Do We Still Need Gazetteers in the Era of LLMs? Chaining Retrieval with a Spatial Neuro-Symbolic Index

2026-10-04

AI Summary

This study investigates whether gazetteers remain necessary for toponym resolution in the era of large language models by examining if text encoders can replace explicit spatial indexing. We construct a spatial-semantic indexing benchmark to compare dense retrieval with neuro-symbolic hierarchical search, and propose a neuro-symbolic hierarchical beam search strategy that integrates gazetteer hierarchies. Experimental results demonstrate that unconstrained dense retrieval is prone to catastrophic spatial errors. While hierarchical constraints significantly improve coarse-grained localization, frozen general-purpose encoders struggle to preserve fine-grained spatial fidelity. This work reveals the spatial limitations of dense retrieval and confirms the irreplaceability of gazetteers even in the age of large language models.

Abstract

Geographic information retrieval (GeoIR) tasks require systems to interpret ambiguous toponyms for downstream applications. Traditionally, toponym resolution relies on gazetteers to provide an explicit index of place entities and spatial relationships. Recently, gazetteer-free approaches seek to reduce dependence on handcrafted searches: dense retrieval utilizes text encoders to capture rich context, moving beyond the limitations of lexical search. However, text encoders implicitly assume that learned representations can function as reliable spatial-semantic indexes. In this paper, we evaluate this assumption through a spatial-semantic indexing setup: given a contextualized toponym mention, we retrieve the corresponding gazetteer entity represented by text derived from a gazetteer knowledge graph. We benchmark five frozen text encoders under two retrieval strategies: brute-force nearest-neighbor retrieval over entity representations, and a neuro-symbolic hierarchical beam search that constrains retrieval (i.e. chaining the search with gazetteer hierarchy). Experimental results reveal a distinct coarse-versus-fine trade-off. Unconstrained dense retrieval frequently incurs catastrophic spatial errors. Conversely, hierarchical constraints improve coarse geographic grounding, but still yield limited benefit for fine-grained localization metrics: vanilla text encoders fail to capture the fine-scale spatial fidelity encoded in gazetteers. Our code is publicly available at: https://doi.org/10.25439/rmt.31094269

Problem

• Geographic Information Retrieval
• Toponym Resolution
• Spatial-Semantic Indexing
• Dense Retrieval
• Gazetteer

Innovation

• GeoIR
• Neuro-symbolic Index
• Hierarchical Beam Search
• Dense Retrieval
• Toponym Resolution