Ancient AI — Translating Yesterday’s Ink with Tomorrow’s Tech
Ancient AI is an interdisciplinary capstone project aimed at making 19th-century Marshallese handwritten letters accessible to historians and researchers. Sponsored by Dr. Monica LaBriola, our team developed a streamlined, user-friendly web application powered by a custom-trained AI model to transcribe scanned documents into searchable, tagged text.
Our solution combined a custom Transkribus AI model, trained specifically on Marshallese archival handwriting, with a Meteor + React web application that allowed users to search and browse transcribed documents by keywords, author, time period, and location.
Key contributions:
- Researched and evaluated transcription tools including Transkribus, Handwriting OCR, and Pen2Txt to assess privacy, legal, and technical suitability.
- Prepared and manually corrected a dataset of handwritten documents to train the custom Transkribus model.
- Implemented an OpenAI-based post-processing model to correct spelling errors while preserving historical grammar and style.
- Integrated the trained model into a web application with a clean UI and robust search functionality.
- Demonstrated the impact of AI in digital humanities by significantly improving the efficiency and accuracy of document transcription for academic research.
This project underscored the potential of AI and computer science to preserve cultural heritage and support interdisciplinary collaboration.