Neural Network Method In Law KIS-L (Knowledge in Space-Law)

• Autore: Vladimir Vrecion, Pavel Vrecion
• Pagine: 37-47

Page 37

@1. The development of ESs in law

Very generally we can distinguish three basic types of possible ESs in the area of law:

1) ESs where the prevailing instrument of formallzation, modelling, algoritmization of legal thinking and/or reasoning is (relatively deep, sophisticated) logicai analyses, logicai formalization.

2) ESs based on a space stractures especially on. neural network methods, where information or knowledge is considered as elemento (points, knots) and their relations in space, where these elements, their positions and relations are permanently changing, These ESs can be in principia more dynamic than ESs sub 1/but they are modelling intellectual and/or information operations ´more roughlyª.

3) ESs for more special legally regulated procedures where izti prescribes some quantifications, quantitative evaluations especially in taz law (see the works of L.T. McCarty), farnily law - determination of allmony, etc. They use mainly ample arithrnetic algebraic procedures at algorithmzation there but mathematically nontrivial precedures can also be used (e.g. from the mathematical theory of graphs see [3]).

Ad 1/Now the time is coming to concentrate our endeavours on developing practical broadly usable ESs for some importatn parts of legal reasoning based onsystematic logical analysis (see especially [1, 2, 4] and the works of H. Yoshino, L. Philipps, E. Fameli and others). Such ESs will Page 38 always create the core or skeleton of sophisticated computerization in the area of law,

Ad 2/Such ESs can serve as aids especially for the creation of better legal empirical experience, for better associaticil of legal and empirical facts. In future they will be Important component of complex legal ESs especially for ´more globalª intellectual operations.

Ad 3/ESs of this type can aiready effectively serve in present practice. They can also be incorporated as a part of complex and/or combined legal ESs in the future.

¿t the present time in the whole area of AI there is an endeavour to combine the strengths and advantages of rule-based ESs (as sub 1) and neural network ESs. Neural network ESs need a lot of samples to shape (´learn, trainª) themselves and they work in a rather transparent but dynamic way. The rule-based ESs need a lot of expert work and the user can know a relatively unchangeable algorithm for the creation of answers,

Usually it is done through bridges which provide interfaces between the ES and neural network. Shells like Nexpert (Neuron Data), Guru (Micro Data Base Systems) enable linking in external routines, There are e.g. C libraries Eke Owl (Hyper Logic) and NeumoSoft (hnc), neural networks like Neural Works Professional II (Neural Ware), Neuro Smarts (Cognition Technology), Explore Net (hnc) available. Some of them can be converted to C source code and connected with an ES.

Now it is already possible to begin a more concrete constructive discussion about the general architecture of mixed and /or combined legal ES. We should like to take part in such discussion (in a separate article for this journal). The creation and experiences with ES of the type sub 2/, now described are needed for such a project (earlier we have worked in the areas'sub 1/, sub 3/and at the creation of ´standardª legal ISs).

@2. Theoretical bases of the ES KIS-L

Our information about certam areas of human activity generally represent a set of elemeetary pieces of information, knowledge and the relation among ilieni: we can cai it information space, in law we cari call it legal Information 5pace; where elements of information mainly are: legal norms and their parts, actual relevant facts and elements of their description and relations amosg them.

It is of course relative what we consider as element of information and/ or knowledge. It depends on the level, scope and aim of the intellectual Page 39 operation (thinking, consideration) in the framework of the given information space, The described representation of the information space of ES KIS-L respects this fact by corresponding dynamic strecturing of information elements and/or information space.

The basic principle of ES KIS-L is to represent information and/or information space in the 3 D net. 3 D can be determined by the definiticil of knots, their connections - edges and by the determination of rules according to which the net is shaped.

The knots represent the objects which consist of header, memo and of the list of connection to other knots. The header is a text information -term, specificaticil which denotes the raeaning of the knot. Memo is full-text placed on the hard disk, describing in more detail information in the knot e.g., legal norm, scientific explanation, etc.

The connections are oriented lines - edges defined by the start knot, end knot and a length. The length of distance between the knots i.e., the iength of the respective edge is a function of probabilit‡ (intensity) of relation...