---
license: cc0-1.0
language:
- en
task_categories:
- text-generation
task_ids:
- language-modeling
tags:
- history
- newspapers
- religion
- ocr
- digital-humanities
size_categories:
- 100K<n<1M
---

---
# Dataset Card for Spiritualist_Newspaper

### Dataset Description

This resource includes plain text transcriptions of *The Spiritualist Newspaper* (1869), created using Transkribus (https://app.transkribus.org/) and manually corrected across the first serialisation's 50 pages. The transcriptions included were generated using a text extraction model, layout analysis model and text region classifier, all trained within the Transkribus environment. The former model is available here for quick transcription drag and drop tests, entitled 'Ghostwriter' - https://app.transkribus.org/models/public/text/421381, and was trained on 158,825 words, achieving a character error rate of 0.86%.
The .txt transcription included in this resource was used as a reference string for calculating character error rates against other OCR engines as part of an internally funded National Library of Scotland (NLS) fellowship, within a CERberus local host environment - https://github.com/WHaverals/CERberus. The engines tested included: OCR4ALL, eScriptorium, Transkribus, Tesseract, EasyOCR, Paddle OCR, Surya OCR, GPT 4o, Claude Sonnet 4.5 and Gemini-Pro-3.
We also include the page images used, with thresholding and additional cropping conducted with R Studio using Image Magick, as well as page-level ALTO XML for interactive image hosting in compliance with IIIF standards (https://iiif.io).


-- ABOUT THE SPIRITUALIST --

*The Spiritalist Newspaper*, published by E.W. Allen (London), forms a key source for how those interested in physical investigations and spiritual forces communicated with the deceased, argued about criticism and issues, as well as publicised gatherings. The Newspaper followed a wake of spiritualist activity in Britain, directed by earlier 1850s American Transcendentalism and a more genral reaction to scientific naturalism.

*The Spiritualist* was chosen as an OCR case study for several reasons:

- Scottish spiritualist histories remain underrepresented as Foot (2023) discusses. In our case, this partially results from historical detail being contained in heterogeneous and hard-to-transcribe material. This makes accurate OCR recognition more challenging, especially without requisite training data. With a fully searchable sample of 50 pages of *The Spiritualist* enabled through our OCR testing and reference, manually corrected, transcription, spiritualist activity outside of Scotland's Central Belt can be further explored. The trained reference model will also be applied to the entire collection as future work.

- In addition, Foot (2023: 73) suggests that seance mediums underwent 'automatic writing' to channel spiritual forces. The latter insight forms a neat parallel to charting the extent of unexplainable OCR processes in the current technical offering to libraries, as well as the potentially opaque behaviour and outputs of AI-enabled transcription models.

Alongside these two primary reasons, the project also aimed to provide metrics for the standardisation of OCR transcription at the NLS, with the Library currently undergoing a reassessment of its Foxit / ABBYY FineReader license. The timings registered through the below trials, alongside the CER of outputs, were used as an evidence base for NLS decision-making around digital collection transcription.

- **Data format**: png. manuscript images, pre-processed (thresholding and additional cropping) in R Studio using Image Magick.
- **Data creation process**: Transformer-based Optical Character Recognition.
- **Data source**: https://data.nls.uk/data/digitised-collections/spiritualist-newspapers/, 50 pages of the first 1869 edition.Provide a longer summary of what this dataset is. 

![Public trained model in Transkribus environment, used to generate OCR text](Ghostwriter.png)

-- ABOUT THE PROJECT --

This transcription output formed part of Dr Joe Nockels's research, 'Recognising Text, Recognising Processes - eXplainable Automatic Text Recognition for Scottish Spiritualist Newspapers', which aimed to establish how far Automatic Text Recognition can coalesce with 'eXplainable' developmental principles, and the extent to which such principles can aid non-technical library users in their understanding of how AI reaches its transcription results. As such, the project's OCR experiments, which ranged from using open-source models to commercial LLMs, occurred alongside a thematic coding of tool's public-facing documentation, including how-to guides, model cards and GitHub README files, as well as other grey literature such as technical papers produced by developmental consortia. This was done through a bespoke web crawler using in-built request delays to ensure server compliance, with eXplainable themes coded via NVIVO (https://lumivero.com/products/nvivo/).

--- Citations ---

Ames, S., Havens, L. (2022) Exploring National Library of Scotland datasets with Jupyter Notebooks, International Federation of Library Associations and Institutions, 48(1): 50-56. https://doi.org/10.1177/03400352211065484
Foot, M. (2023) Modern Spiritualism and Scottish Art - Scots, Spirits and Seances, 1860 - 1940. London: Bloomsbury.
Haverals, W. (2023) CERberus: Guardian Against Character Errors. 2023, v1.0, https://github.com/WHaverals/CERberus.


- **Curated by:** [Dr Joe Nockels, with support from Dr Sarah Ames, NLS Digitisation Unit]
- **Funded by** [NLS internally, National Librarian's Fellow in Digital Scholarship]:
- **Shared by** [Dr Joe Nockels]:
- **Language(s) (NLP):** [Eng]
- **License:** [CC0, Please refer to the National Library of Scotland's terms of use for the original data. Users should comply with NLS's licensing terms when using this dataset.]

### Dataset Sources

https://data.nls.uk/data/digitised-collections/spiritualist-newspapers/

- **Repository:** [https://data.nls.uk]

## Dataset Structure 

- `/images/`: Page images (PNG)
- `/text/`: Corrected plain-text transcription
- `/alto/`: Page-level ALTO XML (to be uploaded)

## Uses

The dataset, manually cleaned from Transkribus modelling, was used as a reference string for measuring Character 
Error Rates in other Automatic Text Recognition (ATR) tools, using Haveral's (2023) CERberus local host: https://github.com/WHaverals/CERberus.

### Direct Use

The aim is to use this dataset as Ground Truth for an ATR Spiritualist model, across the whole 1869 - 1882 serialisation.

## Citation

```bibtex
@misc{nls_spiritualist_newspaper,
  title={Spiritualist Newspaper Dataset},
  author={National Library of Scotland},
  year={1869},
  url={https://data.nls.uk/data/digitised-collections/spiritualist-newspapers/}
}

Transcription generated by Joe Nockels (j.nockels@sheffield.ac.uk)