Eliciting Image Schemas for Urban Digital Twins

The construction of urban digital twins (UDT) is a broad objective that, to be successful, must go beyond structural and functional information to include the human factor with its cognitive and experiential knowledge. In this paper, we investigate this issue by focusing on a typical, yet less complex, element of the urban scenario, namely the square. More specifically, our general aim is to study how human knowledge of the square may be elicited and structured for inclusion in UDTs. To do this in a fairly broad perspective, we study which knowledge characterises squares by considering two quite different real cases: Navona square in Rome and Djemaa el-Fna square in Marrakech. Starting from three images of each square, we use an LLM to obtain a descriptive text which is then used to generate a knowledge graph. Both the text and the knowledge graph are then provided again to the LLM to elicit relevant image schemas. The results on each square are then compared to select the image schemas that are more characteristic of squares. The hope is that this kind of effort can lead to enrich our knowledge of urban spaces and, subsequently, to develop UDT that are more comprehensive representations of our cities by including along with structural and functional information also human perspectives and values.