17-03-2014, 08:18 PM
Abstract
In this paper, we discuss the problem of reasoning with
cases in a domain governed by rules, in particular, the
problems of interpreting the meaning of terms (statutory
predicates) used in the rules and combining case-based reasoning
(“CBR”) with other modes of reasoning, such as
rule-based (“RBR”) and model-based reasoning. Terms
used in statutes are typically underdefined in the statute
and inherently open-textured and thus require precedentbased
reasoning to interpret their meaning for particular
fact situations. In such domains one needs to combine reasoning
about the rules (statutes) with the precedents that
concern them.
We describe our precedent-based case-based reasoner
TAX-HYPO that operates in the statutory domain of tax
law. TAX-HYPO is a derivative system of our earlier CBR
system HYPO which operated in the common law domain
of t d d e secret law. We describe our system CABARET
which is an environment to support (1) building precedentbased
CBR systems like HYPO and TAX-HYPO; and (2)
experimentation with mixed paradigm systems involving
CBR.
1. Introduction
Many domains of interest to AI have the following characteristics:
There are rules, policies or codes setting forth a controlling
model of the domain or action in it. That is, the domain is
rule-governed.
These rules use terms whose meaning is unclear, for instance,
because the term is not adequately defined in the
rule set or because it can’t be or is not intended to be. That
is, the terms are underdefined or open-textured.
There is a body of cases about the past application of the
rules and interpretation of the terms used in them. That
is, there are cases.
The ambiguous terms must be interpreted by appeal to
precedent cases addressing past interpretations. That is,
the reasoning is precedent-based.
A new fact situation must be analyzed by appeal to both
the rules and the cases. That is, the reasoning involves
mixed paradigms.
Copyright 01988. Edwina L. Rissland and David B. Skalak. All rights
reserved. This work was supported (in part) by the Advanced Research
Projects Agency of the Department of Defense, monitored by the Office of
Naval Research under contract no. N00014-87-K-0238th,e Office of Naval
Research under a University Research Initiative Grant, contract no. N00014-
86-K-0764,a nd a grant from GTE Laboratories, Inc., Waltham, Mass.
Tax law is a typical example of such a domain. So is the law
involving the administration of such things as Social Security,
housing regulations, and employee benefits. Non-legal domains
sharing these characteristics are tactical planning, where there
are rules governing courses of action as well as past episodes of
action, and certain fields of design, where there are design methods
and constraints as well as past designs. While mathematics
has both rules (i.e., definitions and theorems) as well as cases
(i.e., examples and counter-examples), the terms used are as far
as one can get from open-textured since they are tightly defined.
(Of course, in a nascent or rapidly changing mathematical area,
things are not as “close-textured” as one might hope [Lakatos,
19761.) Thus in pure mathematical domains the problems of interpretation
of interest to us !me are minimal although the problem
of mixed paradigm reasoning is still central.
The problem of open- textured terms (“predicates”) is central
in legal reasoning, especially in statutory domains. Opentextured
statutory predicates often present hard problems of
statutory interpretation - hard for both attorneys and computer
programs purporting to do legal reasoning.
In this paper, we discuss the problem of reasoning with
cases in a domain governed by rules, in particular, the
problems of interpreting the meaning of terms (statutory
predicates) used in the rules and combining case-based reasoning
(“CBR”) with other modes of reasoning, such as
rule-based (“RBR”) and model-based reasoning. Terms
used in statutes are typically underdefined in the statute
and inherently open-textured and thus require precedentbased
reasoning to interpret their meaning for particular
fact situations. In such domains one needs to combine reasoning
about the rules (statutes) with the precedents that
concern them.
We describe our precedent-based case-based reasoner
TAX-HYPO that operates in the statutory domain of tax
law. TAX-HYPO is a derivative system of our earlier CBR
system HYPO which operated in the common law domain
of t d d e secret law. We describe our system CABARET
which is an environment to support (1) building precedentbased
CBR systems like HYPO and TAX-HYPO; and (2)
experimentation with mixed paradigm systems involving
CBR.
1. Introduction
Many domains of interest to AI have the following characteristics:
There are rules, policies or codes setting forth a controlling
model of the domain or action in it. That is, the domain is
rule-governed.
These rules use terms whose meaning is unclear, for instance,
because the term is not adequately defined in the
rule set or because it can’t be or is not intended to be. That
is, the terms are underdefined or open-textured.
There is a body of cases about the past application of the
rules and interpretation of the terms used in them. That
is, there are cases.
The ambiguous terms must be interpreted by appeal to
precedent cases addressing past interpretations. That is,
the reasoning is precedent-based.
A new fact situation must be analyzed by appeal to both
the rules and the cases. That is, the reasoning involves
mixed paradigms.
Copyright 01988. Edwina L. Rissland and David B. Skalak. All rights
reserved. This work was supported (in part) by the Advanced Research
Projects Agency of the Department of Defense, monitored by the Office of
Naval Research under contract no. N00014-87-K-0238th,e Office of Naval
Research under a University Research Initiative Grant, contract no. N00014-
86-K-0764,a nd a grant from GTE Laboratories, Inc., Waltham, Mass.
Tax law is a typical example of such a domain. So is the law
involving the administration of such things as Social Security,
housing regulations, and employee benefits. Non-legal domains
sharing these characteristics are tactical planning, where there
are rules governing courses of action as well as past episodes of
action, and certain fields of design, where there are design methods
and constraints as well as past designs. While mathematics
has both rules (i.e., definitions and theorems) as well as cases
(i.e., examples and counter-examples), the terms used are as far
as one can get from open-textured since they are tightly defined.
(Of course, in a nascent or rapidly changing mathematical area,
things are not as “close-textured” as one might hope [Lakatos,
19761.) Thus in pure mathematical domains the problems of interpretation
of interest to us !me are minimal although the problem
of mixed paradigm reasoning is still central.
The problem of open- textured terms (“predicates”) is central
in legal reasoning, especially in statutory domains. Opentextured
statutory predicates often present hard problems of
statutory interpretation - hard for both attorneys and computer
programs purporting to do legal reasoning.