OsmT: Bridging OpenStreetMap Queries and Natural Language with Open-source Tag-aware Language Models

OSMT: Bridging OpenStreetMap Queries and Natural Language with Open-source Tag-aware Language Models Zhuoyue Wan∗, Wentao Hu∗, Chen Jason Zhang∗, Yuanfeng Song†¶ , Shuaimin Li‡, Ruiqiang Xiao§, Xiao-Yong Wei∗, Raymond Chi-Wing Wong§ ∗The Hong Kong Polytechnic University, Hong Kong, China †WeBank, Shenzhen, China ‡Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China §The Hong Kong University of Science and Technology, Hong Kong, China Abstract—Bridging natural language and structured query lan- guages is a long-standing challenge in the database community. While recent advances in language models have shown promise in this direction, existing solutions often rely on large-scale closed- source models that suffer from high inference costs, limited transparency, and lack of adaptability for lightweight deploy- ment. In this paper, we present OSMT, an open-source tag-aware language model specifically designed to bridge natural language and Overpass Query Language (OverpassQL), a structured query language for accessing large-scale OpenStreetMap (OSM) data. To enhance the accuracy and structural validity of generated queries, we introduce a Tag Retrieval Augmentation (TRA) mechanism that incorporates contextually relevant tag knowledge into the generation process. This mechanism is designed to capture the hierarchical and relational dependencies present in the OSM database, addressing the topological complexity inherent in geospatial query formulation. In addition, we define a reverse task, OverpassQL-to-Text, which translates structured queries into natural language explanations to support query interpretation and improve user accessibility. We evaluate OSMT on a public benchmark against strong baselines and observe consistent improvements in both query generation and interpre- tation. Despite using significantly fewer parameters, our model achieves competitive accuracy, demonstrating the effectiveness of open-source pre-trained language models in bridging natural language and structured query languages within schema-rich geospatial environments. Index Terms—structured query generation, natural language interfaces, Text-to-OverpassQL, OverpassQL-to-Text, language model I. INTRODUCTION Structured query languages are essential interfaces for man- aging and interacting with complex databases. Establishing effective alignment between natural language and structured queries has emerged as a prominent research focus in both academia and industry, motivated by the need to lower exper- tise barriers and facilitate intuitive database access for non- technical users. Significant research addressing this challenge has been presented in a broad literature, including works such as [1]–[12]. These studies collectively demonstrate sustained scholarly and practical interest in advancing natural language interfaces for structured data and have been widely adopted ¶ Corresponding author. 10 0 10 1 10 2 10 3 10 4 Number of Parameters (Billions, Log Scale) 58 60 62 64 66 Average Performance Score (EM, chrF, KVS, TreeS, OQS) LLaMA-3.1-8B Qwen2.5-72B Qwen3-235B DeepSeek-V3 GPT-4.1 GPT-4o Claude-4-sonnet GPT-4 CodeT5-base CodeT5+ ByT5-small OverpassT5 OsmT-small OsmT-base High Performance Low Parameters OsmT (Ours) Other Models Unknown Params. Fig. 1: Model performance vs. parameter size (log scale). Comparison of OSMT with state-of-the-art open-source and closed-source models on the Text-to-OverpassQL task. Aver- age performance is over five metrics. across diverse real-world applications, ranging from business analytics to scientific data management. Among the various types of structured data, geospatial databases have emerged as particularly critical due to their foundational role in supporting large-scale spatial analysis, complex query execution, and location-based decision-making. These capabilities underpin a wide range of downstream applications, including geospatial knowledge extraction, urban mobility modeling, and spatio- temporal forecasting [13]–[19]. A prominent example of a geospatial database is Open- StreetMap (OSM), a collaboratively maintained, open-access platform that provides foundational infrastructure for spatial data analysis. OSM supports various sophisticated spatial query functionalities, including location filtering, proximity searches, and routing, as exemplified by widely used ap- plications such as OsmAnd1 and Locus Map2. Retrieving structured geospatial data from OSM typically relies on the Overpass Query Language (OverpassQL), a domain-specific language designed for fine-grained spatial data extraction through filters, scoped conditions, and recursive constructs. 1https://osmand.net/ 2https://web.locusmap.app/en/ arXiv:2512.04738v1 [cs.CL] 4 Dec 2025 While OverpassQL offers powerful expressive capabilities, it demands that users possess detailed knowledge of OSM’s schema and tagging structure, creating significant usability barriers for non-expert users. To improve

This content is AI-processed based on open access ArXiv data.