NaviLex: Search and Navigation in a Semi-automatic Content Acquisition Legal Hypertext

• Pagine: 211-234

Page 211

@1. Introduction

The goal of most legal work - seen as a process of text handling - is actually to combine different types of texts in an effective way, Statutes, regulations, cases, precedents, legal literature, contracts are examples of documents that may have to be investigated together in order to solve a, legal problem or even to be able to understand the practical meaning of a legal rule. This well-known phenomenon can be described as legal rule fragmentation: the necessary information is often scattered in different documents or even in different data banks, and the links among the required pieces of information are difficult to establish. This problem, though particularly hard in a strongly text-centered field like the legal domain, is not limited to the legal area.

As a consequence, in the recent years a growing interest has been dedicated to the integration of search functions (typical of the Information Retrieval field) with navigation functions (typical of Hypertext systems). An effective combination of the benefits provided by the two models is expected to provide the best support tools for the localization of scattered information that is of interest for the user.

Among the different models that have been proposed in various fields, in the legal domain the interest for legal databases led to the adaptation for legal data [Agosti et al. 91] [Di Giorgi Nannucci 92] of a general two-level model [Agosti et al. 91a] providing a conceptual layer intended to improve the system transparency. Although a sound conceptual layer is a vital component of any effective model, we believe that a functional limitation of many existing models is the lack of context information, suitable to be combined with concepts to improve the retrieval accuracy (especially the precision component).

Besides this functional limitation, an economical problem which hampers the development of powerful models is the cost for the manual indexing/Page 212 authoring task. In the legal field the importance of the problem can be easily understood by considering the figures that express the rapid growth of the overall document collection: more than one million new cases and statutes per year, according to piafner 90].

The state-of-the-art Natural Language Techniques do not allow to envisage effective solutions to this problem on general domains. Nonetheless, in limited sub-languages the feasibility of an automatic mapping from texts to suitable information structures has been demonstrated in various fields [Liddy et al. 91] [Rama Srinivasan 93], In the legal domain the Nomos project [Graziadio et al. 92; Giannetti et al. 92; Pietrosanti et al. 93] has demonstrated the potential of Shallow' NPL techniques, as opposed to the traditional 'deep' techniques, for effective large-scale text processing in the legal domain.

Moving from the previous considerations, our work in the legal field has conformed to the following guidelines:

* Emphasis on the Modelling of context characterization of concepts, to improve precision;

* Use of semi-automatic tools for the acquisition of the legal information components;

* Definition of a friendly interactive environment, referring to a bock metaphor.

The resulting hypertext model is based on the close integration of the search and navigation dimensions, in a framework which offers multiple views (structural, conceptual and functional) over legal documents. The prototype system NaviLex has been implemented, in order to experiment the potential for advanced applications of the model, dealing with a significant excerpt of the Italian Banking Regulations.

The main focus of this paper is on the description (including an overview of the relevant linguistic tools derived from the Nomos project) of the functionalities of the prototype, along with the underlying retrieval model based on a 'legal electronic book' metaphor.

The structure of the paper is as follows. Section 2 is devoted to the reference model for the integration os search and navigation functions on legal domains. The structural, conceptual and contextual dimensions are presented, together with the fundamental search and navigation functions made available by the model. The presentation provides also (2.4) a detailed overview and discussion of related approaches and an outline (2.5) of the basic assumptions underlying our approach to 'shallow' Natural Language Processing, including a brief description of the NPL modules.

Page 213

The following section 3 describes the motivations and the functionalities of the NAviLex prototype, including aiso (3.2) an extensive demonstration of the functionalities by means of an example of a user session,

@2. Fundamental Components of Legal Documents; A Reference Model for the Structural, Conceptual and Functional Dimensions

The proposed reference model for the content representation of legal texts, is described in this section in terms of the multiple views (structural, conceptual and functional) provided over the legal texts. The presentation emphasizes the issues of the transparency of the model (relying on a "natural" structure for the typical user who is familiar with legal text-books) and of the incrementality of the search process.

The structural dimension of the content representation includes the hierarchical organization of legal documents and the network established by the legal cross-references, which can both be used for navigation, purposes allowing respectively direct access to a given text (through a table of contents) and a hypertext access through the ingoing or outgoing citations.

The conceptual dimension Is based on the definition of complex linguistic terms (namely noun phrases) which constitute more effective content descriptors (as those typicaly included in the index of a book) with respect to word-based indexing [Croft et al. 91] [Evans et al. 91].

Another crucial dimension of the content representation is the functional component, which allows to associate to the concepts specific functional roles which constitute meaningful contexts for the user (e.g. the definiendum of a definition, or the subject of an obligation),

@@2.1. The structural dimension

Legal databases exhibit a general structure characterized by the well-known subdivisions of legal texts (e.g. sections, articles and paragraphs in statutes; chapters, sections and sub-sections in regulations). In addition to constituting the basic text cohesion device, this hierarchical organization is an essential aspect of the structural dimension of legal documents, as it provides also an 'address space' that is extensively used to make explicit reference, in a given legal document, to related parts of the same or other documents. The links established by these references make a complex cross-referring web of documents, that constitute the second component of the structural dimension.

Page 214

Anticipating the description of the document base of the prototype (see the following section 3), Fig. 1 represents both components of the structural dimension.

The document base is composed of documents belonging to two related legal texts, whose hierarchical organizations, correspond in the figure to vertical tree-like structures. The arrows illustrate the cross-reference links arising from two excerpts of the left-side text. The legal sub-domain is that of the Banking Legislation, which includes the Decree (´Decretoª) represented in the right-side structure. The articulation is typical of Statutes (issued by the Parliament): the Decree is composed of numbered Articles (´Articoliª) organized in Titles (´Titoliª) that are in turn made of Sections (´Sezioniª), both indexed by roman numbers.

[ GRAPHICS ARE NOT INCLUDED ]

In order to specify and give the correct interpretation of the whole Banking Legislation - that is scattered in several Laws and Decrees - the national Bank of Italy (Banca d'Italia) issues the Regulations on Bank-Surveillance (´Istruzioni di Vigilanzaª). These Regulations are represented by the structure on the left hand side, that is articulated in Chapters (´Capitoliª), Sections and nested Paragraphs (for instance the upper text excerpt,Page 215 that is discussed in detail in the following section, belongs to the Paragraph 2.c of the first Section of Chapter L). Due to the status of the Regulations, in addition to the internal cross-references pointing to other sections of the Regulations book, many cross-references point the Decree, which contains the original legal rules referred to in the regulations. This is a typical example of the need to navigate the cross-reference links to cope with the mentioned problem of the legal rule fragmentation.

@@2.2. Providing a functional context for the conceptual dimension: modelling the norm-kernel and definitions as functional schemata

The investigation of legal concepts and their relations with other concepts is a crucial goal of much effort in legal work, aimed at identifying legally relevant items of knowledge and relevant relations between such items. The upper conceptual layer of the model (described in Fig. 2) represents the universe of possible usable terms and their relationships. For the purpose of our model, a concept represents a meaningful entity for the domain. Each concept is linked to the documents, in which concept instances are denoted by concepts anchors, that represent the linguistic manifestations of the concepts, In general, a set of possible linguistic expressions (i.e. the Concept Anchors present in the documents) denote a concept identified by a normalized linguistic term that represents the concept name (i.e. a noun phrase). For example, in the figure Ca-A1 (e.g. ´sales of goodsª) and Ca-A2 (e.g. ´to sell goodsª) are two anchors that identify two instances of the same concept Ca (e.g. ´sale of goodsª)

Relations are established among concepts that are semantically linked. In the figure a generic hierarchical structure is depicted, that can be...