Jusletter IT

The development of advanced Internet services is mainly dependent on the implementation of the Semantic Web concept, promoted by the World Wide Web Consortium (W3C) as an architecture of standards and tools for content description where the information can be made accessible and understandable not only by humans, but also understandable and processable by machines.

The actual use of Semantic Web technologies in large scale however is currently facing obstacles due to different reasons: on the one hand for the difficulties to come into an agreement within the Internet communities in sharing general-purpose semantic models, as well as the hardly scalable costs of their implementation; on the other hand for the efficacy of general purpose search engines which usually provide a valuable degree of satisfaction of the users' information needs.

However the legal domain, in particular the legislative one, has specific characteristics, as well as users of legal information systems share peculiar needs, which can make the implementation of the Semantic Web concept more viable.

First of all, legislative documents have well defined features as far as their structure and semantics are concerned; similarly the law-making process, in particular legislative drafting, is characterized by a well structured and defined workflow, which involves legislative offices in Parliaments and PAs. Moreover legal information systems are not only to be conceived for advanced search and retrieval services to the users, but also to maintain and upkeep the legal order, to monitor the impact of new norms on it, to handle document timeline and versioning. Finally users are mainly interested in accessing norms rather than simply documents; they are particularly interested in knowing the relations between norms, having support to legal reasoning and consultancy services, as well as having instruments to check procedures compliance with respect to specific statutes and regulations.

In this context the need of technologies providing users and machines with legal information advanced management services is ever growing. Such need can find an effective answer in Semantic Web standards and technologies which represent a pre-condition for the development of advanced services oriented from the one hand to legislators and legal information providers, as well as to legal experts and citizens from the other.

Such services are mainly addressed to

• improve the quality and accessibility of legal information in different legal systems;
  
• promote the interoperability among applications managing legal information;
  
• provide high quality integrated services (information retrieval, legal reasoning, legal consultancies) for both policy makers and citizens.
  

A complete evolution towards the Semantic Web concept, as conceived by Tim Berners-Lee, has yet to be realized. However in some specific domains, as the legal one, it might be not so far to occur, especially when advantages for different levels of users were undoubtedly clear. In this respect the legal domain, for its technical nature, seems a privileged field in which the Semantic Web concept can be effectively implemented, moreover its specialized profile makes it more sensitive than other fields to the benefits deriving from such implementation.

In this paper standards and tools for a possible implementation of the Semantic Web in the legal domain, on the basis on a semantic model for legislative documents, are discussed. In particular in Section 2 considerations about the semantics of generic texts are discussed; in Sections 3, 4 and 5 a semantic model for legislation, its organization and possible implementation standards are respectively presented; in Section 6 possible strategies and tools to implement this model are shown; in Section 7 advanced services which can be conceived on the basis of such model are sketched; finally in Section 8 some conclusions are discussed.

The semantics of a generic text can be mainly perceived according to 3 meaningful levels:

Managing these three levels is an essential pre-condition to develop effective tools able to address the Semantic Web concept, in terms of document indexing, semantic relations between concepts, documents organization based on subject matters.

In the legislative domain similar considerations can be made; in fact, the semantics of legislative documents can be described according to the previous three levels as well. The bottom level can be defined as the level of domain concepts, describing entities which are regulated by the including act. The middle level, representing aggregations of such concepts, describes the relations between domain concepts in terms of rules. Finally, the top level describes the matter of interest for the act.

The lack of control of the previously described semantic levels causes problems of different nature: from a difficult harmonization and control over the legal lexicon (such as unclear, confusing terminology, incomplete or inconsistent regulations, use of vague terms), to the uncertainty of the impact of new laws in terms of coherency preservation, as well as the difficulties in accessing norms by both citizens and legal experts, till the inability to obtain an analytical/systematic vision of a legal order which creates obstacles to its knowledge and upkeep.

While semantic annotation of concepts and document indexing are effective tools able to handle legal texts semantics at bottom level (concepts) and top level (subject matter), they are usually not sufficient to fully address problems of legislation accessibility and upkeeping.

In the '90 Biagioli [Biagioli, 97], senior researcher at ITTIG-CNR, identified a possible reason of these problems in the fact that while a legislative text is a normative and documentary unit of reference, users and legal experts usually manage, access and refer to the legal order in terms of the contained norms. Therefore a more analytical unit of reference was identified in order to have a more organic view of the legal system, interpreting the middle level of legislative text semantics in terms of legalprovisions .

To elaborate his theory Biagioli combined the Searlian theory of rules preceived as «speech acts»' with the Raz's lesson [Raz, 80] to perceive laws and regulations as a set of rules orprovisions , rather than of laws, carried by speech acts.

Following a provision-centric point of view, Biagioli underlined two views orprofiles according to which a legislative text can be perceived:

• a structural orformal profile , representing the traditional legislator habit of organizing legal texts in chapters, articles, paragraphs, etc.;
  
• a semantic orfunctional profile , considering legislative texts as composed byprovisions , namely fragments of regulation [Biagioli, 97] expressed by speech acts.
  

In this perspective, fragments of a legislative text are, at the same time, paragraphs and provisions, according to whether they are seen from a formal or functional view-point. Basically the formal profile represents the physical structure of a legislative text; on the other hand the functional profile represents the logical structure of it and it is related to how the semantics of a legislative text is organized.

The functional profile of a legislative text, in particular, can be more analytically analysed to distingush two sub-profiles representing as many specific semantic roles: theregulative profile and thethematic profile . The first one reflects the lawmaker directions, the second one the peculiarities of the regulated field. The regulative profile can be described in terms ofprovision types . A provision can be of different types asTerm definition ,Duty, Redress , as well as more technical ones asInsertion ,Abrogation ,Substitution , etc. The thematic profile can be described by the so-calledArguments of the provisions (for example theBearer of a Duty).

Provision types and relatedArguments represent a semantic model able to describe the functional profile of legislative texts; such a model has been calledModel of Provisions [Biagioli, 97]. Provision types and arguments can be considered as a sort of metadata scheme able to analytically describe fragments of legislative texts. For example, the following fragment of the Italian privacy law:

besides being considered as a part of the formal profile of a legislative text (aparagraph ), can also be viewed as a component of its functional profile (aprovision ) and qualified as aprovision of typeDuty , whose arguments are:

A specific classification of legislative provisions was carried out by analysing legislative texts from a semantic point of view, grouping provisions into two main families:Rules (introducing and defining entities or expressing deontic concepts) andRules on Rules (different kinds of amendments). Rules are provisions which aim at regulating the reality considered by the including act. Adopting a typical law theory distinction, well expressed by Rawls, they consist in:

• constitutive rules : they introduce or assign a juridical profiles («empowering norms») to entities of a regulated reality;
  
• regulative rules : they discipline actions («rules on actions») or the substantial and procedural defaults («remedies»).

On the other hand, Rules on Rules can be distinguished into:

• content amendments : they modify literally the content of a norm, or their meaning without literal changes;
  
• temporal amendments : they modify the times of a norm (come-into-force and efficacy time);
  
• extension amendments : they extend or reduce the cases on which the norm operates.
  

As discussed in the previous section, in Biagioli's model each provision type has specific arguments describing the roles of the entities which a provision type applies to (for example theBearer is argument of aDuty provision) [Biagioli, 08].

Other models of legal concepts have been proposed in the literature, from the traditional Hohfeldian theory of legal concepts [Hohfeld, 78] until more recent legal philosophy theories due to [Rawls, 55], [Hart, 61], [Ross, 68], [Bentham, 70], [Kelsen, 91]. With respect to them, Biagioli's model has been conceived to describe legislative text semantics: its specific textual anchorage represents, in our point of view, its main strength.

Specific tools can be used to describe the semantics of legislative texts at different levels of abstraction.

The semantics of legal concepts (bottom level) is usually described by controlled vocabularies, thesauri or domain ontologies. They give an unequivocal meaning to the terms used in legislation, as well as a possible harmonization of legal concepts in a multilanguage domain, as the European one [Francesconi, 09]. Domain ontologies, in particular, can provide relationships between concepts describing the scenario of the domain they pertain to¹ .

The semantics of a whole act (top level) is usually described by general metadata, whose content are usually thesauri descriptors able to identify the subject matter which the act deals with. Several metadata standards can be used, however, especially in the Web context, the Dublin Core metadata standard, in its RDF implementation, is probably the most widely used.

The semantics of provisions (middle level) can be effectively implemented in terms of a set of metadata, able to analytically describe norms. Other possible implementation of the Provision Model can be envisaged: for example the NormeInRete [Francesconi, 07] standards provide an XML implementation of the model for the Italian legislation. On the other hand a more general, standard neutral, RDF implementation of the model is currently under development.

The Model of Provision can be implemented in bottom-up or top-down strategies.

When implemented in a bottom-up strategy it can support the documentalistic activities of annotating legislation from a detailed semantic point of view. Particularly the bottom-up strategy can be carried out manually or automatically.

The manual bottom-up detection of provisions within an existing legislative text consists, essentially, in an analytic effort in which all the possible distinctions among the elements that go to make up a legislative text are identified, singling out the nature and function of each. Basically, on the basis of the Model of Provisions, this activity consists in classifying portions of legislative texts according toprovision types and in defining roles of text fragments in terms ofprovision arguments , within a coherent functional vision of the legal system. The automatic (or semi-automatic) bottom-up detection of provisions consists in using automatic tools able to detect and classifying provisions, as well as to extract their arguments. Both the manual and the automatic bottom-up approaches have been implemented within xmLegesEditor (Fig. 1) [Agnoloni, 07], the open-source legislative editor, developed by ITTIG-CNR, able to implement legislative XML standards² .

Figure 1: xmLegesEditor legislative drafting environment

The manual bottom-up approach consists in a number of forms by which qualifying portions of legal texts, usually paragraphs, in terms of provision types, and to fill the related arguments with concepts, possibly coming from domain ontologies [Antoniou, 99] [Hoekstra, 08] [Francesconi, 09].

For the automatic bottom-up approach two tools calledxmLegesClassifier andxmLegesExtractor [Francesconi, 07a] [Francesconi, 09] using machine learning and NLP techniques, have been developed: they are respectively able to classify paragraphs in terms of provision types and to identify the related arguments.

When implemented in top-down strategy the Provision Model can provide facilities to draft new bills according to a semantic point of view [Branting, 99] [Biagioli, 07]. Using planning facilities the legislative drafter will be able to describe and organize firstly the functional profile (logical structure) of a text by managing provisions instances; on this basis, the system will be able to generate the correspondent formal profile (physical structure) of a text and wording the related provisions [Biagioli, 07].

This strategy implements anex-ante (Fig. 2) management of semantics [Hoffmann and Lachmayer, 07]: the drafter is required to manage semantic objects in terms of Provisions Model and domain entities, to establish the necessary relations, to organize the semantics according to functional criteria of aggregation. Such criteria can be either thematic, depending on provision arguments content, or regulative, deriving from the types of provisions and their typical structure.

For instance, according to a common criterion, followed at least by the Italian legislative drafter and in the European directives, theDefinitions can be grouped in a single article at the beginning of the text. Another typical criterion is the aggregation of the provisionsDuty ,Procedure ,Derogation , related to a specificaction [Biagioli, 07].

At the end of this process, the formal partitions of the act will contain semantically correlated components (provisions), and a qualified formal skeleton of the new act can be generated. The basic assumption is that a «paragraph», the basic component of the formal profile, usually contains a «provision», basic component of the functional profile, assumption which is widely observed by the legislator. Finally, partition wording can rely upon the user, or proposals of partitions wording can be generated on the basis of the semantics of the provision associated to each partition.

This conceptual, model-driven, legislative drafting activity (also calledmeta-drafting [Biagioli, 07], [Biagioli, 08]) is intended to enhance quality of legislative texts from both structural and linguistic point of views, since the formal profile of a legislative text can be obtained as a result of the organization of its semantics. On the contrary, in the traditional activities of drafting a bill (ex-post strategy, Fig. 2), at the end of the drafting process the formal structure of the document may not be the best one to express its semantics, since a semantic annotation is usually carried out at the end of the drafting activity, when legislative text structure and wording are already defined.

Figure 2: Traditional vs. Semantic (model-driven) legislative drafting

Using a model-driven legislative drafting approach, on the contrary, the traditional drafting process is reversed, providing facilities to firstly express legislative text semantics in terms of provision instances, and only in a second phase, to organize such semantic objects in a well-suited formal structure (ex-ante strategy, Fig. 2), as well as to implement partitions wording (manually or automatically) on the basis of provisions semantics. This process will contribute to well-structured, standardized, more accessible texts, where the formal profile fits well the defined functional profile.

Legislative text semantics can be expressed, as mentioned, in terms of Provision Model, usually added to existing texts as metadata (ex-post ), chosen by documentalist or by specialized software. In theex-ante drafting strategy they play a different role: firstly, they are chosen during the legislative drafting process itself, before document structuring and wording, helping and standardizing such a process; secondly they are chosen by the drafter, therefore they are 'authentic' metadata, as chosen by the legislator: this point is a very crucial issue, recently discussed within the legislative XML community, as pointed out in [Lachmayer, 05].

Advanced applications and services based on the Model of Provisions can be conceived.

The description of the amendments using the related provision types paves the way to the development of applications for automatic consolidation of legislative texts [Spinosa, 09].

A corpus of laws and regulations entirely qualified according to the Provisions Model allows to develop advanced search and retrieval services for legislation able to retrieve not just documents but also the contained norms [Biagioli and Turchi, 05], as well as services for legal assessment and reasoning.

Moreover, the Provision Model applied to legislation can effectively support analyses concerning the coherency of the legal system. Similarly regulatory impact analysis procedures can be implemented, able to assess the impact of new norms on the legal systems on the one hand, as well as on the public opinion on the other, making legislation more accessible and understandable, thus stimulating the democratic participation of citizens in the legislative process.

Finally, as discussed in Section 6, the Model of Provisions can be used to provide law-makers with tools to plan organic and well structure bills according to a conceptual point of view, enhancing the quality of legislative documents.

For its technicality and specialized nature, the legal domain, in particular the legislative one, seems to be a privileged field in which the implementation of the Semantic Web concept can be effective and viable. A deep analysis and implementation of models to describe legislative texts semantics can provide instruments to make legislation more understandable and processable by both humans and machines. The Model of Provisions, in particular, seems to be well suited to this aim for its ability to describe the semantics of norms as expressed in legislative texts.

The top-down or bottom-up, manual or automatic implementation of such a model paves the way to a set of services able to enhance the quality of legislative texts, making legislation more accessible and manageable for both Public Administrations and citizens.

Biagioli, Carlo , Towards a legal rules functional micro-ontology, in Proceedings of workshop LEGONT ’97 (1997).
Raz, Joseph , The Concept of a Legal System. Oxford University Press (1980).
Biagioli, Carlo and Grossi, Davide , Formal aspects of legislative meta-drafting, in Proceedings of the Jurix Conference: Legal Knowledge and Information Systems, pp. 192-201 (2008).
Hohfeld,Wesley Newcomb , Some fundamental legal conceptions. Greenwood Press (1978).
Rawls, John, Two concepts of rule, Philosophical Review, vol. 64, pp. 3-31 (1955).
Hart, Herbert, The Concept of Law. Clarendon Law Series. Oxford University Press (1961).
Ross, Alf, Directives and Norms. London: Routledge (1968).
Bentham, Jeremy and Hart, Herbert, Of Laws in General. London: Athlone, (1970 (1st ed. 1872)).
Kelsen, Hans, General Theory of Norms. Clarendon Press, Oxford (1991).
Agnoloni, Tommaso, Bacci, Lorenzo, Francesconi, Enrico, Peters, Wim, Montemagni, Simonetta and Venturi, Giulia, A two-level knowledge approach to support multi-lingual legislative drafting, in Law, Ontologies and the Semantic Web, J. Breuker, P. Casanovas, M. Klein, and E. Francesconi, eds., IOS Press, vol. 188 of Frontiers in Artificial Intel ligence and Applications, pp. 177-198, (2009).
Francesconi, Enrico, Technologies for European Integration. Standards-based Interoperability of Legal Information Systems, European Press Academic Publishing (EPAP), ISBN 978-88-8398-050-3 (2007).
Agnoloni, Tommaso, Francesconi, Enrico and Spinosa, Pierluigi, xmlegeseditor: an open-source visual xml editor for supporting legal national standards, in Proceedings of the V Legislative XML Workshop, pp. 239-251, European Press Academic Publishing (EPAP) (2007).
Antoniou, Grigoris, Billington, David, Governatori, Guido and Maher, Michael, On the modeling and analysis of regulations, in Proceedings of the Australian Conference Information Systems, pp. 20-29 (1999).
Hoekstra, Rinke, Breuker, Joost, Bello, Marcello and Boer, Alexander, LKIF Core: Principled ontology development for the legal domain,” IOS Press (J. Breuker, P. Casanovas, M. Klein, and E. Francesconi, eds.), in Legal Ontologies and the Semantic Web (2009).
Francesconi, Enrico and Passerini, Andrea, Automatic classification of provisions in legislative texts, International Journal on Artificial Intelligence and Law, vol. 15, no. 1, pp. 1-17 (2007).
Francesconi, Enrico, An approach to legal rules modelling and automatic learning, IOS Press, in Proceedings of the Jurix Conference (G. Governatori, ed.), pp. 59-68, (2009).
Branting, Karl, Lester, James and Callaway, Charles, Integrating discourse and domain knowledge for document drafting,” in Proceedings of the Seventh International Conference on Artificial Intel ligence and Law, (Oslo, Norway), pp. 214-220, 1999.
Biagioli, Carlo, Cappelli, Amedeo, Francesconi, Enrico and Turchi, Fabrizio, Law making environment: perspectives, European Press Academic Publishing (EPAP), in Proceedings of the V Legislative XML Workshop, pp. 267-281, (2007).
Hoffmann, Harald and Lachmayer, Friedrich, Provisions as legislative elements: Annotations to Biagioli’s theory, European Press Academic Publishing (EPAP), in Proceedings of the V Legislative XML Workshop, pp. 137-148 (2007).
Lachmayer, Friedrich and Hoffmann, Harald, From legal categories towards legal ontologies, in Proceedings of the International Workshop on Legal Ontologies and Artificial Intelligence Techniques, pp. 63-69 (2005).
Spinosa, Pierluigi, Giardiello, Gerardo, Cherubini, Manola, Marchi, Simone, Venturi, Giulia and Montemagni, Simonetta, NLP-based metadata extraction for legal text consolidation, in Proceedings of the International Conference of Artificial Intelligence and Law, pp. 40-49 (2009).
Biagioli, Carlo and Turchi, Fabrizio, Model and ontology based conceptual searching in legislative xml collections,” in Proceedings of the Workshop on Legal Ontologies and Artificial Intelligence Techniques, pp. 83-89 (2005).

---

Enrico Francesconi, Researcher, Italian National Research Council, Institute of Legal Information Theory and Techniques, via de' Barucci 20, 50127 Firenze, IT,francesconi@ittig.cnr.it ;www.ittig.cnr.it

---