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 diferent 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 efort 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.
The study and construction of Digital Twins (DT) is a live and central research goal of the last 20 years
or so [
This paper presents a step within a longer research project aiming to analyse how to understand and exploit Urban Digital Twins. Our focus goes beyond the structural and functional aspects that are typically addressed in DT studies, to concentrate on how to build DTs that comprise the cognitive and experiential perspective typical of humans. We address this problem by focusing on a common, if not typical, element of cities, namely the square. This focus is not a restriction in interest but a way to deal with the complexity of the goal without constraining the generality of our project.
In this contribution, we posit the question of how to elicit cognitive-based information in a way suitable for enriching DTs. Generally speaking, we believe that the elicitation of this kind of information should be based on cognitive representation approaches like the image schemas. However, extracting and analysing image schemas on a varied and complex structure, which includes both static and dynamic information, is a particularly complicated process which is too dificult and error prone to be done manually. For this reason, we investigate the use of AI software over a set of sources (text, graphs, images). The results of this analysis highlight the variety and utility of cognitive-centred knowledge and the kind of knowledge that today’s structural and functional UDTs clearly lack.
This work should be read within our previous eforts on the application of diferent research paths
to provide a rich, unified and consistent framework for the modelling of complex environments like
the urban territory [
In the literature there are only a few software solutions for managing image schemas that could
produce outputs useful for our goal, i.e., knowledge that comes into play during planning decision
processes or in scenario testing. Within this research line, there are some works on pipelines for
extracting image schema from vision to support robots to operate more eficiently within spatial
contexts [
Overall, our approach seeks to incorporate nuanced and dynamic elements of spatial experience and everyday urban life, extending beyond conventional sensor data and standard environmentalgeographical information. The focal point of our recent work has been the urban square - a complex spatial node characterized by diverse patterns of movement and occupation, ranging from habitual pedestrian routes that have evolved through repeated use, to the qualities of light that characterize the square atmosphere at particular hours of day.
By combining the methodologies of ontological analysis and applied ontology with the reading of places given by image schemas we would like to develop new tools that can manage heterogeneous knowledge. We think that the use of image schemas to “reverse engineering” human ways to inhabit and understand places can lead to develop better representation tools for the urban planner, making her more aware of intrinsic cognitive aspects and dynamics that make spaces and places unique to their inhabitants.
As anticipated, in this paper we focus on urban squares because they are important spaces within
cities whose meaning goes clearly beyond their structure, are hard to characterize, and yet reflect
the habits and social behaviours of city inhabitants. A city square is often the place where meeting
and crossing events occur—one of the semantically densest places in the urban fabric, where diverse
meanings, events, and objects create a richer substance [
Following this introduction, §2 briefly describes the tools we use while the following, §3, uses them to study two paradigmatic squares and elicit the image schemas that characterise them. The last section, §4, draws some further conclusions.
Our proposal is to investigate a three-stage procedure to systematically transform visual spatial information into image schemas. This is the strict aspect of interest we intend to explore in this contribution, a partial slice in a broader research scenario. Image schemas can be considered a format for formal knowledge representation, i.e., a way to capture knowledge that is suitable for enriching UDTs (e.g., in the form of knowledge graphs).
The pipeline we use is depicted in Fig. 1. We decided to start from a set of photos of the urban space of interest because these are rich sources of heterogeneous information and overcome, at least to some extent, some constraints of verbal language (like imposing the choice of terminology and enforcing a structural linearity of knowledge), so that these aspects intervene only later in the elicitation process and are mitigated by the integration of knowledge from large language models, knowledge graphs, and ontological structures (even though the latter is not really exploited in this paper). The process begins with the selection of photographic perspectives of the spatial environment of interest. In the next step we submit these to a large language model for the generation of comprehensive textual descriptions. These descriptions are aimed to capture not only the visible architectural and environmental features but also the implicit functional and experiential qualities of the space.
The resulting textual output is processed through the AMR2FRED tool, which converts natural language descriptions into structured knowledge graphs. This transformation represents a crucial transition from the contextual, distributional representations characteristic of neural language models to the explicit, relational and conceptual structures required for formal reasoning. The generated graph provides a structured representation of the spatial relationships, entities, and properties identified in the original descriptions.
The next stage involves the analysis, again via the large language model, of both the textual description and of the knowledge graph to extract relevant image schemas in their role of fundamental spatial and experiential patterns motivated by embodied cognition theory. For instance, the generated description of an instance of “walking” is now formally connected to the source-path-goal image schema, while the spatial configuration of filled and empty areas is now mapped to container and support image schemas as a cognitive counterpart.
The integration of these ‘ontologized’ (thanks to FRED) image schemas serves a dual function: they provide formal constraints that guide subsequent language model generation while simultaneously grounding the abstract spatial concepts in established patterns of human spatial cognition. This creates a hybrid computational system that combines the contextual understanding capabilities of large language models with the inferential rigour of formal knowledge representation, resulting in a framework capable of representing and reasoning about implicit spatial knowledge through explicit formal structures.
Below we describe the procedure we followed for the elicitation of image schemas related to the two squares.
The procedure, as illustrated in Fig. 1, consists in providing to the LLM Perplexity1 three photos (daytime image, nighttime image, aerial image) of each square considered for the comparison: Navona square and Djemaa-el-Fna square. You see here the pictures we used, figure 2 the ones concerning Piazza Navona, in figure 3 the ones about Djemaa-el-Fna. [figure 2] [figure 3]
The descriptive text generated by Perplexity was provided to AMR2FRED, which transformed the text into a graph. Subsequently, both the graphs and the original texts were provided to Perplexity to elicit the image schemas related to each square under examination. The procedure involved comparing the resulting image schemas derived from text and graph representations of both Navona and Djemaael-Fna, and ultimately determining which image schemas better characterize each square in order to acquire the needed variety that a general DT for urban squares should contain.
The analysis of the graph produced by FRED regarding the flows and spatial relations within Navona square, allows us to identify diferent image schemas to support the understanding of space, movement and relations between elements in the urban space. As a result of this step, the main emerging image schemas are: • PATH: the lines drawn in the graph suggest paths and directions, representing the movement through the square. • CONTAINER: the square appears as a delimited space that collects events, people and movements. • LINK: connections emerge between distinct points of the square, visible in the lines that join the nodes. • CENTER-PERIPHERY: where elements are concentrated in central areas and more marginal zones, the distinction between center and periphery emerges and is clear. • FORCE: curves and directions of the graph suggest dynamic forces, such as flows of people or spatial attractions.
• BOUNDARY: the limits of the square are marked, distinguishing the inside from the outside.
The comparison between the graph and a descriptive text of the square (of a historical, social and architectural nature) makes evident some substantial diferences. The graph makes immediately visible the patterns related to movement and spatial structure, such as PATH, AXIS, CONTAINER and BOUNDARY. The text, on the other hand, emphasizes the lived experience, the social function and the narration of the space, bringing out patterns such as CONTAINER, AXIS, CENTER-PERIPHERY, LINK and FORCE, with the PATH being less evident. When the two approaches – visual and textual – are combined, some main image schemas emerge, here ordered by representativeness according to a weighted estimate that gives equal weight (50/50) to their evidence in the graph (in terms of recurrence, spatial centrality and visual structure) and to their relevance in the text (based on frequency, narrative role and descriptive function): 1. PATH – dominant pattern (30%), since it reflects the longitudinal character of the square and the strong presence of crossing flows. 2. CONTAINER (25%) – the square is perceived as a large enclosed space that hosts events and interactions. 3. AXIS (15%) – the central axis, marked by the arrangement of the fountains, is a structuring element. 4. CENTER-PERIPHERY (12%) – the square has a clear center (the Fountain of the Four Rivers) that attracts activity, contrasted with less active edges. 5. LINK (8%) – the relationships between the architectural and social elements define a network of connections. 6. BOUNDARY (5%) – the physical limits, given by the buildings that surround the square, delimit the urban container. 7. FORCE (5%) – the dynamic forces are less constant but present - for example in the movement of the crowd or in the water that invades the square in summer.
Assuming a single image schema could be used to best summarizes Navona square, the most likely choice would be PATH, which has an estimated probability of 40%, followed by CONTAINER (30%) and AXIS (15%). This estimate is based on a criterion of relative representativeness, which proportionally combines the visual centrality of the image schema in the graph (structural presence, directionality, frequency) and its semantic relevance in the text (narrative function, recurrence, descriptive role). The other schemas appear less central in comparative terms. A mutual hierarchical evaluation (pairwise) between the image schemas confirms this preference: PATH is dominant compared to all the others. Therefore, in decreasing order of importance, other schemas follow: CONTAINER, AXIS, CENTERPERIPHERY, LINK, BOUNDARY and finally FORCE. As a final note, the VERTICALITY image schema was not included among the main ones because, although vertical elements are mentioned in the text (the obelisk, the dome of the church), the graph representation and the experience of the square appear predominantly horizontal. There are no dominant vertical diferences or developments in the perception of the urban space, which remains strongly axial and traversable along a longitudinal direction.
The graph of interactions and flows relative to the Djemaa el-Fna square highlights multiple image schemas that can shape the possible experience of the urban space. Among these, the dominant image schema is undoubtedly PATH, highlighted by the multiple lines that represent movements, paths and lfows of people and activities. Alongside this, the CONTAINER image schema emerges, which describes the square as an enclosed space, with well-defined areas within it. The connections between elements stalls, shows, thematic areas - reflect the LINK image schema, which underlines the relationships between actors and spaces. The presence of a particularly active central area, from which the activities branch out towards the outside, recalls the CENTER-PERIPHERY image schema. Finally, the intertwining and simultaneous coexistence of multiple activities and paths brings out the concept of MULTIPLICITY, typical of chaotic and dynamic spaces like this one. Comparing this analysis of the graph with the text analysis significant diferences emerge in the perception of the same image schema. In fact, the text transmits information in a narrative and sequential way, forcing the reader to mentally reconstruct the spatial relationships. However, even in the text diferent image schemas are clearly recognizable. The most evident is once again CONTAINER: the square is described as the heart of the medina, a space that contains activities, people, noises, lights and transformations. CENTER-PERIPHERY follows, since the square is the focal point from which the souks and streets of the city branch out, and around which social and symbolic activities revolve. The PATH image schema is implicit in the movements of visitors who cross the spaces, passing from one function to another, while LINK shows up in the relationships between functions: trade, entertainment, food and rituality. More subtle patterns can then emerge, such as SOURCE-PATH-GOAL, which is recognized in the daily transformation of the square – from a daytime market to an evening theater – and UP-DOWN, suggested by the presence of the minaret that dominates from above. Finally, BALANCE/FORCE can be perceived in the tension between order and chaos, between structure and improvisation. Combining graph and text, it is possible to order the image schemas by degree of representativeness. The most important seems to be CONTAINER, which collects about a third of the occurrences and reflects the essence of the square as a space that welcomes, contains and organizes. PATH and the related SOURCE-PATH-GOAL follow, which together represent a quarter of the spatial dynamics, describing the flow of bodies, goods and meanings. The CENTER-PERIPHERY pattern covers about 18% of the occurrences, testifying to the central position of the square in urban and symbolic life. Less representative but still relevant are LINK, UP-DOWN and BALANCE/FORCE, which complete the layered perception of space. Assuming that only one image schema is to be identified to represent Djemaa el-Fna square, CONTAINER would have the highest probability of being chosen, equal to 40%, because of its double evidence, visual and narrative. PATH/SOURCE-PATH-GOAL follows with 25%, and CENTER-PERIPHERY with 15%. The others would be placed below 10%. A mutual comparison between the image schemas (according to a “one against one” logic) confirms this hierarchy. CONTAINER prevails over all the others due to its spatial and conceptual centrality. PATH is placed second, which is more evident than LINK, CENTER-PERIPHERY and UP-DOWN. The latter, linked to the vertical perception given by the minaret, is present but secondary. BALANCE/FORCE, finally, is the most transversal and least structuring schema. As for the theme of VERTICALITY, it is present but not central. Although the Koutoubia minaret visually dominates the square, its function remains symbolic rather than organizational. Neither the text nor the graph seem to suggest a vertically hierarchical spatial or functional structure. Djemaa el-Fna is perceived and experienced as an essentially horizontal, open, dynamic and reticular space. In short, Djemaa el-Fna appears to be configured as a container space, crossed by multiple paths, the vital center of the city and a point of connection between activities, functions and urban identities. The main image schemas not only help to describe it, but also to understand its symbolic power and its role in the social space.
The CONTAINER image schema is strongly represented in both squares, but with diferent nuances. In Navona square, the notion of container is shown through the clear architectural boundaries and the role of the square as a closed and organized urban space. The text describes it as a “large urban theater”, accentuating its aesthetic and scenographic function. In Djemaa el-Fna, however, the concept of container takes on a more dynamic and social value: the square is described both in the graph and in the text as a place that welcomes and gathers in a fluid way a multiplicity of activities - from commerce to shows - and is perceived as the beating heart of the medina. For this reason, while in Navona the container is more static and architectural, in Djemaa el-Fna it is alive and transformative. The concept of AXIS is instead clearly present only in Navona square. Here the longitudinal axis is clearly perceptible both in the visual arrangement (graph) and in the textual description, which highlights the ordered sequence of the three main fountains along a central path. In Djemaa el-Fna, on the contrary, this image schema is practically absent. The Moroccan square is not organized along an axis, but shows up as a reticular and polycentric space, in which the activities are distributed without a prevailing orientation. Concerning CENTER–PERIPHERY, both urban spaces present a significant centrality, but with diferent functions. In Navona square the center is symbolically represented by the Fountain of the Four Rivers, which acts as a visual and architectural node. In the graph, this centrality is underlined by the greater density of connections that start from this point. In Djemaa el-Fna, instead, the centrality is essentially functional and social: the square constitutes the heart of the medina, the nerve center of urban life, from which the streets branch of towards the souks and other neighborhoods. This centrality is less linked to a single element and more to the overall function of the square.
The image schema LINK is present in both squares, but with diferent intensity. In Navona, the connections are mainly spatial and architectural: they are relationships between fountains, churches and other landmarks, often visually aligned. In Djemaa el-Fna, on the other hand, the links are dense and varied: connections between diferent functions (market, entertainment, gastronomy), social, cultural and symbolic relationships. In this case, the square acts as a node of exchange and interaction between diferent languages and practices, not only as a physical space.
The notion of BOUNDARY is more evident in Navona square. Its limits are marked by baroque buildings and historic palaces, which frame the space in a clear way. In the graph, the edges are clearly represented, while the text underlines the defined perimeter of the square. On the contrary, in Djemaa el-Fna, the boundaries are more fluid, permeable and less emphasized. The square merges with the souks and adjacent streets, and there is no clear separation between “inside” and “outside”.
The image schema SOURCE–PATH–GOAL is particularly present in the textual description of Djemaa el-Fna, and almost absent in that of Navona square. The Moroccan square is described as a space that transforms during the day: a market in the morning, a theater in the evening. This transformation could suggest a functional and temporal path that starts from an initial condition (source), goes through a change (path), and reaches a final goal (goal), such as the experience of a show or a collective ritual. Navona square, on the contrary, maintains a greater stability of function and structure.
As regards the UP–DOWN image schema, linked to verticality, here too a contrast is highlighted. In Djemaa el-Fna the minaret of the Koutoubia mosque visually “dominates” the square, introducing an evident, albeit symbolic, vertical dimension. This aspect is recognized both in the text and in the spatial perception of the place. Navona square, on the other hand, features vertical elements – such as the obelisk of the central fountain or the domes of the churches – but these do not seem to structure the perception or use of space, thus appearing accessories to the prevailing horizontal configuration.
The image schema BALANCE/FORCE is more strongly perceptible in Djemaa el-Fna. The square is a dynamic place, where order and chaos, stability and improvisation coexist. The text highlights the simultaneous presence of organized structures (such as markets) and more spontaneous forms of sociality (shows, crowds). This tension between opposing forces represents a precarious but vital balance, which characterizes the identity of the place. In Navona square, however, this dimension is less marked. The movement seems more regulated, the spatial composition more harmonious and contained.
Finally, the image schema MULTIPLICITY seems to be absent in Navona square, but central in Djemaa el-Fna. The Moroccan square simultaneously hosts a plurality of events, people, languages and gestures. This multiplicity is perceptible both in the graph, which shows a dense and polycentric network, and in the textual description, which describes the square as a kaleidoscope of activities. Navona square, while ofering diferent functions (tourism, art, sociality), appears to be experienced in a more sequential and orderly way, along a traversable axis. Therefore, multiplicity appears to be a structuring quality in Djemaa el-Fna, but only an accessory quality in Navona.
This paper presents an approach to the use of image schemas as a methodological contribution to elicit cognitive and human knowledge and make it accessible to UDTs for planning and urban decisionmaking. The approach is preparatory to the integration of structural and functional information with personal and cognitive information that together characterize urban spaces.
Our experiment concentrated on two diferent squares and revealed that while both share certain fundamental image schemas (such as container and path), their hierarchical organization difers significantly in ways that reflect deep cultural and functional logic. Navona square emerges as a space primarily structured by the path scheme (30%), probably reflecting the longitudinal character and the Baroque tradition of organized space sequences, with a prominence of the AXIS image schema (15%) which underlines an emphasis on directional clarity and hierarchical organization. The function of the square as a well-defined CONTAINER (25%) suggests architectural traditions that favour the creation of borders and the scenic framework. On the contrary, Djemaa El-Fna reveals a more characterized space logic characterized by the CONTAINER image schema (30-40%) but in a dynamic and transformative sense-probably reflective urban traditions of adaptable public space, with SOURCEPATH-GOAL patterns (25%) that capture the temporal transformations of the ‘market square’ from market to performative space.
The comparative analysis suggests direct implications for planning and decision-making processes,
revealing the importance of an urban design sensitive to cultural diferences in space cognition and
social organization. For example, the presence of the MULTIPLICITY image schema in Djemaa El-Fna
which is absent in Navona square can indicate how diferent urban traditions organize complexity
and social interaction. This suggests that planning approaches that traditionally emphasize order and
hierarchy can be inadequate for diferent contexts and communities. The time dimension captured
through the SOURCE-PATH-GOAL image schema pattern also suggests that successful public spaces
could be designed for adaptive use rather than a fixed function - a particularly relevant principle for
today’s urban contexts characterized by rapid changes. The intuition we explored in this contribution
is this that in the context of Urban Digital Twins (UDTs), Image Schema information could serve as
a cognitive-semantic layer that enriches the representation of spatial and experiential aspects of the
city. For example, schemas such as CONTAINER or PATH may be used to structure how movement,
accessibility, or boundaries are modeled in digital replicas of urban environments. Integrating such
schemas could improve the interpretability of simulations, support citizen-centered analyses of urban
space, and complement data-driven techniques by grounding them in embodied conceptual structures
together with a knowledge analysed and represented via applied ontology.A follow-up could involve
the extraction of image schemas directly from the participation of a space’s users or the inhabitants
of a place (a direct form of knowledge from which it is often dificult to extract consistent knowledge,
as mentioned). The research evokes important perspectives for applying the Urban Digital Twin
technologies beyond their current limitations in capturing cognitive and social dimensions, ofering
a path towards more knowledge aware and inclusive urban modelling systems. The methodological
potential for simulating future scenarios could allow decision makers to explore not only future physical
impacts but also cognitive and social impacts – so matching very crucial emerging needs nowadays. As
a model of knowledge management, this hybrid approach looks particularly suitable to be integrated
into system architectures to support the interactive and real-time urban decision, allowing systems that
adapt dynamically to upcoming events and occurrences while maintaining methodological consistency.
Recent contributions demonstrate how to build a novel framework that bridges embodied cognition
theory and agent systems by leveraging a formal characterization of image schemas [
During the preparation of this work, the author(s) used Claude 4 for: Grammar and spelling check, Paraphrase and reword; and Perplexity for: automatic image description generation. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the publication’s content.