Decoding the Past: Professor Emily Kadens uses Artificial Intelligence to bring 16th-Century Law to Life

What happens when 16th-century legal documents meet artificial intelligence? Northwestern University Pritzker School of Law’s Edna B. and Ednyfed H. Williams Memorial Professor and Vice Dean Emily Kadens is answering that very question through her current research project, which blends centuries-old court equity records with the illuminating capabilities of modern AI.

For the past three years, Kadens and a small research team have worked to build a machine-learning model that can decipher English secretary hand —a notoriously difficult handwriting style used in legal documents during the 16th and 17th centuries.

“Our goal was to create a model that transcribes documents written in English secretary hand,” Kadens says. “There are hundreds of thousands of extant documents that, through this project, we will be able to make more accessible.”

To build the model, Kadens and her team have used a program called Transkribus, an AI platform that automates transcription of handwritten text. Kadens affectionately nicknamed her model Egerton after Thomas Egerton, the longtime Lord Chancellor who presided over the English Court of Chancery during the era she is studying.

Transkribus: An Accidental Goldmine

As a legal historian, Kadens focuses her research specifically on commercial fraud during the time in which these documents were written. She works primarily with the records of equity courts—royal courts of discretionary jurisdiction that were especially popular in the sixteenth century. She notes that equity court documents—primarily documented in secretary hand—contain much more detail about the disputes compared to the records of the common law courts, which used a largely oral procedure.

“These 16th- and 17th-century equity court documents offer a goldmine of information about early modern English life,” she says. “The records recount stories of commerce, family disputes, property conflicts, and more.”

With minimal prior use or knowledge of artificial intelligence in her work, Kadens says she came to this research juncture by accident. Prior to discovering Transkribus, she would transcribe court documents manually during her regular trips to the National Archives in London. Having previously transcribed more than a thousand pages of documents, she began looking for a way to share her work online. That’s when she found Transkribus.

“Instead of simply making the files available, I learned that I could use my manual transcriptions to seed a model for the public to use,” she says.

Currently, Kadens and her team have put nearly one million words into the model, most coming from 16th- and early-17^th-century equity court documents. Once they hit the million-word mark, they plan to make the model available to the public. Their goal is to do so later this year.

The Process and the Challenge

While the team is well on their way to completing their work, the journey hasn’t always been easy. One such bump in the road came in the form of handwriting variances. 

“Many scribes had very idiosyncratic forms of secretary hand that, to the lay eye, seem nothing alike,” Kadens says. “We have sometimes struggled to read the worst hands ourselves, much less teaching the model to decipher them.” 

To train the model, Kadens and her team developed a 17-page transcription convention guide to address myriad situations that arose in the documents, from abbreviations and punctuation to Latin phrases.

“It is amazing to see what the technology can do,” Kadens says. “Egerton can correctly transcribe words where the top half of the letters are cut off or some of the letters are completely obscured. Sometimes it can read scripts better than I can now.”

The result has been extraordinary. The model now achieves an average character error rate (CER) of just 2.9 percent—in other words, its transcriptions are better than 97 percent accurate. Kadens stresses, however, that even the error rate contains caveats.

“Many of the mistakes Egerton makes now are insignificant errors, such as spelling “defendant” with two ff’s. We have gotten the model to the point of producing transcriptions that are usable right out of the box without needing to correct much.”

Kadens says she hopes this model will open the door to greater efficiency in working with handwritten documents for other historians and legal scholars.

“Egerton will reorganize workflow for historians of early modern England. You can’t skim this stuff. You have to actually read it,” she says. “If scholars can run their a document through Egerton in a matter of minutes and obtain an accurate transcription, that’s a huge time-saver!”

Egerton Model Brings History to Life for Students

Bridging research and pedagogy, last spring Kadens had the opportunity to teach a course called the Legal History Workshop, in which she used Egerton to introduce students to equity court documents. During the course, students studied the equity courts and learned to read secretary hand. Then, during Spring Break, they traveled to the National Archives in London, where they sorted through boxes of court documents from the Court of Chancery to find a case from the 1590s, transcribe the documents using Egerton, and write a paper about the case.

For Kadens, the joy in bringing students to the archives comes from watching history become accessible for them.

“Every time I take students to the archives, I see a light bulb come on as they hold these documents in their hand for the first time. Students discover that these cases were about real people with complex problems, not dissimilar to their own. It gives a more intimate and tangible perspective to the conflicts, testimonies, and sentiments happening at that time.”

As the Egerton Model inches closer to one million words and public release, Kadens’s work stands as a beacon for what’s possible when deep scholarship meets cutting-edge technology. In decoding these legal hands of history, she’s not only preserving the past, but also making it highly accessible for like-minded historians moving forward.