University of Twente
Public Administration and Public Policy
University of Limburg
Faculty of Law
In this paper it is argued why legal reasoning should be analysed as dialogue games,
rather than as logical proofs. This argument is complemented by a set of dialogue rules
which define these dialogue games, and an extended example in which the working of (part
of) these rules is illustrated.
Traditional rule-based legal expert systems are usually built around a theorem prover
that deduces conclusions from a set of facts and a set of rules. This approach to legal
problem solving is unsatisfactory for a number of reasons.
First, the focus on providing a logical proof as the solution to a presented problem is
too limited. In many cases, rules are not decisive [Cf. Gardner 1987]. If rules do not
decide the solution of a case, theorem provers that work with rules will not be able to
generate all legally valid conclusions. Second, even if one can logically deduce a
solution for a case, this does not provide certainty about the legal correctness of the
conclusion. In the law it is often possible to provide arguments against the conclusion of
a logically valid argument. As a consequence, legal arguments are defeasible [Cf. Hage
1987; Gordon 1988; Sartor 1991; Prakken 1993], and this makes the use of a traditional
theorem prover less attractive.
Thirdly, the knowledge base of traditional systems consists (most often) of undisputed
correct legal knowledge on the basis of which it operates. This is not very realistic in
the law, because there are very few domains which consist of rules that are not disputed
[De Vey Mestdagh et al 1991; Gordon 1993b].
Considerations like these, have led researchers to investigate the possibility to model
(legal) argument as dialogues. [Habermass 1973; Alexy 1978; Bench-Capon et al. 1992; Hage
et al 1992; Gordon 1993; Loui et al. 1993; Nitta et al. 1993]. This research resulted in
several models of dialogical reasoning. In this paper we formulate a dialogue model for
legal reasoning that is based on Reason Based Logic (RBL) [Hage 1993, 1994; Verheij 1994].
This logic is an attempt to overcome several of the shortcomings that face traditional
logic in dealing with legal reasoning. In particular, reasoning with rules, and with
reasons based on rules, is given special attention. RBL has a dialogical variant which is
characterised by having dialogue rules in addition to the rules which define valid
inference [Hage et al. 1992]. In this paper we explore the use of dialogical RBL by means
of a dialogue game.
At the University of Limburg, dialogue games are used as the basis for an intelligent
tutoring system. In this tutoring system the computer is a party in a debate with a law
student. We believe that dialogue games in which students have to construct legal
arguments form a useful method to teach legal skills [Lodder 1993].
The paper is organised as follows. First we present some previous work on legal dialogue
systems. Second, we give our motivation for taking a dialogical approach to legal
reasoning. Then we illustrate our dialogue model by means of an extended example and we
formulate a number of dialogue rules.
The use of dialogue models is gaining attention in the AI and law community.
Bench-Capon et al. [1991, 1992] are primarily interested in improving the way traditional
expert systems provide explanation. It appears that Toulmin's [1958] model of arguments
can be used to achieve this aim. The labeling of elements in the knowledge base on the
basis of a modified version of the Toulmin schema enables a meta-interpreter to select
relevant information for presentation to the user. If the user asks further explanation,
he is provided with more extensive information. Because the user can specify his interest,
he is given the information he wants, instead of a standard explanation.
Gordon [1993a, 1993b] constructed a system that simulates civil pleading: the Pleadings
game. The game is about two players who have a dispute about some civil case concerning a
priority conflict between two secured transactions. The purpose of the game is to identify
the issues of the case. The dialogue between the parties is governed by rules, which are
based on Alexy's discourse theory of legal argumentation [Alexy 1978]. The parties in the
dispute present arguments and backing for their arguments. Conflicts between arguments are
solved by arguing about the validity and priority of rules. MacKenzie [1979] aimed to
learn more about the problem of question-begging in dialogues. To explore this, he
constructed a dialogue game which he calls DC. Central to DC is that both players have a
commitment store which contains the locutions, both statements and challenges, a player is
committed to. Commitment is governed by commitment rules. The dialogue itself is guided by
dialogue rules, comparable to the rules mentioned by Alexy [1978]. Nitta et al. [1993]
describe HELIC-II: a parallel knowledge-based system for trial reasoning. In this model,
dialogues are between two agents (plaintiff, defendant) that have different goals, views
and reasoning strategies. A debate starts if the plaintiff gives his point of view. The
defendant chooses the best reasoning strategy in order to reach his goal: the refutation
of the plaintiff's viewpoint.
Our motivation for a dialogical approach to legal reasoning has the same background as the work of Gordon's. It is based on a procedural view of legal arguments. This view finds its inspiration in the work of John Rawls on procedural justice [John Rawls 1972, p. 85f] {1}. In particular we hold the opinion that the rightness of the outcome of a legal procedure depends solely on the correctness of the procedure followed. Legal procedures aim at, what might be called, 'pure procedural law'. To explain this expression, we must say a little more about the work of Rawls.
In practical affairs, we often adopt a particular procedure to obtain just results. In
this connection we can distinguish between two cases. Either there is an independent
standard for judging the outcome of the procedure, or there is no such standard. According
to Rawls, dividing a cake and a criminal trial are examples of the first case, gambling
and elections are examples of the second case.
If three persons divide a cake, a good procedure would be to have one of them cut the cake
in three pieces and let him choose a piece as the last person. Because of this procedure,
he will normally try to divide the cake into three equal pieces, thereby conforming to the
available independent standard, which says that each person is entitled to an equal piece
of the cake. In such a case, Rawls speaks of perfect procedural justice, because the
procedure more or less guarantees an outcome conform the pre-existing standard. The
outcome of perfect procedural justice is substantive justice.
According to Rawls, in a criminal process there is also a pre-existing standard for the
right outcome. If the defendant is guilty, he should be convicted, and otherwise not. A
criminal procedure should accomplish that as many guilty defendants as possible are
convicted and as many innocent defendants as possible are released. But there is no
independent way to establish whether the procedure succeeds in its aims. Therefore, even
if the procedure is made as just as possible, there is no guarantee that its outcome is
substantially just. That is why a criminal procedure is an instance of imperfect
procedural justice.
In the case of pure procedural justice, there is no independent standard to evaluate the
outcome of the procedure. Instead, the only standard to judge the outcome is the course of
the procedure. This is the case with elections, where the best way to evaluate the outcome
is to investigate whether the elections were held according to a (procedurally) just
procedure. (In this connection, Rawls uses the example of gambling.)
In our opinion, judging cases should be compared to elections or to gambling, rather
than to dividing a cake or to a criminal procedure. There is no way to establish the
correct outcome of a legal decision making process, even if we had perfect knowledge. The
only way to arrive at the right outcome is to follow a legally correct procedure. Judging
cases is an example of something perfectly procedural. But the outcome is in this case not
justice, but law. It may appear odd that judging cases should not be compared to a
criminal procedure, which seems to be a paradigm of judging a case. To make this clear, we
must relate to the distinction between factual questions and legal questions. The criminal
procedure, as meant by Rawls, concerns a factual question, namely whether the defendant is
guilty. In contrast we are primarily concerned with legal questions which assume that some
factual questions have already been answered in the affirmative.
In an argument, one party tries to convince (or persuade) another party of the correctness
of some thesis. The argument ends if both parties agree to the (denial) of the thesis, if
they agree to disagree, or if the argument ends because of some institutional reason (e.g.
time is over). With some qualifications because of the often institutionalized nature of
legal arguments, this characterization of arguments and how they end holds true for the
law too. Each party tries to convince the judge of his position. They can do this by
legitimate and by illegitimate means (e.g. flirting with the judge). Ideally, only
legitimate means are used or taken into account. A procedure of this kind is rhetorical in
character [Witteveen 1988]. Characteristic for such a procedure is that there is no fixed
outcome. By presenting reasons, each party tries to draw the outcome in his direction, but
the final result cannot be determined logically. Even if some reasons should logically win
(no matter how this is defined), it is not guaranteed that they will win in fact.
Moreover, if the procedure is followed correctly, there is nothing wrong when logically
'compelling' arguments do not win. There is no outcome which is better or worse in view of
the legal procedure. There is no other correct outcome for a legal case, than the one
arrived at by following the legal procedure in a correct manner. Therefore rhetorical
legal procedures are purely procedural{2}.
There is a classical objection to the view of law as a rhetorical procedure. This
objection has a modern form in the theory of law as institutional fact [MacCormick and
Weinberger 1987]. According to this theory legal conclusions are institutional facts,
which are obtained by applying legal rules to cases. These facts exist by virtue of rules,
but are nevertheless independent of the procedures by which they are established. In other
words, institutional legal facts are cognition-independent.
This theory of institutional legal facts fits well in established legal ideology according
to which legal decision makers ought to give their decisions on the basis of what is
legally the case, independent of their decision making process. This view of legal
decision making was frequently been criticized by Legal Realists [e.g. Frank 1930], but
nevertheless continues to dominate legal theorising. And, of course, it is clear that
legal rules guide legal decision making to a large extent. Legal decision makers normally
feel themselves bound to rules of law.
It turns out that the view of law as a rhetorical procedure can be combined with the
recognition that legal decision makers are guided by rules of law. Rules of law can be
seen as rules which specify how legal procedures are to be conducted. A procedure that
violates a clearly applicable rule of law is not properly conducted. As a consequence the
outcome of that procedure is not law, just as the outcome of wrongly conducted elections
is invalid. In theory it is imaginable that legal rules fully determine the outcome of a
particular procedure. In that case, the procedure can lead to only one outcome, and the
situation cannot be distinguished from the one in which legal rules establish
independently existing institutional facts. But, firstly, such a situation occurs less
frequently than some might want us to believe. And, secondly, even if this situation would
occur almost constantly, this would not disprove the difference between pure procedural
law and imperfect procedural law; it would only make the difference more difficult to
recognize. To avoid possible misunderstandings, we want to point out that although
material rules of law can be seen as procedural rules, they do not exhaust the rules that
govern legal procedures. There are, for example, also rules that govern the burden of
proof, which determine under what circumstances particular dialogue moves are allowed, and
which conditionally disallow particular argument moves (for instance adducing illegally
obtained evidence).
Another possible misunderstanding would be that if legal rules are dialogue rules, they
are not capable of generating normal legal consequences such as obligations. Traditional
rule application may be considered as a trivially simple interior dialogue. By means of
such an interior dialogue, a legal subject may (non-authoritatively) establish the legal
consequences of a particular case.
If law is purely procedural, this has implications for computational models of legal
reasoning. One implication is that it is not feasible to take fixed sets of premises
(facts and rules) as a starting point. Rather one should allow that rules and premises are
introduced in the course of a dialogue. Which ones are finally accepted as a basis for the
legal conclusion depends on the behaviour of both participants in the dialogue, rather
than on the truth or the validity of the premises. (We will see later on that this
position has to be amended somewhat.)
Moreover, there are characteristics of legal dialogues which are not easily translated
into features of classical proofs. In this connection we think not only of the widely
acknowledged phenomenon of the division of the burden of proof [Sartor 1991; Gordon 1993b,
p. 4], but also of the less recognized phenomenon that sometimes it is not allowed to
adduce arguments for one's position. In this connection one can think of illegally
obtained evidence in criminal cases, of the case where particular reasons may only be
adduced for a higher court provided that they were also adduced at previous proceedings,
and of the situation where one has contracted not to sue for particular causes of damage.
Phenomena like these influence the actual proceedings of legal debates, and consequently
the outcomes of dialogues and the answers to what the law is in concrete cases. To account
for them in computational models of legal reasoning, it is desirable to base these models
on dialogues.
On this basis we have developed a dialogical model of legal reasoning which aims to take
the typical characteristics of legal procedures into account. In particular the model
allows for the so-called 'refusal' of claims. Such a refusal is the dialogical equivalent
of not allowing to adduce particular arguments in real life debates. Our framework is
based on Reason Based Logic (RBL), a non-monotonic logic that is especially developed to
deal with legal rules, and with reasons based on rules [Hage 1993]. RBL is characterized
by the fact that application of a rule only leads to a reason in favour of the rule's
conclusion, which may have to weighed against other reasons. If the conditions of a rule
are fulfilled, this only is a reason for applying the rule. Our framework is the
dialogical version of RBL, and is therefore called 'Dialogical Reason Based Logic' (DRBL).
In DRBL, we consider legal disputes to be dialogue games between two parties about a
specific proposition, which is called the thesis of the dialogue game. One party, the
proponent, claims and defends the thesis, the other party, the opponent, does not accept
and maybe even denies the thesis. Both parties are permitted to make argument moves [cf.
Skalak and Rissland 1992], such as adducing reasons and asking for justifications. In a
dialogue each party seeks to make such moves, as to commit the other party to his point of
view. Moreover, they themselves try to avoid to become committed to their opponents point
of view. The dialogue is the process of making argument moves which in the end leads to
some result concerning the thesis. A dialogue game is a set of one or more dialogues.
There is one top level dialogue, which is about the thesis of the dialogue game. The top
level dialogue , and also other dialogues, often contain sub dialogues, the thesis of
which is a claim in the dialogue one level higher. The rules for the top level dialogue
also govern the sub dialogues. The structure of dialogues is recursive.
Some moves in a dialogue commit the party which makes them to accepting particular
sentences. A dialogue ends when the opponent is committed to the thesis, or when the
proponent is committed to the denial of the thesis, or revokes the thesis.
In a dialogue minimally two commitment stores are involved. A commitment store is a set of
sentences to which a party is committed. Since both parties are committed to the sentences
in their own set (which may be empty), each dialogue counts two commitment stores. It is
often useful to have a set of sentences to which both parties are committed from the
beginning of the dialogue. These sentences form a basis from which one can argue, and
which is not in need of being justified itself. Candidates to be included in the
commitment stores of both parties are generally accepted rules of law, and facts which are
assumed to be beyond dispute.
As mentioned before, in a discussion both parties will try to adduce reasons for their
positions and against their opponents position. In this section we pay attention to the
various possible argument moves. It should, however, be noted that such a presentation
necessarily presupposes dialogue rules. The framework of DRBL itself however does not
commit to specific dialogue rules. Moreover, it even permits discussions about the rules
of the dialogues. This means that the following presentation about the process of
reasoning is amendable at a number of points.
A dialogue game begins when a party claims some sentence. For instance, a party claims
that the other party (B) owes him ten thousand dollars. By making a claim, one starts a
dialogue with the claimed sentence as its thesis. The party who makes the claim becomes
the proponent in this dialogue. Since a dialogue game may contain more than one dialogue,
it is possible that the party who is the proponent in one dialogue, is the opponent in
another dialogue within the same dialogue game.
In principle, the opponent has four ways to react to a claim. The claim can be accepted,
it can be questioned, it can be refused, and it can be denied. In fact, denial is not an
independent type of argument move; it is just another claim, namely a claim of the
negation of sentence claimed by the proponent. The refusal of a claim comes down to making
a claim that the previous claim was not allowed. If a claim is accepted, the dialogue
which was started by the claim immediately ends. After acceptance both parties are
committed to the claimed sentence. If a claim is questioned, the proponent has the choice
between withdrawing the claim and defending it by giving reasons in the form of further
claims.
Normally a claim is defended by adducing reasons for it and arguing that the reasons
adduced for the claim outweigh the reasons adduced against it. We will discuss the normal
procedure in the following section. To commit the opponent to the sentence that particular
facts are a reason for a thesis, the proponent must commit the opponent to the sentence
that the rule which makes these facts into a reason applies in this particular case. This
sentence is a sentence like any other, so that it can be argued for by adducing reasons
for it. To these reasons, the same procedure applies as to any other reason. In this
connection, the fact that a rule is applicable counts as a reason why the rule applies.
Notice the difference between the sentence that a rule is applicable on the one hand, and
the sentence that a rule (actually) applies on the other hand. The former is a reason for
the latter, whereas the latter constitutes a reason for the conclusions drawn from the
rule{3}.
To commit the opponent to the sentence that a rule is applicable, the proponent must
commit him to the validity of the rule and to the facts that fulfill the rule's
conditions. If the opponent is committed to the validity of a rule and to the facts that
fulfill the rule's conditions, he is committed to the applicability of the rule, unless he
commits the proponent to the sentence that the rule is excluded.
This corresponds to a shift in the burden of proof. If the facts that fulfill a rule's
conditions and the rule's validity are established, the rule is by default applicable. If
the opponent thinks that the rule is nevertheless not applicable, he has the burden to
prove that the rule is excluded. If he does not succeed in this proof, the opponent
becomes committed to the applicability of the rule.
Before introducing the dialogue rules, we give an extended example of a dialogue which
our system should be capable of generating. The dialogue is pictured schematically in the
following table. Note that the example dialogue is stated rather informally to facilitate
its reading, and that it does not exhaust the possible reasoning steps of DRBL.
This dialogue is represented in the following table. The first two columns indicate,
respectively, the number of the dialogue move and the level of nesting of dialogues. The
string 'C>' indicates that the following sentence is claimed.
Table belonging to the example dialog is missing at this time!!!
A dialogue game is defined by dialogue rules. It this paper we use a common distinction
between dialogue rules and commitment rules [MacKenzie 1979; Loui 1992]. For the purpose
of the game we have added termination rules. The commitment rules describe under what
circumstances a player becomes committed to sentences, and under what circumstances
sentences are removed from a player's commitment store. The dialogue rules describe the
way in which a player can use dialogue moves. The termination rules define the termination
of a dialogue, and ultimately, the game.
There are also derived move types:
Commitment rules
Dialogue rules
rule 10
Termination rules
rule 16 A dialogue about the sentence S ends if and only if either:
rule 17 If a dialogue ends, all of its sub-dialogues end.
The dialogue rules as described in this paper permit the modeling of legal reasoning as
dialogue games. In this way the pure procedural nature of legal reasoning can be done
justice to. The rules presented in this paper not only allow 'normal' reasoning with
rules. They also make it possible to deal with the exclusion of rules, and with the
weighing of reasons. As a consequence they have the full power of the monological version
of Reason Based Logic. As an additional feature, they allow for the refusal of claims, a
feature which can not, or only with difficulty be accounted for in monological approaches
to legal reasoning, and is, to our knowledge, not dealt with by other dialogical
approaches.
We would like to thank D.W.P. Ruiter for his careful proofreading of this paper. The research of the third author was partly financed by the Foundation for Knowledge Based Systems (SKBS).
{1}Similar ideas of Rawls were (indirectly) also the inspiration for the work of Gordon
on legal dialogues. Cf. Gordon 1993a and b.
{2} Cf. Rorty 1980 in connection with the alleged procedural nature of all knowledge.
{3} In order to avoid the use of technical terms in dialogues, the notion of the
applicability of a rule as such does not occur in the dialogue rules. However, the rules
are formulated in such a manner that the effects are the same as when the notion of
applicability would have occurred.