The project harnessing data and AI to revive ancient manuscripts

The use of Artificial intelligence (AI) is becoming increasingly prominent in projects, but human oversight remains essential, as the STEMMA project shows.

The STEMMA project (Systems of Transmitting Early Modern Manuscript Verse) uses an innovative data-driven approach to study the circulation of early modern English poetry in manuscript form. Using network analysis, graph theory and large language models (LLM), the project has managed to build a significant database that enables scholars to trace how stories, histories and legends have changed over time and across regions.

The genesis of the project came out of the work of Prof Erin McCarthy (pictured, right), a researcher at the University of Galway. Initially envisioned as a side project, it gradually grew into a more significant venture.

She said: “As I started fleshing it out, I found that it was going to take a lot more time and resources. We’re now finishing our second year with a team and contractors. If I had tried to do it myself, it would have been a much smaller project. I’m thankful to have had the European Research Council supporting this work in a time when other governments are not investing in frontier research.”

Leading the project

As an academic, Erin didn’t have any project management training or knowledge. Spearheading the STEMMA project successfully meant that she had to learn new skills and develop quickly in her role.

Erin continued: “Learning to communicate, set expectations and navigate interpersonal relationships was a steep learning curve. It's been an adjustment for me, but it’s also been for the better. It’s allowed me to find new ways to approach problems and tasks. I'm a little like a conductor now, making sure everyone else is doing their bit and just trying to tie it all together.”

As a newcomer to project management, Erin credited the following as vital elements that helped her successfully lead the STEMMA project:

• Finding good mentors - Learning from a group of people who had been in similar situations proved invaluable, allowing them to brainstorm solutions together and learn from each other’s experiences. Erin took a six-month leadership course that gave her the opportunity to meet people who could offer their wisdom.
• Seeking out expert knowledge – Erin proactively sought out advice and insights from experts of areas she knew little about. In the early stages of the project, she would speak to other principal investigators and the project manager from her database development contractors to learn more about the ways of working and to gain practical advice she could apply herself. Despite being an academic, it was important to recognise where the knowledge gaps were and to try and fill them in.
• Being open – Having a positive mindset and being open to new ways of thinking and working was vital. Erin was able to navigate the new challenges she came across because of her attitude and willingness to try different things. She stressed that if she had approached the project purely as a researcher, it wouldn’t have become the success it has become.

“There's the benefit of synergy: it’s greater than that the sum of the parts. It’s a totally different way of working for me. It's what can I learn from whom and how can we put it together.”

Despite the challenges of new ways of working, Erin has enjoyed the experience so much that she is already looking for her next project. She said: “There have been difficult moments, but I’m already thinking about what's going to be next and starting to feel out collaborators. Overall, it’s been a rewarding experience.”

Using AI in the project

For the STEMMA project to achieve its goals, LLMs were a necessity. LLMs were used to search through documents and records, alerting researchers when they found a potential match. This allowed the researchers to concentrate on other more high value tasks. While this has clearly been beneficial for the STEMMA project, allowing it to progress at a speed it might not have without AI, Erin stressed that the STEMMA project has shown here that AI is best used within projects in a restrained way and human oversight is still crucial.

"I'm more open to AI than most, but we've not just turned it loose on the project. It's not perfect, but I look at it as an iterative process. The alternative would be to go through every single record manually, which would not only be tedious, but I also think we'd miss things because our brains can't hold as much information as a computer. We do still need a level of supervision because AI doesn't know everything, but I think that LLMs are effective when used in this narrow way."