19GB of IIIF manifests with transcriptions) is hosted on Hugging Face Datasets for free, fast access:

🤗 Hugging Face: huggingface.co/datasets/adoistic/voc-data

Download the dataset:

from huggingface_hub import hf_hub_download

# Download the combined manifest (7GB)
file = hf_hub_download(
    repo_id="adoistic/voc-data",
    filename="combined_all_manifests.json",
    repo_type="dataset"
)

# Or download individual manifest files
file = hf_hub_download(
    repo_id="adoistic/voc-data",
    filename="merged_1111.json",
    repo_type="dataset"
)

# Load and use the data
import json
with open(file, 'r') as f:
    manifest = json.load(f)

List all available files:

from huggingface_hub import list_repo_files

files = list_repo_files("adoistic/voc-data", repo_type="dataset")
print(f"Available files: {len(files)}")

Data source and license

• Manifests: https://data.globalise.huygens.knaw.nl/manifests/inventories/{id}.json
• Images: IIIF Image API services referenced within each canvas
• Rights: https://creativecommons.org/publicdomain/mark/1.0/ (as present in manifests)

Notebook overview

The notebook performs these steps:

1. Read a .tab file listing transcription TXT URLs (one per line).
2. For each URL:
   • Download the TXT file; infer the document identifier (e.g., 1111) from the filename.
   • Construct the corresponding IIIF manifest URL via a template.
   • Parse the TXT into sections keyed by canvas label IDs (e.g., NL-HaNA_1.04.02_1111_0001).
   • Merge the matched text into the manifest by setting a text field on each canvas.
   • Save as merged_{id}.json under merged_manifests/.
3. Create combined_all_manifests.json that concatenates all canvases from the successfully processed manifests.
4. Zip the merged_manifests/ folder for convenience.

Output schema (IIIF Presentation 3 with additions)

Each merged_*.json is an IIIF Manifest with a non-standard addition: a text field on each canvas. Key fields:

• Manifest
  • @context: http://iiif.io/api/presentation/3/context.json
  • id, type: Manifest
  • label, metadata, rights
  • items: Array of canvases
• Canvas (per page)
  • id, type: Canvas
  • label.en[0]: Canvas identifier, e.g., NL-HaNA_1.04.02_1111_0001
  • height, width
  • items[0].items[0].body: IIIF Image resource
  • text: String of merged transcription for this canvas (may be empty if no match)

Example (truncated):

{
  "type": "Manifest",
  "label": {"en": ["Inventory 1111"]},
  "items": [
    {
      "type": "Canvas",
      "label": {"en": ["NL-HaNA_1.04.02_1111_0001"]},
      "items": [
        {
          "type": "AnnotationPage",
          "items": [
            {"type": "Annotation", "motivation": "painting", "body": {"type": "Image", "format": "image/jpeg"}}
          ]
        }
      ],
      "text": "... merged transcription text ..."
    }
  ]
}

combined_all_manifests.json is also a Manifest whose items array is the concatenation of the items from all successful merged_*.json files. Its label and metadata summarize the collection.

How to Generate the Dataset Yourself

If you want to generate the dataset from scratch:

1. Install dependencies:

pip install requests huggingface_hub

2. Run the notebook: Open VOC_Data_Downloader.ipynb in Jupyter and run all cells, or call main(...):

main('globalise_transcriptions_v2_txt.tab')

Default concurrency is 15 workers; adjust via max_workers if needed:

main('globalise_transcriptions_v2_txt.tab', max_workers=20)

Outputs will be written to:

• merged_manifests/ (individual manifests and combined_all_manifests.json)
• merged_manifests.zip

3. Upload to Hugging Face (Optional):

To share your generated dataset:

a. Create a free account at huggingface.co b. Get your API token from Settings > Access Tokens c. Run the upload script:

python3 upload_to_huggingface.py \
    --token YOUR_HF_TOKEN \
    --repo-id your-username/voc-data

This uploads all files in merged_manifests/ to your Hugging Face dataset repository.

Environment requirements

The notebook depends on standard Python 3 libraries plus requests. Minimal setup:

pip install requests

Recommended: Run inside a virtual environment. The notebook uses network access to fetch TXT and JSON files.

Notes and caveats

• Canvas text is added locally and is not part of the IIIF Presentation 3 standard; downstream tools should treat it as a custom extension.
• TXT-to-canvas matching relies on labels like NL-HaNA_1.04.02_{id}_{page}; if labels or TXT formatting differ, some canvases may have empty text.
• combined_all_manifests.json may be large; use streaming parsers for programmatic consumption.

Citation

If you use this data, please credit the NL-HaNA VOC Archives and the GLOBALISE project data services. Include a link to the manifest endpoints and the public domain mark indicated in the manifests.