From Archives to AI: Using Old Maps and Gazetteers to Improve the Geoparsing Capabilities of Large Language Models

Large language models (LLMs) reveal new opportunities for automated geoparsing of text and speech data, facilitating the collection of spatial data in settings where map literacy is low. However, the geoparsing capability of LLMs depends on the coverage of its training data and consequently will not perform well in regions without detailed map data. This research demonstrates how computer vision techniques can be used with old maps and gazetteers as a complimentary data source to improve the geoparsing capability of LLMs, demonstrating how patient journeys can be extracted from interview transcripts collected in a Ugandan medical facility.