From Text to Map: AI-Based Graphic Translation of Information

Abstract

Abstract. This study investigates the potential of artificial intelligence in translating historical textual registers into graphic reconstruction, exploring interdisciplinary approaches that integrate historical studies and GeoAI. The case study focuses on the “Calcato” of Gandino, a mid-18th-century textual register accompanied by a large-scale territorial map. Produced by land surveyors “treading the ground”, the register describes property boundaries as closed polylines, whose vertices correspond to fixed landscape markers and whose segments are defined by length and orientation. As the map represents a later graphic transposition rather than a direct survey output, it provides the basis for this research: replicating that translation process through computational tools. Following an initial exploratory phase to validate the concept, conversational AI models were used to transcribe the historical text and compute vectors to generate local coordinate tables necessary for reconstructing the geometries of the described areas. The method was tested on cases with varying levels of complexity, comparing the results with the corresponding map representations. A final phase extended the analysis to registers from different historical periods to enable further testing and comparison. The results demonstrate the feasibility of transforming archival textual data into spatial representations, supporting qualitative analysis and the study of historical and territorial transformations.