Bridging National and International Legal Data: Two Projects Based on the Japanese Legal Standard XML Schema for Comparative Law Studies

This paper presents an integrated framework for computational comparative law that connects Japanese statutes to international standards via a JLS-to-Akoma Ntoso conversion pipeline and employs multilingual semantic techniques to identify and visualize cross-jurisdictional legal correspondences.

The Gist

Imagine you are trying to compare the rulebooks of three different sports teams: one from Japan, one from France, and one from Germany. Each team writes their rules in a different language, uses different formatting, and organizes their chapters in unique ways. Trying to find out, "Do they all have the same rule about fouling a player?" is a nightmare for a human researcher. You'd need to be a polyglot expert who knows the history and culture of all three sports just to make sense of it.

This paper describes a project that builds a digital translator and a smart search engine to solve this problem, specifically for laws. The researchers, led by Makoto Nakamura, built a two-step system to connect legal data from Japan with the rest of the world.

Here is the breakdown of their "magic trick" using simple analogies:

The Big Problem: The "Tower of Babel" of Laws

Laws are like massive libraries. In Japan, the books are organized one way (using a system called JLS). In the rest of the world, most libraries use a different, international standard called Akoma Ntoso (AKN).

• The Issue: Even if you speak the language, you can't easily compare the books because the "shelves" and "labels" don't match. A "Chapter 1" in Japan might look like "Section A" in France.
• The Goal: Build a bridge so computers can read Japanese laws and compare them with French, German, or Korean laws automatically.

---

Step 1: The "Universal Adapter" (Project 1)

The Analogy: Think of Japanese laws as a phone charger with a unique Japanese plug. The rest of the world uses a standard international socket (AKN). You can't plug the Japanese charger into the world's power grid without an adapter.

What they did:
The team built a conversion pipeline (a digital adapter). They took the Japanese legal data and rewrote it so it fits the international standard.

• They mapped every Japanese "Law," "Article," and "Paragraph" to its international equivalent.
• The Result: Suddenly, Japanese laws look exactly like the laws of other countries to a computer. They are now "structurally interoperable." The Japanese library has been re-shelved to match the global library's layout.

Step 2: The "Smart Librarian" (Project 2)

The Analogy: Now that the books are on the same shelves, you still need to find the content that matches. Imagine you are looking for a specific recipe. You don't just look for the exact same words; you look for the same idea. Maybe one book says "Add a pinch of salt," and another says "Season lightly." A human knows these are the same. A computer usually needs help to realize that.

What they did:
They built a Semantic Search Engine using advanced AI (specifically, a type of "brain" called BERT).

• The Brain: This AI doesn't just count words; it understands meaning. It knows that "unmarried fathers" in Japanese law and "fathers of children born out of wedlock" in French law are talking about the same concept, even though the words are totally different.
• The Process:
  1. Vector Search: The AI turns every law article into a "mathematical point" in a giant 3D space. Laws with similar meanings end up close to each other, like friends standing in a circle.
  2. The Filter (FAISS): Since there are thousands of laws, the AI quickly scans the crowd to find the top candidates that are standing near the Japanese law.
  3. The Referee (Cross-Encoder): The AI takes a second look at the top candidates to make sure they are actually a good match, acting like a strict referee checking the rules.
  4. The Map: Finally, it draws a network map. Imagine a spiderweb where the Japanese law is in the center, and lines connect it to the matching laws in Korea and France.

Why This Matters

Before this, comparing laws was like trying to compare two different languages by hand, relying on a few experts who spoke both. It was slow, expensive, and limited.

This project creates a foundation for "Computational Comparative Law."

• It's not replacing lawyers: It's not saying the AI is smarter than a human judge.
• It's a super-tool: It acts like a high-powered telescope. It scans millions of laws instantly and says, "Hey, look! These three laws from three different countries seem to be talking about the same thing."
• The Benefit: This helps researchers, policymakers, and lawyers spot patterns, understand how laws spread across borders, and find similarities they might have missed.

The Current Status

Think of this system as a prototype car. It's not a fully finished luxury vehicle yet, but it proves the engine works.

• It successfully connected the "plugs" (Project 1).
• It successfully found "meaningful matches" between Japanese, Korean, and French laws (Project 2).
• Next Steps: The team needs to test it more, get human experts to verify the matches, and expand it to cover more countries and more types of laws.

In a nutshell: They built a universal translator for legal structures and a smart AI librarian that can find the "soul" of a law, regardless of what language it's written in. This turns the chaotic, isolated world of national laws into one giant, connected, searchable network.

Technical Summary

1. Problem Statement

Traditional comparative legal research faces significant barriers due to:

• Linguistic and Cultural Diversity: Differences in legal terminology, language, and historical context make direct text comparison difficult.
• Structural Incompatibility: Legal systems use different document structures and schemas (e.g., Japan's Japanese Legal Standard vs. international Akoma Ntoso), preventing automated integration.
• Scalability Issues: Comparative analysis traditionally relies on manual, expert-driven interpretation of selected provisions, which is not reproducible or scalable to large legislative corpora.
• Data Silos: National legal databases (like Japan's e-LAWS) often lack interoperability with international standards, hindering cross-jurisdictional quantitative analysis.

The paper addresses the need for a unified infrastructure that combines structural interoperability (standardizing document formats) with semantic interoperability (understanding meaning across languages) to enable computational comparative law.

---

2. Methodology

The research is divided into two consecutive projects that form a unified pipeline:

Project 1: Schema Conversion and Structural Interoperability

• Goal: Convert Japanese legislative data from the domestic Japanese Legal Standard (JLS) XML schema to the international Akoma Ntoso (AKN) schema (LegalDocML).
• Pipeline:
  1. Input/Parsing: Ingest JLS XML files from Japan's e-LAWS.
  2. Structural Analysis: Map JLS hierarchical elements (e.g., <Law>, <Article>, <Paragraph>) to AKN equivalents (<act>, <article>, <paragraph>).
  3. Transformation: Apply approximately 50 XSLT mapping rules to transform the XML structure.
  4. Metadata Enrichment: Assign FRBR (Functional Requirements for Bibliographic Records) identifiers (Work, Expression, Manifestation) and metadata (jurisdiction, version, language).
  5. Validation: Verify converted files against the LegalDocML AKN schema to ensure structural integrity.
• Outcome: A validated conversion pipeline that allows Japanese statutes to be parsed and integrated into international AKN-compliant databases without losing logical hierarchy.

Project 2: Semantic Mapping and Cross-National Correspondence

• Goal: Automatically identify corresponding legal provisions across different jurisdictions (Japan, Korea, France, Germany) using NLP.
• System Architecture:
  1. Data Ingestion: Store AKN-encoded Civil and Commercial Codes from multiple jurisdictions.
  2. Semantic Representation: Use the multilingual-e5-large embedding model (based on RoBERTa) to convert legal articles into dense vector representations in a shared semantic space.
  3. Candidate Retrieval (FAISS): Utilize FAISS (Facebook AI Similarity Search) for efficient approximate nearest-neighbor search. For each Japanese article, retrieve top candidates (e.g., top 120) from foreign codes.
  4. Multi-Stage Reranking:
     • Apply country-level quotas to balance candidates.
     • Use a Cross-Encoder model to jointly encode provision pairs and re-rank them based on precise semantic correspondence (doctrinal consistency).
  5. Visualization: Construct a bipartite network graph where nodes represent legal provisions and edges represent high-confidence semantic correspondences.

---

3. Key Contributions

• Integrated Framework: Proposes a novel architecture that unifies structural normalization (JLS $\to$ AKN) and cross-lingual semantic modeling (BERT/STS) into a single pipeline for computational comparative law.
• JLS-to-AKN Conversion Pipeline: Successfully developed and validated a technical bridge allowing Japanese legal data to participate in the global LegalDocML ecosystem.
• Prototype Cross-National Extraction System: Built a working prototype that generates candidate correspondences between Japanese, Korean, and French Civil Codes using a hybrid retrieval-reranking approach.
• Graph-Based Visualization: Introduced a method to visualize inter-jurisdictional relationships as a network, allowing researchers to explore structural proximity and "skewering" correspondences (cross-cutting links) across legal systems.

---

4. Results

• Structural Validation: The JLS-to-AKN conversion pipeline successfully transformed representative Japanese laws. Manual verification of 10 random laws confirmed 100% schema compliance and preservation of logical integrity.
• Semantic Mapping Performance (Previous Experiments):
  • Domestic (Japanese to Japanese): Mapping between the Electricity Business Act and Gas Business Act achieved an F1 score of 0.768.
  • Cross-Jurisdictional (Japanese to German): Mapping between Japanese and German Civil Codes (family law sections) achieved an F1 score of 0.348. The lower score highlights the complexity of cross-lingual and cross-tradition alignment.
• Prototype Output: The system generated a correspondence network for the Civil Codes of Japan, Korea, and France containing:
  • 2,000 vertices (776 Korean, 232 Japanese, 992 French).
  • 3,545 edges representing candidate correspondences.
  • The system successfully filtered candidates using similarity thresholds (e.g., 0.95 for Korea, 0.80 for France) to maintain analytical balance.

---

5. Significance

• Methodological Shift: Moves comparative law from a qualitative, expert-dependent discipline toward a quantitative, data-driven science. It enables the processing of entire legislative corpora rather than just selected provisions.
• Infrastructure for Global Collaboration: By standardizing Japanese data to AKN, the research removes technical barriers, allowing Japanese statutes to be compared directly with European and other international legal databases.
• Augmentation of Expertise: The system does not replace legal experts but serves as a tool to generate candidate correspondences and visualize structural patterns, expanding the scope of human analysis and revealing latent connections.
• Future Potential: The framework lays the groundwork for a "Global Comparative Law Graph," facilitating large-scale studies on legal evolution, concept diffusion, and the influence of one legal system on another.

Limitations: The current system is a prototype. The correspondence candidates require expert validation (gold-standard datasets are not yet fully constructed for the new prototype), and the accuracy of cross-lingual mapping remains lower than intra-lingual mapping due to translation nuances and conceptual divergence.