Beyond Toponyms: Nominal Spatial Entities for Understanding Movement from Hiking Descriptions

Abstract

Movement descriptions do not only indicate where people go, they also reveal how routes are understood, followed, and explained through actions, landmarks, directions, and distances. While many computational approaches to mobility rely on GPS traces or named place recognition, textual itineraries contain richer spatial cues that are often expressed through ordinary landmarks such as paths, bridges, churches, or intersections. This work studies French hiking descriptions as a case study for extracting structured spatial information from human-written route instructions. We build an annotated corpus of route descriptions and develop neural methods to identify the linguistic elements that support movement understanding, including named places, nominal landmarks, movement actions, spatial relations, and measures. Particular attention is given to nominal spatial entities, which are frequent in route instructions but difficult to match directly to geographic databases. The long-term objective is to support an explainable route reconstruction system able to connect textual movement descriptions with geographic evidence while preserving the way humans naturally describe routes.