A Computational Model for Trial Reasoning Abstract - CiteSeerX

2 downloads 0 Views 740KB Size Report
lawyer qualification exanli- nation. We have also had lawyers evaluate the .... the street, where. Bill froze and died. In this case, it is to difficult to judge. Tom's.
A

Computational

Katsumi

Nitta,

Institute

for

4-28,

Model

Stephen

New

Mita

for

Wong,

Generation

l-chorne,

Trial

Yoshihisa

Minato-ku,

Technology

Tokyo

108,

the domain

of criminal

HELIC-11 model

for

legal

soning). seven

of this

This key

facts

of a new

components:

various

meta

and

The

and

The

the

situation,

and

selects

the

best

one for

ends

when

any

one

agent

Certain

debate

in this tional

paper model

system

for

trial

the

make are

agent

The

claim,

tain

a move.

con~puta-

output

and

of old

on the

defendant’s

ity

form

to the

root

and

matching.

other

opinions.

of a new case.

In addition,

lawyer

We have also had lawyers drew

Cer-

(prosecutor’s)

arguments

arguments.

cases in the Japanme

Each

is the conclusion

on the plaintiff’s

while

They

based

is a set of arguments.

resultant

of HELIC-11.

rules

to the case base

of case rules,

facts

a

rule

of logical

by applying

refers

tree whose

cases

and

to the

form

by similarity

are based

inlpleIt con-

refers

in the

cases in the

system (PI M).

base engine

conclusions

leaves are the initial

opinions

nation.

reasoning.

- a rule

rules

of HELIC-11

arguments

several

based

base engine

hypotheses

is an inference

assigned

knowledge-based

legal

92].

Machine

case base engine old

legal

and whose

game

design

contains

argument

rule legal

draws The

generates

eugines

The

contains

and

which

illustrated

the

used in the

a parallel

and

recognizes

The

which

formulas

et al.

Inference

inference

deductively.

in this

One

move.

In addition,

-

goals

to refute

model

has been

of HELICII

agents,

agent

no longer

base,

[Nitta

is a logic-programming

case base engine.

re-

law

on the Parallel

sists of two

cases,

process game.

next

of this

examples.

presented

development

can

strategies with

the

old

explicit

other

candidates

rea-

matching

opposing

to a two-agent

mented

contains

case,

two

with

trial

model

argumentation

an argument. generates

(i.e.

similarity

the defendant,

be likened

forward

rules,

implications,

strategies.

can

puts

law

based

lations,

model

in criminal

a computational

logic-programming

knowledge,

the plaintiff

is to describe

reasoning

domain

reasoning

paper

Japan

@icot.or.jp

Abstract purpose

Ohtake

Computer

{nitta,wong,ohtake}

The

Reasoning

are

based

no priority HELIC-11

qualification

evaluate

is

solved exanli-

the practical-

two observations

from

their

evaluation. First,

Introduction

1 The

primary

is statutes. rules,

source Since

the mode

deduction. discretionary fixed

until

is known study and

of legal rules,

legal they

of legal

a statute

Legal

with

statutes

however,

often

contain

and their to actual

we at ICOT reasoning

countries

of a set of legal

have system,

problem.

been

To

HELIC-11,

in

all or parL of iks malenal is granted provided distribumd for direct commercial advantage, tie tide of the publication and its date appear, and by permission of the Association for Computing or to refubfish, requires a fee and/or specific

1993 ACM

a comparison

Such

the

the weakness

refutation

reasoning (i.e.,

In court,

parties

both

individual aim

to secure

sible

in court.

The

interpretation

that

agent

is, thus, this

simpler,

biased

goal-driven

more

is obtained

reasoning

during

is important

if we and to

parties.

Second, instead

facts

of

a new

plaintiff,

case).

arguments for

serious

of legal

towards

manageable

and ar-

be goal-driven

the most

A comp!ex

it

to present

The

initially

formation 2 ()

are trying

would

cal and dynamic.

$1.50

the

goals.

ref-

standpoints

of the

should

through

sLIch as the in arguments,

of each argument

strategies proc=~

of data-driven

into 0-89791 -606-9/93/0006/0020

premises.

Moreover,

pnnlssion,

@

gument

to achieve

designing

is included

the different

the legal

This

information,

was not eaay to compare

predict

and

are not

cases in court.

relevant

cases and opinions,

are to understand

is usually vague

meaninga

as the open texture

a legal

Penmsnon to copy withouz fes *z! the copies are not made or AC\l copyrighl no:ice and :k notice is given [hat copying k >fachmery. To copy othcmuse,

in most

consists

reasoning

are applied

a problem,

developing

knowledge often

concepts

in the literature

such

although

erenced

example, crime

pos-

knowledge

achieving

this

process

goal is usually

by goal.

is hierarchidecomposed

subgoals,

and

as new

the debate,

the

agent

inmay

shift

the initial

To take

into

HELIC-11

are

of legal

modeling le,

the

less severe.

these two observations,

notion

many

good

debate,

but,

of legal

HYPO

to something

account

with

There ject

goal

of two-agent research

projects

few of them

agents

[Rissland

in their

et al.

we extend

on

involve

prototypes.

87] [Ashley

the

by comparing

and contrasting

old cases.

HYPO

can treat

change

points

extent,

it does not discuss

cepts.

Furthermore,

set of good computational dress

viewpoints

and

The

provide

model.

key

debate

paper

3 explains

into

and Section

a

to adincorpo-

legal

argumen-

Section

computational

strategies

4 shows

5 presents

a

for debate.

of the

debate

Section

to select

attempts

1

con-

we introduce

is as follows.

components

components,

Though

concepts,

and opinions

of this

the

Section

on these

of legal

a set of strategies

organization

2 introduces

difiicult

that

that

build

an example

of a

the conclusion. Figllre

2

Computational

In this

section,

guides

the

Figure

1 presents

The

model,

and this

M,

]: The

a computational

development architecture

consists

model

of

new

where

Y

[Ueda

and Chikayama

Guarded

of HELIC-11.

(L),

as domain

Z is meta

hypotheses, lation,

rules

*

consists

Horn

fendant

goal

of the

(D)

When

and meta

in the form To and

ccmeri The

are

(the in

the

and

rules

rules

Z.

Then,

repeated

they

agent

and

until

there

generate

the

is

meta no

goal

is Guarded

knowledge

bases

of

be

programming

in a new case is a ground

the

Flat

following

subsec-

Facts predicate.

Each

predi-

“objectID”

to the

form:

objected,

list-o

instance

is an identifier of the

of “attribute= For example,

D.

These

pro-

f.slots)

which

predicate,

variable”

and

is used “list

to refer

of slots”

is a

pairs.

the predicate

orun(objectID, where

generated.

language

object-oriented

of name

own

is defined

as

follows.

goal

are

has the following

where

and

other

Horn

HELIC-11

A fact

list

a

an argument

to

of M

logic

(+C

rules.

Case

predicate.name(

de-

is sent to agent

language

parallel

on

and

P generates

D generates

and

in the

New

such

(+’),

(P)

agent

which

each component

2.1

cate

re-

relations

it to implicators

tree,

argument,

is based

making

matching

implication

in 7,

and sends

it to irnplicators

object

logic ten

case is given

this

sends

includes legal

as interpretation

/ prosecutor

of an inference

attack

which and

agents.

a new

which

case, C is a set

(H)

and rule

plaintiff

to be achieved

+’)

the new knowledge

of a set of similarity

(+=)

90]

clause).

tions.

of a set of implication

as case implication S consists

S >>,

postulates

such

= consists

+-,

about

cases, V is domain

such

HELIC-11

of a 7-tuple:

C, D, T, =,

is a set of facts

of old legal entities

F,

of new

that

HELIC-11.

We describe