Kaspar Beelen

Heritage Weaver investigates the use of multimodal AI to link and explore museum data across collections. Through a series of experiments, from zero-shot learning and information retrieval to record linking, we demonstrate the value of fine-tuning multimodal models on digital heritage. The paper elaborates on various evaluation strategies, leveraging existing metadata or using expert annotations, to measure improvements in the model’s “understanding” of often complex and messy historical materials.

This article implements a critical method for assessing bias in large historical datasets that we term the “Environmental Scan.” The Environmental Scan sheds new light on newspaper collections by linking newly available “reference metadata” gathered from historical sources to existing full-text and catalogue metadata. The rise of computational methods in history and the social sciences, in tandem with newly “datafied” source materials, creates a challenge for researchers to adapt their existing critical practices to the increasing scale and complexity of computational research. To help address this challenge, the Environmental Scan situates big historical datasets in much greater context, including estimating what materials are missing, thereby revealing the ways digital collections can be “oligoptic” in nature. Using the British Newspaper Archive (BNA) as a case study, we diagnose the biases and imbalances in the digitised Victorian press. We determine which voices are under- or over-represented in relation to the political composition of the collection as well as its content and we trace the origins of these biases in the digitisation process. This article informs future interdisciplinary discussions about data bias and offers a conceptual model adaptable to diverse historical datasets. The Environmental Scan provides a more nuanced and accurate understanding of how newspaper data reflects past societies, making it a valuable tool for researchers.

MapReader is an interdisciplinary software library for processing digitized maps (Hosseini et al., 2022a) and other types of images with two tasks: patch classification and text spotting. Patch classification works by 'patching' images into small, custom-sized cells which are then classified according to the user's needs. Text spotting detects and recognizes text. MapReader offers a flexible pipeline which can be used both for manual annotation of small datasets as well as for computer-vision-based inference of large collections. As an example, in Hosseini et al. (2022a), we annotated 62,020 patches, trained a suite of computer vision models and performed model inference on approximately 30.5 million patches.

List of Contributors was published in Digitised Newspapers – A New Eldorado for Historians? on page 437.

This article introduces the ‘Environmental Scan’ as answer to the question of hidden biases in digital heritage collections. Its substantive focus is digitized nineteenth-century British provincial newspapers, and in particular the JISC corpus, a popular, publicly-funded resource for scholars. While multiple papers have meticulously investigated the genesis of such newspaper collections, in the process highlighting the often unacknowledged politics of collection, preservation and dissemination via microfilm and now digitization, our aim is to explore questions of representativeness and bias in new ways by enriching computational analysis of digital corpora with the historical insights that can be derived from a contemporaneous reference source: namely, the Victorian newspaper press directories.

Mariona Coll Ardanuy, Federico Nanni, Kaspar Beelen, Kasra Hosseini, Ruth Ahnert, Jon Lawrence, Katherine McDonough, Giorgia Tolfo, Daniel CS Wilson, Barbara McGillivray. Proceedings of the 28th International Conference on Computational Linguistics. 2020.