Statutory Article Retrieval from Arabic Judicial Narratives: Family Law and Law of Evidence in Saudi Arabia

Zarah Shibli; Areej Alhothali; Lulwah Alharigy

cs.CLOct 28, 2026

Statutory Article Retrieval from Arabic Judicial Narratives: Family Law and Law of Evidence in Saudi Arabia

Authors: Zarah Shibli, Areej Alhothali, Lulwah Alharigy

Organizations: Department of Computer Science King Abdulaziz University Jeddah, Saudi Arabia

Abstract

Reliable legal information retrieval must ground legal reasoning in authoritative statutory sources. In Arabic judicial settings, long factual narratives must be mapped to concise articles that express abstract legal rules, creating a narrative-to-legal-rule abstraction gap. Using a benchmark of 200 synthetic Saudi Arabic judicial cases and 381 articles from the Family Law and the Law of Evidence, we compare word-level and char-4 BM25, zero-shot dense encoders, per-law retrieval, dense-sparse reciprocal rank fusion (RRF), and a proposed Decompose-and-Refine (DaR) pipeline. In DaR, a large language model splits each narrative into disputed legal issues and generates, for each issue, three retrieval channels: a statute-style paraphrase, legal keywords, and a candidate taxonomy chapter. On the whole corpus, the instructed Qwen3-Embedding-8B dense encoder reaches an MRR@10 of 0.849, the best result of any non-decomposed system. Under per-law scoring, this advantage reflects Family Law only: restricted to Evidence articles, the same encoder drops to an MRR@10 of 0.087, and fusion with BM25 raises it only to 0.207. DaR changes this picture, raising Evidence MRR@10 to 0.696 and Family MRR@10 to 0.864. Without decomposition, the two laws favor different strategies; under DaR, the same retriever achieves the best MRR@10 on both.

Figures & tables

Cases (queries)
Statistic	Value
Total cases	200
Sarih (explicit) / kinaya (implicit) / muallaq (conditional)	66 / 66 / 68
Mean reference-set size (range)	$\sim$ 5 (1–12)
Cases with $\geq$ 1 Family citation	200 (100.0%)
Cases with $\geq$ 1 Evidence citation	161 (80.5%)

TABLE I: Dataset and corpus statistics.

Fig. 1: The Decompose-and-Refine (DaR) pipeline: an LLM decomposes a case narrative into $m$ issue entries, each retrieved through three channels and fused by RRF, then merged across issues by a per-article max score.

Panel of Figure 3 — Fig. 2: Decompose-and-Refine prompt template (both routes).

	Whole corpus			Family		Evidence
Model	R@10	R@20	MRR@10	R@10	MRR@10	R@10	MRR@10
Sparse (BM25)
BM25 (word)	0.258	0.379	0.361	0.351	0.352	0.049	0.025
BM25 (char-4)	0.202	0.434	0.134	0.249	0.110	0.082	0.033
Arabic-specific encoders
AraBERT-v2	0.003	0.003	0.004	0.001	0.001	0.004	0.004

TABLE II: E1–E2 whole-corpus retrieval: sparse baselines and sixteen zero-shot encoders.

n{=}200

(whole corpus, Family),

n{=}161

(Evidence). Bold: best per column.

Components
System	R@10	R@20	MRR@10	Hit@10
BM25 (word)	0.258	0.379	0.361	0.780
Qwen3-8B-instr.	0.508	0.635	0.849	0.990
Qwen3-4B-instr.	0.543	0.616	0.836	1.000
RRF combinations
word + char-4	0.261	0.428	0.258	0.780

TABLE III: Whole-corpus RRF fusion (

n{=}200

) compared with its components. Bold: best per column.

Dense
	Family			Evidence
System	R@10	R@20	MRR@10	R@10	R@20	MRR@10
Qwen3-8B-instr.	0.723	0.880	0.850	0.196	0.314	0.087
Qwen3-4B-instr.	0.764	0.864	0.836	0.300	0.507	0.130
Fusion
RRF, 8B + word BM25	0.613	0.783	0.744	0.220	0.310	0.207

TABLE IV: Per-law baselines (

n{=}200

Family,

n{=}161

Evidence). Bold: best per column.

Route	System	R@10	R@20	MRR@10
Evidence	DaR, BM25 (char-4)	0.752	0.865	0.696
	DaR, BM25 (word)	0.707	0.842	0.661
	DaR, Qwen3-4B-instr.	0.750	0.923	0.469
	DaR, Qwen3-8B-instr.	0.674	0.853	0.386
Family	DaR, BM25 (char-4)	0.726	0.928	0.864
	DaR, BM25 (word)	0.749	0.926	0.848

TABLE V: DaR systems per route (Evidence: 161 cases, Family: 200 cases). Bold: best per route and column.

	Family		Evidence
System	MRR@10	R@10	MRR@10	R@10
BM25 (word)	0.412	0.399	0.048	0.088
Dense, Qwen3-8B-instr.	0.850	0.723	0.087	0.196
Dense, Qwen3-4B-instr.	0.836	0.764	0.130	0.300
RRF, 8B + word BM25	0.744	0.613	0.207	0.220
DaR (char-4 BM25)	0.864	0.726	0.696	0.752

TABLE VI: Best-performing systems from each retrieval method group, selected by MRR@10 for each law. (

n{=}200

Family,

n{=}161

Evidence).

	MRR@10			R@10
Retriever	Base	DaR	$\Delta$	Base	DaR	$\Delta$
Family ( $n{=}200$ )
BM25 (word)	0.412	0.848	$+0.436$	0.399	0.749	$+0.350$
BM25 (char-4)	0.409	0.864	$+0.455$	0.456	0.726	$+0.270$
Qwen3-4B-instr.	0.836	0.720	$-0.116$	0.764	0.778	$+0.014$
Qwen3-8B-instr.	0.850	0.716	$-0.134$	0.723	0.702	$-0.021$

TABLE VII: Effect of DaR with the retriever held fixed. Base: raw Fact query with per-law restriction;

\Delta

= DaR

-

Base. Green : improvement; red : decline.

Fig. 3: Baseline vs. DaR retrieval on one case, both routes. Green IDs are reference articles; DaR recovers one additional Family article at $K{=}10$ that the baseline misses entirely, and both paths fail to rank the case’s digital-evidence article inside the top 10, though DaR alone recovers it by rank 28 of 50.

Legal IR

Legal information retrieval finds and ranks relevant statutes, cases, patents, and other legal materials for queries and evidence-grounded systems.

See all 55 papers

Reading a Legal Question Word by Word: Embedding Trajectories of 2,144 Vietnamese Legal Headlines
Sep 8, 2026Tran Minh QuanText Embeddings Word Embeddings
Legal-R1: Agentic Retrieval for Evidence-Based Legal Reasoning
Oct 13, 2026Huaiwen Pang, Jiale Han, Junying Chen +2Legal QA Legal Reasoning
The Chinese University of Hong Kong, Shenzhen · Shenzhen Loop Area Institute · Shenzhen Shenten Digital Technology Co., Ltd.
GreekBarRetrieval: A Benchmark for Greek Statutory Retrieval
Aug 19, 2026Ernest Beta, Odysseas S. Chlapanis, Dimitrios Galanis +1Legal IR Query Reformulation
Department of Informatics, Athens University of Economics and Business, Greece · Archimedes, Athena Research Center, Greece · Institute for Language and Speech Processing, Athena Research Center, Greece

Sep 8, 2026J/K move · Enter open · S save

Reading a Legal Question Word by Word: Embedding Trajectories of 2,144 Vietnamese Legal Headlines

Tran Minh Quan

Text Embeddings Word Embeddings Legal IR Information Retrieval Dense Retrieval

Open paper AskPDF arXiv

A dense retriever encodes a question as one vector, but the question arrives one word at a time. We read 2,144 held-out headlines from Thu Vien Phap Luat (Vietnamese legal library) word by word with Nemotron-3-Embed 8B/1B and Qwen3-Embedding 8B/0.6B, encoding 65,444 prefixes against 20,034 articles, plus every prefix of 3,438 sub-questions from 1,112 multi-question headlines and of 168 answers. (i) The gold article becomes rank 1 after a median of 6-7 content words in every encoder, before the interrogative frame is read, and stays there to the end in 78-85% of cases. (ii) In a multi-question headline the lock is inside the first sub-question 94-98% of the time; the second leaves rank unchanged in 89-95%; encoded alone, the second reaches rank 1 in 42-58% vs 91-96% for the first, at the same lock word (95-97% identical). (iii) Numbers, dates and instrument identifiers move the embedding twice as far as content words and four times as far as interrogative words; 72-78% of steps move toward the gold article, and the closing interrogative frame moves against that direction in 95-99% of headlines. (iv) Rank/cosine clustering yields six archetypes (instant, typical, unstable, late, never-locking) that differ by legal area and form (chi-squared p < 1e-8): real-estate and litigation headlines never lock on a number; environmental and accounting headlines do so a third of the time. (v) An answer read word by word retrieves its article after 8-16 words and addresses the sub-questions in order asked in 83-89% of cases. (vi) A word's step keeps a consistent direction across headlines (cosine 0.25-0.33; 0.44-0.60 for numbers); a preceding question rotates that step by about 60 degrees and a greeting by about 30 degrees; steps shrink as i^{-0.8}; and a two-question headline is within 12-17 degrees of a linear mix of its two questions. We call this a context-modulated additive walk.

Abstract

Legal IR

Explore similar work

Reading a Legal Question Word by Word: Embedding Trajectories of 2,144 Vietnamese Legal Headlines

Legal-R1: Agentic Retrieval for Evidence-Based Legal Reasoning

GreekBarRetrieval: A Benchmark for Greek Statutory Retrieval