Thursday, January 31, 2013

Using Computers to Date Medieval Manuscripts

A recent article in the Journal of Applied Statistics by two computer scientists and a medieval historian at the University of Toronto details their work using computer algorithms to date previously undated charters from the post-Conquest period in England. The researchers used the archive of dated early English charters at the University of Toronto to devise algorithms that can offer additional tools for dating medieval documents. 

By using a data set that has dates attached, the researchers were able to compare their results with existing (and, presumably, accepted) data. They based the algorithms on specific phrases that appear in charters and that vary over time. A charter that had been dated by a medievalist as having been written between 1235 and 1245 was compared using the data from the training set, and was dated by computational methods to 1246. 

It's fantastic to see another instance of humanists and scientists working together to solve historical problems. Archaeologists, paleo-epidemiologists, and biological anthropologists have been using scientific methods to shed more light on the bacterial cause of the Black Death and the reasons for its rapid spread in the 14th century. Yet the use of computational algorithms suggests several questions that remain to be answered. A specific question has to do with forgeries. Many medieval charters are forgeries. The most famous is the Donation of Constantine, which was discovered to be a fraud by the fifteenth century. It's not yet clear how a computer algorithm could help identify forged charters. Some may betray themselves with anachronistic phrasing, but others were written with enough care or close enough to the time that they purported to be from that looking for particular words and phrases would not uncover them.

Another, larger, question has to do with the way that text-based analysis and scientific methods can work together to shed new light on old questions. The Donation of Constantine was discovered to be a fraud by using similar methods: checking the language of the document itself. Two versions of the history of the Trojan War, widely accepted throughout the Middle Ages as being eyewitness accounts of the conflict (and more accurate than Homer's poetry), were unmasked as frauds in the early modern period--again, due to the efforts of scholars who painstakingly compared the texts with other, external data. It's not yet clear if computational methods are more accurate than what we might call scholarly expertise (though, of course, the scientists writing the algorithms are experts themselves), or computational methods offer more speed, and thus the ability to examine larger data sets. Both benefits can be enormously useful to medievalists. I can imagine that scholars working on projects that rely on charters as evidence might be grateful to have tools that could confirm (or call into question) their own chronology.

No comments: