Detecting laboratory objects in early modern illustrations
Early modern technical handbooks increasingly draw on and re-use illustrations of laboratory equipment that are investigated using computer vision methods following the Distant Viewing paradigm.
The computer vision subproject investigates alchemical laboratory objects through computer vision methods applied to a corpus of early modern handbooks and technical treatises. These works incorporated increasingly more sophisticated illustrations of laboratory equipment and experimental setups. The project builds on an annotated computer vision dataset (Lang, 2025) and explores the possibilities of computational image analysis in this context. At the same time, it critically examines the limitations of computer vision approaches for early modern materials (Lang, 2026). The primary challenge is not the visual style of the images themselves but the difficulty of annotating and analysing concepts that contemporary algorithms have not been trained to recognise (Lang et al., 2023). This reveals forms of bias that often remain hidden within computational approaches commonly associated with distant viewing in the computational humanities.
References
2026
(Doing) Computational History: On the Role of Data Work in Computational Approaches
Computational methods have become increasingly prominent within the historical sciences, generating significant enthusiasm among some scholars. Yet their practical demands, epistemic limits, and ethical implications are less often critically examined than praised. This article explores what it means to do computational history today, arguing that it is not primarily defined by algorithms but by datasets. It is methodologically specific, resource-intensive, selective in scope, labour-heavy, and dependent on pre-digitised sources, specialised infrastructure, and interdisciplinary collaboration. These dependencies limit the scope of research questions and can produce narrow outcomes despite substantial effort, lending some validity to the concern over whether the field yields sufficient historiographical return for the labour invested. Corpus construction and data work lie at the epistemic core of computational history. These often undervalued tasks are not merely technical precursors to analysis, but interpretive and epistemic acts. Data are shaped by digitisation politics, historical bias, and institutional power. They shape the questions asked, the answers produced, and the legitimacy of findings. Recognising and valuing data work is essential, both to embed critical perspectives into computational humanities and to counteract the privileging of certain forms of labour over others. Due to the association of quantification with rigour and scholarly prowess, algorithmic work receives more credit, creating a two-tier system in this division of labour in which those who develop algorithms are elevated above those who curate data, despite their symbiotic interdependence. Computational history, when done well, requires deep engagement with our sources, be they historical or data. For computational history to stabilise as a meaningful discipline, it must prioritise building better datasets over pursuing increasingly complex algorithms on an unstable basis of data.
@article{Lang2026DoingComputationalHistory,author={Lang, Sarah},title={(Doing) Computational History: On the Role of Data Work in Computational Approaches},journal={Histories},volume={6},number={2},pages={26},year={2026},url={https://doi.org/10.3390/histories6020026},doi={10.3390/histories6020026},keywords={computational history; digital humanities; computational humanities; data work; knowledge work; corpus criticism; data ethics},}
2025
Fine-Tuning Machine Learning with Historical Data: An Alchemical Object Detection Dataset for Early Modern Scientific Illustrations
Sarah Lang
Zeitschrift für digitale Geisteswissenschaften, 2025
This paper introduces a dataset designed to improve object detection for laboratory equipment in historical handbooks. Derived from digitized sources at the Herzog August Bibliothek, the dataset includes pixel-level annotations of laboratory apparatus like cucurbitae, ampullae, and furnaces and allows researchers to study how chymical knowledge was transmitted visually. Adapting an Iconclass-based taxonomy revealed discrepancies between theoretical classifications and practical image annotation, as some alchemical vessels appear nearly identical but were labeled differently according to their functions by early modern authors, which complicates annotation for models that focus on visuality rather than concepts. This underscores that historical data requires tailored computational approaches as training data significantly shapes how machines interpret historical materials.
@article{Lang2025ZfdG,author={Lang, Sarah},title={Fine-Tuning Machine Learning with Historical Data: An Alchemical Object Detection Dataset for Early Modern Scientific Illustrations},journal={Zeitschrift für digitale Geisteswissenschaften},volume={10},year={2025},doi={10.17175/2025_002},url={https://zfdg.de/2025_002},}
2023
Toward a Computational Historiography of Alchemy: Challenges and Obstacles of Object Detection for Historical Illustrations of Mining, Metallurgy, and Distillation in 16th–17th Century Print
Sarah A. Lang, Bernhard Liebl, and Manuel Burghardt
In Proceedings of the Computational Humanities Research Conference 2023 (CHR 2023), 2023
This study explores the use of modern computer vision methods for object detection in historical images extracted from 16th–17th century printed books containing illustrations of distillation, mining, metallurgy, and alchemical apparatus. We found that the transfer of knowledge from contemporary photographic data to historical etchings proves less effective than anticipated, revealing limitations in current methods like visual feature descriptors, pixel segmentation, representation learning, and object detection with YOLOv8. These findings highlight the stylistic disparities between modern images and early print illustrations, suggesting new research directions for historical image analysis.
@inproceedings{LangLieblBurghardt2023,author={Lang, Sarah A. and Liebl, Bernhard and Burghardt, Manuel},title={Toward a Computational Historiography of Alchemy: Challenges and Obstacles of Object Detection for Historical Illustrations of Mining, Metallurgy, and Distillation in 16th--17th Century Print},booktitle={Proceedings of the Computational Humanities Research Conference 2023 (CHR 2023)},editor={{\v{S}}e{\c{l}}a, Artjoms and Jannidis, Fotis and Romanowska, Iza},year={2023},publisher={CEUR-WS},address={Aachen},pages={29--48},keywords={computer vision, object detection, alchemy, chymistry, early-modern print, metallurgy, mining, distillation, annotation}}