Computational Models of Large Scale Systems of Legai Rules

• Autore: B, Intrigila/M. Cianci/P. Damiani/P. Di Salvatore/A. Centanni
• Pagine: 7-35

Page 7

B, Intrigila, M. Cianci, P. Damiani Dipartimento di Matematica Pura ed Applicata - Università di L'Aquila - Via Vetoic - 67010 Coppito (L'Aquila) - Italy.

P. Di Salvatore Istituto di Teoria dell'interpretazione ed Informatica Giuridica - Università «La Sapienza», Roma,

A. Centanni Ministero di Grazia e Giustizia: Direzione Generale per l'Organizzazione Giudiziaria,

@Introduction

In the present paper, we consider computational models of systems of legai rules, that is «rules» (suitably represented) which are, in a sense, the abstraction of existing legai rules, resembling them in many relevant aspects5 e.g. in shape (see later, Section 2), The exposition is organized as follows.

Sectìon 1 is devoted to give the main motivations for constructieg such kind of models, and to expose the generai setting in which this task is here accomplished.

In Section 2, we discuss some requirements on formai models of legai rules. Such requirements come from the judìciai theory, which points out the formai structure of norms, and also from the computer science needs,Page 8 since we are looking for computationaliy treatable models, which determines a suitable representation of legai rules.

Sectìon 3 illustrates the use of Puzzy Logic in our computational model; moreover, to facilitate the unexperienced reader, we have included some basic notions concerning the use of Fuzzy Logic in control systems. TMs Section ends with some worked examples.

Section. 4 describes the future work to be done to achieve the task of building up complete computational models, and in particular the generai tool we are setting up to automatically generate the kernel of a coraputational System of rules.

@1. The need for Computational Models

Here we consider legai systems in the most wide sensef that is any infonnation System, composed by rules, on which base some specific case gives rise to some legal decision. So, we are adopting a Normative Model, according to the classification of [GS1] (Chapter III.1). By a legai decision, we mean the assignment of some deontic value; we prefer here to relay on the intuitive meaning of «legai decision» rather than commit ourselves to any specific formai System of deontic logic.

@@1.1. Interaction Problems

Our starting point is the fact that interaction with existing legai systems is a difficult task [ht]. Moreover, the construetion of computer-based tools for significaetly improve such interactions is difficult top.

There ìs a generai agreement on the need of a better understanding of sueh difficulties. To be more concrete, let us enlist some of them, without any claim to be exhaustive or complete:

(i) at a logicai level we find that it is eot clear how far the existing formai systems of (deontic) logic are able to correctly represent or simulate the logicai inference mechanisnis of actual legai decisions; in saying this, we are not referring to paradoxes of deontic logic (which are, in any case, of great importance), rather, to practical problems in raodelling valid legai inferences [GS1, GS2, McC, DT, AV];

(ii) a related problem is the representation of legai rules, in this respect classical logic seems not to be suitable as a language to formally represent legai knowledge, since we have often to encode incomplete, uncertain or even contradictory information; classical logic (and therefore logic prò-Page 9 gramming) needs to eliminate ueceitainty, so making a sequence of doubtful choices, at least enforcing sharp interpretations and discarding other acceptable ones (e.g. [CB]);

(iii) the elicitation of legai rules silice actual rules are embedded in texts, written in naturai languages, in a much more complex way than, say, mathematical propositions are embedded in mathematical books or articles; moreover according to [GS2] legai languagef though partially technical, borrows the termifiology of the field it intendi to regolate. Therefore to accomplish the elicitation task we must often refer to common sense, and this is not at ali a mechanical step. «In this respect legislation can be viewed as programs expressed in human language to be executed by humans rather than computer» [K].

(iv) moreover we find the interpretation problems, since, even if we assume that the extraction of the rales from the texts has beee made in successful way, we are faced with ambiguities not depending on naturai language, but related to the very structure of the niles themselves («Like ambiguity, vagueness is also a form of imprecision in the law, but one that is usually intended. À draftsman will often make a conscious decision to leave some concepts undefined and vague, because he prefers that meaning of the concepts should be determmed later in the context of real cases when all the individual circumstances can be given proper consideration. À draftsman could never anticipate ali the possible combinations of circumstance that might arise in the future and make explicit provision for them. Vagueness in the law is essential. It enables the law to adapt to unanticipated circumstances and to adjust gracefully to changing needs» [KS]);

(v) as in most cases, such ambiguities are to be solved referring to other legai rules, we are led to the complex System of relationships betweee legai rules, which includes crossreferences, hierarchical and temporal relations, etc. [PDGMM];

(vi) as far as we want to apply rules to spedfic cases, we have analogous problems: case representation formalisms, relevance of a rule to a specific case, finding all relevant rules, etc.

@@1.2. Computer-based Interaction Tools

As previously stated, in the construction of such tools a number of problems have not received a completely satisfactory solution.

We can distinguish, in order of difficulty, the following applcation fields:

- Information Retrieval [ht];

- Computer Àided Scardi and Navigation in Legai Databases [pgdmm]; Page 10

- Computer Alded Decision Making;

- Legai Drafting [bms];

- Expert Systems [av].

We limit ourselves to discuss this last item, where ali problems are to be faced.

First of al, we remark that an enormous work has be done on constinering formai logicai systems, to represent, at a more or less high level of abstraction, the inferential legai mechanisms; also ranch work has been done ie representing legai knowledge as well as the conditions of applicability of legai norms or rules to specific faets or actions if we have:

Representation of Knowledge

Representation of Facts

Inferential Mechanism

then, as it is well known, we are in position to obtain an experi System, that is a Knowledge Data Base, which acts on (representations of) Facts, via the Inferential Mechanism.

Despite some successes5 it is commonly recogniEed that the reaHzation of a completely satisfactory Expert System in some law field is a tremendously difficult task, which can be partially accomplished only in very restricted domains. The reason of this difficulty can be found in the mentioned phenomena, which in an expert System perspective act as foflows. Considering, for instance, the Knowledge Elicitation process, we have to put the legai rules in the chosen representation System, mainly extracting them from texts. Here we have the following dilemma:

(a) an automatic treatment of texts to this aim? is out of the scope of actual Naturai Language Processing techniques;

(b) on the other hand, if we use human experts to do the job, we then find two main problems:

- high costs for lttle benefits, since legai mles are rapidly changing «U.S, case law comprises roughly 50 gigabytes of text and grows by 2 gigabytes per year» [ht];

- high risks to commit the System with some subjective inierpretattons of the law.

@@1,3. Mathematical and Statistkal Models

While such difficulties are well known, what seeras to be stili lacking is a suitable modelling of such phenomena. Put in different words, we wouldPage 11 like to have formally precise models which represent the various forms of interaction between legai systems and the users (of different kinds) of these systems,

To this aim, we may ask what Mnd of appraach Is sultable to improve our comprehension:

(A) the theoretical, i.e. mathematical, approach;

(B) the empirical, Le. statistical, approach;

(C) the computational, Le. computer simulation, approach.

Às an example in the Information Retrieval field [smc], the construction of more effective tools to interface text databases strongly relies on mathematical and statistical modelling of phenomena such as occurrences of words or combinations of words in texts. So, we can often give a proper assessment of what happens when a database user formulates a query of such a forni. This corresponds to the abovementioned approach A,

Silice this it not completely reliable, we need an evaluation of empirical performance of existing ir tools on colections of texts or documenta specifically built up as test benches, as the trec Test Collection.[TREC] This corresponds to approach B.

Coming back to the Legai Field, the following considerations are in order.

(A) It is well known that there are not reliable models of the abovementioned interactions between users and legai systems,

(B) Moreover, it seems not convenient to cope with-such problems via extensive statistical investigations.

In fact:

(1) it is likely that any conclusion obtained by investigating a specific field of law, cannot be applied to other fields without introducing (large) interpretations errors of (desperately) difficult assessment;

(2) a pool of experts is required to evaluate the interactions users-system, and the results will be strongly biased by the choices of experts;

(3) the enormous number of relevant norms requires a very large (and consequently very expensive) sampling; and in many cases a reliable sampling would be exceeding large;

(4) the legai knowledge quickiy changes as new laws, statutes, regulations supersede the previous ones. Therefore it is very difficult to evaluate how far our resuits will...