Update: This work is something that I am committed to pursuing but was not able to work on in time for the DH2024 conference, unfortunately. I am still sharing the proposal and project information in case it is of interest and to continue to signal my work. And, I think it’s useful to be honest about moments of – not failure – but adjustment that acknowledge when we aren’t able to complete our ambitions as we hoped because work and life get in the way.
ADHO Digital Humanities Conference
August 5-9, 2024
Alexandria, Virginia
Abstract
Digital humanities projects benefit from sharing their underlying data because doing so creates opportunities for new research. The Roman de la Rose Digital Library (RDL) is such a project, and this poster will show the results of applying computer vision to the manuscript corpus and comparing it with human-created metadata.
Proposal
Digital Humanities projects require immense commitment and resources, and many of them endeavor to make their data, materials, and outputs available for reuse. While some work does take on further life through this open process, many projects end intentionally or lose momentum and languish. There is great opportunity for digital humanists to work with data and existing projects, to further our own research interests and to continue the work and honor the value of past projects.
The Roman de la Rose Digital Library (RDL) was begun in 1996 by Stephen G. Nichols, and reached a milestone in 2017 with a special issue in Digital Philology, celebrating its development into the Digital Library of Medieval Manuscripts (McWebb and Walters 2017). It continues today as a resource for the study of the most popular secular work of the European Middle Ages, providing access to 146 digitized manuscripts in IIIF format and with additional datasets describing the manuscripts in the corpus.
I have been working with this material to learn about patterns of book production and to experiment with digital humanities methods of visualization and analysis on a pre-existing set of material. I am interested in creating teaching materials and interactive avenues for students and scholars to explore both the digitized manuscripts and the research into them.
The codicological and descriptive information required cleaning and supplementation, but I have built on the spreadsheets provided by the RDL to create interactive visualizations to allow a user to engage with the information in new ways. I also gathered images of 28 of the digitized manuscripts (digitized copies from the Bibliothèque nationale de France that had more than one illustration) to experiment with color metrics, visuals using Imageplot, and developing a pilot recommender system.
In this poster, I will report on the next phase of the project, which is two-fold: (1) to work with the IIIF API to pull all images from all digitized copies available in the RDL; and (2) to use the Distant Viewing Toolkit across the digitized corpus to develop a computer vision driven recommender system and to complement the existing human-created metadata in the RDL.
The use of computer vision in the digital humanities has developed immensely in the past several years, and there are several approaches that computational image analysis can take (van Noord 2022). Taylor and Arnold’s theoretical framework of Distant Viewing (2023) distinguishes the visual rhetoric and affordances of the image in contradistinction to text, helping to distinguish the work of computational image analysis from distant reading, while pointing to some of the intriguing possibilities of both digital humanities methods.
The Distant Viewing Toolkit will apply facial and object detection across the digitized manuscripts in the corpus, creating a recommender system for exploring all images/pages in the corpus . The RDL has datasets of Illustration Titles and Narrative Sections, providing insight into characters and their frequency across the many variations of the corpus. Bringing the RDL datasets into comparison with algorithmic analysis of characters may reveal new understandings of the consistency of figures in illustrations.
Applying new methods to this corpus will open new avenues of research for medieval studies scholars interested in the history of the book, illustration transmission, and more. The process of developing these systems, and then others’ potential use of them, also provide opportunities to craft training materials and example sets for classes, in digital humanities and beyond.
In addition to the research and teaching materials developed through this project, I see this work as an example of how digital humanities projects benefit from sharing their material and process openly. Because the RDL makes their spreadsheets available, others can create new work and pick up the research. By making the manuscripts publicly visible and accessible through a IIIF Mirador instance, the RDL has enabled new forms of computational research, which hopefully will be able to feed back into the RDL itself, allowing for project reinvention and development in itself.
Works Cited
Arnold, Taylor and Lauren Tilton. (2023) Distant Viewing: Computational Exploration of Digital Images. MIT Press.
McWebb, C., & Walters, L.J. (2017). Introduction: The Roman de la Rose and Christine de Pizan in the Digital Age. Digital Philology: A Journal of Medieval Cultures 6(1), 1-14. https://doi.org/10.1353/dph.2017.0000.van Noord, N. (2022). A survey of computational methods for iconic image analysis. Digital Scholarship in the Humanities, 37(4), 1316-1338.