At University Libraries, a new AI-enhanced tool is making the historical record more accessible – and giving student workers an opportunity to provide feedback that makes the tool even better.
The Libraries recently became top tier charter members of JSTOR Digital Stewardship Services and have been using their new tool, Seeklight, to assist with rare collection digitization and processing. Seeklight is an AI-enhanced tool that has proven especially valuable in the transcription of digitized handwritten documents, says Digital Collections Librarian Katie Hoskins.
“We have over 100,000 pages of handwritten documents that are digitized but not transcribed,” Hoskins notes. “While these documents are technically available to anyone through our digital collections, they’re neither discoverable nor, in most cases, legible until they’re transcribed. OCR tools can’t read handwritten materials, and the cursive script the writers used in past centuries is illegible to many people today.”
Even in its current early developmental stage, Seeklight has proven to be impressively good at transcribing handwritten documents with a high degree of accuracy. “The student workers who are using the tool are, on average, only having to edit three words out of every 100,” Hoskins says.
Each semester, anywhere from six to ten students work in the Libraries’ digital collections department, scanning and transcribing documents and generating the descriptive terms, or metadata, which help library users find relevant items. Seeklight can also generate metadata so that the documents it has transcribed become discoverable. But transcription and description on a large scale brings another major advantage to researchers that they have never had before: a searchable corpus of text from letters, diaries, and other historical documents.
As beta users, the department’s student workers have also shaped the evolution of the tool. “We’ve been able to offer a lot of feedback,” Hoskins says. “JSTOR has been very receptive to the pain points we’ve pointed out. For example, initially we could only download transcripts one at a time, and when we told them it would be more convenient to download in bulk, they added that capacity.”
The Libraries’ role on campus as a repository for vast amounts of information and a hub for research, much of which is increasingly data-driven, has made the value of AI tools particularly apparent to many librarians. In addition to trialing tools like Seeklight, librarians are participating in monthly AI lunch and learn sessions where they share their experiences deploying different AI tools in different ways to facilitate growth in awareness of how AI can make them both more efficient and more effective in the work they do to provide research support to the campus community. The Libraries’ Data Services team, which includes AI Specialist Vandana Srivastava, can also help faculty and students across campus optimize their own use of AI and machine learning tools.
Ultimately, says Dean of Libraries David Banush, “Tools like Seeklight are adding value and helping us be much more productive. We can open new insights for scholars and students, and it’s allowing us to make more of our voluminous distinctive collections accessible. Such work is also getting student workers genuine experience with AI tools, with insights into their use that they can carry into the workplace, whatever future paths they pursue.”