LD4 2025 - Evaluating AI-Generated Linked Data

Records from historical anthropologists’ research are scattered across many archival repositories, making it difficult for communities represented in those records to access them. Tracing the records is difficult because many archival collections are not described or digitized, or have only been minimally described. The ability of linked data to connect primary and secondary sources across different institutions and collections offers a promising approach to addressing the challenges posed by scattered anthropological records. Additionally, the process of representing records as linked data offers communities an opportunity to describe the records themselves, rectifying misrepresentations and reclaiming data ownership to a degree, even if the records are held in colonial collections. However, the communities that would most benefit from undertaking this work often do not have the resources to do so. Given the increased access to generative AI that freely available chatbots offer, our research team has been investigating whether this technology could reduce the barriers to entry for representing records as linked data. In this talk, we will present the results of our qualitative and computational evaluation of ChatGPT-generated linked data for unprocessed (i.e., neither digitized nor described) archives, a continuation of the project we presented at last year's LD4 conference.