Controlling and Augmenting Legal Inferencing: YSH, A Case Study
Proceedings of the 4th Int. Conf. on Artificial Intelligence and Law (Amsterdam 1993), ACM Press, New York, pp162-166
5 Pages Posted: 16 Jun 2017
Date Written: June 1, 1993
[This 1993 paper is the earliest papers to discuss the inclusion in the DataLex Project of the YSH (y-shell) rule-based inferencing component, and a quasi-natural language knowledge representation.]
If legal inferencing systems are to be used for immediate practical application, they are best constructed by embedding them in other technologies which can assist in augmenting and controlling the course of inferencing. Adoption of a (quasi) natural language knowledge representation assists easier development of user interpretative facilities, user control of the course of inferencing and explanation facilities. The paper explains how the DataLex Workstation Software, particularly its inference engine, ysh, implements these approaches.
There are two broad reasons for asserting that to develop legal inferencing systems in isolation from other technologies for representing and manipulating legal information is unlikely to yield systems of immediate practical application.
(i) Augmenting inferencing – The first reason arises from the inherently and endemically open textured nature of legislation and case law. The lack of fixed meanings of the language of cases and statutes means that it is not possible, even in theory, for an application developer to anticipate all possible factual circumstances which may come within the meaning of predicates used in system dialogues. Inferencing techniques to resolve unanticipated open texture problems are not yet commercially viable. A legal inferencing system must therefore constantly require the user to make significant interpretative decisions. This makes it necessary to give the user effective and open-ended access to a wide variety of textual interpretative materials. The overall goal is to build an inferencing system that most effectively supports the user’s interpretative activity, allowing control of a problem’s solution to alternate between a semi-expert system and a semi-expert interpretive agent, the user.
(ii) Controlling inferencing – The second, related, problem is that the depth or extent to which a user will want, or need, to use aspects of a legal inferencing system to solve problems will vary. This may depend upon either the extent of the user’s domain expertise, the facts of the problem at hand, or both. Effective resort to textual materials will be far more efficient than running an inferencing session. If users are always forced to pursue tedious inferencing dialogues to the bitter end they are unlikely to find inferencing systems efficient, attractive to use, or ‘intelligent’. This second problem, user control of depth of inferencing, is an instance of a broader principle that, for a knowledge-based system to be of the greatest practical utility, the user needs to be able to exercise various types of control over its operation.
This paper goes on to demonstrate how these problems are resolved in the DataLex Workstation software and applications, using a quasi-natural language knowledge representation and the YSH rule-based inferencing shell combined with hypertext and text retrieval software.
Keywords: DataLex, AI & Law, artificial intelligence, expert systems, Australia, hypertext, text retrieval
Suggested Citation: Suggested Citation