Spatial Language Representation with Multi-Level Geocoding
This work addresses the challenge of spatial language understanding for applications like mapping and localization, offering a method that can generalize to toponyms not in knowledge bases, though it is incremental in improving accuracy over existing approaches.
The authors tackled the problem of associating texts with geographic locations by introducing a multi-level geocoding model (MLG) that uses space-filling curves to represent the Earth's surface hierarchically, achieving state-of-the-art results for toponym resolution on three English datasets without dataset-specific tuning.
We present a multi-level geocoding model (MLG) that learns to associate texts to geographic locations. The Earth's surface is represented using space-filling curves that decompose the sphere into a hierarchy of similarly sized, non-overlapping cells. MLG balances generalization and accuracy by combining losses across multiple levels and predicting cells at each level simultaneously. Without using any dataset-specific tuning, we show that MLG obtains state-of-the-art results for toponym resolution on three English datasets. Furthermore, it obtains large gains without any knowledge base metadata, demonstrating that it can effectively learn the connection between text spans and coordinates - and thus can be extended to toponymns not present in knowledge bases.