The
role of context in case-based legal reasoning:
teleological,
temporal, and procedural
CAROLE
D. HAFNER and DONALD H. BERMAN
Northeastern
University, Boston MA 02115, USA
E-mail:
hafner@ccs.neu.edu
Abstract. Computational models of
relevance in case-based legal reasoning have traditionally been
based on algorithms for
comparing the facts and substantive legal issues of a prior case to those
of a new case. In this paper we
argue that robust models of case-based legal reasoning must also
consider the broader social and
jurisprudential context in which legal precedents are decided. We
analyze three aspects of legal
context: the teleological
relations that
connect legal precedents to the
social values and policies they
serve, the temporal
relations between
prior and subsequent cases in a
legal domain, and the procedural posture of legal cases, which defines
the scope of their precedential
relevance. Using real examples
drawn from appellate courts of New York and Massachusetts, we
show with the courts own
arguments that the doctrine of stare decisis (i.e., similar facts should lead
to similar results) is subject
to contextual constraints and influences. For each of the three aspects of
legal context, we outline an
expanded computational framework for case-based legal reasoning that
encompasses the reasoning of
the examples, and provides a foundation for generating a more robust
set of legal arguments.
1. Introduction
Prediction is a fundamental
component of legal reasoning, as the legal realists
demonstrated (Holmes 1897) and
as practicing attorneys well understand. When
an attorney drafts a will for
a client who has children but who wishes to leave
all of his property to his
wife, the attorney predicts that a court will enforce the
clients written intention.
When advising a client on a tax deduction, the attorney
predicts how a tax court will
interpret the relevant statutes and regulations. In
advising a client whether to
settle a case or accept a plea bargain, the attorney
predicts whether a trial will
result in a decision more favorable than the deal that
lies on the bargaining table.
From the standpoint of case-based legal reasoning,
when deciding which cases to
cite in a legal argument before a judge, the attorney
makes a prediction whether the
court is likely to follow a prior case or not.
Research in case-based legal
reasoning has led to significant advances in modelling
the way lawyers use previously
decided cases. Some models generate legal
arguments that to some degree
simulate human adversarial discourse (Ashley 1990;
Skalak and Rissland 1992;
Aleven and Ashley 1994). Others represent case-based
techniques for interpreting
open-textured statutory predicates (Gardner 1987; Mc-
20 CAROLE D. HAFNER AND DONALD H.
BERMAN
Carty and Sridharan 1981;
Sanders 1994; Rissland and Skalak 1991; Branting
1991). Taken as a whole, this
body of research models the process of adversarial,
precedent-based legal
reasoning and argumentation with an elegance and precision
not found in any comparable
legal text.
However, the ability of most
of these models to simulate the reasoning of practicing
attorneys is limited by their
exclusive focus on the factual similarity between a
prior case and a new case.
Real-world legal precedents are embedded in a complex
jurisprudential context that
includes the level and jurisdiction of the prior court
(Berman 1989), the procedural
posture of the prior case (i.e., what formal claim
or motion was before the
court), and the influence of dissenting opinions. Legal
precedents are also embedded
in a political context, where competing policies
and values are balanced by the
courts, and where legal doctrines evolve to accommodate
new social and economic
realities. Contextual factors such as these
are considered by skillful
attorneys when they make predictions about whether a
precedent is likely to be
followed, and whether a legal argument is likely to prevail.
As a result, computational models
that do not take account of context fall short of a
robust analysis of the
case-based legal reasoning performed by a practicing lawyer.
This paper, in which we
consider the influence of the larger jurisprudential
and social context on
precedent-based legal reasoning and argumentation, includes
research presented at three
successive meetings of the International Conference
on Artificial Intelligence and
Law (Berman and Hafner 1991, 1993, 1995).1 We
demonstrate that comparing the
facts and issues of a new case with those of prior
cases to determine which cases
are the most on point, while essential, is not a
sufficient basis for robust
models of legal case-based reasoning. We analyze three
aspects of legal context: the teleological structure of a legal domain (Section 3),
the temporal relationships between prior and subsequent
cases in a legal domain
(Section 4), and the procedural posture of the prior case (Section 5).
Using real
examples drawn from appellate
courts of New York and Massachusetts, we show
with the courts own arguments
that the doctrine of stare
decisis (i.e.,
similar facts
should lead to similar
results) is subject to contextual constraints and influences.
For each of the three types of
context, we propose an expanded computational
framework for case-based legal
reasoning that encompasses the reasoning of the
examples.
By incorporating context into
case-based legal reasoning, we are able to explain
several important phenomena in
case-based legal reasoning that have not been
captured by earlier studies:
the use of policy arguments to choose between
competing on-point precedents.
the ability to recognize when a factually
on-point precedent is procedurally
inapt.
the ability to recognize when the most on-point
precedent is unlikely to be
followed because more recent
decisions weaken its authority.
The representation framework
developed in this paper also points the way toward a
model of legal reasoning
capable of recognizing and exploiting analogies between
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 21
different legal domains, both
factual (such as duck hunting and education) and
doctrinal (such as tort and
contract law).
2. Models of case-based legal
reasoning
The formalization of law and
legal reasoning is not a new enterprise, but rather
one that has challenged
philosophers of law throughout the 20th century. Wesley
Hohfeld (1913), a legal
positivist, proposed that all legal states of affairs can be
described in terms of four
fundamental relations: rights, duties, powers and liabilities
(with inverse relations:
no-rights, privileges, disabilities and immunities).
In contrast, legal realists
such as Jerome Frank (1948) argued that law cannot be
characterized as a formal
system, because the unlimited variety of fact patterns in
human affairs make legal
decision-making inherently unpredictable. The problem
of open-textured concepts
(i.e., legal concepts that do not have clear definitions to
determine their applicability,
but which depend on experience and common sense,
such as the concept of
recklessness) is the focus of a famous debate between H. L.
A. Hart and Lon Fuller (Hart
1958; Fuller 1958), in which a legal rule stating No
vehicles are permitted in the
park is used to demonstrate that even an apparently
clear concept such as
vehicle can be legally indeterminate, as in the case where
a veterans group wants to
drive a tank through the park as part of a Memorial
Day parade. Fuller explained
the difficulty of creating fixed definitions for legal
concepts by observing that the
terms in a legal rule cannot be interpreted without
taking into account the
purposes the rule is intended to serve.
With the advent of computers
and artificial intelligence (AI), both the range
of techniques available and
the practical motivation for creating legal reasoning
models have grown. In
particular, researchers have made significant advances in
modeling the way lawyers
classify, analyze and argue with cases, using a variety
of computational approaches
including frames, transition nets, semantic networks,
discrimination trees, and
neural networks (Ashley 1990; Aleven and Ashley 1994;
Branting 1991, 1994; Gardner
1987; Hafner 1981, 1987; Rissland et al. 1996; Rose
1994; Rissland and Skalak
1991). The legal reasoning model employed in this
paper is based on a
computational framework (originating in the HYPO project
described below) that adapts
and extends general AI techniques for case-based
reasoning to the domain of
legal precedents. In this section, we briefly review
the HYPO framework, focusing
on the use of a case knowledge base to support
the adversarial process of
legal argumentation. Other frameworks for legal argumentation
based on formal logic (Prakken
and Sartor 1997; Gordon 1994) and
on explicit models of legal
discourse (Hage et al. 1994; Loui and Norman 1995)
are not reviewed here,
although our analysis of the influence of context on legal
decisions could equally well apply
to those frameworks.
The field of case-based
reasoning in artificial intelligence (AI) is concerned with
the use of remembered prior
experiences in analyzing new situations and solving
problems (Leake 1996). The
distinguishing characteristic of case-based reasoning
22 CAROLE D. HAFNER AND DONALD H.
BERMAN
is that prior experiences are
stored as distinct histories or cases, in contrast to
other AI reasoning frameworks
where knowledge (however acquired) is integrated
into a unified structure of
logical rules, frames, neural nets, or combinations of
these. Thus, case-based
reasoning uses knowledge that is more contextualized than
that used by other reasoning
models, and relies more on powerful and effective
memory mechanisms and less on
powerful and effective inference algorithms
(Riesbeck and Schank 1989).
Research in case-based legal
reasoning, as practiced in the field of AI and law,
has a more restricted meaning:
formal judicial decisions (that is, legal cases) play
the role of remembered
experiences, and the problem-solving goal is the construction
of precedent-based adversarial
legal arguments.2 The thrust of this research
has been aimed at creating
models of relevance based on the similarity between the
facts and legal issues of a
prior case and those of a new case. In these models, a
set of domain-dependent,
legally-relevant factors is defined, and each prior case is
considered more or less on
point depending on the similarity between its factors
and those of the new case.3 Implicitly, such models embody
the following prediction:
the more similar the factors
of a prior case are to those of a new case, the
stronger the prior cases
precedential value
that is, the more likely it is that a
court will follow it in
deciding the new case.
Computational models of
case-based legal reasoning are concerned with the
following problems:
i. Representation of cases,
including what information should be included (facts,
issues, holdings, results, and
justification) and what formalism(s) should be
used.
ii. Organization of the case
base (including indexing attributes and index structure)
and retrieval of cases that
are potentially relevant to a new case.
iii. Similarity ranking of the
retrieved cases to decide which are the most relevant
to the new case.
iv. Creating a legal argument
for each side, using the most relevant cases retrieved.
One of the earliest AI systems
to represent legal cases, the Legal Research System
(LRS) (Hafner 1981) was a
retrieval system for cases in negotiable instruments law.
LRS used a framelike
slot-filler structure to encode the main elements of cases:
Plaintiff, Defendant, Cause of
Action, Facts, Hypotheticals, Legal-Criteria, Legal-
Effects, and Holdings. The
fillers of the slots were symbolic descriptions of legal
concepts: those concepts that
appeared in the last three slots comprised the legal
issues of the case. The
justification (or ratio
decidendi) of
the decision given by
the court was encoded as part
of the Holdings slot. The LRS system retrieved cases
with matching facts, issues,
and ratios based on queries rather than a
description
of a new case; it did not
perform similarity ranking, nor did it construct legal
arguments.
The first complete case-based
legal reasoning model, one that addressed all four
of the problems above, was
embodied in HYPO (Rissland and Ashley 1987; Ashley
1990),4 a computer program for
analyzing cases and constructing legal arguments
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 23
in the domain of trade secrets
law. HYPO used a multi-level description of cases:
facts and factual predicates represented the underlying
circumstances of a case,
and a library of dimensions (selected by the systems
designers, based on a survey
of relevant legal authorities)
were used to index and retrieve cases.
A dimension in the HYPO model
is a domain-specific factor, representing a
fact pattern that influences
the outcome of cases in the domain (Rissland et al.
1984). Dimensions form an
intermediate level of abstraction, between the facts of
a case (defendant hired an
employee who previously worked for plaintiff), and
the top-level legal issues
(defendant misappropriated plaintiffs trade secrets)
whose resolution determines
which side wins. Dimensions in HYPO can be binary,
multi-valued, or quantitative;
for example, in the trade secrets domain a
binary factor is Generally-Known, indicating that the alleged
secret information
is generally known within the
industry, and a quantitative factor is Secrets-
Voluntarily-Disclosed, representing the number of
people to whom the plaintiff
told the information.
HYPOs legal domain knowledge
includes, for each dimension, which side of
a dispute would be favored by
the factors having particular values (e.g., a true
value for binary dimensions,
or a larger value for quantitative dimension), and how
the factors value is related
to a cases underlying facts. For example, in the trade
secrets domain, a true value for Generally-Known favors the defendant; a false
value favors the plaintiff.
When a new case is presented
to HYPO, the dimensions applicable to the new
case are computed. Then, HYPO
constructs a claim lattice a similarity network
that has the new case as its
root node, and prior cases that share a maximal set
of dimensions with the new
case as its immediate successors. These immediate
successors are called the most-on-point cases. The claim lattice would be
further
populated with other on-point
cases (those sharing at least one factor-value with
the new case), ordered by the
subset relation (i.e., if the set of shared dimensions
of one prior case properly
includes the set of shared dimensions for another prior
case, then the first case is
more on point and is ranked higher than the second
case.)
Figure 1 illustrates an
imaginary HYPO-style knowledge base; Figure 2 illustrates
a claim lattice for a
hypothetical new case Smith v. Jones. There are three
most-on-point precedents, each
sharing two dimensions with the new case. Precedents
whose shared dimensions with
the new case are a proper subset of another
precedent are its successors
in the claim lattice. Since Smith v. Jones has both
factors favoring the plaintiff
(F1 and F6) and factors favoring the defendant (F3
and F5), it represents a
typical hard case in which legal arguments can be made
for either side.5
After building the claim
lattice to determine the most-on-point precedents and
the best cases for each side,
HYPO constructs a 3-ply argument for each side in
the new case. Letting Side1
represent the party on whose behalf HYPO is arguing,
first a Point for Side 1 is produced making a claim
that Side1 should win, next a
24 CAROLE D. HAFNER AND DONALD H.
BERMAN
Figure 1. A HYPO-style case knowledge
base for a legal domain.
Figure 2. Claim lattice for Smith v.
Jones.
Response for Side2 is produced, countering the
claim, and then a Rebuttal
for Side
1 is produced to answer the
response. To create the initial claim for Side1, HYPO
identifies the most on point
case(s) whose result favored Side1, and justifies the
claim by noting the similarity
between the new case and the most on point cases.
Responses and rebuttals use
the legal argument techniques of distinguishing and
presenting counter-examples.
(Ashley 1990, p. 71).
For example, in Smith v. Jones
the most on point cases for the plaintiff are
Arthur v. Beale and Isaac v.
Jackson. The most-on-point cases for the defendant
are Carson v. DeWitt and Grantham
v. Hodge. The following 3-ply argument could
be generated from this claim
lattice:
Point for Plaintiff as Side1:
Where F1 and F3 are present,
plaintiff should win its claim. Cite: Arthur v.
Beale. Also, where F6 is
present, plaintiff should win. Cite: Isaac v. Jackson.
Response for Defendant as Side2:
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 25
Arthur v. Beale is
distinguishable because: in Arthur v. Beale F5 was not
present. Cite: Grantham v.
Hodge, which held for Defendant when F5 was
present.
Carson v. DeWitt is a
counter-example to Isaac, since it held for defendant even
though F6 was present.
Therefore, defendant should
win.
Rebuttal for Plaintiff as Side1:
Grantham v. Hodge and Carson
v. DeWitt are distinguishable, since they
include the factor F4, which
is not present in Smith.
In addition to the
capabilities summarized above, the HYPO system included
algorithms for constructing
hypothetical cases and making arguments using
hypotheticals.6 The theory of arguing with
hypotheticals (Rissland 1985) was
based on identifying
dimensions that could be manipulated (often by changing just
one aspect of the fact
situation) to create a hypothetical case that could be used in
a legal argument.
Later systems extended the use
of HYPO-style factors in several ways. In the
GREBE system, Branting (1991)
developed a hierarchical representation of the
ratio decidendi of cases, which was used to
apply portions of a prior cases holdings
to a new case, even when other
portions of the prior case were not on-point. Brantings
model addressed a well-known
weakness of the pure factor-based approach
to case-based reasoning: its
failure to represent the reasoning that connects the
factors with the results of a
case. The logic underlying a legal decision is implicitly
encoded in GREBE by grouping
the factors of a precedent according to the
holdings they influence. In
the CABARET system, Rissland and Skalak (1991)
described a mixed strategy for
legal argumentation using both rules and cases.
Dimension-based analysis of
precedents is one of the methods CABARET employs
to guide its selection of
argument moves: the broadening of a rules interpretation,
the discrediting of a rule, or
some other strategy.
Legal reasoning models that
combine rules and cases include (Gardner 1987), in
which cases were used to
handle the problem of open-textured concepts legal concepts
that do not have clear
definitions to determine their applicability, but which
depend on experience and
common sense, such as the concept of recklessness.
Gardner used a decision tree
to model the application of legal rules to a case,
with the leaves representing
either factual predicates that could be observed, or
open-textured concepts. The
open-textured leaves were linked to cases comprising
positive and negative
examples. Other approaches to combining rules and cases
include Groendijk and Oskamp
(1993) and Vossos et al. (1991). The issue-case
discrimination trees described
by Hafner (1987) combine legal rules with cases in
a conceptual retrieval system.
The use of fact-based
dimensions to evaluate the relevance of precedents, as
exemplified by HYPO, GREBE,
CABARET and other systems (Cuthill and Mc-
26 CAROLE D. HAFNER AND DONALD H.
BERMAN
Cartney 1994) appears to be
well-suited to modeling legal reasoning in common
law jurisdictions, since the
rule of stare
decisis applies
in these jurisdictions. However,
despite the important role of
fact similarity, the application of stare decisis,
as understood by skillful
attorneys and applied by the courts, is both more complex
and less absolute than can be
accurately described using fact similarity alone.
3. Teleological context:
Modeling the purposes of legal concepts
In this section, we examine
the teleological aspects of case-based legal reasoning:
the purposes and policies that
often determine which precedents are found to be
most relevant, and which of
several competing precedents will be followed. While
the need for teleological
structure in rule-based systems has previously been noted
by Smith and Deedman (1987),
their deep structure model of purposes did not
stipulate any necessary
correspondence to the doctrines or concepts articulated in
judicial opinions (op. cit.,
p. 87). In contrast, below we propose a model of legal
purposes in case-based legal
reasoning systems that is based on the teleological
arguments actually made (or
likely to be made) by skilled advocates. Thus, we
model teleological context in
order to provide support for legal argument moves,
as described in Rissland
(1985).
3.1. THE ROLE OF TELEOLOGY IN CASE-BASED LEGAL REASONING
From their first day at U.S.
law schools, students learn that the relevance of facts
cannot be divorced from the
purposes behind legal rules. Consider the following
three cases, which are studied
by many American law students. In Pierson v. Post7
the plaintiff was fox hunting
on open public land. While the plaintiff with horse and
hound pursued the fox, the
defendant well knowing the fox was so hunted . . . did,
in the sight of Post, to
prevent his catching the same, kill and carry it off. The
Court ruled that in order to
recover, the plaintiff had to have gained possession of
the fox. In ruling that the
plaintiff had not gained possession, which would require
either capturing the fox or
mortally wounding it, the Court explained:
[We so hold] for the sake of
certainty, and preserving peace and order in
society. If the first seeing,
starting, or pursuing such animals . . . should afford
the basis of actions . . .
itwould prove a fertile source of quarrels and litigation.
The Court articulated a vision
that, if they did not establish strict guidelines for
obtaining title to wild
animals, they might be swamped with disgruntled hunters
arguing over who first saw the
animal.
A dissenting judge, in an
effort to justify a contrary result, argued:
When we reflect . . . that the
interest of our husbandmen, the most useful of
men in any community, will be
advanced by the destruction of a beast so
pernicious and incorrigible,
we cannot err in saying that a pursuit like the
present . . . confers . . . a
right to the object of it.
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 27
The dissenter had the vision
of many additional foxes ravaging local farms unless
hunters are protected in their
pursuits.
Since factual characterization
takes place within a broad philosophical and jurisprudential
context, judges may not adhere
to a uniform definition of possession in
all wild animal cases. Given
the dissenters justification, he might have agreed with
the majoritys result had the
hunter been pursuing a quail rather than a fox. Similarly,
the majority might have
reached a contrary decision were violence among
sportsmen a frequent result of
such confrontational behavior.
Students next read Keeble v. Hickeringill.8 There the plaintiff owned a
duck
pond upon which he placed duck
decoys. The defendant, intending to injure the
plaintiffs livelihood, used
guns to scare away the ducks. Even though the plaintiff
had neither wounded nor
captured the ducks, the Court found for the plaintiff by
reasoning,
[W]here a violent or malicious
act is done to a mans occupation, profession
or way of getting a
livelihood, there an action lies in all cases. But if a man
doth him damage by using the
same employment; as if Mr. Hickeringill had
set up another decoy on his
own ground near the plaintiffs, and that had
spoiled the custom of the
plaintiff, no action would lie . . . This is like the case
of 11 H. 4,47. One
schoolmaster sets up a new school to the damage of an
antient school, thereby the
scholars are allured from the old school to come to
his new. (The action was held
there not to lie.) But suppose Mr. Hickeringill
should lie in the way with his
guns, and fright the boy from going to school
. . . sure that schoolmaster
might have action . . .
***
And when . . . decoy[s] have
been used . . . in order to be taken for profit
of the owner of the pond . . .
and whereby the markets of the nation may be
furnished; there is great
reason to give encouragement thereunto . . .
Once the purpose of the rule
is understood, analogous cases setting forth the right
of school-masters become more
relevant than cases dealing with foxes.
In dealing with these cases,
law students also develop the ability to isolate and
match the relevant factors.
Obvious similarities exist between Pierson and Keeble:
both cases involved unfettered
wild animals, and both involved defendants motivated
by malice. However, students
easily distinguish Pierson
from Keeble on
the
grounds that the plaintiff in Pierson was hunting the fox on open
land while the
plaintiff in Keeble set out the decoys on his own
land, and the plaintiff in Pierson
was hunting for sport while
the plaintiff in Keeble
was pursuing
his livelihood.
The importance of this type of
symbolic matching becomes evident when the
students move to an
examination of Young
v. Hitchens.9 In Young, the plaintiff,
a commercial fisherman, spread
a net of 140 fathoms in length across a portion
of open ocean. After the
plaintiff had closed the net to a space of a few fathoms,
the defendant went through the
opening and spread its net and caught the fish.
Students, focusing on the
plaintiffs need to make a living, match to Keeble, which
would portend a plaintiffs
victory. Students also map to the open land in Pierson
28 CAROLE D. HAFNER AND DONALD H.
BERMAN
Figure 3. Factors and claim lattice in
wild animal domain.
which suggests that the
defendant should prevail. In Young, the court found for the
defendant.
Figure 3 shows the relevant
factors in the wild animal domain, and the relationship
among the three cases based on
those factors, following the HYPO framework
described in Section 2. This
data structure can be used to generate a 3-ply argument
of the type described in
Ashley (1990, pp. 7071):
Argument I.
Point for Defendant
Where: Plaintiff was seeking
capture of a wild animal and did not mortally
wound or capture the animal,
the defendant should win a claim for trespass.
Cite: Pierson v. Post. The open land in Pierson is like the open sea in the
current case.
Response for Plaintiff
Pierson is distinguishable because in Pierson the plaintiff did not make his
living from capture of wild
animals.
Counterexample:
Keeble v. Hickeringill held for the plaintiff even
though the plaintiff was
seeking to capture wild
animals, and did not mortally wound or capture the
animals.
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 29
Rebuttal for Defendant
Keeble is distinguishable because in Keeble the plaintiff was conducting
his
business on his own land.
Keeble is also distinguishable
because in Keeble
the defendant
was acting
maliciously while here the
defendant was engaged in business competition.
Though accurately transcribing
many law school class discussions, the model is
incomplete because at some
point the student must answer the question which
precedent should govern and
why? Even though attorneys wield their adversarial
foils with considerable
dexterity, judges, when studying legal arguments, do not
act as referees at an
intellectual fencing match. Rather, they make rules that significantly
affect human lives and their
decisions necessarily embody their views
(or prejudices) as to which
rules improve the quality of life in society. Therefore,
lawyers, in addition to
arguments based on factors, must suggest to judges persuasive
policy arguments for choosing
the side they represent. Here are some of the
teleological (i.e., policy)
arguments that the court probably considered in Young,
reconstructed from the
reported opinion10
Argument II:
Point for Defendant
You should apply the rule of Pierson because the uncertainty about
what constitutes
property rights to fish
swimming in the open sea will cause endless
controversy.
Response for Plaintiff
You should apply the rule of Keeble because it is important that
people earn
a living and the defendant has
interfered with the plaintiffs ability to make a
living.
Rebuttal for Defendant
Keeble is distinguishable
because the defendant in Keeble was acting maliciously,
while here the defendant was
merely engaged in vigorous competition
like the schoolmaster who sets
up a competing school. Society benefits from
such competition.
Surrebuttal for Plaintiff
Society permits vigorous
competition but not unfair competition. The actions
of the defendant will force
the plaintiff to take wasteful actions to protect his
catches when the defendant
could be pursuing other fish, which would increase
the amount of fish available
to consumers.
30 CAROLE D. HAFNER AND DONALD H.
BERMAN
Surrebuttal for Defendant
Guidelines for determining
whether competition among fishermen is fair or unfair
should be left to the
legislature, otherwise there could be endless lawsuits
attempting to establish what
fishermen may or may not do.
Lord Denmans opinion in Young
indicates that he was moved by the plaintiffs
argument, but denied relief because
of a procedural error.11
Judge Pattesons
concurring opinion suggests
that he was influenced by the thrust of defendants
surrebuttal. It would appear
that Judge Wightmans concurring opinion accepted
the endless quarrels
rationale.
3.2. A COMPUTATIONAL MODEL OF
TELEOLOGICAL CONTEXT
To develop an adequate model
of legal purposes and their role in legal reasoning
is one of the long-term goals
of the AI and Law field; in this paper, we take a
small step in that direction.
We show how to make a case-based reasoning system
a little smarter by attaching
some information about the purposes involved in its
case knowledge base. To do
this, we consider what these purposes are, how they
are related to each other in
the domain of interest, and how they can be used within
the framework of case-based
legal reasoning and argument.
Discovering the relevant legal
purposes in a domain is the goal (along with finding
the relevant authorities and
identifying the relevant factors) of a skillful legal
researcher. By reading
judicial opinions and consulting appropriate commentary, it
is possible (although
sometimes difficult) to understand what purposes the courts
are trying to advance. Our
approach to discovering purposes in a legal domain will
be the same one used to
discover factors or dimensions in the HYPO-style systems
described in Section 2. That
is, a knowledge engineering approach is used, where
the human expert determines
what concepts to include in a domain model. For
the wild animal examples, the
purposes we will discuss are those mentioned in
Argument II above:
to define possession in a way that promotes
certainty and discourages
lawsuits;
to protect a persons livelihood from
interference;
to protect free enterprise and competition;
for the judiciary to respect the powers of the
legislature;
to protect the rights of property owners;
to protect the rights of all citizens to use
and enjoy public lands.
Many if not most interesting
legal cases involve some underlying teleological
issues, and in order to fully
appreciate these issues, it is necessary to understand
how different legal purposes
are related to each other. There are several types of
relationships among legal
purposes, which are illustrated in Figure 4, including two
adversarial relations (c and
d):
a. the taxonomic relation
(ISA) relates a general purpose, such as the achievement
of certainty and
predictability in the law, to a more specific one, such
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 31
Figure 4. Knowledge structure for legal
purposes.
as the unambiguous definition
of what it means to be in possession of a wild
animal. This ISA relation also
connects the general purpose of encouraging
useful activities to the more
specific ones of encouraging fox hunting, duck
hunting, and fishing.
b. the instrumental relation
(INST) relates a general purpose to a purpose that
contributes to achieving it
(thus, SUPP is a causal link). For example, assuming
there is a general goal of
promoting economic welfare of citizens, a more
specific goal that supports
this is the protection of a persons livelihood from
interference.
c. a limitation relation
(LIMIT) expresses the situation where the degree to which
one purpose can be pursued is
limited by another purpose. For example, the
personal freedom of
individuals is limited by the prohibition against doing
harm to others; the exercise
of the right of free speech is limited by the purpose
of protecting the public from
obscenity; and in the wild animal example, the
purpose of preventing unfair
competition imposes a limit on the purpose of
protecting free enterprise.
d. a relationship of direct
opposition (OPP) sometimes exists between pairs of
legal purposes in the
extreme case, for each purpose advanced by one side
of a legal dispute, there is a
competing and equally worthy purpose advanced
by the other side. The
adversarial structure of legal argument can be analyzed
in terms of such competing
purposes. (Kennedy, 1991, p. 101) Figure 4 shows
three pairs of opposing
purposes:
32 CAROLE D. HAFNER AND DONALD H.
BERMAN
i. certainty of legal rules
competes with context sensitivity (for example, the
need for consistency in
criminal sentencing competing with the need to
consider each individual
defendants circumstances);
ii. protection of livelihood
competes with protection of economic competition;
iii. prevention of unfairness
by the courts competes with respect for legislative
prerogatives.
Note that the structure of
legal purposes shown in Figure 4 (and legal argument
II, which is derived from the Young opinion) leads to argument
points that were
not explicitly raised in the
earlier wild animal cases points involving unfair competition
and judicial/legislative competency.
Teleological knowledge allows a legal
reasoning system to go beyond
fact-based comparisons with precedents to include
these broader jurisprudential
concepts.
There are several approaches
that might be used to incorporate legal purposes
into the computational
framework of legal case-based reasoning systems, as described
in Section 2. One could create
knowledge structures that package the
reasoning underlying
explanations [of cases] (Schank and Leake 1990, p. 354).
Such explanations might
constitute a slot in a case frame and these slots could be
indexed in numerous ways to
retrieve both similar and analogous cases. So the
explanation of Pierson might be not-possession to
avoid unnecessary disputes
while the explanation of Keeble might be possession to
protect the plaintiffs
livelihood. Such a scheme
would have problems, since the explanations would
need to be broken up and
recombined when only some of the dimensions of the
current case match the prior
case, as is done in the GREBE system (Branting 1991),
which combines partial
arguments from several prior cases.
We have chosen a different
scheme, illustrated in Figure 5, which utilizes the
factor-based model of
case-based legal reasoning, but adds teleological knowledge
to it. Prior systems using the
factor-based approach (described in Section 2) have
encoded each factor as either
favorable to the plaintiff or favorable to the defendant.
A limitation of this approach
is the lack of explanatory power as to why a factor
favors the plaintiff or
defendant, and why that factor is legally relevant. In our
scheme, each factor is linked
to the legal purpose(s) which it affects, and each legal
purpose in turn specifies
whether it favors the plaintiff or defendant. Finally, each
legal purpose is linked to
other purposes that may compete with it (adversarial
Links), so that the
case-based reasoning system knows when purpose A is advanced,
that purpose B may be
endangered, providing a foundation for making
policy arguments such as those
we have illustrated above.
3.3. IMPLICATIONS FOR FUTURE SYSTEMS
Using the knowledge
representations proposed here, a system for case-based legal
reasoning would generate
several sophisticated types of arguments. For the Young
case discussed in Section 4.1,
an argument can be generated that resembles Argu-
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 33
Figure 5. Incorporating legal purposes
into case analysis structure.
Figure 6. Teleological representation of
wild animal cases.
34 CAROLE D. HAFNER AND DONALD H.
BERMAN
ment II more than Argument I.
In Argument I, the defendant points out that the fish
had not been captured or
mortally wounded, just as in Pierson. However, using the
knowledge representation for
the wild animal domain shown in Figures 4 and 6, an
argument can be created whose
first steps would be:
Point for Defendant
This case is like Pierson v.
Post, since the plaintiff did not capture or mortally
wound the fish. It is
important to define possession in a way that promotes
certainty and avoids
unnecessary disputes.
Response for Plaintiff
This case can be distinguished
from Pierson
v. Post,
since in that case the
plaintiff did not make his
living from the hunting of wild animals.
This case is more like Keeble v. Hickeringill, since in that case the
plaintiff was
prevented from pursuing his
livelihood. It is important to protect people who
are pursuing their livelihood,
since productive work benefits society.
Rebuttal for Defendant
This case can be distinguished
from Keeble, since in Keeble the interference
was not done for economic
competition, while in this case the defendant was
merely competing vigorously
with the plaintiff. The law should not discourage
free enterprise and
competition.
In distinguishing a case (as
the plaintiff in Youngmay attempt to do), the
adversarial
link from certainty to
encourage useful enterprise can be used to help the casebased
reasoner create policy
arguments. For example, the plaintiff in Young might
bolster his response by
stressing the importance of fishing as a more useful activity
than fox hunting:
Response for Plaintiff (additional point)
If fishermen who spread their
nets must bear the additional risk that anyone can
come along and take the fish
out of the net, this will hurt our economy. Fishing
will become such a risky
occupation that people will avoid it altogether.
The limitations of this scheme
are several: first, the policies motivating a legal rule
may change due to a shift in
societys values or a change in other social institutions
(such as insurance) that make
existing rules unnecessary. (In Section 4, we explore
this limitation and propose
further computational mechanisms for overcoming it.)
Second, attorneys and courts
conceive and express purposes at varying levels of
abstraction. An advocate for
the defendant in Young
v. Hitchens who
believed
that a judge (possibly
influenced by the law and economics movement) would be
persuaded by policies based on
maximizing market efficiency might argue:
Point for Defendant
The defendant should win
because this type of competition will encourage the
most efficient methods for
trapping fish.
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 35
Or, believing that a judge
(possibly influenced by the work of critical legal theorists)
would be persuaded by
arguments based on the inequities resulting from the
influence of large
corporations, the attorney for the defendant might argue:
Point for Defendant
The defendant should win
because protecting the plaintiff is protecting established
big business interests at the
expense of smaller entrepreneurs.
Though incomplete, the arsenal
of teleological arguments described here more
closely approximates real
world advocacy in common law jurisdictions than those
produced by case-based
reasoning models that do not include a representation of
legal purposes. Although our
model is far from being able to replicate the full range
of teleological argument moves
employed by a gifted attorney, we anticipate that
case-based legal reasoning
systems incorporating even limited teleological knowledge
will prove useful both for
legal practice systems and for computer-assisted
legal education (Aleven and
Ashley 1994).
The representation framework
developed in this Section also points the way
toward a model of legal
reasoning capable of recognizing and exploiting analogies
between different legal
domains, such as wild-animal hunting and education. By
representing knowledge about
the competing legal purposes at stake in a case such
as Keeble, the relevance of analogous
cases involving the same competing purposes
(such as the hypothetical case
of two competing schools (11 H. 4) distinguished
by the court in Keeble) can be recognized. In the
next Section, we also consider
the potential of teleological
structures to support analogies between cases decided
under different legal
doctrines (e.g., contract and tort).
4. Temporal context: Modeling
the evolution of legal doctrine
In this section we consider a
second aspect of context that profoundly affects the
reasoning of lawyers and
judges about previously-decided cases: their relationship
to other, more recently
decided cases. We demonstrate that legal precedents are
embedded in a temporal context
of evolving legal doctrine, which can result in a
strong precedent becoming
weaker over time, to the point where a skillful attorney
could reasonably predict that
it will no longer be followed. We propose computational
mechanisms that would permit
such trends in judicial thinking to be taken
into account in a case-based
legal reasoning system.
Similar processes of change
occur in science, and many readers will notice
similarities between the
evolution of legal doctrine described here and the process
of change in scientific
theories described by Kuhn (1970). The desirability
of giving greater weight to
more recent decisions (Ashley and Rissland 1987) and
utilizing legal doctrine to
assess relevance has been noted by others (Branting 1994,
Rissland et al. 1996).
However, to date no one has attempted to represent precisely
and computationally the
methods used by lawyers to assess the continuing strength
of legal precedents over time.
36 CAROLE D. HAFNER AND DONALD H.
BERMAN
The standard approach to legal
case-based reasoning selects precedents by comparing
each prior case to the current
fact situation, and preferring the prior case(s)
that are most similar.
Unfortunately, this approach does not take into account how
a prior case has been treated
by other (more recent) prior cases, which may change
the picture radically. Such
temporal analysis of case law to recognize evolving
legal doctrine is taught from
the first year of law school; case-based models of
legal reasoning would be much
improved by incorporating even limited abilities
for temporal analysis of a
case knowledge base.
In Section 4.1, we analyze a
series of cases from New York to show how the
territorial view in conflict
of laws was gradually eroded and finally overturned
between 1945 and 1963. Our
problem scenario involves a passenger who is injured
in an automobile accident, and
later sues the driver for negligence in a state other
than the one where the
accident occurred. The conflict of laws issue is whether
to apply the substantive law
of the place where the accident occurred (the situs)
or the place where the case is
being litigated (the forum) regarding the liability of
drivers to passengers. In
Section 4.2 we identify patterns of legal reasoning that
occur over a series of cases,
and consider how a skillful attorney might recognize
these patterns and reasonably
predict that a precedent which was once strong will
no longer be followed.
In Section 4.3 we propose
further extensions to the extended case knowledge
base structures described in
Section 3, which could support the kind of legal reasoning
observed in the New York
conflict cases. In Section 4.4 we introduce an
algorithm for evaluating the
precedential strength of a prior case in its temporal
context, and consider the
implications of this kind of analysis for case-based legal
reasoning and argumentation.
4.1. GUEST PASSENGERS: AN EXAMPLE OF EVOLVING LEGAL DOCTRINE
What follows is a summary of a
series of New York cases dealing with the right
of a guest passenger in an
automobile to sue a host driver when both the driver
and passenger reside in New
York, the car is registered and ensured in New York,
and the accident takes place
in a jurisdiction (other than New York) where a statute
limits the guest passengers
right to sue the driver.
Two considerations led some
jurisdictions to adopt guest passenger statutes
which are exceptions to the
generally accepted policy that injured victims should
receive compensation for
injuries resulting from the defendants negligent behavior.
First, some jurisdictions like
Ontario, Canada evidently felt that if the host
does the guest a favor by
providing transportation, gratitude demands that the guest
not sue the host for the
hosts negligence. This would explain Ontarios absolute
bar to such law suits.12 Other jurisdictions evidently
believed that too often the
guest passenger and the host
driver (perhaps good friends or relatives) colluded
to defraud the defendants
insurance carrier. Some states, such as Montana and
Virginia, adopted statutes
requiring a showing of gross negligence in order for the
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 37
guest to recover, which would
require a host driver to admit to engaging in conduct
that could lead to a loss of
license or criminal sanctions.
Our problem case is Babcock v. Jackson.13 Babcock and Jackson were both
residents of New York, the
guest-host relationship had been formed in New York,
and a few hours after their
trip began in New York, Jackson lost control of his car
in Ontario, Canada. Babcock
was injured and in 1961 sued in a New York court
alleging negligence in the
operation of the vehicle. New York law would allow
Babcock to recover on a mere
showing that Jackson drove negligently.
A straightforward matching
with prior cases would lead an attorney to conclude
that the law of Ontario
applies and therefore the plaintiff loses. In 1938 New Yorks
highest court in Smith v. Clute14 had to decide whether the
Montana guest statutes
applied to New York residents
traveling through Montana on their way to California.
The court held that the
accident occurred in Montana, and the law of that state
governed the plaintiffs right
of action. However, the court found for the plaintiff
anyway, holding that because
the guest had contributed money toward the expenses
of the trip, he did not meet
the criteria of a guest passenger according to Montana
law.
Seven years later in Kerfoot v. Kelley,15 the intermediate court of
appeals in
New York made a similar
determination of applicability of Virginia law, finding for
the defendant in a case of
wrongful death resulting from an automobile accident.
In both Smith and Kerfoot, the guest passenger law
required a showing of gross
negligence in order for the
guest to recover. Thus, in 1945 a practitioner would
predict that Babcock had a
losing case. In fact, all cases involving guest passengers
in New York from 1945 until
1963 left little doubt that the law of the state in which
the accident occurred governed
the case.16
The decisions in Smith and Kerfoot were based on the traditional
territorial
choice of law doctrine that
applies the law of the place where the events that gave
rise to the dispute occurred.
The justification for the territorial rule was to assure
that a case will be treated
the same way regardless of whether the plaintiff brought
the case in New York or
Montana. Applying the rules of law irrespective of the
forum appears fairer because
the rights of the parties do not depend on which state
decides the case. In legal
parlance, the territorial view minimizes the advantages of
forum shopping. Another
rationale for the territorial rule is based on the principle
that parties should be subject
to predictable laws. A visitor to Montana understands
and implicitly agrees that
while in Montana he is subject to the laws of that state.
In the 1940s this territorial
theory of conflict of laws had been applied by the
highest courts of New York to
most choice of law cases. In 1959 the highest court
of New York in Kaufman v. American Youth
Hostels17 decided that an Oregon law
that immunized charities from
tort liability barred a wrongful death suit by a New
York plaintiff against a New
York defendant who died in a mountain climbing
accident in Oregon. Therefore,
even a New York practitioner who took a somewhat
broader view of relevant cases
would probably predict that Ontario law would
apply since the accepted rule
in New York in 1962 was that "the substantive rights
38 CAROLE D. HAFNER AND DONALD H.
BERMAN
and liabilities arising out of
a tortious occurrence are determinable by the law of
the place of the wrong (Babcock, p. 473)
However, the territorial rule
had been subject to much scholarly criticism based
primarily on the fact that the
territorial view of conflict of laws did not take into
account the social policies
underlying the substantive law of the concerned jurisdictions
(Babcock, footnotes 3 and 4). The
policies behind these statutes were
either to protect hosts from
ungrateful guests or to prevent fraud on insurance
companies. But these policies
would not apply to a case where both the plaintiff and
defendant are from New York, a
state that believes the potential for ingratitude or
fraud does not outweigh the
policy of compensation for injured plaintiffs. A simple
hypothetical example
demonstrates why so many legal scholars criticized the territorial
rule. If the plaintiff in Babcock were injured several feet
before crossing
into Ontario, then the
plaintiff recovers and New Yorks policy of compensation
would be fulfilled; but if the
accident took place several minutes later after the
parties had crossed into
Ontario, then the plaintiff loses and the New York policy of
compensation would be thwarted
without furthering any interest of Ontario, since
the defendant was from New
York.
A knowledgeable attorney would
have appreciated that the New York courts
had begun to question the
territorial rule in areas other than personal injury law.
In 1954 New Yorks highest
court in Auten
v. Auten18 abandoned the territorial
approach in the field of
contracts. Under the territorial doctrine, the law governing
contracts required that
matters concerned with validity be governed by the law
where the contract was
executed (in this case, New York) while matters related to
performance were governed by
the law where the contract was to be performed
(mostly England). Rather than
rely on this traditional view, in Auten the court,
interpreting a child support
agreement, adopted a center of gravity rule, holding
that the execution of the
agreement in New York was merely fortuitous and the
most important contacts like
the marriage, separation, and domicile of the wife and
children were in England.
Though representing a partial
rejection of the territorial rule, Auten could not
be relied upon by an attorney
trying a case in the area of personal injury. First, the
opinion made clear that the
same result would have been reached had the territorial
rule been utilized, since the
issue involved the wifes performance and that performance
occurred in England. Thus, the
articulation of a new legal doctrine could
be seen as merely dictum.
Secondly, and more to the point, Auten had been decided
five years prior to 1959 when
the highest court of New York decided Kaufman,
which affirmed the territorial
rule in the area of tort law.
In May of 1961 the New York
court decided Haag
v. Barnes,19 which affirmed
the center of gravity approach
adopted in Auten
in an area of
law similar to Auten.
After Haag, an attorney could
reasonably conclude that a paradigm shift had occurred
in choice of law in the field
of contracts. This, in combination with the
scholarly criticism of the
territorial rule in tort law, might engender a prediction
that a similar paradigm shift
would eventually take place in torts as well. Kilberg v.
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 39
Northeast Airlines,20 decided in January 1961,
provided evidence to support such
a prediction. Kilberg, while
in the state of New York, purchased a ticket to fly from
New York to Massachusetts.
Kilberg died when the plane crashed in Nantucket,
Massachusetts. A Massachusetts
statute limited the damages from wrongful death
to $15,000, while New York had
no such limit. Under the territorial rule, the Massachusetts
law would govern. However, the
court ignored the precedents and held
that the Massachusetts
wrongful death statute applied to the question of whether
the plaintiff had a right to
sue, but that New York law applied as to the amount of
damages the plaintiff could
recover.
Four of the seven judges,
while not rejecting the territorial rule, invoked exceptions
to the general rule that the
law of the place of the wrong would apply. The
court relied primarily on the
public policy exception:
Modern conditions make it unjust
and anomalous to subject the traveling citizen
of this State to the varying
laws of other States through and over which
they move. An air traveler
from New York may in a flight of a few hours
duration pass through several
. . . commonwealths. His plane may meet with
disaster in a State he never
intended to cross but into which the plane has flown
because of bad weather or
other unexpected developments . . . The place of
injury becomes entirely
fortuitous. Our courts should if possible provide protection
for our own States people
against unfair and anachronistic treatment
of the lawsuits which result
from these disasters. (Kilberg, p. 34)
Two judges, though concurring
with the result on procedural grounds, rejected the
majoritys analysis:
Questions relating to such
defenses as contributory (comparative) negligence
. . . charitable immunity . .
. incapacity of wife to sue . . . have all been regarded
by this court as regulated by
the law of the place of injury rather than our own
law . . . In each of these
cases we applied a foreign rule although such rule was
clearly contrary to the law of
our own State . . . The majority would apply our
own law of damages because the
place of injury is entirely fortuitous. The
same argument may be made with
respect to each of the cases just referred to.
We should not overrule
well-established principles . . . The position adopted
by the majority may result in
the situation where, in a single airplane crash
in which numerous passengers
from various States are killed, a different law
will be applied in each action
resulting therefrom. (Kilberg, p. 71)
Although Kilberg did not explicitly reject the
territorial rule, it represents an implicit
paradigm shift that casts
doubt on the rule. An attorney reading Kilberg (a tort
case) in combination with Auten and Haag might reasonably predict that
the New
York court was ready to extend
the new choice of law theory to tort law in general.
This is precisely what
occurred in Babcock, where the court held for the
plaintiff,
rejecting the territorial view
in favor of the center of gravity approach (also called
grouping of contacts) which
the court had adopted in Auten:
40 CAROLE D. HAFNER AND DONALD H.
BERMAN
Comparison of the relative
contacts and interests of New York and Ontario
in this litigation, vis--vis
the issue here presented, makes it clear that the
concern of New York is
unquestionably the greater . . . The present action
involves injuries sustained by
a New York guest as the result of the negligence
of a New York host in the
operation of an automobile, garaged, licensed and
undoubtedly insured in New
York, in the course of a week-end journey which
began and was to end there. In
sharp contrast, Ontarios sole relationship
with the occurrence is the
purely adventitious circumstance that the accident
occurred there. (Babcock, p. 458)
4.2. TEMPORAL PATTERNS SIGNIFYING A
WEAKENING OF PRECEDENT
How would the attorney in
Babcock predict that the on-point and formerly solid
precedents of Smith, Kerfoot
and Kaufman might not be followed? We have identified
five temporal patterns of
judicial behavior which could lead a skillful attorney
to that prediction (ordered
from the strongest evidence to the weakest):
1. If the precedent is explicitly
over-ruled, the prediction that it will not be followed
in the future is an obvious
(although far from certain) consequence.
However, the frequency of
explicit over-ruling is low compared to other signs
of weakening.
2. A precedent is implicitly over-ruled
when the result in a newer case clearly
contradicts an older case with
very similar facts, even if the court does not
explicitly say the prior case
is over-ruled. For example, one can argue that
Babcock, once decided, implicitly
over-ruled Kerfoot, and Kilberg implicitly
over-ruled earlier cases which
applied the situs limitations on wrongful death
claims made by New York
residents.
3. The ratio decidendi of an older case has been
over-ruled, even though the
facts may differ in
significant respects. For example, in explicitly rejecting
the territorial view in favor
of the center of gravity view, one can argue that
Babcock over-rules Kaufman, even though Kaufman was not a guest passenger
case.
4. A shift in the relative
priority of competing purposes is in evidence from the
courts tendency to make a
rule increasingly narrow by finding exceptions and
distinguishing away
situations where one might have expected the rule to
apply. The finding of a
public policy exception in Kilberg can be viewed in
this light, suggesting that
the court was straining for reasons to avoid applying
the territorial rule.
5. A general shift in the
relative priority of competing purposes is in evidence
from cases from other legal
domains. The Auten
and Haag cases, coming from
the contracts domain, could
provide evidence that the purposes advanced by
the territorial rule were
losing ground to the competing purposes advanced by
the center-of-gravity rule.
Babcock extended this shift to the guest passenger
domain.
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 41
Table I. Case knowledge base for guest
passenger cases
Smith v. Klute (1938) Kerfoot
v. Kelley (1945)
Result: Passenger wins (Side1)
Result: Host wins (Side2)
Relevant Factors Relevant
Factors
Claim: driver negligence
Claim:Driver Negligence
Same-residence: yes
Same-residence: yes
Relationship-loc: forum
Relationship-loc: forum
Non-forum-law-limits-recovery:
yes Nonforum-law-limits-recovery: yes
Basis of limits:
gross-negl-standard Basis of denial: gross-negl-standard
Passenger is guest: no
Passenger is guest: yes
Factor Purpose served Favors
Claim:driver negligence PD
Side1
Same-residence: yes PB Side1
Relationship-loc: forum PB
Side1
Nonforum-law-limits-recovery:
yes PA Side2
Basis of denial: gross-negl-standard
PC Side2
Passenger is guest: {yes, no}
{PC, PD} {Side2, Side1}
PA = Territorial Rule (law
should be determined by location of events giving rise to the dispute);
PB = Center of Gravity Rule
(law should be determined by location of parties overall
relationship); PC = Avoid
fraudulent claims; PD = Compensate accident victims.
Figure 7a. Legal purposes for choice of
law doctrines.
42 CAROLE D. HAFNER AND DONALD H.
BERMAN
Figure 7b. Legal purposes for guest
passenger preclusion statues.
In the next two sections, we
consider how these patterns can be recognized and
taken into account by a
case-based legal reasoning system. (A possible sixth pattern
signifying legal change would
be the existence of a substantial scholarly attack on
a given legal doctrine;
however our model of case-based reasoning does not include
recognizing trends in the
scholarly literature.)
4.3. A COMPUTATIONAL MODEL OF
TEMPORAL CONTEXT
In this Section, we consider
how a case-based legal reasoning model can support
the kind of legal analysis
that can explain, and possibly predict, the results in the
New York conflicts cases. The
jumping-off point for our analysis is the model of
teleological case-based
reasoning proposed in Section 3. In that model, we extended
the HYPO-style factor
comparison model described in Section 2, by linking
each judicially-recognized
factor in a legal domain to one or more legal purposes
or policies that define its raison detre. The various legal purposes
were in turn
organized into a semantic net,
with links representing both supporting and competing
relations between pairs of
purposes. Whereas in the factor comparison model
the computation of relevance
was based on which prior cases had a maximal set of
shared factors with the
current case, in our model prior cases could also be found
relevant if they involved the
same structure of competing purposes, even if they
shared few (or even no)
domain-specific factors.
Applying this approach in a
straightforward way to the guest-passenger cases,
we can create a network of
legal purposes as shown in Figure 7, and index the Smith
and Kerfoot cases as shown in
Table I. Figure 7a shows a legal policy network in
the choice of law domain: at
the lowest level, the policy that says a dispute should
be resolved using the law of
the location of the events that gave rise to the dispute
competes with the policy that
says avoid applying laws of a jurisdiction that plays
only a fortuitous role in the
dispute and has little or no interest in the outcome.
Figure 7b shows competing
purposes in the torts domain that are directly relevant
to guest passenger statutes
the accepted legal policy of compensating victims of
negligence vs. the states
desire to minimize fraudulent claims.
Table I shows how Smith and
Kerfoot would be presented using the teleological
case representation developed
in Section 3. (Side 1 in all examples is the party
whose recovery or compensation
is limited by the non-forum law e.g., the guest
passenger.) Factors that might
influence the court include whether the parties are
from the same state (other
than the state where the accident occurred) and whether
the parties had a prior
relationship formed outside of the state where the accident
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 43
occurred. If both of these
factors are present, this makes the location of the accident
appear more fortuitous.
Factors that affect the application of a statutory preclusion
(or limitation) of recovery
include whether the guest passenger has paid something
to the driver (and was
therefore not completely gratuitous) and whether the claim
before the court is one to
which the preclusion can apply.
In considering Smith and Kerfoot as precedents for Babcock, we immediately
see that the relevant holding
is not the final result of the case but rather the intermediate
holding of whether to apply
New York law or the law of the situs.
Although different sides won
in these two cases, as precedents for Babcock they
are on the same side they
would both be precedents for the defendant. As Branting
(1991) has pointed out, most
cases have several holdings which depend on
different subsets of the
relevant factors, and which can be used as precedents even
in cases where factors outside
the relevant subset do not match. In a later paper,
Branting (1994) presented a
formal definition of the ratio decidendi of cases, and
showed that the precedential
significance of a case cannot be understood without
considering the ratio as well as the facts and the
final result.
Therefore we propose to extend
our case knowledge base design in the following
way: we define the results
of a case as the set of all holdings that are part
of the ratio as defined by Branting. The
extended results are represented in a slot-
filler structure, so that in
guest passenger cases the winning-side slot would have
fillers guest or host while the apply-forum-law slot would have fillers yes or
no. With this extension, we can
represent the relevance of cases such as Kaufman
and Kilberg to guest passenger cases such
as Smith and Kerfoot.
However, we are still not
satisfied with this characterization of the results of a
case as the set of all
holdings, because it does not mention the territoriality doctrine
that was criticized in Kilberg and finally rejected in favor
of the center-of-gravity
doctrine in Babcock. It is difficult to say that
we have represented the result
of Babcock if we do not include this
aspect of the case, since it weakens every
earlier precedent that cites
the territorial rule as part of its ratio. To make the
representation more accurate
we need to include a representation of the legal rules
used by a court to justify its
holdings. In so doing, we are not intending to model
the process of rule
application, but merely to include a reference to the legal rules
cited by the court, as one of
the attributes on which cases can be compared. For
the sample cases, we will add
a result slot: rules-followed, whose fillers include
territorial and center-of-gravity. (Note: after Babcock, this
simple story of
two competing doctrines was
complicated by the rejection of the center-of-gravity
doctrine in Tooker v. Lopez21 in favor of a greater
interest approach, but we will
not consider that distinction
here.)
It is interesting to note that
the two contracts cases that first rejected the territorial
rule failed to apply the law
of the forum, finding instead that some other
jurisdiction was the center of
gravity for the contractual relationships. Thus, as
in the case of Babcock, the correct interpretation
of these cases by a case-based
44 CAROLE D. HAFNER AND DONALD H.
BERMAN
Table II. Extended case knowledge base
for choice of law cases
Smith v. Klute (1938) Kerfoot
v. Kelley (1938)
Holdings: Holdings:
Winning-side: passenger (Side1)
Winning-side: driver/owner (Side2)
Forum-law-controls: no
Forum-law-controls: no
Choice-of-law rule: territorial
Choice-of-law-rule: territorial
Relevant factors: Relevant
factors:
Claim: driver negligence Claim:
driver negligence
Same-residence: yes
Same-residence: yes
Relationship-loc: forum
Relationship-loc: forum
Nonforum-law-limits-recovery:
yes Nonforum-law-limits-recovery: yes
Basis-of-limit:
gross-negl-standard Basis-of-limit: gross-negl-standard
Trip-origin: forum Passenger-is-guest:
yes
Length-of-non-forum-contact:
brief
Passenger-is-guest: no
Auten v. Auten (1954) Haag v.
Barnes (1961)
Holdings: Holdings:
Winning-side: Supported-spouse
(Side1) Winning-side: Supporting-spouse (Side2)
Contract-upheld: yes
Contract-upheld: yes
Forum-law-controls: no
Forum-law-controls: no
Choice-of-law-rule:
center-of-gravity Choice-of-law-rule: center-of-gravity
Relevant factors: Relevant
Factors:
Claim: enforce support
agreement Claim: reopen support agreement
Same-residence: unknown Same-residence:
no
Relationship-loc: nonforum
Relationship-loc: nonforum
Forum-law-limits-recovery: yes
Nonforum-law-limits-recovery: yes
Basis of limit: contract Basis
of limit: contract
Length of forum visit: brief
Length of nonforum visit: months
Kaufman v. American Youth
Hostels (1959) Kilberg v. Northeast Airlines (1961)
Holdings: Holdings:
Winning-side: Vendor (Side2)
Winning-side: Passenger (Side1)
Forum-law-controls: no
Forum-law-controls: yes
Choice-of-law-rule: territorial
Choice-of-law-rule: public-policy overrides
territorial
Relevant Factors: Relevant
Factors:
Claim: travel guide negligence
Claim: common carrier negligence
Same-residence: yes
Relationship-loc: forum
Relationship-loc: forum
Nonforum-law-limits-recovery: yes
Nonforum-law-limits-recovery:
yes Basis of limit: wrongful death damage limitation
Basis-of-limit: charitable
immunity Trip-origin: forum
Length of nonforum visit: days
Length of non-forum visit: brief
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 45
reasoning system requires the
enunciated legal rules to be included in a cases
representation.
Table II shows the extended
representation of the six previously-described cases
that the Babcock attorney
might have considered. New factors introduced after
Kilberg include the length of time
spent in the situs jurisdiction, and the place
where the trip originated.
This information has been added to prior cases where it
can be determined from the
written opinion.
4.4. INTEGRATING TEMPORAL CONTEXT
INTO CASE-BASED LEGAL
REASONING: THE RED FLAG ALGORITHM
In this section, we consider
how the temporal structure of legal decisions can be
used by a case-based reasoning
system to influence the evaluation of relevance.
We assume there will be an
initial comparison of the prior cases with the current
case based on the traditional
factor-comparison model, and a similarity network for
on-point cases will be
created. For each on-point case, we then create a temporallysequenced
list for each of the holdings of that case, containing later
cases that had
holdings on the same issue.
The list elements are labeled in a manner similar to the
treatments used in legal
citators (i.e., followed, distinguished, criticized, over-ruled,
etc.) (cf. Ashley and Rissland
1987).
Using this structure, it is
possible to evaluate, for the on-point cases from the
similarity network, the degree
to which their holdings (on issues relevant to the
current case) have been
weakened by subsequent cases. We now consider how a
case-based reasoning system
should represent and calculate this weakening measure.
It is clear that temporal
evaluation is a form of evidential reasoning, so we
propose a numerical measure of
weakening ranging from 0 (no weakening) to
1 (complete over-rule). This
value, like the numbers attached to conclusions in
expert systems, should be used
for comparison purposes for example, although
the precise numbers are not
very meaningful, assigning a weakening factor of 0.9
to a precedent indicates a
high degree of risk to the party who relies on it.
We have considered two
approaches to calculating the weakening factor: the
first approach is based on
comparing the average age of cases that treated a precedent
positively with the average
age of those that treated it negatively. A more
recent (i.e., lower) average
age indicates more current doctrine. In considering the
details of such an algorithm,
we concluded that this approach would pose signi-
ficant difficulties. One
difficulty would arise when a decision pattern that has been
solidly entrenched for a long
time has not been litigated very much in recent years,
but is still good law. For
example, if the court had retreated from its reasoning in
Auten and reinstated the territorial
rule (as would seem quite plausible in 1959 after
Kaufman was decided), the average age
approach would not give a correct result,
since the cases that followed
the territorial rule would be much older than Auten.
Another difficulty would arise
when, in the course of a legal doctrine undergoing
change, the new doctrine is
applied to one narrow sub-domain after another, and
46 CAROLE D. HAFNER AND DONALD H.
BERMAN
eventually is accepted as a
new general rule. If a recently-decided case happened to
be in a sub-domain where the
change has not yet occurred, this would appear to be
a retreat from the new
doctrine when it really is simply a characteristic of normal
evolution.
A different approach is based
on the recognition of clues or red flags suggesting
that an established legal
doctrine is under attack. When a legal decision either
refuses to follow prior cases
(as in Kilberg) or explicitly rejects the ratio of prior
cases (as in Auten), this acts as a red flag
that may lead a skillful attorney to question
whether the holdings of the
prior cases (and others) that rely on the same legal rules,
will still be followed in the
future. We have developed an algorithm based on this
conception of legal reasoning.
We define five types of red
flag cases, corresponding to the five weakening
patterns described in Section
4.2. For a red flag of type i, we
define a weakening
factor Wi between 0 and 1, representing the evidential
strength supplied by that
type of red flag for the
prediction that a precedent is no longer reliable. (The actual
values of theWi should be empirically determined; or they could
be provided by the
user of a case-based reasoning
system.) We use the traditional rule for combining
evidence in expert systems
(Hayes-Roth et al. 1983): if there are two red flags
that weaken the same precedent
P with weakening factors Wx and Wy
, the overall
weakening of P would be calculated as Wx + Wy - (Wx .
Wy).
An algorithm to compute the
overall estimate of how much a holding H of
precedent P has been weakened by later cases C1, C2,
. . . , is outlined below. The
algorithm is recursive: once a
red flag case C has been identified, before
its effect
on P can be estimated, the weakening of C itself by even later cases much be taken
into account.
RED FLAG ALGORITHM: Given a
precedent P, a holding of P on an issue J ,
and a temporally-sequenced
list of later cases that had holdings on the same issue,
compute the degree of
weakening Wpj that Ps
holding on issue J has undergone
since it was decided.
Step 1. Identify the red
flags in the sequence of later cases (that is, later cases
providing evidence that Ps holding on issue J has been weakened).
Step 2. For each type of red
flag i identified in Step 1:
i. Let C be the most recent case that is a red flag of
type i.
ii. Compute (recursively) the
weakening factor Wcj , based on even more recent
cases.
iii. Let the weakening effect
of C on P (relative to issue J) be Wi . (1 - Wcj ).
Step 3. Combine the evidence
from the various types of red flags, as computed in
Step 2.
Step 2 provides that if a
precedent P applies the territorial rule,
and the territorial
rule is weakened by a red flag
case C, the later case may itself be
weakened by a
return to the territorial rule
subsequently. Assuming Wi is the weakening factor for
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 47
the type of red flag
represented by C, if Wcj is close to zero indicating Cs holding
is still strong, the full
weakening effect Wi of C will
be applied to P. But if Wcj
is close to 1, indicating Cs holding has been eroded by second-order
weakening,
then very little weakening
will be applied to P on account of C.
If there are several red flag
cases representing the same pattern, we use the most
recent one to stand in for the
set. This approach has the advantage of simplicity:
when legal doctrines undergo
change, it is not uncommon for decisions to shift
back and forth between the old
approach and the new approach during a period
of great uncertainty. In those
situations, an algorithm of this type would become
very inefficient if each red
flag case were fully analyzed. By using the most recent
red flag of each type, we lose
the ability to combine the influence of several similar
departures from the
previously-established doctrine, but we gain the ability to focus
quickly on the end result.
The approach described here,
in order to be implemented, requires additional
algorithms to identify the
five types of red flag patterns. The first two patterns
could be readily identified by
comparing the holdings of P with
the holdings of
later decisions (assuming the
case knowledge base is structured as described in
Section 4.3). The third
pattern could also be readily identified, now that the results
of a case have been extended
to include elements of the ratio.
The fourth pattern,
in which new dimensions or
rules are introduced to avoid following a precedent,
poses more difficulty, since
we probably would not want to assume that every time
a case is distinguished, this
represents a red flag.
The fifth pattern would be the
most difficult to identify. In our example case
base, in order to predict that
Auten and Haag weakened the precedents of Smith
and Kerfoot, it would be necessary first
to find a connection between them, so that
the contracts cases could be
included in the data being searched for red flags. One
possible approach to this
problem is the following: whenever a case is decided
that breaks new ground, a
marker is inserted into the network of legal purposes
indicating a shift in
priorities between two competing purposes. Since each holding
of a case is identified with
the legal purposes advanced by that holding, such
markers can be retrieved
during Step 1 of the Red Flag Algorithm. This still leaves
a question, however, about how
to relate the holding of the analogous case to the
holding in the area of
interest. In the choice of law domain, this would not be
difficult since both domains
invoke a territorial rule. However, this would not
always be true, and further
study is needed to understand how results are used
analogously across domains,
whether for supporting or weakening a precedent.
4.5. IMPLICATIONS FOR FUTURE SYSTEMS
The prospect of adding
temporal analysis to a case-based legal reasoning system
gives rise to a number of
questions, the most important being: how would a model
such as we have proposed
affect the overall operation of a case-based legal argumentation
system? We can see two ways in
which the use of weakening factors
48 CAROLE D. HAFNER AND DONALD H.
BERMAN
could be used to improve the
quality of arguments generated: first, if there are
several equally on-point cases
(a frequent occurrence when a HYPO-style similarity
network is used), the system
can present an argument based on the strongest
precedent first. Second, a new
argument move can be developed to discredit a
precedent used by the opposing
side, based on the claim that the precedent has
been weakened by subsequent
decisions.
The evolution of legal
doctrine that we illustrate above using the choice of law
domain is widespread within
common law legal systems. In the United States it
can be found in the evolution
of personal injury law from a negligence standard
toward strict liability; in
the evolution of landlord and tenant law from a property
conveyancing doctrine to a
contract law doctrine; in the evolution of contract law
from a paradigm of
offer-and-acceptance toward the treatment of some cases under
a theory of promissory
estoppel; and in the evolution of real-estate law toward
the imposition of new duties
of disclosure on brokers and sellers. Thus the need
to integrate temporal analysis
into case-based legal reasoning models arises in
many legal domains that affect
large numbers of people, and presents a significant
challenge for those wishing to
make case-based legal reasoning systems more
robust.
5. Procedural context: Modeling
the process of legal decision-makin
In this section, we demonstrate
that a robust model of legal cases must include a
representation of the
procedural context of legal precedents. By procedural context,
we mean how a legal systems
decision-making machinery was invoked to decide
a case. This includes the
jurisdiction in which a legal action was heard; the level
and type of court that
adjudicated it; and the formal process(es) of decision making
employed (which we call the
procedural posture of the case). Such meta-level
legal concepts (that is,
concepts that address the process of decision making rather
than its substance) permeate
appellate court decisions and profoundly influence the
analysis and use of
precedents.
Below, we show the importance
of procedural posture in accurately modeling
case-based legal reasoning,
focusing on the large subset of appellate cases decided
on the basis of either the
sufficiency of pleadings or the sufficiency of the evidence.
Using examples from
Massachusetts property law, we show that evaluating
a precedent based on factual
similarity without considering its procedural posture
can result in unpersuasive or
even erroneous legal arguments. We develop a formal
model of procedural posture
that generates 10 procedurally-distinct types of legal
results, five favoring the
plaintiff and five favoring the defendant. We then propose
rules describing, for each
result type, which other result types function best as
precedents, and outline a
method for incorporating these procedural distinctions
into case-based reasoning
systems.
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 49
5.1. A HYPOTHETICAL EXAMPLE
Consider the following
hypothetical case used in a Property Law class:
In 1978 Ida Peat conveyed to
Max Schnell an unimproved wooded area of
approximately five acres
located in Boxtown, Massachusetts. Max was a
world renowned speed skater
and traveled frequently abroad to compete and
train. In Spring 1979, Max had
posted land marks at each corner of the lot,
and he paid taxes on the lot
between 1978 and 1985. During the cold months
of 19781985 Max cleared a
large area of the lot by removing trees and
bushes, and he flooded the
cleared area for a practice rink. In June 1985, Max
conveyed the lot over to
Sheila Spark and moved away. In July 1985, Sheila
moved onto the lot, put up fences
along the borders and built a barn so she
could raise sheep. Since 1985
Sheila has continuously housed eight to twenty
sheep on the lot.
In December 1999, John Doe,
who possesses document showing a valid
chain of title, has threatened
to commence an action to establish his title to
the five-acre tract. No doubt
exists that John has the superior record title.
You represent Sheila. John has
offered to sell his interest in the land for
$50,000. Sheila values the
land at $120,000 and you have estimated the costs
of litigation to be $20,000.
Evaluate the merits of Johns settlement offer.
The students were aware of the
doctrine of adverse possession, which permits an
occupant without formal title
to land to obtain title to that land when there is:
(1) an actual entry giving
exclusive possession that is (2) open and notorious, (3)
adverse and under a claim of
right, and (4) continuous for the statutory period
of 20 years (Dukeminier 1988,
p. 100). Most students recognize two questions of
continuity: first, can Sheila
combine her 14 years with Maxs 7 years in order to get
the required 20 years of
continuous possession? Second, did Maxs frequent trips
abroad and his failure to use
the land throughout the year interrupt the continuity
of possession? From approximately
100 prior Massachusetts adverse possession
cases available to the
students, we present below two cases in a frame-like summary
format often used by law
students (Wren, p. 161), which resembles the frame-based
knowledge representations used
by AI and Law researchers (Ashley 1990, p. 220;
Hafner 1987, p. 38).
The law students dutifully
mapped each of the facts and legal issues of Sheilas
case to Kershaw. They reasoned that speed
skating resembled circus performing;
land marks resembled pipes;
both claimants by adverse possession cleared the land;
both adverse possessors paid
the taxes; both had continuity problems since they
left the country; both cases
involved combining the interest of a second adverse
possessor (who more fully
developed the land) with that of a prior possessor.
Therefore, they concluded,
Sheila should reject Johns settlement offer.
50 CAROLE D. HAFNER AND DONALD H.
BERMAN
Name: Kershaw v. Zecchini (Kershaw)
Citation: 342 Mass. 318, 173 N.E.2d 624
Date: 1961 Vote: 5-0 Judge: Spalding
Procedural Setting: Bill in equity by holder of
record title is brought in the Land Court to obtain
a declaratory decree
establishing title to a parcel of real estate. From a decree adjudging that the
defendant was the owner of the
parcel by adverse possession, the plaintiffs appealed.
Facts: Plaintiffs had superior record
title to a 5,000 square foot lot. The defendant claimed title
under a deed from Andresen and
his wife. The Andresens were circus performers who travelled
extensively abroad. After his
purchase represented by a deed, Andresen and the defendant went
onto the property, which had
previously been unimproved land, and cleared the brush and put down
pipes at the corners of the lot
and kept it clear. The land was assessed to him for taxes from 1937 to
1942. From 1936 until 1943
Andresen and the defendant exercised and practiced stunts on the lot
when they were in town. In
consideration for payment of several years of back taxes on the land,
Andresen conveyed the property
to the defendant in 1943. Thereafter the defendant erected a wall
and built the house which is
now on the land.
Issues: Did the Land Court err in
finding that the defendants obtained title to the parcel by adverse
possession.
Result: Decision of Land Court in favor
of defendant is affirmed.
Reasoning: The Land Court was justified in
deciding that as a matter of fact the actions of the
defendant were sufficient to
constitute adverse possession. The defendant had marked the property;
the visits away from the
property did not, as a matter of law, disrupt the continuity of possession.
Name: Senn v. Western Mass. Electric
Co. (Senn)
Citation: 18 Mass.App. 992, 471 N.E.2d
131
Date: 1984 Vote: 3 Judge: Rescript
Procedural Setting: Defendant took an easement on a
56 acre parcel by eminent domain. The
plaintiffs who claimed
ownership of the parcel by adverse possession sought damages. The
plaintiffs appeal the trial
judges setting aside a jury verdict which awarded damages.
Facts: The plaintiffs and their
predecessors had used, in some fashion, part of the parcel for at
least 20 years. The main
activity during that time involved the cutting of timber for family use and
for sale to others. They used
the logging roads traversing the parcel, pastured some animals over
a portion of the parcel, and
paid real estate taxes in the years 1967 through 1970. The plaintiffs
conceded that they never
enclosed the parcel by walls or fencing.
Issue: Did the trial judge correctly
rule that as a matter of law the plaintiffs had failed to establish
ownership of the parcel by
adverse possession?
Result: The decision of the trial judge
setting aside the verdict for the plaintiffs is affirmed.
Reasoning: To establish adverse possession
of open woodlands, as a matter of law the claimant
must show cultivation or
enclosure.
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 51
But, as the language in Kershaw makes clear, such a conclusion
was unwarranted
because the appellate court
merely affirmed a lower courts factual determination:
Whether, in a particular case,
these elements [are sufficient to acquire title by
adverse possession] is
essentially a question of fact. This is because . . . [t]he
nature and the extent of
occupancy required to establish a right by adverse
possession vary with the
character of the land, the purposes for which it is
adapted, and the uses to which
it has been put. (173 N.E.2d 624, 626)
As explained in more detail
below, although the appellate court in Kershaw upheld
the trial courts finding in
favor of the adverse possessor, the same court could
(without contradicting itself)
uphold a trial courts finding to the contrary. The
courts decision does not say
that Kershaw was decided correctly, only
that there
are not sufficient legal
grounds for overturning it. Thus, from Kershaw one can only
infer that if (1) a trial
court were to find for Sheila and (2) Sheilas case is similar
to Kershaw then (3) that
finding would not be overturned. It says little about how a
trier of fact would decide
Sheilas case.
A more sophisticated mapping
strategy would have led students to perceive the
potential relevance of Senn, a subsequent case that
purported to follow Kershaw.
In affirming the trial courts
ruling that, as a matter of law, there was insufficient
evidence of adverse possession
to go to the jury, the court in Senn held that:
[a]ccording to the cases
decided inMassachusetts . . . these acts are not in their
nature acts of exclusive
possession, and therefore do not constitute proof of
disseisin . . .Kershaw v. Zecchini is not to the contrary. (471
N.E.2d 131, 132)
But why isnt Kershaw to the contrary, since both
parcels were undeveloped and
neither was enclosed or
cultivated? Perhaps the distinction, although unstated, lies
in the relative sizes of the
tracts. The parcel of land in Kershaw was about 5,000
square feet, roughly 1/8 of an
acre, whereas the parcel in Senn was
over 56 acres.
On a 1/8 acre parcel, almost
any significant use could be reasonably viewed as
exclusive, open and notorious
possession, while on a 56 acre tract the same
activity may have no obvious
impact. Therefore, one should not, based on the
material presented in these
two cases, conclude that a Massachusetts court would
treat Maxs five acre open
tract of land like the 1/8 acre in Kershaw rather than like
the 56 acres in Senn.
Most experienced litigators
would engage in a process of decision analysis in
applying these two cases to
these specific facts. First, the litigator must decide
whether Sheila, as a matter of
law, will be barred from gaining title by adverse possession
byMaxs failure to take those
acts required for adverse possession of a tract
of open woodlands. To answer
that question the lawyer must decide whether a
judge will, as a matter of
law, characterize this five acre tract as open woodlands.
If the land is so
characterized then the lawyer must decide whether the judge, as
a matter of law, will rule
that Maxs actions failed to constitute cultivation or
enclosure. Before predicting
how a court would decide these questions of law,
most experienced practitioners
would obtain additional facts about the size of the
52 CAROLE D. HAFNER AND DONALD H.
BERMAN
area that Max cleared for use
as a skating rink, and the visibility of the markers and
Maxs activities to an owner
who casually surveyed his property. In the absence
of facts showing that Max used
most of the five acre tract and that the markers
and Maxs activities were
highly visible, many practitioners would, given Senn,
recommend accepting Johns
offer.
If the lawyer decides that
Sheilas claim is not barred as a matter of law, then
her lawyer must further assess
the chances of persuading a trier of fact to find in
her favor. This assessment
would be based on the attorneys sense of what a given
judge or jury thinks is the
just result in the given case. Since Sheila believed she
had good title, used the
property for many years, paid the taxes and made a heavy
investment in the property,
most litigators would probably conclude that the factual
issue would be resolved in
Sheilas favor.
5.2. A COMPUTATIONAL MODEL OF
PROCEDURAL CONTEXT
In most common law
jurisdictions, legal issues are characterized three ways. First,
a judge or jury may decide an
issue as a question of fact. If Sheila claims that
Max practiced every day and
John claims that Max practiced only occasionally,
the question of how often Max
practiced is a question for the trier of fact to decide.
If a trial court, after
hearing all the testimony, decides as a finding of fact that
Max practiced every day, this
finding can only be overturned on appeal if it is
clearly erroneous, a most
difficult standard to satisfy.
Second, an issue may be resolved
as a question of law. In these situations, even
if one assumes all disputed
facts are decided in a manner favorable to one party, that
party still cannot win. The
disposition of cases as matters of law fall into several
categories. In some situations
(e.g., judgment on the pleadings) the rule is clear and
the lawyer can do nothing to
change the result by building a richer fact record. If
Massachusetts had a rule that
no one can obtain title to registered land by adverse
possession, then Sheila loses
if the disputed parcel was registered.
In other situations (e.g.,
summary judgment), cases are resolved as a matter
of law because a partys
evidence, even if believed, is not sufficient to satisfy
the requirements of their
claim a decision that may involve open-textured legal
concepts. The court in Senn, following earlier
precedents, held as a matter of law
that title by adverse
possession of open woodlands requires either enclosure or
cultivation. After reading Senn, Sheilas attorney would
undoubtedly try to find
witnesses or photographs
showing that Max used a portion of the land for a garden,
or built a fence around his
practice rink. Or the lawyer might try to argue that a
5-acre lot does not qualify as
open woodland within the meaning of Senn.
Third, there are mixed issues
of fact and law. These are cases where legal rules
are applied to the facts to
determine who wins the case. In Sheilas case the trier
of fact, be it judge or jury,
must decide whether Maxs actions of cutting bushes,
placing markers, paying taxes,
etc., amounted to adverse possession. Significant
indeterminacy arises in the
process of characterizing mixed questions of law and
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 53
fact and deciding who should
decide these questions. In Kershaw the court was
comfortable leaving the
decision to the trier of fact while in Senn the court, in
order to protect the interest
of owners of large tracts of open woodlands, had the
case turn on a question of
law.
To allow certain issues to be
decided as questions of law, legislatures and courts
have crafted formal
procedures. These procedures are described below, along with
a formal representation
consisting of:
a. a three-stage model of legal
decision making,
b. a classification of legal
decisions into 10 procedure-based result types (5
favoring each side), and
c. rules that describe the
procedural relevance of a prior case to a current case,
depending on the procedural
categories of the two cases.
5.2.1. A three-stage model of legal
decision making
We begin by classifying the
results of litigation into the pleading stage (PL), the
pre-verdict stage (PV) and the
trial verdict stage (TV). Both the PL and PV stages
represent the disposition of a
case as a matter of law, while the TV stage represents
the disposition of a case as a
matter of fact, or a mixed question of law and fact
which is decided by the trier
of fact.
In the pleading stage (PL),
cases can be dismissed after the plaintiffs complaint
on the grounds that the
complaint fails to state a cause of action. If John failed to
allege that he had a valid
title to the parcel, then Sheila could seek immediate
dismissal on this ground.
Second, cases can be decided on the pleadings. If Sheila
in her answer alleges facts
that amount to only 18 years of possession, then John
could seek a judgment based on
the pleadings. The PL stage is concerned with the
rules that set forth the
required elements of legal claims and defenses; and verifying
that each side alleges a
minimal set of facts that satisfy the requirements of these
rules.Modeling this stage is
more an exercise in rule-based reasoning because these
requirements are set forth in
statutes or rules that may be inferred from the holdings
of decided cases.
In the pre-verdict stage (PV),
after completing the process of discovery and
examining depositions,
interrogatories, plot plans, photographs, etc., a court might
grant Johns motion for
summary judgment on the grounds that as a matter of law
the parcel was open woodland
and Max failed to enclose or cultivate the land. Or,
after the presentation of all
the evidence at a trial, a judge might direct a verdict
in Johns favor on the same
grounds. Or, the trial judge might allow the case to go
to a jury, and after a jury
finding in favor of Sheila might grant Johns motion for
judgment notwithstanding the
jury verdict on the same grounds. Judges use judgments
notwithstanding the verdict
for reasons of judicial economy. Often, one side
may have a very weak case and
the judge believes a jury will find for the stronger
side on issues of fact,
thereby making an appeal very difficult to win. Second, if
the trial judges directed
verdict is overturned on appeal, then the case must be
re-tried; if the trial judge
instead grants a judgment notwithstanding the verdict
54 CAROLE D. HAFNER AND DONALD H.
BERMAN
then on appeal the jurys
verdict can be re-instated. Although subtle differences
exist among summary judgment,
directed verdict, and judgment notwithstanding
the verdict, these three
methods of disposing of cases can be treated the same for
purposes of our analysis,
since in all three situations the court rules as a matter of
law that the evidence offered
is insufficient to create an issue for a trier of fact to
decide. The trial verdict
stage (TV) involves a decision by the trier of fact, be it a
judge or jury. If the two
sides present witnesses who give conflicting testimony, the
trier of fact must decide
which testimony is most convincing. The trier of fact must
also decide such mixed
questions of fact and law as whether the activities of Max
from 1969-1975 amounted to
adverse possession.
5.2.2. Classification of legal
results
The formal decision-making
procedures described above define a complex array
of legal questions that come
before appellate courts, and the significance of an
appellate decision can only be
understood in its procedural context. Synopses found
in case reporters and digests
typically begin with the procedural context of the case,
as illustrated by the case
descriptions presented in Section 5.1. The rows of Table
III represent legal decisions
under a variety of procedural scenarios; although they
constitute only a subset of
the actual scenarios that confront the expert attorney,
they are sufficient to
demonstrate how procedural context influences legal casebased
reasoning.
Following the three-stage
model of legal decision making, we can (assuming
the equivalences of certain
dispositions) represent legal results using three basic
predicates:22
1. PL(Side1, C) Side 1 of
case C is entitled to judgment at the pleading stage.
2. PV(Side1, C) Side 1 of
case C is entitled to judgment at the pre-verdict stage.
3. TV(Side1, C) A trial
verdict in favor of Side1 in case C should be upheld.
Assuming Side1 and Side2
represent opposing parties, and the result of a case
must be a positive or negative
instance of one of the three basic predicates, there are
12 legal propositions to
choose from in characterizing the result of a case. However,
the equivalence:
TV(Side1, C).PV(Side2, C)
(i.e., reversal of a trial
verdict in favor of Side 1 is legally equivalent to a directed
verdict for Side 2) reduces
this to only 10 distinct legal results. Table 3 enumerates
the 32 common legal scenarios
that arise from our 3-stage model of legal decision
making; the rightmost column
shows the legal result type associated with each
scenario. (Due to the
equivalence shown above, there are no negative instances of
the TV predicate.)
Let us consider how the cases
described in Section 5.1 would be categorized.
Since the party claiming by
adverse possession may be either the plaintiff or defendant,
we will use AP and RT to
represent the two sides instead of the traditional
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 55
P and D. In Senn, the plaintiff (AP) won a
trial verdict, but the trial judge granted
RTs motion to set aside the
verdict, and this was affirmed on appeal. This case
corresponds to scenario 4 in
Table III ( equivalent to a directed verdict for the
defendant), so the legal
result is: PV(RT, Senn). In Kershaw, the defendant (AP)
won a trial verdict and it was
upheld, which corresponds to scenario 1 and legal
result TV(AP, Kershaw).
5.2.3. Procedure-based rules of
relevance
To complete our analysis of
procedural context in case-based legal reasoning, we
consider how the precedential
value of a prior case depends on the result type of the
prior case and the result type
we want to obtain in the current case. For example,
Kershaw is a highly relevant precedent
in arguing against a directed verdict in favor
of John, since it holds that
adverse possession is a question of fact. However, if a
jury trial occurred and John
won, Kershaw would not be a highly relevant
precedent
in arguing that the verdict
should be set aside. So, it is not sufficient to classify
Kershaw as a result that favors the
adverse possessor, as would be done in most
legal case-based reasoning
models. In some procedural contexts, such as an appeal
by Sheila of a jury verdict
favoring John, Kershaw
with its
strong statement about
adverse possession being a
question of fact could by cited by John in arguing that
the verdict favoring him
should be upheld!
Of course, procedural
relevance alone cannot make a strong precedent; it is also
important that the cases have
similar fact situations. In the discussion that follows,
we assume that the prior case
is factually similar to the current case, and consider
only the effect of procedural
context on the strength of the precedent. In Section
5.3, we consider how
procedural and factual relevance judgments can be combined
in a single algorithm for
determining the most relevant prior case.
Table IV shows procedure-based
rules of relevance. Assuming factual similarity,
the procedural relevance of a
prior case to a new case will be classified as
strong, weak, or not relevant.
An example of a strongly supportive precedent would
be the use of Kershaw by Sheila in a scenario where
John appeals a trial verdict
in Sheilas favor. An example
of a weakly supportive precedent would be the use
of Kershaw by John in a scenario where
Sheila appeals a trial verdict in Johns
favor. We claim that a case
which upholds a trial verdict in favor of one side may
(depending on the language of
the decision) have precedential value for upholding
a trial verdict in favor of
either side.
5.2.4. Integrating procedural
context into case-based legal reasoning
To date, models of case-based
legal reasoning have concentrated on representing
and matching the substantive
facts and issues, and have not taken account of procedural
variations. As described in
Section 2, in the HYPO system (Ashley 1990;
Rissland and Ashley 1987),
cases are described and indexed by legally relevant
56 CAROLE D. HAFNER AND DONALD H.
BERMAN
Table III. A catalog of procedural
scenarios
Procedural scenario Precedent
type
1. Side2 wins jury verdict
Side1 appeals Verdict upheld TV(Side2, c)
2. Verdict reversed PV(Side1,
c)
3. Side1 wins jury verdict
Side2 appeals Verdict upheld TV(Side1, c)
4. Verdict reversed PV(Side2,
c)
5. Judgment NOV for Side2 Side1
appeals judgment upheld PV(Side2, c)
6. Judgment reversed TV(Side1,
c)
7. Judgment NOV for Side1 Side2
appeals judgment upheld PV(Side1, c)
8. Judgment reversed TV(Side2,
c)
9. Side1 moves for Directed
Verdict Granted Affirmed PV(Side1, c)
(after trial)
10. Reversed. PV(Side1,
c)
11. (equivalent to scenario 1)
Not granted Affirmed TV(Side2, c)
12. (equivalent to scenario 2)
Reversed PV(Side1, c)
13. Side2 moves for Directed
Verdict Granted Affirmed PV(Side2, c)
(after trial)
14. Reversed. PV(Side2,
c)
15. (equivalent to scenario 3)
Not granted Affirmed TV(Side1, c)
16. (equivalent to scenario 4)
Reversed PV(Side2, c)
17. Side1 moves for summary
judgment Granted Affirmed PV(Side1, c)
(before trial)
18. Reversed. PV(Side1,
c)
19. Not granted Affirmed. PV(Side1,
c)
20. Reversed PV(Side1, c)
21. Side2 moves for summary
judgment Granted Affirmed PV(Side2, c)
(before trial)
22. Reversed. PV(Side2,
c)
23. Not granted Affirmed. PV(Side2,
c)
24. Reversed PV(Side2, c)
25. Side1 moves for judgment on
pleadings Granted Affirmed PL(Side1, c)
26. Reversed. PL(Side1,
c)
27. Not granted Affirmed. PL(Side1,
c)
28. Reversed PL(Side1, c)
29. Side2 moves for judgment on
pleadings Granted Affirmed PL(Side2, c)
30. Reversed. PL(Side2,
c)
31. Not granted Affirmed. PL(Side2,
c)
32. Reversed PL(Side2, c)
.Remanded for additional lower-court proceedings.
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 57
Table IV. Procedure-based rules of
relevance
Prior results Desired result
Procedural relevance
PL(Side1, PC) PL(Side1, CC)
Strong
PV(Side1, PC) PV(Side1, CC)
Strong
TV(Side1, PC) TV(Side1, CC)
Strong
PL(Side1, PC) PL(Side1,
CC) Strong
PV(Side1, PC) PV(Side1,
CC) Strong
A prior holding of a basic
result is a strong precedent for the
same result in a later case.
PV(Side2, PC) TV(Side1, CC) Strong
TV(Side2, PC) PV(Side1, CC) Strong
A prior holding that Side2 is
not entitled to a directed
verdict is a strong precedent
for upholding a trial verdict
for Side1 and vice versa.
PV(Side2, PC) PV(Side1, CC) Strong
PL(Side2, PC) PL(Side1, CC) Strong
A prior holding that Side2
should win as a matter of law is a
strong precedent against Side1
achieving the same result.
PL(Side1, PC) PV(Side1, CC)
Strong
PV(Side1, PC) TV(Side1, CC)
Strong
PV(Side1, PC) PL(Side1,
CC) Strong
PV(Side2, PC) PL(Side1, CC) Strong
TV(Side2, PC) PL(Side1, CC) Strong
A stronger prior result is a
precedent for a weaker result
favoring the same side. For
example, a case that rules Side1
is entitled to summary
judgement can be used to argue it is
entitled a fortiori to have a verdict in its favor
affirmed.
PV(Side2, PC) PV(Side1,
CC) Weak
TV(Side2, PC) TV(Side1, CC)
Weak
TV(Side1, PC) PV(Side1, CC) Weak
PV(Side1, PC) TV(Side1, CC) Weak
A prior holding that Side2 is
not entitled summary judgment, and
that moves the case forward to
trial, may be useful to argue that
Side1 is also not entitled to
summary judgment, especially if it
says certain issues must be
decided by the trier of fact. The same
argument applies when a trial
verdict is upheld. However these
precedents are not as strong as
a prior case whose holding favors
the same side as the current
case.
58 CAROLE D. HAFNER AND DONALD H.
BERMAN
factors or dimensions, which
are used to search for relevant precedents, and to
construct legal arguments.
Above we have described the
complexity of actual reported cases, in which
holdings on fact-based issues
are intermingled with procedural rulings, and we
have argued that there are at
least 10 distinct types of results or holdings that need
to be represented in a
case-based legal reasoning system. Assuming the existence
of a system along the lines of
HYPO for creating a similarity ranking of cases using
factors, it is reasonable to
ask how our model of procedural context could be used
to improve it.
One simple approach would be
to continue to build the similarity ranking,
without regard to procedural
context, select the factually on point cases, and
then use the legal result type
as a filter to select those that are procedurally relevant
as well. However, from the
standpoint of efficiency, in a practical system with large
numbers of precedents, ranking
the precedents based on factual similarity requires
a great deal of computation.
In circumstances where many of those cases may not
be procedurally relevant, it
would be more efficient to apply the filter first, and then
rank those cases known to be
procedurally relevant.
Following is an outline of an
algorithm for selecting procedurally relevant
precedents to be included in a
fact-based similarity ranking.
1. Assign one party to Side1,
and determine the result type(s) to be advocated by
that party.
2. Select rows from Table IV,
where the value of the center column, Desired
Result, is one of the result
types from Step1.
3. Make a list of the values
in the leftmost column of the selected rows, and mark
them to indicate whether they
provide strong or weak support.
4. Select cases from the
database whose result type is a member of the list created
in Step 3, and mark them
according to whether they provide strong or weak
support for Side1.
Steps 58. Repeat Steps 14,
but use the negations of the result types from Step 1.
The selected cases comprise
the input to the fact-based matching process that results
in a relevance ranking of
prior cases. For example, if Sheilas attorney wants
to find precedents that
support a motion to dismiss Johns claim at the pleading
stage, the basic result
corresponding to this motion (assigning Side1 to Sheila) is
PL(Side1, CC) and the opposing
result is PL(Side1, CC). In Step 2, the
following
row would be retrieved:
PL(Side1, PC) PL(Side1, CC)
strong
In Step 6, the following rows
would be retrieved:
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 59
PL(Side2, PC) PL(Side1, CC) strong
PV(Side1, PC) PL(Side1, CC) strong
PV(Side2, PC) PL(Side1, CC) strong
TV(Side2, PC) PL(Side1, CC) strong
The results of filtering would
restrict substantive analysis to six of the 10 result
types, since the other four result
types would not be relevant precedents for this
motion. For example, the fact
that a trial verdict was directed in favor of an adverse
possessor in a prior case
(PV(Side1, PC)) does not imply that in a similar case
Sheila should (or should not)
prevail at the pleading stage, since a judgment at the
pleading stage is a stronger
result than a directed verdict. However, a case where
the court refused a motion for
a directed verdict in favor of the adverse possessor
(PV(Side1, PC)) implies that in a similar case
Sheila should not prevail at the
pleading stage, so Johns
attorney could cite such as case in arguing against Sheilas
motion.
Some procedure-driven
differences would also occur in the process of arguing
with cases once their factual
similarity with a current case has been evaluated.
In existing legal case-based
reasoning systems, the result or holding of a case is
represented as a binary
choice: case favors plaintiff or case favors defendant. This
may lead to the incorrect use
of precedents in some procedural contexts; for example,
a case such as Kershaw, whose result favored the
adverse possessor, would
be useless to the adverse
possessor (and weakly supportive to the other side) in
arguing a motion by the
adverse possessor for a summary judgment.
Finally, there is a need to
consider the implications of using a strong v. weak
classification of procedural
relevance, and its interaction with ranking of precedents
by factual similarity. If two
prior cases are equally on point (which often occurs
with a partial ordering
framework such as HYPO uses), then if only one has strong
procedural relevance, that
case is clearly a better precedent. However, further empirical
study is needed to analyze how
a case should viewed which is more on point
factually but only weakly
relevant procedurally.
5.3. IMPLICATIONS FOR LEGAL
ARGUMENTATION SYSTEMS
Lawyers spend many waking
hours reasoning from cases, with the goal of developing
persuasive legal arguments. To
contribute to the basic education of lawyers, or
to realize Ashleys dream of
mak[ing] a prototype for a Brief Writers Assistant
(Ashley 1988, p. 216)
computational models of cases must accurately represent
the procedural posture of the
case. For example, a system based on the three-stage
model proposed above could,
replicating a HYPO-style argument, generate the
following procedurally-correct
argument for Sheila:
POINT for adverse possessor (AP) in appealing a
directed verdict granted
to the record title holder
(RT):
60 CAROLE D. HAFNER AND DONALD H.
BERMAN
WHERE: AP and APs predecessor
entered under color of title, used the land
continually for a 20 year
period, paid taxes on the land, and placed markers at
the boundaries of the land, AP
should not be barred by law from a claim of
adverse possession.
CITE: In Kershaw, the court held that whether
entering undeveloped land,
paying taxes, and placing
markers at the boundary could amount to adverse
possession was a question of
fact.
RESPONSE for RT:
Kershaw is distinguishable because in Kershaw the tract of land was 1/8 of
an
acre. Max entered onto a 5
acre parcel.
CITE: Senn holds that as a matter of law,
to gain title of open woodlands by
adverse possession the
claimant must cultivate or enclose the land and AP did
not cultivate or enclose this
parcel.
REBUTTAL for AP:
Senn is distinguishable because in
Senn the tract of land was 56 acres. Max
entered a 5 acre tract, which
is not, as a matter of law, open woodlands within
the meaning of Senn.
Although the robustness of
case-based legal reasoning is enhanced by the model
of procedural context proposed
above, there are still many issues that must be
addressed to create an
adequate model of the procedural posture of cases. The
parties to a dispute dont
always fall into the standard opposing categories: for
example, some cases involve
two adverse possessors, neither of whom can claim
record title ownership. The
relief being sought must also be considered: often
plaintiffs seek a mix of
declaratory relief, damages, and a specific relief such as
injunctions, ejectment or
mandamus. The most serious impediment to an accurate
computational model of
procedural context is the lack of a binary set of trial court
and appellate court
dispositions. Appellate courts can choose from an array of
dispositions by affirming,
reversing, remanding, affirming in part and remanding
in part, reversing in part and
affirming in part, or remanding in part and reversing
in part. And appellate courts
can reverse or affirm on grounds that have nothing to
do with the sufficiency of the
pleadings or the evidence: for example on the grounds
that inadequate instructions
were given to the jury.
A truly robust system for
case-based legal reasoning must be able to precisely
represent these complex
variations, and draw the correct inferences from them
when considering the
likelihood that a prior case will be judged relevant by a later
court, and if it is judged
relevant, how it will influence the courts rulings.
6. Conclusions
We have argued that the
contextual aspects of legal decisions - procedural, teleological
and temporal are important
components of case-based legal reasoning,
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 61
and we have begun here to
explore ways of incorporating context into case-based
reasoning systems. We have
shown how these extended computational models can
be used to make legal analysis
more accurate and legal arguments more realistic.
Though incomplete, our
proposed context-informed arguments more closely
approximate real world
advocacy in common law jurisdictions. Granted, there is
a limit to the levels of
abstraction that one could incorporate into a case-based
reasoner, so we cannot
replicate the full range of argument moves found in the
arsenal of a truly gifted
advocate. Nonetheless, we anticipate that case-based
reasoners incorporating
contextual knowledge will prove useful to less experienced
attorneys, and to legal
educators who want to enhance the advocacy skills of their
students (Aleven and Ashley
1994).
The formal representations and
algorithms proposed in this paper have the potential
to enhance the robustness of
legal case-based reasoning systems; however
they are only a start. There
is much basic research remaining to be done, to understand
fully the contextual
influences on the use of legal precedents and their
potential for computational
realization. The analyses of procedural, teleological,
and temporal dimensions of
context need to be elaborated and tested using other
legal domains. The influence
of other contextual dimensions also needs to be considered,
for example, the influence of
legal jurisdiction is increasingly significant
in the global economy. The
social and economic characteristics of the parties to
a legal dispute is another
aspect of context that cannot be ignored when trying to
understand the significance of
legal precedents.
The analysis presented in this
paper raises many more questions than it answers
questions that ought to be
the subject of continuing research by the artificial
intelligence and law
community.
Notes
1 The material in this paper has
been edited and expanded by the first author after the untimely death
of Donald Berman in 1997 (see
the Editors Introduction and other articles in this memorial volume
for tributes to Don and
discussions of his many contributions). The three conference papers are
presented in Sections 35
below, with no substantive additions, but with some modifications for the
sake of clarity, brevity, and
editorial unity. This Introduction (Section 1), Section 2, and Section 6 are
new. For each of the three
conference-paper-based Sections, a revised Introduction puts the material
to follow into context and a
few selected additional (and later) references are included besides those
that appeared in the original
version. There is no attempt, however, to provide a general review of the
state of the art in case-based
reasoning legal reasoning since 1995. The 1993 paper, which generated
the most comment and interest
from the AI and Law community, is presented first in Section 3.
That paper has been shortened
by removing a second set of example cases, which analyzed the
open-textured concept of an
event occurring in the course of employment from U.S. workers
compensation law.
2 Some more recent legal CBR
systems, notably BANKXX (Rissland et al. 1996) include other
forms of remembered experience
(such as argument moves) in addition to judicial opinions in their
knowledge bases.
3 The algorithm for computing
similarity need not be numerical; in HYPO, for instance, it is based
on inclusiveness of sets of
shared dimensions, as described below.
62 CAROLE D. HAFNER AND DONALD H.
BERMAN
4 HYPO evolved over the time period
of its development (approximately 19841990); the description
below and other references to
the HYPO model in this paper are based on the doctoral thesis
of Kevin Ashley (1990). A
historical analysis of HYPOs development and its influence on future
systems appears in Rissland and
Ashley (2002) in this volume. That article also points out some
differences between HYPOs
model of dimensions and the approach used in Sections 3 through 5 of
this paper.
5 The classification of factors
or dimensions as favoring one side of a legal dispute was not part of
earlier (pre-1990) versions of
HYPO (see Rissland and Ashley 2002 in this volume). In this paper,
when we say a factor favors a
side, we mean that a value of true (for a binary factor) or a larger value
(for a quantitative factor)
favors that side.
6 See Ashley (1990, pp. 147154.)
7 3 Cai. R. 175, 2 Am Dec 264
(Supreme Court of New York 1805).
8 11 East 574, 103 Eng. Rep 1127
(Queens Bench 1707).
9 1 Dav. and Mer. 592, 6 Q.B. 606
(1844).
10 We have inferred from the
reported opinions the policy arguments that were considered in Young
because cases were argued
orally, and written briefs are not available. Also, since opinions were
delivered orally and
transcribed by reporters, the opinions tended to be rather short. Finally,
common
law judges used a form of
verbal shorthand that embodied their policy determinations. As evidence
that these kinds of arguments
are routinely considered in law school classes, see Dukeminier and
Krier (1988, pp. 820).
11 This is an example of how the procedural context of a legal dispute can
determine the result, as
discussed in Section 5 below.
12 Ontario Re. Stat. (1960), ch.
172, Sec. 105.
13 12 N.Y.2d 473, 191 N.E.2d 279,
473 N.Y.S.2d 279 (1963).
14 277 N.Y. 407, 14 N.E.2d 455
(1938).
15 294 N.Y. 288, 62 N.E. 2d 74
(1945).
16 Ward et al. v. Strobel, 300 N.Y. 294, 90 N.E.2d 467
(1949), Fortuna
v. Sangster,
59 N.Y. S. 2d 190
(1945), Harry Turnowski v. Aaron
Turnowski, 33
Misc 2d 864, 226 N.Y.S.2d 738 (1962).
17 5 N.Y.2d 1016.
18 308 N.Y. 155, 124 N.E.2d 99
(1954).
19 9 N.Y.2d 554, 175 N.E.2d 441,
216 N.Y.S.2d 65 (1961).
20 9 N.Y.2d 34, 172 N.E.2d 526,
211 N.Y.S.2d 133 (1961).
21 24 N.Y.2d 569, 301 N.Y.S.2d
519, 249 N.E. 2d 394 (1969).
22 In applying this model to a
particular legal domain, Side1 and Side2 can be replaced by more
specific descriptions such as
Plaintiff and Defendant (P and D), or Adverse-Possessor and Record-
Title-Holder (AP and RT).
References
Ashley, K. D. (1990).Modeling
Legal Argument: Reasoning with Cases and Hypotheticals. Bradford
Books/MIT Press: Cambridge, MA.
Ashley, K. D. (1988). Arguing
by Analogy in Law: A Case-Based Model. In Helman, D. (ed.),
Analogical Reasoning, 205224.
Kluwer Academic Publishers: Dordrecht, The Netherlands.
Ashley, K. D. and Rissland, E.
L. (1987). But, See, Accord: Generating Blue Book Citations in
HYPO. In Proceedings of the
First International Conference on Artificial Intelligence and Law,
6774. ACM Press: New York.
Aleven, V. and Ashley, K. D.
(1994). An Instructional Environment for Practicing Argumentation
Skills. In Proceedings of the
10th National Conference on Artificial Intelligence (AAAI-94),
485492. AAAI Press: Menlo
Park, CA.
THE ROLE OF CONTEXT IN
CASE-BASED LEGAL REASONING 63
Berman, D. H. and Hafner, C. D.
(1991). Incorporating Procedural Context into a Model of
Case-Based Legal Reasoning. In
Proceedings of the 3rd International Conference on Artificial
Intelligence and Law, 1220.
ACM Press: New York.
Berman, D. H. and Hafner, C. D.
(1993). Representing Teleological Structure in Case-Based Legal
Reasoning: The Missing Link. In
Proceedings of the 4th International Conference on Artificial
Intelligence and Law, 5060.
ACM Press: New York.
Berman, D. H. and Hafner, C. D.
(1995). Understanding Precedents in a Temporal Context of
Evolving Legal Doctrine. In
Proceedings of the 5th International Conference on Artificial
Intelligence and Law, 4251.
ACM Press: New York.
Berman, D. H. (1989). The Need
for a Medium Grain Choice of Law Filter in Certain Legal Expert
Systems. In Martino, A. (ed.)
Pre-Proceedings of the Third International Conference on Logica
Informatica Diritto, 6581.
Consiglio Nazionale delle Richerche: Florence.
Branting, K. (1991). Reasoning
with Portions of Precedents. In Proceedings of the 3rd International
Conference on Artificial
Intelligence and Law, 145154. ACM Press: New York.
Branting, K. (1994). A
Computational Model of Ratio Decidendi. Artificial Intelligence and Law
2(1): 131.
Cuthill, B. and McCartney, R.
(1994). Issue Spotting in CHASER. Artificial Intelligence and Law
2(2): 83111.
Dukeminier, J. and Krier, J.
(1988). Teachers Manual, Property Law, 2nd ed. Little, Brown: Boston,
MA.
Frank, J. (1930). Law and the
Modern Mind. Brentanos: New York.
Fuller, L. (1958). Positivism
and Fidelity to Law A Reply to Professor Hart. 71 Harvard Law
Review 630
Gardner, A. (1987). An
Artificial Intelligence Approach to Legal Reasoning. Bradford Books/MIT
Press: Cambridge, MA.
Gordon, T. F. (1994). The
Pleadings Game: An Exercise in Computational Dialectics. Artificial
Intelligence and Law 2(4):
239292.
Groendijk, C. and Oskamp, A.
(1993). Case Recognition and Strategy Classification. In Proceedings
of the 4th International Conference
on Artificial Intelligence and Law, 125132. ACM Press:
New York.
Hafner, C. D. (1981). An
Information Retrieval System Based on a Computer Model of Legal
Knowledge. Ph.D. Thesis, The
University of Michigan. UMI Research Press: Ann Arbor, MI.
Hafner, C. D. (1987).
Conceptual Organization of Case Law Knowledge Bases. In Proceedings of
the First International
Conference on Artificial Intelligence and Law, 3542. ACM Press: New
York.
Hart, H. L. A. (1958).
Positivism and the Separation of Law and Morals. 71 Harvard Law Review
593.
Hayes-Roth, F.,Waterman, D. A.,
and Lenat. D. B. (eds.) (1983). Building Expert Systems. Addison-
Wesley: Reading, MA.
Hage, J. C., Leenes, R., and
Lodder, A. N. (1994). Hard Cases: A Procedural Approach. Artificial
Intelligence and Law 2(2):
113167.
Hohfeld, W. N. (1913).
Fundamental Legal Conceptions as Applied in Judicial Reasoning. 23 Yale
Law J. 16.
Holmes, O. (1897). The Path of
the Law. Harvard Law Review 10: 457.
Kennedy, D. (1991). A Semiotics
of Legal Argument. Syracuse Law Review 42(1): 75116.
Kuhn, T. (1970). The Structure
of Scientific Revolutions. University of Chicago Press: Chicago, IL.
Leake, D. B. (ed.) (1996).
Case-Based Reasoning: Experiences, Lessons and Future Directions.
AAAI Press/MIT Press:
Cambridge, MA.
Loui, R. P. and Norman, J.
(1995). Rationales and Argument Moves. Artificial Intelligence and Law
3(3): 159185.
64 CAROLE D. HAFNER AND DONALD H.
BERMAN
McCarty, L. T. and Sridharan,
N. S. (1981). The Representation of an Evolving System of Legal
Concepts II: Prototypes and
Deformations. Proceedings of IJCAI-81, 246253.
Prakken, H. and Sartor, G.
(1997). A Dialectical Model of Assessing Conflicting Arguments in Legal
Reasoning. In Prakken, H. and
Sartor, G. (eds.), Logical Models of Legal Argumentation, 175
212. Kluwer Academic
Publishers: Dordrecht, The Netherlands.
Riesbeck, C. K. and Schank, R.
C. (1989). Inside Case-Based Reasoning. Lawrence Erlbaum:
Hillsdale, N.J.
Rissland, E. L. (1985).
Argument Moves and Hypotheticals. In Walters, C. (ed.), Computing Power
and Legal Reasoning, 129143.
West Publishing Co.: St. Paul, MN.
Rissland, E. L., Valcarce, E.
M., and Ashley, K. D. (1984). Explaining and Arguing with Examples.
Proceedings of the Fourth
National on Artificial Intelligence (AAAI-84), 288294, Austin, TX.
Rissland, E. L. and Ashley, K.
D. (1987). A Case-Based System for Trade Secrets Law. In Proceedings
of the First International
Conference on Artificial Intelligence and Law, 6167. ACM Press:
New York.
Rissland, E. L. and Skalak, D.
B. (1991) CABARET: Statutory Interpretation in a Hybrid Architecture.
International Journal of
Man-Machine Studies (34): 839887.
Rissland, E. L., Skalak, D. B.,
and Friedman, M. T. (1996). BankXX: Supporting Legal Arguments
through Heuristic Retrieval.
Artificial Intelligence and Law 4(1): 171.
Rissland, E. L. and Ashley, K.
D. (2002). A Note on Dimensions and Factors. Artificial Intelligence
and Law 10.
Rose, D. E. (1994). A Symbolic
and Connectionist Approach to Legal Information Retrieval.
Lawrence Erlbaum: Hillsdale,
NJ.
Schank, R. and Leake, D.
(1990). Creativity and Learning in a Case-based Explainer. In Carbonnell,
J. (ed.), Machine Learning:
Paradigms and Methods, 354385. MIT Press: Cambridge, MA.
Sanders, K. (1994). CHIRON:
Planning in an Open-Textured Domain. Ph.D. diss., Technical Report
94-38, Computer Science Dept.,
Brown University, Providence RI.
Skalak, D. and Rissland, E. L.
(1992) Arguments and Cases: An Inevitable Intertwining. Artificial
Intelligence and Law 1(1): 34.
Smith, J. C. and Deedman, C.
(1987). The Application of Expert Systems Technology to Case-Based
Law. In Proceedings of the
First International Conference on Artificial Intelligence and Law,
8493. ACM Press: New York.
Vossos, G., Zeleznikow, J.,
Dillon, T., and Vossos, V. (1991). An Example of Integrating Legal Case
Based Reasoning with Object
Oriented Rule-Based Systems: IKBALS II. In Proceedings of the
3rd International Conference on
Artificial Intelligence and Law, 3141. ACM Press: New York.
Wren, C. and Wren. J. (1986).
The Legal Research Manual: A Game Plan for Legal Research and
Analysis. Abrams and Ambrose:
Madison, WI.
Retirado de: http://www.kluweronline.com