In his lecture “Specialization and Diversity in Dutch and Flemish Printmaking: A Computational Approach,” PhD candidate Matthew Lincoln explored two separate but related questions pertaining to sixteenth- and seventeenth-century Dutch and Flemish printmaking and painting. As part of the Frick Collection’s Digital Art History Lab lecture series, both parts of the lecture addressed how the study of art history can be aided and enhanced by computational analysis.
The first part of the lecture was concentrated on part of Lincoln’s dissertation research in the field of Dutch and Flemish printmaking. He posed the question of whether printmakers gravitated toward specializing in a genre as painters did during the sixteenth and seventeenth centuries. This question is complicated by a number of factors—not least of which is attribution, in part because so many etchers and engravers worked in workshops, sometimes reproducing other artists’ work, and their prints were not always attributed to specific people. Lincoln used three databases to gather his datasets: two contemporary databases, The Rijksbureau voor Kunsthistorische Documentatie (Netherlands Institute for Art History or RKD) and the Rijksmuseum collection, and an archival database, the Montias Database of 17th Century Dutch Art Inventories. The Montias Database differs from the other two databases in that it was built using seventeenth-century inventories of goods, including art, that contain information about who sold which paintings or prints to whom, as well as where and when.
Once Lincoln had established and defined the ten genres he would use and assigned them to specific prints or tags already used by the databases, he used Shannon’s diversity index to measure artists’ diversity of genre. Lincoln did not get into too many details about his process, which seemed appropriate given the time constraints and audience for the lecture, but it was clear enough from the visualizations he did show what trends he was able to find. Painters did, indeed, tend to start specializing over the selected time period and, perhaps predictably, printmakers did not to the same extent. Lincoln was able to infer that almost all printmakers who specialized also made their living as painters. Printmakers who made their living solely from that trade—those who owned their own presses, made reproductions of famous works, etc.—tended to be generalists. He discussed some reasons behind this: commercial motivation and contemporary attitudes toward printmaking, as well as the collectability and accessibility of printmakers’ renditions of paintings.
The second part of the lecture focused on the specific genre of seventeenth-century Haarlem tabletop still life painting. In this, Lincoln collaborated with his colleague Quint Gregory, who had previously developed a database of paintings in the genre, describing what objects were on the tables, the orientation, viewpoint, dimensions, symbolic motifs, and other significant details. The database included a custom-made vocabulary to describe the contents of the paintings as specifically as possible. Using this data from 600 paintings by 10 artists, Lincoln ran random forests of decision trees to test whether the model could predict an artist based on the variables in the database. Moreover, Lincoln wanted to examine why the artists were picked and when the prediction was incorrect, why that might be the case. He looked at which artists tended to be mistaken for one another and why. In some cases the artist was “after” another artist—a member of his workshop, a student, etc.—and were more or less predictable than the artist himself. The model allows one to trace how later artists may have built off their predecessors in a clear, variable by variable way. In this methodology the how and why behind the computation is what is most illuminating.
Lincoln ended his lecture by making the observation that on a macro level, this sort of study brings into question how art historians talk about painting—which words are used and how. The questions that Lincoln posed—of specialization over time and model predictability—both rely heavily on vocabularies and how they are assigned to artists and their art. He suggested that using computing to study ontologies in art history might force a larger conversation about problems and inconsistencies that exist in the field when it comes to language.
Though his methodology was more transparent in the second part of the lecture, Lincoln’s use of computational analysis throughout his work with Dutch and Flemish printmaking and painting is certainly digital humanities scholarship in the realm of art history (especially taking into account his use of networks in other parts of his dissertation research). Examining the long-term trends in printmaking over the sixteenth and seventeenth centuries via subject genres rather than specific prints brings to mind Moretti’s approach to studying literature by examining the rise and fall of various genres in Graphs, Maps, Trees. Both use genre as his lens, but more importantly, it is the objective, data-driven industry-wide trends that are made apparent by their studies.
The tabletop still lifes are an example of a more minutia-based project, but nonetheless demonstrate how digital humanities can be applied to art history. Breaking the paintings down into the variables used as data in this project is similar to so many literary projects that use words and phrases as their data. There is not yet a way to get around the initial human involvement needed to assign words to variables, but once that is done, the computational analysis is more or less the same. That human involvement, however, and the decisions that need to be made about vocabulary are, as Lincoln suggested, what can be problematic. Still, using digital tools to study art history and thinking critically about why a computer gets specific results has the potential to lead to interesting new avenues of analysis.