This paper details the creation of an oficial graph of Italian law, which is exposed using the Linked Data 5 stars standard and provides a SPARQL endpoint for users to query the data. The graph is constructed using a network analysis of legal documents, with each entity fully dereferenced to provide rich and comprehensive information. The graph also maps oficial descriptors to facilitate understanding and allows users to navigate the data visually using an intuitive interface. The paper discusses the technical implementation of the graph and the potential benefits for legislators and citizens.
Graph for Enhanced Legislative © 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License 6https://huggingface.co/nlpaueb Attribution 4.0 International (CC BY 4.0).
In recent years, the adoption of Semantic Web technologies has provided a new framework for accessing and tion Framework) triples, ontologies, and Linked Data principles have enabled data integration from diferent sources, making it possible to create a web of interconnected data. In the legal domain, the adoption of Semantic Web technologies has the potential to revolutionize ing ontologies and RDF triples can enable the integration of legal data from diferent sources, making it possible to create a unified and interconnected legal knowledge applications and services that can improve the access to legal information for citizens and professionals alike. formation are Knowledge Graphs (KGs), initially conceived to connect documents by means of machineunderstandable semantic links between the entities to improve data retrieving and access. The success of knowledge graphs is indicated by the results of the Linking guage models [10], such as Chalkidis et al. [11]6, Strubell et al. [12], and Liu et al. [13]7, to enable additional automated tasks that are infeasible for a single human being. However, Large Language models and RDF technologies are not enough to guarantee data is actually “linked”. 1https://lod-cloud.net/datasets 2https://searchengineland.com/library/bing/bing-satori 3https://blogs.bing.com/search/2013/03/21/ understand-your-world-with-bing
introducing-knowledge-graph-things-not/
under-the-hood-the-entitiesgraph/10151490531588920/ base. This can facilitate the development of new legal ing due to the overwhelming amount of unstructured text
The de-facto standard for representing structured in- the various laws. These challenges have sparked the sharing information on the web. RDF (Resource Descrip- lowing reasoning, inference, and access to widespread the way legal information is accessed and processed. Us- ated several advancements in the tasks of recommendaItalian Legal Archive
IPZS APIs
Law Description
Extractor
Law Description
Converter
RDF Graph
Triplestore Dereferencer Graph Browser
SPARQL Endpoint Navigable Legal Graph
Dashboard Indeed, an ontological alignment was still missing. • it is not designed to automatically correlate acts (e.g.,
To this extent, the European Union published the Eu- in order to easily identify impacts deriving from the roVoc thesaurus8 to classify legal documents. EuroVoc is repeal, or promulgation, of a law). a multilingual thesaurus maintained by the Publications To overcome this limitations, Politecnico di Bari and Ofice of the European Union 9, which contains more than Istituto Poligrafico Zecca dello Stato are cooperating to 7,000 concepts referring to various activities of the EU and create an ecosystem capable of representing the graph its Member States. The thesaurus is used as a classifica- of Italian regulations. We have realized an experimention schema for the two most well-known legal datasets, tal pipeline that extracts relevant information from the JRC-AcquisV3 [14] and EURLEX57K [15], which contain documents currently produced by IPZS. Currently, this legal documents from the legal information system of process is performed manually, and the graph is deduced the European Union (Eur-Lex). EuroVoc provides all its only by the appointed expert. Therefore, the project is terms in the oficial language of the EU member states part of a set of initiatives aimed at innovating the techto enable a multilingual search. The EuroVoc thesaurus niques used by IPZS and improving service delivery. has been introduced to harmonize the classification of The project’s contributions are significant as they indocuments in the communications across EU institutions. troduce automation to a process that was previously man
The Oficial Gazette of the Italian Republic 10 is the of- ual, reducing the risk of errors and inconsistencies. Furifcial source of knowledge for the laws in force in Italy. It thermore, the use of containerization technology allows has the mission of disseminating to all citizens the infor- for easy deployment and scaling of the system, making it mation published in the Gazette in its series, the general more eficient and flexible. Moreover, the project uses an and the special ones. The Gazette is published by the approach to document classification based on the extracIstituto Poligrafico e Zecca dello Stato (IPZS) in collab- tion of relevant information, including the text, hyperoration with the Ministry of Justice, which provides for links, and connections to other regulations using three drafting the laws and directing their publication. Accord- main sources: the previously annotated correlations, the ing to the Law of 13 July 1966, n. 559, IPZS provides for textual references to other laws, and the semantic simithe printing and management, also with IT tools, of the larity between the laws.
Oficial Gazette. Legislative information has not been For what regards the framework to expose the graph, structured in electronic format since the beginning. The it is composed of the three technological pillars, a triplemanagement system was conceived to take solely care store, a dereferencer, and a navigator. These tools enable of the editorial aspect of the publication process of the the construction, representation, and navigation of the Oficial Gazette. This approach caused some limitations: RDF graph, making it more accessible and user-friendly. • the juridical collections are defined on an editorial The triplestore is used to store the files generated by the basis and not on a reliable and detailed classification; processing of the normative corpus, while the dereferencer provides a user-friendly interface to display the 8https://eur-lex.europa.eu/ information stored about a specific regulation. The navi9https://publications.europa.eu/en/web/eu-vocabularies gator allows users to explore the graph visually, enabling 10https://www.gazzettaufficiale.it them to understand the relationships between diferent regulations.
Overall, the project’s contributions are manifold, as the project introduces an automate approach to the creation of the graph of Italian regulations, which was previously a manual process. The use of containerization technology and the integration of tools to store, represent and navigate the RDF graph makes it more accessible and user-friendly. The system’s deployment is improving the quality of the service provided by Istituto Poligrafico Zecca dello Stato, and it is an important step towards the innovation of the techniques in the regulation industry. 2. The composition of the titles of the documents is carried out and a scanned text is attached to each title. A publication of the Oficial Gazette is assigned; 3. Digitization of text also using OCR techniques; 4. The act is impaginated and undergoes a quality check; 5. A security glyph is generated and applied for each
page of the Gazette; 6. Following several checks, the paginated law is digi
tally signed to create the act certified version. 7. The act is classified, enriched with metadata and rela
tionships with other acts. The act is published.
ure 1, designed to analyze the normative corpus provided by the Istituto Poligrafico Zecca dello Stato and process it The process of publishing a law involves the Italian Min- to construct a turtle file that can be used to build the RDF istry of Justice and the Istituto Poligrafico e Zecca dello graph indices in the triplestore. We treated the graph Stato (IPZS). From the beginning, this process has un- preparation phase as an independent development phase, dergone several variations. The technologies that have as a future replacement of the norm extraction parameter followed one another over time have been diferent, and model will require a diferent data preparation process. even today, there are manual activities that introduce the The turtle files are then used to build a graph in the risk of performing rework. Figure 2 shows the process triplestore, which is finally served by exposing a query of publishing of a law in the general series. function. This functionality activates a dereferencing tool that allows for a human-friendly and concise visuSTART alization of the stored information regarding a specific norm. This tool provides access to another service that 1 2 is properly configured to navigate the normative graph
Title processing using transparent SPARQL endpoint queries. Therefore, Issue of acts sacnadnnteinxgt the human operator will have four possible interfaces to use, depending on the activity to be performed and their 4 3 technical preparation.
TQeuxatlliatyyocuotnatrnodl Text digitization definTihnegpursoejeccatsiensvaonlvdedthseevaecrtaolrsdeinsivgonlvpehda,sleosa,dinincglutdhineg database and transforming it to match the necessary data 5 6 model for training, identifying data cleaning activities geGnelyrpahtion Digital signature for the provided data type, researching and identifying methods for integrating the machine learning system with the rest of the software architecture. 7
Data enrichment and publication
END The system for the representation, visualization, and querying of the graph of Italian laws is designed as a Figure 2: The process of publishing an Italian law: (i) The tool that not only assists operators in identifying regulaMinistry of Justice issues the law, (ii) the metadata are as- tory impacts but also innovates and simplifies a burdensigned to the law, (iii) irrespectively to the original format, a some decision-making process for human operators. In digital version of the law is generated, (iv) the law is laid out production, this system will be useful in simplifying all (avni)dtphaegdinigaitteadl,l(avw)aissescigunrietdy,g(lvyipi)ht hisegleanweriastceodmabnidneadppwliietdh, other activities involving potential connections between metadata and published. the regulations of the Italian legal system. The system includes the following modules: (i) Law Description Ex1. The act is issued by the Ministry of Justice in paper tractor, (ii) Law Description Converter, (iii) Triplestore, or digital format; (iv) Dereferencer, (v) Graph Browser, (vi) SPARQL End- data in RDF format following the Linked Open Data stanpoint, (vii) User-friendly Dashboard. dards through a SPARQL endpoint. The tool guarantees Law Description Extractor Module. The Extractor a good user experience by providing access to RDF reloads the entire regulatory database available to the Is- sources through web pages. Additionally, the module is tituto Poligrafico Zecca dello Stato. Despite the current fully customizable, enabling the modification of the web storage of regulations in the form of files, the module interface, page style, and all the content that needs to allows to retrieve such information remotely through be displayed. In addition to dereferencing resources, the appropriately prepared APIs. The module is currently module must allow for publishing data in RDF, guaranable to read and load the text of each regulation and all teeing diferent solutions for serialization. For example, metadata provided by IPZS. The module includes the fol- it is possible to access the same resource by requesting lowing features: (i) Read regulation and its descriptors; a response in HTML (web page), JSON, turtle, n-triples. (ii) Collection of regulations into an interoperable data The module includes the following features: (i) Access to structure suitable for subsequent conversion into a graph; individual entities (norms) of the graph; (ii) Reporting all (iii) Recognition of currently in-force regulations. information related to the entity; (iii) Saving information related to an entity.
Law Description Converter Module. The Converter transforms the data structure of regulations into a file suitable for loading into a triplestore. Currently, the used format is turtle. However, as the graph representation system used may vary, so may the data processing activities. Therefore, the module provides a modular structure, where each transformation phase can be activated and deactivated if necessary. The module includes the following features: (i) Construction of the regulation skeleton in the form of a graph; (ii) Identification and use of prefixes and predicates currently in use in Linked Open Data; (iii) Composition of the descriptor triples of a regulation; (iv) Saving the database in an interoperable format for graph representation.
Graph Browser. This module allow for navigating data exposed in the form of Linked Open Data, requiring only a SPARQL endpoint to be contacted. Starting from a node within the graph, it lets navigate all the relations starting from it, thus reaching all connected entities. Proceeding recursively, it allows for analyzing and navigating the knowledge graph without any limitation. The module includes the following features: (i) Visualization of the entity and its related predicates; (ii) Visualization of the objects of triples (information about the norm or other norms); (iii) Possibility to continue navigation through new encountered norms.
Triplestore Module. The Triplestore loads the triples functionalities on the graph using the standard SPARQL previously produced into a system capable of construct- language. Query results could be obtainable also by using ing the RDF graph and its respective indices. The graph APIs and can be returned using diferent interoperable can be divided into multiple subgraphs, and the system formats. Since the methods for using the model could must allow individual subgraphs to be loaded and re- vary depending on the system of classification used, the moved. The module includes the following features: (i) module provides a unified interface for the use of such Loading files representative of the graph; (ii) Identifica- functions. The module includes the following features: tion of individual subgraphs; (iii) Querying of subgraphs; (i) APIs for accessing the query system; (ii) Graphical (iv) Updating of subgraphs. interface for writing queries; (iii) Possibility to choose the response format, also through content negotiation.
User-friendly Dashboard. The Dashboard makes could cause unwanted impacts on other laws in force. available several ready-made queries for the user who The legislative corpus is wide and not devoid of entropy. does not have familiarity with the SPARQL query lan- Sometimes, the only solution to make a set of laws more guage. Since future queries may difer from those cur- understandable is to group them into a consolidated text. rently available, the module is divided into modular tasks Paul needs an assisted system that can make the acthat can be easily maintained and extended. The module tivities of identifying and verifying legislative impacts. includes the following features: (i) Graphical interface The system under study includes the existing impacts that masks the actual complexity of queries; (ii) Possibil- between laws and is able to receive a draft of a new law as ity to choose the response format; (iii) Modular encapsu- input, understand its natural language, and carry out an lation of queries. assisted preliminary verification of the impacts that this would have on other laws in force after its promulgation.
mented following this study, the “personas model” was employed. The model describes the behaviors, dificulties and needs of the two main typical users identified by the study: the legislator and the lawyer/judge.
Persona 1 - Violet (lawyer): Profile description and opportunities of the system in relation to her needs. Violet is a lawyer who consults online legislative collections daily to search for normative references to use for her hearings. The consultation activities bring good results but usually take several hours of careful work.
The dificulties in carrying out her work are due to the fact that the laws generally refer to many other laws, and often the connections between them are challenging to ifnd. The legislative database is wide, and a satisfactory search often requires many hours of work. Moreover, consultation systems require precise text inputs, not being able to understand natural language.
Violet needs to reduce his research times without compromising the quality of his work. She would like to carry out smart queries, which accept more flexible inputs and return all the material she needs. The output should not be fragmented way and easily navigable.
The system under study provides a knowledge graph of the entire legislative database and recognizes active and passive impacts between the laws. Following a query, the system isolates and returns a subgraph. This representation greatly simplifies Violet’s work; now, she can navigate within the subgraph and expand it in the depth that she prefers.
Persona 2 - Paul (legislator): Profile description and opportunities of the system in relation to his needs. Typical operations in charge of Paul are the introduction, the modification, the repeal of a law, and the transposition of a European directive. He deals with the gold-plating processes of European directives and with simplifying and merging laws when they are too fragmented.
These and other operations are never risk-free and
Zecca dello Stato developed an experimental pipeline that automates the creation of the graph of Italian regulations. The project reduces the risk of errors and inconsistencies by introducing automation to a previously manual process. Furthermore, the use of containerization technology and the integration of tools to store, represent, and navigate the RDF graph makes it more accessible and user-friendly. The project is an important step towards the innovation of the techniques used in the regulation industry, and it improves the quality of the service provided by Istituto Poligrafico Zecca dello Stato. The framework enables the construction, representation, and navigation of the RDF graph, making it easier for users to explore the relationships between diferent regulations. Overall, the project represents a significant achievement in the ifeld of regulation management.
Acknowledgements