Towards
a Legal
Knowledge
Analogical
Representation
and
Hajime
Faculty
Minat~ku,
Haraguchi,
Department
Reasoning
Gakuin
Tokyo 108, JAPAN
Sakurai
of Systems Science, Tokyo Institute
Midoriku,
Sigeru
of Technology
Yokohama
Kagayama
Toyonaka
560, JAPAN
Abstract
content
is often incomplete,
it is necessary at
properly.
He may apply the rule, though
originally
been deemed related
it may not have
to such an event, on the
basis of some similarity
between the event of the case
and the requirement
of the relevant legal rule. This type
of reasoning is called legal analogy. This paper analyzes
an actual case of legal analogy in the field of Japanese
civil law in order to clarify
in analogy,
it will
in a symbolic
and standard
the reasoning
as well a knowledge
methods
to justify
used
Figure
is ut~lzed
system both in terms of inverse
requirement
Structure
of
Legal Analogy
of a hypothetical
statue
10
B to generate
rule for a give case, provided
are similar
with
This logical
Fig.
respect
structure
1 for a while.
these two
to some imporcan be shown
Given
a current
case subsumed by B, suppose that a lawyer has in his
mind a target goal, which is an intended legal conclusion
never derived from the case under the present domain
theory. In other words, the domain theory is too weak
to get the desired conclusion.
Then the problem is to
make the hypothesis B + X in Fig. 1 in order to obtain
X for the present case. The logic used to justify such
an analogy can be stated as follows [9]: The legal rule
vtild
becau8e
~ +
x is ~ld
B + X is considered
and because A + a. Moreover, since B + a, we can
conclude that the same effect X is derived from B under
the existence of common a. It should be noted here that
all or pafi of this material is granted provided
distributed
for direct commercial
advantage, the
title of the publication
and its date appear, and
by permission
of the Association
for Computing
or to republish,
requixes a fee and/or specific
0-89791-606-9/93/0006/001
This type of rea-
rule A + X by other requirement
Let us examine
Analogy has many important
functions
in the domain
of law.
Since the number of legal rules is restricted
and their content is often incomplete,
it is necessary at
times for a lawyer to opt for an analogical application
of a legal rule to a given case in order to decide the case
properly.
He may apply the rule, though it may not
have originally
@ 1993 ACM
legal rule.
is conceived as a process of generation
rule, which supplies the lack of law
tant legal aspects.
following Fig. 1.
Legal
Analogy
Permission
to copy without fee
that the copies are not made or
ACM
copyright
notice and the
notice is given that copying
is
Machinery.
To copy otherwise,
permission,
of the relevant
soning is called legal analogy.
a hypothetical
resolution.
Logical
schema
been deemed related to such an event, on the basis of
some similarity
between the event of the case and the
requirements
1
1: analogy
for a certain particular
case. In other words legaJ Analogy is said to be an act of replacing a requirement
A of
the analogy.
be shown how the knowledge
reasoning
B
B+x
important
functions
in the domain
number of legal rules is restricted
times for a lawyer to opt for an analogical application
of a legal rule to a given case in order to decide the case
F&.lly
A-.
A+X
has many
Since the
and their
227
of Law, Osaka University
Kaneyama-cho,
Analogy
of law.
Methods
University
Seiichiro
4259 Nagatsuta-cho,
Faculty
System:
Yoshino
of Law, The Meiji
Shirakartedai,
Makoto
Reasoning
the intermediate
hypothesis cc + X represents the heart
of the law in a sense. The situation
is more precisely
illustrated
$1.50
110
by the diagram
shown in Fig. 2.
Article
I: a-+x
A+a
94(Fictitioua
declaration
of intention):
B-+a
1 A fictitious
declaration
of intention
made in
collusion with the other party is null and void.
Clause
Clause
2: analogy
Figure
Fig.
2
consists
diagram
of two sub-diagrams
with
means a deduction to get the final hypothesis
is the result of analogical interpretation.
As we have just
analogical
eralization
observed,
the primitive
interpretation
and deduction.
Figure
In this
operators
LES [7] performs
deductive
to
un-
as Horn
clauses.
The
system is therefore
in resolution-based
de-
Since the reasoning
deductive
system
we expect
to execute
we choose an absorption
various
operator
should
types
be basi-
2 and 3, we first
the knowledge
present
p-id(idl,
qid(id2,
representation
The Telation-identified
is used ss a reference
function
similar
<
relation~denti
<
list
non-compound
clause is:
2. There
object
to
the
to a pointer
relation.
Hence
f ier >,
As a result,
it
p is useful for readability
of program
as:
Al, ....An
suppose we have the following
and X
rules as
predicate
and
has
instance
is a relation
pred.
p([agt
and, pred*(Slots)
Slots) for each
The
meaning
of each
: a, obj : c]) holds.
instance
of q whose agent and
cases are b and p([agt
: a, obj : c]), respec-
3. If we have a relation instances of q whose agent and
object are Agt2 and a relation
instance (p([agt :
Agt, ...]) of p, respectively, then we also have a relation instance of r whose agent and object are Agt2
and Agt 1, respectively.
of slots >),
in a frame
ier, SlotJist).
tively.
of rules
called a
is a name of the relation
+ pid(Identif
“-” denotes an anonymous
variable
is an abbreviation
of pred>d(r(Slots),
In Section
representation
> (
clause
where
LES is a logical system to perform deductive reasoning
under a set of Horn clauses. This means that both legal
< predicateaarne
p-id, we assume
p-id(P, [agt : Agtl, obj : .]), q([agt : Agt2, obj : P])
+ r“(([agt:
Agt2, obj : Agtl]])),
of represent-
rules and other rules to interpret
requirements
are encoded to Horn clauses. Each predicate,
compound predicate, has the following form:
is denoted
by a constant
[agt : a, obj : c]).
[[agt : b, obj : idl]).
1. A relation
Knowledge
definition
Each rule is described
For example,
language, and try to represent Article
94 of Japanese
civil code for which analogy was applied.
2
predicate
name p followed
well as facts:
and deduction,
knowledge.
as
Code
where Aj is an atomic formula of our language
should be an instance of compound predicate.
in a resolution-based
deductive engine. A more detailed
description
on absorption,
deduction,
and analogy can
be found in [2] which defines the analogical
reasoning
related
Civil
for each predicate
X t
[4] to realize our gen-
ing legal rules and other
Moreover,
Such predicate
clauses,
of knowledge,
Now we are ready to discuss a method
each compound
predicate
p(SlotJist)
eralization.
This is because the absorption
is a kind of
inverse resolution
and is therefore essily implemented
we use here in terms of absorption
of intention
94 of Japanese
that p has the unique
ductive engine, especially in Prolog. On the other hand,
generalization
has been studied in the field of machine
learning, and various computational
operators for generaUzation
have been proposed by many researchers.
cally
paper,
word “id”.
systems.
reasoning
der a set of legal rules encoded
implemented
3: Article
by a word p-id,
to say, deductive
engines are necessary for most legal reasoning
For instance,
of declaration
H, which
of legal rules are gen-
Needless
nullity
in the preceding clause cannot be set
a good faith (bona fide) third person.
the same
shape V. The left V is to obtain an intermediate
hypothesis 1: a + X by a generalization.
The other V
realize
2 The
mentioned
up against
we have r([agt
: b, obj : a]) whose identifier
is T([agt : b, obj : a]) itself.
a
Now
based system.
we are ready
to encode
legal
norm
sentences
in
slot is a pair of slot name and its value, and is written
the form of compound predicates.
Clause 2, Article 94 of
Japanese Civil Code, which is one of the most frequently
as slot-name
applied
: value.
111
rules in legal analogy,
is shown in Figure
3,
The Clause 1 means that the declaration
is invaEd if
the parties negotiated
that the declared intention
was
different from the real intention.
The Clause 2 means
that the invdldity
cannot
Case of a petition
Case: After
“A” bought a house, which
from “O”, he approved the registration
2 is to protect the right of a good faith person who believed the declaration
of intention.
The Clause2, for instance, can be encoded in the form
predicates
Rule 1 (Artic1e94,
contract.id(Ctrctl
Y], obj:Obj,
ownership
contract-id(ctrct2,
[paTties:/Y,z]],
obj:Falsityl)])
-+ cannot~et-up”
(/agt:
obj: Obj, time:-),
X, goa:Z,
but compiled
with
denotes logical
variables.
denotes some contract.
In other words, the
the declaration
intention
of intention
in its object’s
cannot~et-up
cate,
and
is
the
the
predicate
Figure
of the
knowJact
by an identifier
are initially
is
not
4: Example
defined
so
his ownership
that
of analogical
“A”,
Art.
of
application
ranging
right.
over
without
who was the real owner,
However,
the judge
its
claimed
approved
that
right by legal analogy.
by a set of approved facts,
and an approved fact is also represented by a compound
predicate.
An example set is shown in Facts 1. In Fact
1, the pa, pb, p= and p. represent the agents of the case.
of
The first
argument
of each compound
predicate
is an
identifier
for reference and the second argument is a list
of pairs of case and value. The name, “agt:”, represents
an agent case and “obj:”
represents an object case. The
predicate Teg-f
title” and “reg”
Goa/s).
know(Agt,
Goal) : –clause(Goal,
Goals),!,
(Goals = true– > know-f actzel(Agt,
; know(Agt,
Goals)).
sold the house to “C”
“C” had the real ownership
Legal cases are formalized
in our fact database,
know(Agt,
(Goal, Goals)) : –!,
know(Agt,
Goal), know(Agt,
“B”
real ownership,
predi-
as shown in Fact 1. According
to such a representation
of facts, the predicate know is eazily realized as a kind
of meta-interpreter:
know-f act_re/(Agt,
know-f act(Agt,
(falsity)
sentation of facts, a leading case of analogical application of Clause 2, Article
94 is shown in Figure 4. In
All the possible instances
presented
of Clause2,
the nullity
In legal reasoning, it is important
to represent facts
which constitute
the case. Before describing the repre-
as follows:
Prolog’s
know
Id.
application
Code,
where < comp-pred
> is a meta variable
every compound predicate.
in spite
know(Agt,
Content)
succeeds iff the relation
instance
Content
is proved under the set of facts Agt knows.
We use an auxiliary
predicate
know-fact,
where
knowJact(Agt,
Id) means that Agt knows a relation
indexed
Civil
claim the invalidity
is defined
standard
of the house.
the real
the case, since
not
and oth-
of the registration
good-f aith” ([agt : Agt, obj : Content)
+
not(kmow([agt
: Agt, obj : Comtent])
where
do
of” A“, so that A cannot claim that “ C“ must do
cancellation
procedure of passage of title and others
contract
because of the invalidity
good-faith
must
the registration
of ‘B” cannot be set up against the
good faith” C“, who did not know the real intention
The predicate
contract, Cntrct 1. Note that this rule is applied
of the legal invalidity
of its requirement
part.
The predicate
“C”
means that
from
Cntrctl.
X cannot
Cntrct2,
falsity
is different
value,
means that
of the contract,
The predicate
The predicate
that
of passage of title
By analogical
94 of Japanese
definite clauses corresponds to a rule interpreted
by a
lawyer.
In Rule 1, symbols, which are preceded by a capital
letter,
claimed
procedure
have the ownership
obj:Ctrct2])
encoded from legal sentence
legal knowledge.
“A”
ers because the ownership of the house in the case
should belong to “A” , and “B” and “C” did ~ot’t
Judgment:
The rule was not directly
title.
cancellation
Ctrctl]),
good- faith([agt:Z,
“O’’owned,
of passage
not know the real intention
of “A% and only know
“C” registered the passage of
that “B” registered.
time:-]),
falsity.id(Falsity,[obj:
bench,
of title from “O” to “B” without
his real intention of the p~sage. Registered the paasage of title,
“B” sold the house to a good faith, “C”, who did
as follows:
Clause2)
,/parties:[X,
of passage of a
((0) No.107-1951~udgment
of the second pretty
Aug. 20th 1951))
be set up against a good faith,
who do not know the fact that the declaration
is different from the real intention.
The purpose of the Clause
of compound
against registration
house’s title
Other
_ptitle means “registration
means “registration”.
knowledge
of passage of
is needed to reahze legal reasoning,
and it is also represented
by rules.
Goal)
Facts
< comp+red
> (Id, X)) :Id), < comp+red
> (Id, X),
1 (Facts of the case in Figure
4)
sale-o f-immovables-id(idl,[
parties:[p-o,p-a],
obj:imm_X,
ownership-id(id8,
112
time: tO]).
[agt:p.a,
obj:imm_X,
time: tO/).
owner8hip.id(id2,
[agt:p.a,
obj:imml,
Caee in Figure 4
tirne:tl]).
0
[parties:[p-o,p-b],
reg-of-ptitle.id(id3,
time:tl]).
obj:imm~,
obj:imtn3,
reg-id(id.#)[agt:p-b,
time: tl]).
approvaLid(id5,[agt:p
_a, obj:id3,
8ale.of.irnrnovable8-id(id6,/parties
obj:imrnl,
reg-of.ptitle.id(id7,
obj:imml,
time:t2]).
[parties:[p-b,p-c],
time: t2]).
know..act([agt:p-c,
obj:id3]).
time:tl]).
:[p-b,p-c],
Clause 2, Article
94 of Japaneee Chil
Code
know- fact([agt:p-c,obj:i~]).
know. fact([agt:p-c,obj:id6]).
Jmabz
x
know. fact([agt:p-c,obj:id7]).
Chummgthe Inval ldlty
[1
3
Analysis
tion
of analogical
and
realize
knowledge
legal
Let us consider
applicaneeded
Figure
to
4, again.
between
the case in Figure
4 and
94
0
analogy
the case in Figure
5: Relation
Clause 2, Article
Obviously,
if the good faith, “C”, did not appear, the ownership of
“A” would be approved.
Then, what inference is done
in this case? What
knowledge
is needed to realize such
legal analogy?
To clarifY the reason why the rule can be applied to
the csse in Figure 4, let us consider the difference between Clause 2, Article
tion is shown in Figure
both
of them
94 and the case. Their rela5. If O in Figure 5 is omitted,
can be seen as a trinary
one-to-one partial correspondence,
can be found.
However, although
tion
between
relation.
between
A and B. What
If a content
is different
contract,
the case
relation
his common-sense
In Figure
or legal common-
sense knowledge. Therefore, legal analogy requires some
knowledge in order to transform
the csse description
so
that a correspondence
between a csse and a rule can be
ure 6, two virtual
duced.
of title
case is shown in Figure
relations
of passage of title
of the representation
they were not registered
6. In Fig-
then it is concluded
7, to represent
that the contract
are intro-
the variable
content,
a slot
tract”.
However,
more knowledge
7 is assumed,
to introduce
a relation
the relation,
may be
falsity,
must be assumed since the rule is not
about
the registration
to fill
the gap between
but about
the contract.
the registration
and
In order
the con-
formally,
If the case is transformed
as shown in Figure 6, can
a relation,
we apply Clause 2, Article 94? Obviously,
which corresponds to the falsity in Clause 2, Article 94,
is missing. In legal reasoning, a missing relation disables
the application
of a rule, such a relation must be fulfilled
contTactM(Ctrct,
repr_of_ctTct([obj
scwof ztmt([obj
Reprcts
Figure
of such a rule.
Declaration)
as:
113
-)
: Ctrct, AtT : RqK%?]),
: Ctrct, Atr : R.ealCts])),
# Realct8
+ fa18ity* ([obj : Ctrct])
by some knowledge before the application.
We can assume theories of interpretation
as such a knowledge.
A
rule of a theory of interpretation
is also written
as a
compound predicate rule. Figure 7 shows an example
The rule is paraphrased
of a contract
variable, Atr, is used. The predicate, reprmf _ctrct, represents the “representation
of contract”
and the predicate soa_0f_ctTct7 represents
“state of affair of con-
introduced.
In other words, we can assume that the passage
from O to A and the passage of title from A to
B, although
4
from that of the state of affair of the
If the rule in Figure
found easily.
The transformed
Csse in Figure
has falsity.
corre-
spond to the contract between X and Y or the falsity
of the contract?
In such a case, a judge seems to transform the case with
6: Transformed
If so,
{A-X, B-Y, C-Z, ...}.
the rule has a rela-
X and Y as a legal requirement,
has no relation
Figure
7: A Rule
of A Theory
of Interpretation(l?alse
4
Utilization
by
Human
Act
Act
generalization
Act
Registration
of Classification
pa claimed
Knowledge
that
in Section
pc should
Cts)
juristicact-id(Id,
+
juristicsctid(Id,
Cts)
+
+
CHs-).
registrationtid(
~d, Cts).
Cts)
registrationJd(Id,
Cts) + quasi.juristic-actid(
quasi~uri8tic-actAd(Id,
Commonsense
p, with
to the passage of title.
respect
of
set up
Hence we first
goal to be explained:
Cts)
+
cannot_set.up([agt
Id, Cts).
law ful_act-id(Id,
%gp, ~ariies
: [-, Agt], obj : Obj, time
~eg” ( [obj : Regp, agt : Agt]).
repr-of
regal f -ptitle~d(
&gp,
~arties
: .] )
the goal 1. with
Resolving
94, we have the following
ownership([agt
: Agt, obj : Obj, time : Time]),
+ soumf _reg* ([obj : Regp, agt : Agt]).
contracttid(Ctwtl,
Rule
ma~f&rct-id(Id,
repraf
-Tegli?ast(rd,
soaaf~egid(~d,
Cntu ) +
reprJd(Id,
: ~ti, Y], obj : Obj, time
~arties
Cnta)
~
soa_id(Id,
Cnts)
+ repr-id(Id,
Cnts).
The
Cnts).
with
first
goal
any rules.
of Classification
some another
Knowledge
generdlze
Control
The knowledge
knowledge may be needed as such a
8 shows an example of classification
in Figure
8 is a textbookish
edge. In order to fill the gap, another
tion knowledge is needed.
of such a knowledge.
The classification
Figure
knowledge
in the goal list cannot
The standard
knowl-
for
9 shows an example
goes
On the contrary,
such a failure
Strategy
our system tries to
goal, provided
there is no other
can be resolved with
1 ([3])
and
Let
some rules.
A be a selected goal 2
suppose
we fails
in
resolv-
another
goal in the goal list.
In
other
words,
our
system tries generalization
for some goals in a goal list
only when every goal in the list fails.
important
The goal (3) can be resolved with the legal theory rule
in Figure 7. Thus our system prefers the resolution
for
‘ Precisely
resolved
with
and is con[agt : pa, goa : pc, obj : id7]).
verted to a goal carmot~et-up-id(Id,
denote the same
Since, for each predicate pred, pred and pretid
relation at conceptual level, and since the process of converting
the former predicate instance to the latter one is a trivial one step
resolution, we identify them unless there is confusion,
2 Norm@
the left most gozd is selected.
cannot~etmp(Slots)
is very huge. To reduce the number and obtain an adequate pairing, the classification
knowledge can be used.
For example, we can obtain the pairing between the contract and the passage of title, since the upper class of
is the lawful
be resolved
interpreter
ing it with some rules.
If there is no other goal Aj
which can be resolved, then try to generalize A. Otherwise the generalization
is delayed in order to suppress
over generalization,
and the standard resolution is tried
role in legal analogy. Even if virtual
relations are correctly introduced
with theories of interpretation,
the
number of possible paring between the case and the rule
concepts
Prolog
goal (1) and tries to resolve it with
in a goal list Al, .... An,
type of classifica-
plays another
(3)
(5)
(2)
rule.
: Ctmtl]),
(2)
: p., obj : Fa/sity]).
goal in a goal list that
tract, classification
knowledge.
Figure
knowledge.
[obj
: Tl]),
(4)
back to the previous
9: Description
Article
~m%ies : [Y, p=], obj : Obj, time : 2’2]),
good-iaith([agt
Cnts).
-+ soa_isnt(ld,
Cnts)
contract_id(id7,
Cnts).
?? of Clause2,
goal list as the next goals:
falsityJd(Fa/sit~,
repr~fztrctJd(Id,
(1)
interpreter
with some additional
reasoning functions, it
tries to resolve the goal with some rules as well as facts.
: Time]),
: _ obj : Obj, time
: pa, goa : p=, obj : id7])
where id7 is the identifier
of the registration
of passage
of title of C.
Since our system is basically an extension of Prolog
Cts).
Knowledge
reg-o f -ptitletid(
the both
procedure
it suffices to show that he cannot
C%).
law f uLact-id(Id,
regaf@tZeJd(Id,
Figure
Since the plaintiff
knowledge
contractJd(Id,
+
3, 4.
do cancellation
passage of title,
make the following
Textbookish
deduc-
reasoning system.
First suppose that we try to give a legal explanation
to protect the right of good faith person p= appeared in
the case presented
8: Example
and
Now we are ready to show how to utilize legal knowledge and how to carry out legal reasoning in a symbolic
Quasi Juristic
Con’tract
Figure
knowledge
tion
Act
111-ful
.Juristic
of legal
act.
114
speaking,
+
this
goal
cannot~et-tipJd(
ie
firstly
Id, Slots),
(3)
above
goal
list:
Replacing
generalization.
(3)
with
the
Now our system tries to resolve the new goal list (11),
sub-
(6), (7), (8), (9), (4), (10). The goal (11) can be resolved
with the classification
rule
Cent?’actJd(ctTctl,
Tepr-0f4trct([obj
: C%ctl,
soaa~dmt([obj
: C%ctl,
RepreCtsl
the goal fist consequently
_),
(6)
juristic-act~d(ld,
atr : ReprCtsl])
(7)
atr : RealCtsl]))
(8)
(9)
# RealCtsl
becomes [(2), (6), (7), (8), (9),
(4), (10), where
Ct) + contracttid(ld,
need a test called a subsumption
Control
faith([agt:
p., obj : fatsity([obj
: Ctrctl]).
(10)
Strategy
and
that
ization
is an instantiated
check: 4
3 (Subsumption
goal of (5).
goal in the goal list 3. We therefore
some goal. The generalization
tive goal reduction”
tion.
According
, which
try
is carried
is a kind
to the standard
to any
to generalize
of inverse
resolution,
a head of
in the goal list. Then, remove the subsumed goals from
the goal list, and add the head of clause to the goal list.
description
Generalization
by
Abductive
ezists a rule A *
Goal
be a current
BI, ...Bk
If there
and a substitution
@ such
that (Bl, .... Bk)e C Al,,.., An, then r’ernooe Bjd
Al, .... An fov each j, and add A% to the goal.
relation
from
speaking,
from
given a goal list, there exists sev-
erai subsets of goals for which the genertilzation
ductive
goal reduction
or generalization
by
G to G’,
according
check, Since there exists no juristic
pa
in our fact
database,
fails.
We have now goal list in which
Repeating
similar
the goal (11)
each goal is a failure
generalizations
lawfuLactJd(Ctmtl,
~artie.
larufuLaettid(Cktl,
-)),
repr([obj
aoaaf
: Ctretl,
atr
-ctmt([obj
several
times,
ReprCta
by ab-
can be applied.
: ~=, Yl, obj : Obj, time
atr
: RealCts]),
# RealCts,
contTaetid(id7,
~aTtiea
gooaLf
: pc, obj : fal.sity([obj
aith([ag$
: [Y, pc], obj : Obj, time
from
: Ctrctl,
: T2]),
: Ctretl])
atr : R@rCts]),
: id3, atr : p-b]) can be provable
succeeds since Tepr([obj
our facts and rules, where id3 referring
tration
: Tl]),
: ReprCts]),
: Ctretl,
Now the goal ~epr([obj
Generally
from
we have a goai fist:
Reduction[2]
goal list.
by a general-
G’ is subsumed
cancel the reduction
by the subsumption
goal
is given ss follows:
let G : Al, A2, .... An
[3])
Thus, the goai (11) and the rule (12) is never resolved
resolu-
rule is unified with a goal. On the contrary, the inverse
resolution tries to subsume body of clause to some goals
A formal
goal list,
Whenever
and try another resolution
to Control Strategg 1.
out by “abduc-
Check
goal list G : Al, .... An
a next goal list G’ is obtained
or a resolution.
a previous
Now any rule in the rule base is not applicable
(12)
However the resolution with this rule results in the original goal (2). This causes a cyclic reasoning.
In order
to avoid such a generalization
- specialization
cycle, we
Suppose that we have a current
good.
Ct)
of paasage of title
the regis-
from p_o to p-b. The next goal
list ix
Control
Given
Strategy
a goal
2 (Generalization
Order)
list
such that every goal is a failure
choose an arbitra~
$ubset of goal list,
should be an instance
goal,
where the subset
of a body part of some rule in the
rule base,
Suppose we choose a goal (2) to be generalized.
Ct) + jrwistic~ct-id(Id,
~Tties
: ~a, Y], abj : Obj, time : Tl]),
-)),
ma~f ..etrct([obj : id3, atr : ReaK%]),
ph # RealCta,
contract-id(id7,
~rtiea
: [Y,p=]]),, obj : Obj, time: T2]),
goodJaith([agt
: p=, obj : jalsity([obj
: id3])
Then
Now
by the rule
contractJd(Id,
Iawf&act4d(id3,
/awfuLact&J(Wt,
goodJaith
can be evaluated
and in fact
suc-
ceeds. The goal soa~f-ctrct([obj
: id3, atr : l?ealCts])
to soa([obj
: id3, atr : RealCts])
is once generalized
= p~
as well as
that results in the solution RealC’ts
C%!)
Similarly
contract~d(id7,
...) is general# p_a.
ized to law f ul~ct~”d(id7,
...) which is true under our
p-b
the goai (2) is replaced
with
databaae.
jrwisticactJd(Ctretl,
@ties
: ~.,
Y], obj : Obj, time
: 2’1])
(11)
~We resume that
the predicate
good.~aith
is freezed.
means that it is evaluated ordy when all the argumentz
stantiated to some ground terms.
Thus we have the final goal list:
lawf tdact(id3,
lawf 7d4et(id3,
@wties
-)),
, ~-a,p-b]]),
TNIE
are in-
115
‘Similarly
“specialization
- generaHzation”
cycle can occur.
Our srrbsumption cheek is applied to both types of cycles.
No generalization
is possible for this goal list.
a case, our system
output,
This
result
presents
us the goal list
is due to the fact that
legal relationship
between
pa
the
system
Default:
Normally
a sale
of immovable
tration.
~egmf-ptitle(~am%es
therefore
assumable.
Assumption
ership
regis-
: ~-o, pa],
can generate
facts:
registered
is also
Teg4~@tle(~an5es
interpret
...]) is
legal rules.
The
recorded
realized
by
: ~m,p~],
own-
reasoning:
assumption
From
the
default
in
: ~-a,p~],
and
the
real
the
case,
[2]
...]) is substan-
However
it still
M. Haraguchi
A form
In Proc.
Learning
remains
5
as a future
as well as reasoning
virtual
relations
present system.
Some
that
support
Theory,
work
method
the final
comments
on
Workshop
Intelligence,
law
Press, 1974.
as an abductive
on Algorithmic
Japanese
Society
1991.
to develop
[3]
to reason such
output
of our
conflicts
[4]
rules
various
types of knowledge
legal rules analogically
mechanism
eralization
to perform
and analogy
preting legal rules. As
analogy are based on a
the
give
long
legal
used for
M. Haraguchi
What kinds of knowledge and inferences aTe needed to realize
legal reasoning?
(in
S. Muggleton.
and W. Buntine. Machine invention
of fh%t-orde?’ predicates by inverting
resolution,
In
352,
and a fundamental
[5]
interpretation.
Both genus possible ways of interas our generalization
and
conceptual hierarchy with
in generfllzing
of view, this problem
rules.
Ftom
a com-
of confllcts
can be
[6]
pages 339-
S. Muggleton.
Ist
Inductive
WoTkshop
iogic
programming,
of Algorithmic
Learning
in
The-
H. Yoshino, M. Haraguchi,
S. Kagayama, Y. Matsumura. Foundation
of systematization
of analcgy
conference
in law (in Japanese) , In Proc. National
of Japanese Association
foT Artificial
Intelligence,
pages 219-222, 1991.
[7]
H. Yoshino
Lecture
[8]
Z.
[9]
Aoumi.
Introduction
Koubunn-dou,
C.:
However it is true that we cannot decide what constrains
der Theories”,
tive Learning
are needed before reasoning,
C@elo,
116
Computer
Science,
in SpringeT
Logic
march
Pro-
1987.
to Philosophy
of Law (in
1989.
H. Gasyuu Analogy in law (in Japanese),
lished lecture note, Legal Expert Systems
tion, Meiji Gakuinn Univ., Tokyo, 1986.
[10] Rouveirol
maintenance
in
’86 pages 34-45,
Japanese),
The first one is to introduce
semantic constraints
by
which some inappropriate
generalizations
are rejected
due to the inconsistency
with the constraints.
This
is also a well known approach to check the appropriateness of generdlzations,
and can be implemented
in
an ATMS-like
truth maintenance
system in principle.
Legal ezpert sgstem LES-2,
Notes
gramming
considered as the problem of nondeterministic
choice of
generalizations.
Roughly speaking we have in our mind
two ways to cope with this problem.
So the truth
Learning,
ory, pages 42–66, 1990,
might contain several rules to generalize legal rules in
different ways. Some lawyers agree with some generalizations
with which another lawyers disagree.
Thus
there exist conflicts
on Machine
1988.
Proc.
which most lawyers agree, a hypothesis produced by our
system will be approved.
However our knowledge base
point
of analogy
2nd
pages 266-274,
PTOC. Workshop
We have described
putational
of Tokyo
Japanese), PTOC. 6th sgmposium on knowledge repsystem, Legal Exresentation
and legal reasoning
pert System Association
in Japan, 1992.
in generalizing
interpreting
to
paper will make the
to the study of positive
University
inference,
assumable.
Introduction
(in Japanese),
for Artificial
knowledge
A forthcoming
showing
generalization
. ..]).
facts
supporting
reg_of_ptitle(~aTties
tially
explanation
References
assuming
[1] H. Tanaka.
Substantial
a case-based
the reason why it chooses a particular
idea more clear,
supporting
pJ
implies
of
by one of authors. The key idea is to find past cases for
which the same or similar genertilzations
are used. In
a word, this is a case-baaed justification
of generalization and analogy, According
to thh approach, no new
generalization
is found by our system.
However, the
As we have ex-
seems to support
the appropriateness
The second way to cope with the problem of nondeterministic
choice of generalizations
is now investigated
plained in the previous sections, judges seems to have
as8umed a virtual
relation
law f ul~ct (id*, ~aTties
:
~-a, p-b]]). In fact the followings
virtual relation:
for checking
hypotheses.
there is no dhect
and p~.
is used only
system
In such
ss a final
unpub-
Associa-
“ITOU:
Induction
of First
OrProceedings of the j%st InducProgramming
Workshop, Vlana de
in
1991a