Recent developments in artificial intelligence (AI) hold the potential for a seamless transformation and integration between visual maps and textual representations of space. To investigate new approaches for unlocking this potential, this Dagstuhl Seminar seeks to bring together experts in geoinformatics, cartography, spatial computing, geographical information science, and artificial intelligence, aiming to foster entirely new user experiences and modes of expression. Within the broader seminar theme, three specific challenges will be addressed: the transformation from maps to text, the transformation from text to maps, and the integration of maps and text. To tackle these three challenges, a large potential is seen in recent methods of natural language processing, specifically large language models (LLMs) and large vision language models (LVLMs). Other recent technologies that will be discussed include but are not limited to cartographic foundations models, agentic AI, and cartographic knowledge graphs.

Translating textual prompts into maps is an ongoing challenge in automated cartography. Initial image generation methods yielded map-like images with little grounding in real-world geography. A fundamentally different approach is the generation of a map by applying an image style transfer model to a given data set. However, so far, this approach has only been used for the generation of topographic maps in historical style. Therefore, the prompt-based creation of trustworthy, accurate, editable, and legible maps with generative AI is largely an open challenge, which we aim to discuss in the seminar. Related topics to be discussed are, e.g., the generation of maps based on travel reports or social media.

Transforming maps into text enables new forms of understanding and accessibility. As predominantly visual artifacts, maps are often inaccessible to people with visual impairments, but AI can generate textual descriptions compatible with screen readers, making them more inclusive. Sighted users can also benefit, for example through descriptions of spatial patterns or answers to geospatial questions. This is especially valuable for historical maps, which contain rich information about past environments. Achieving reliable map-to-text transformation requires methods that handle variations in quality and complexity, account for differences in precision and uncertainty, and extract key concepts and relationships, even in the absence of legends or metadata.

Integrating maps and text remains challenging. Unlike natural images, maps encode spatial relationships, symbolic abstractions, scale, and legends, requiring alignment between visual, geometric, and linguistic representations. While map and spatial text understanding have been studied separately, joint map–text reasoning is still emerging. Applications include summarizing map-rich reports, answering spatial questions, and enabling interactive map systems. Key research questions involve representing maps as multimodal data, grounding spatial relations in language, and evaluating performance with spatial question answering benchmarks. The seminar will also discuss ethical and epistemological issues, where text and map depictions jointly shape meaning and trust.

The seminar is meant to initiate collaborations aiming at various types of measurable scientific outcomes, such as scientific publications, new training data sets, benchmarks, and machine-learning models for cartographic tasks. The program will be structured by plenary presentations and discussions as well as work in groups of approximately 5 to 10 participants. These groups will be formed after a plenary session in which the participants will briefly describe their research interests, propose focus topics or concrete open problems to work on. Equally important will be the informal parts of the seminar that will contribute by serendipity to new inspirations and collaborations.

Creative Commons BY 4.0

Yao-Yi Chiang, Amy Griffin, Jan-Henrik Haunert, and Monika Sester

This seminar qualifies for Dagstuhl's LZI Junior Researchers program. Schloss Dagstuhl wishes to enable the participation of junior scientists with a specialisation fitting for this Dagstuhl Seminar, even if they are not on the radar of the organizers. Applications by outstanding junior scientists are possible until Friday, May 29, 2026.