New “ArchAIval” Practices: Using GPT for OCR and Historical Narration of Index Cards

This study presents an innovative approach to archival digitisation and narrative generation using OpenAI’s Generative Pre-trained Transformer (GPT) models. It focuses on the photographic index card catalogue of the Documentation Library Davos, digitised by the Fotostiftung Graubünden. Traditional archival retrieval systems often suffer from inefficiencies related to the manual handling and interpretation of historical records. Our project addresses these challenges by employing GPT models for optical character recognition (OCR) and automated narrative generation, transforming how index card content can be accessed and utilised. By prompting GPT models, we convert the analogue records into digital formats through accurate OCR with an average bag-of-characters F1 score of 93.5% and integrate this data with other historical sources to generate cohesive and engaging historical narratives about documented buildings in Davos. To access the mentioned buildings, persons and locations in a structured way, we develop a domain-specific NER model by bootstrapping it from GPT-generated automatic pre-annotations. Our methodology not only increases the accessibility and interpretability of archival materials but also enriches the cultural heritage by providing a more comprehensive understanding of historical contexts. Preliminary results of this pilot study demonstrate the potential of this approach to enhance the public’s engagement with history and improve archival practices through technological innovation.