Spatialthat

Spatialthat is a theoretical framework and prototype software approach within geospatial information science and computational linguistics. It centers on modeling how spatial references in natural language—such as that, there, here, and near—are anchored to real-world locations and spatial relationships. By treating deictic and demonstrative cues as pointers to spatial frames rather than fixed entities, Spatialthat aims to improve the interpretation and execution of geospatial queries in context-rich environments.

The core idea is a layered model: a linguistic annotation layer that marks referential expressions; a spatial frame layer that encodes the current or asserted reference frame (viewport, temporal window, user context); and a spatial graph layer that represents entities and relationships (points, lines, regions, adjacency, containment). Disambiguation is performed by aligning phrases with candidate frames, then selecting the best match using context, prior interactions, and spatial constraints. Outputs can be translated into standard geospatial queries and data formats such as GeoJSON or Well-Known Text (WKT).

Implementation can take the form of a software library that exposes APIs to map phrases to spatial queries, with support for common GIS standards and a query language for spatial relations. Potential applications include natural-language GIS search interfaces, interactive cartography, digital humanities projects dealing with place descriptions, and augmented reality navigation systems. The approach emphasizes explainability by attaching the reasoning path that led to a particular spatial interpretation.

As a concept, Spatialthat has influenced discussions on declarative spatial querying and context-aware GIS design but remains largely experimental with limited publicly available implementations. Further work focuses on formalizing the frame alignment process and evaluating across multilingual corpora.