Legal Reasoning Models
Carole D. Hafner
College of Computer Science
Northeastern University
Boston, MA 02115
hafner@ccs.neu.edu
1. Introduction
Research in legal reasoning models investigates formal and
computational theories of how legal experts analyze problems, create arguments,
and make decisions. 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 infinite variety of particular facts on which
legal decisions turn make legal reasoning 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.
With the advent of computers and artificial intelligence, both the range
of techniques available and the practical motivation for creating legal
reasoning models have grown. Rule-based systems using logical inference techniques
have been developed and used successfully in a number of legal domains,
especially those dominated by complex regulations, such as taxation and social
benefits administration. (Johnson and Mead 1991). A unique challenge in
developing legal rule-based systems is the need to model the relationship
between the logical representation and its textual sources (Bench-Capon and
Coenen 1992). This need arises first, because legal experts (and legal
expert systems) must be able to cite the authoritative texts on which their
conclusions are based, and second, to enable legal knowledge bases to be
efficiently updated when laws and regulations are amended.
Despite the successful adaptation of rule-based reasoning models to the
legal domain, it is the view of most researchers that the essence of legal
reasoning is its open-textured and indeterminate nature (Berman and Hafner
1987). Two essential cognitive abililties for legal reasoning models include:
-- reasoning with cases to interpret open-textured or conflicting legal
rules and concepts
-- adversarial reasoning: the ability to create persuasive
arguments for both sides of an issue
2. Case-based Legal Reasoning Models
Case-based reasoning research in artificial intelligence (see also
Problem solving:Case-based) is concerned with the use of remembered prior
experiences in analyzing new situations and solving problems. The
distinguishing characteristic of case-based reasoning (CBR) 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.
Research in case-based legal reasoning 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 application of a prior case to answer legal questions about
a new case. In legal CBR models, a set of domain-dependent, legally-relevant
features is defined, and a prior case is considered more or less on point depending on the degree of
match between the features of the prior case and those of the new case. Computational
models of legal CBR are concerned with the following problems: representation
of cases, including their facts, issues, holdings
(the judicial conclusions they contain), and results; organization of the case
database, including the indexing structure and retrieval algorithms; ranking of
retrieved cases by their similarity to a new case; and creation of case-based
legal arguments.
The first legal CBR model that addressed all of these problems was
embodied in HYPO (Rissland and Ashley 1987, Ashley 1990), a computer program
for analyzing cases and constructing legal arguments in the domain of trade
secrets law. In HYPO, legally-relevant features called dimensions are used to index and retrieve cases. HYPO’s legal
domain knowledge includes, for each feature, which side of a dispute would be
favored by the feature’s having particular values (e.g., a true value for predicates, or a larger value for number-valued
features), and how the feature’s value is related to a case’s underlying facts.
For example, in the trade secret domain, one such feature is Generally-Known, indicating that the alleged
secret information is generally known within the industry. When a new case is
presented to HYPO, the features 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
feature-values with the new case as its immediate successors. These immediate
successors are taken to be the most-on-point
cases.
Later research extended the use of HYPO-style CBR in several ways. In
the GREBE system (Branting 1991), a representation of the ratio decidendi (the
rationale or justification) of cases
was used to identify portions of a prior case’s holdings that were relevant to
a new case, even when other portions of the prior case were not on-point. Branting’s
model addressed a well-known weakness of the pure feature-based approach to
CBR: its failure to represent the reasoning that connects the features with the
results of a case. The logic underlying a legal decision is implicitly encoded
in GREBE by grouping the features of a precedent according to the holdings they
influence. Berman and Hafner (1993) addressed another aspect of legal CBR: the
need for legal reasoning models to include the societal goals and policies that
motivate legal rules and decisions. They proposed an extended representation of
cases which incorporates knowledge about legal purposes.
Several researchers have developed models of legal reasoning in which
rules and cases are used in combination. Gardner (1987) used a logic tree to
model the top-down, rule-based analysis of a legal situation. At the leaves of
the tree are undefined concepts, linked to cases providing positive and
negative examples. In the CABARET system, Skalak and Rissland (1991) described
a mixed strategy for legal argumentation using both rules and cases. When a
rule-based strategy produces a result favoring the opposing side, CABARET uses
precedents to support argument moves such as the broadening of a rule’s
interpretation or the discrediting of a rule.
3. Models of Legal Argumentation
Argumentation is considered by many law teachers and legal scholars to
be the quintessential element of a lawyer’s cognitive skill. Thus, many models of legal argumentation
have been developed, falling into three broad categories: case-based,
logic-based and legal discourse models.
Case-based models of legal argumentation focus on the process of arguing
with precedents, using techniques such as analogizing, distinguishing and
trumping. A case-based model of legal argumentation was developed as part of
the HYPO system described above. In HYPO, after the claim lattice is created to
determine the most-on-point precedents, HYPO constructs a 3-ply argument for
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 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 point for Side1, HYPO identifies the
most on point case whose result favored Side1, and makes an argument by
analogy, noting the similarity between the new case and the prior case. Responses
and rebuttals use the legal argument techniques of distinguishing by focusing on differences rather than similarities,
and trumping by presenting more
on-point counter-examples. (Ashley 1990, p. 71).
Logic-based models of legal argumentation begin by expressing arguments
as formal proofs within a system of logic, and then focus on the common
situation in law where two logical arguments exist which reach contradictory
results. In order to resolve such contradictions, it is necessary to adopt a
framework of defeasible logic (Reiter 1980), where rules may have exceptions,
and beliefs may be defeated by the acquisition of new, conflicting knowledge. For
example, a legal rule “No vehicles are permitted in the park” may co-exist with
another legal rule “Emergency vehicles are permitted to violate traffic
laws”. Two logical arguments can then
be constructed regarding an ambulance that entered the park: one concluding
that the ambulance violated the law, the other concluding that the ambulance
acted legally.
A logical argumentation framework is described in (Prakken 1993), where
two conflicting arguments, represented as formal proofs, can be compared using
one or more ordering principles applied to the legal premises of the two
arguments. (Traditional ordering principles found in legal texts include Lex Superior (e.g., federal laws trump
state laws), Lex Specialis (more
specific laws trump more general laws), and Lex
posterior (newer laws trump older laws); however Prakken’s framework allows
for any criterion whatever that can prefer one argument to another.) Given a set of ordering principles,
Prakken’s framework classifies legal arguments into three categories: justified
arguments (which cannot be defeated), defeated arguments, and defensible
arguments (which are neither justified nor defeated).
Legal discourse models are concerned with the structural and semantic
characteristics that give a legal argument legitimacy. A number of
argumentation models have been developed
based on Toulmin’s (1958) theory of justified arguments. In
Toulmin’s theory, an argument’s conclusion is called the claim; the grounds are
the facts asserted to support the claim; the warrant is a bridge-like statement (e.g., a legal rule) that
provides the logical connection between the grounds and the claim; and the backing is the authority for the warrant
(e.g., a statute or regulation).
Loui and Norman (1995) have developed a discourse-based model of legal
argumentation that focuses on defining categories of rationales (i.e. warrants)
used in adversarial legal arguments, and the forms of attack appropriate for
each. They identify several types of legal rationales: for example, the compression rationale justifies a simple
and general rule as a normal-case simplification of a more complex and specific
one. An example of a compression rationale described by Loui and Norman (p.
161) is the rule “vehicles are not
allowed in the park”, which (one might argue) is a normal-case simplification
of the more complex rule: “vehicles used for private transportation are not
allowed in the park”, since normally vehicles are used for private
transportation. A straightforward strategy for attacking the “no-vehicles” rule
(as in the case of the Memorial Day tank display) is to re-state the rule in
its uncompressed form, and argue that the case at hand falls outside the normal
category of private transportation.
4. Conclusions and Future Directions
The field of law offers a fertile domain for the study of important
problems in artificial intelligence and cognitive modeling, such as using prior
experiences to solve new problems, reasoning with open-textured concepts, and
making persuasive arguments. Promising new directions for research include the
development of formal legal ontologies and their use in knowledge-based
systems (Visser and Bench-Capon) and the use of explicit legal concepts and
models (such as contracts) to govern the behavior of systems for electronic
commerce.
5. Bibliography
Ashley K D 1990 Modeling Legal
Argument: Reasoning with Cases and Hypotheticals. Bradford Books/MIT Press,
Cambridge, MA
Bench-Capon T J M, Coenen F P 1992 Isomorphism and Legal Knowledge Based
Systems. Artificial Intelligence and Law 1:
65-86
Berman D H, Hafner C D 1987 Indeterminacy: A Challenge to Logic-based
Models of Legal Reasoning. Yearbook of Law Computers and Technology Vol.
3:1-35. Butterworths, London.
Berman D H, Hafner C D 1993 Representing Teleological Structure in
Case-Based Legal Reasoning: The Missing Link.
Proc. 4th Intl. Conf. on Artificial Intelligence and Law, 50-60.
ACM Press, New York
Branting K 1991 Reasoning with Portions of Precedents. Proc. 3rd Intl. Conf. on Artificial Intelligence and Law,
145-154. ACM Press, New York
Frank J 1930 Law and the Modern
Mind. New York, Brentano’s
Fuller L 1958 Positivism and Fidelity to Law - A Reply to Professor
Hart. 71 Harv. L. Rev. 630
Gardner A vd L 1987 An Artificial
Intelligence Approach to Legal Reasoning.
Bradford Books/MIT Press, Cambridge, MA
Hart H L A 1958 Positivism and the Separation of Law and Morals. 71 Harv. L. Rev. 593
Hohfeld W N 1913 Fundamental
Legal Conceptions as Applied in Judicial Reasoning. 23 Yale Law J. 16
Johnson P, Mead D 1991 Legislative knowledge-based systems for public
administration – some practical issues. Proc.
3rd Intl. Conf. on Artificial Intelligence and Law, 108-117. ACM
Press, New York
Loui R P, Norman J 1995
Rationales and Argument Moves. Artificial Intelligence and Law 3(3):
159-185
Prakken H 1993 A logical framework for modelling legal argument. Proc. 4th Intl. Conf. on Artificial Intelligence and Law, 1-9.
ACM Press: New York.
Reiter R 1980 A logic for default reasoning. Artificial Intelligence
13:81-132
Rissland E L, Ashley K D 1987 A Case-Based System for Trade Secrets Law. Proc. 1st Intl. Conf. on
Artificial Intelligence and Law, 61-67. ACM Press, New York
Skalak D B, Rissland E L 1991 Argument moves in a rule-guided domain. Proc. 3rd Intl. Conf. on
Artificial Intelligence and Law, 1-11. ACM Press, New York
Toulmin S 1958 The Uses of
Argument. Cambridge University Press, Cambridge UK
Visser, P R S, Bench-Capon, T J M 1998 A comparison of four ontologies
for the design of legal knowledge-based systems. Artificial Intelligence and Law 6:27-57
Retirado de: http://www.aaai.org/AITopics/html/law.html