Artificial Intelligence for Security and Defence Applications IV · 12 Nov 2026
LLM4NL2BQS: using generative AI to transform natural language in BQS
Matthias Jooß, Daniel Haferkorn, Roland Rodenbeck
Citations by database
Each database indexes a different set of journals, preprints and books, so they are shown side by side rather than blended.
- OpenCitations0
- Crossref0
Citations per year
The current year is still counting.
- 2026 so far0
Abstract
Complex query languages such as Boolean Query Syntax (BQS) are powerful tools for information retrieval within NATO’s Coalition Shared Data network. Their effective use generally requires substantial technical and domain specific expertise, restricting high quality query formulation to a limited number of specialised analysts and creating bottlenecks in information gathering and decision making. In this paper, we investigate three complementary approaches to translating natural language information needs into syntactically correct and semantically appropriate BQS queries. These are fine-tuning a small, open source language model, few-shot prompting and Retrieval-Augmented Generation (RAG) using schema context drawn from the Allied Engineering Documentation Publication AEDP-17 specification of the Standardization Agreement(STANAG) 4559. Training data for fine-tuning is synthesised directly from the specification and validated both grammatically and against a live Coalition Shared Data (CSD)-Server. All three approaches are evaluated on a separate benchmark stratified by difficulty under three cumulative prompting conditions and compared against several general purpose language models, including commercial ones. The fine-tuned model achieves statistically significantly higher semantic accuracy than every comparison model at baseline prompting and higher syntactic and execution accuracy at every prompting condition, with the advantage widest on the most structurally complex queries. Fine-tuning already captures most of this benefit before any few-shot examples or RAG are added, while the commercial models depend far more on prompting strategy to reach their best results. This indicates that for a narrow query language tied to a specific schema, targeted fine-tuning of a small model is a more effective and more robust strategy than relying on few-shot prompting or RAG of a larger general purpose model, while lowering the barrier to entry for users without specialised expertise and reducing organisational reliance on highly specialised staff.
What the data shows
- No database records a citation yet, as is usual for a paper this new.
Graph
This paper in the middle, its authors and their most-cited papers below, the papers it cites to the left and the papers citing it to the right. The timeline lays every paper out by year.
Select a node for its details. Drag to pan, scroll to zoom, drag a node to move it.
Start with
Cited by 0 papers
Who cites this paper, from open citation data, with Crossref's record of each; the 100 newest when there are more.
References (0)
What it cites, from open citation data, with Crossref's record of each.