Where the Cat Sat: A Multilingual Framework for Spatial Language Understanding

Spatial language understanding is fundamental to tasks from navigation and direction, robot control, to document understanding for a multitude of languages and tasks, yet current work exhibits biases toward English and prepositional marking. We present a multilingual framework and benchmark decomposing spatial relations into surface elements (figure, ground, predicate, markers) and semantic components (dynamicity, stasis). Evaluating frontier large language models (LLMs) on Spanish, Basque, and Chinese with text-only input, we find high performance on figure and ground identification but persistent gaps in two areas: semantic classification of topological and projective relations, and surface identification of morphological spatial markers-Basque case affixes proving most challenging with recognition of spatial elements as low as 15.3%. These results suggest that surface parsing does not entail spatial understanding, and that evaluation must include spatial marking strategies used across typologically diverse languages.