Japan Advanced Institute of Science and Technology

Japan Advanced Institute of Science and Technology

JAIST Repository

https://dspace.jaist.ac.jp/

Title

Author(s)

Citation

Issue Date

Type

Text version

URL

Rights

Description

Temporal Relations

Yosuke Koreeda

Scho ol of Information Science,

Japan Advanced Institute of Science and Technology

February 14, 1997

Keywords: case-based reasoning, legal reasoning, similarityassessment,temporal

relations among aairs.

In legal reasoning systems, case-based reasoning (CBR),as wellas rule-based reason-

ing (RBR), plays an imp ortant role. CBR can nd the most similar precedent from a

numb er of cases. Precedents are the important sources of the law like statute, custom,

jurisprudentialtheoriesandsoforth. Inaddition, theyare increasingmoreandmore,and

wehaveagreatnumb erofprecedentsto day. Thereforeitishopedthat CBRcansupport

the pro cessing of them.

As the experts do not organize their knowledge as a rule, we must cope with many

diculties in consructing RBR systems. On the other hand, cases are easy to extracted

fromgreatexperienceoftheexp erts. Butthereare otherproblemsinstead. In particular,

similarity assessment isa fundamental issue inCBR.

Rules are expressedintermsof\open-textured word"that canbedenedonlywithin

specic contexts in law. Because precedents illustrate the meanings of \open-textured

word", itis importantto determinethe similarityb etween new legalsituation and them.

In legal reasoning systems, the similarity should be assessed by causal relations in key

factors, and those causalities are strictly related to temporal relations. Because these

temporal relations are determined objectively, they are easy to pro cess with computer.

This paper deals with the system assessing similarities using temp oral relations among

aairs. To accomplish this purp ose, we focus ontwoissues:

1. Representing and indexing cases

We can pick up many aairs from legal cases. To represent these aairs, we prop ose

a classication of aair types by their temporal features as: State, Durative event, and

Punctual event. State is stative aair and holds for a time interval. Durative event is

Copyright c

1997byYosukeKoreeda

does not keep it. The former event occurs over a time interval and the latter do es at

timepoint. Inaddition, theseaairsare relatedbysuchpredicates as: start,end, during,

overlap and meet. Theyare dened interms of Allen's logic.

Temp oral relations among aairs are generated automatically by rules. Legal cases

consist of these temporal relations.

pp

2. A framework of similarity assessment

Todeterminethedegreeofsimilaritiesanddierencesb etweentwocases,weexaminetwo

matchscores:

(1) the match scorebetween twotemporal relations

(2) the match scorebetween twocases

On the basis of the classication of aairs and the relation between aairs, (1) is

computed. Next we dene the numerical evaluation function according to (1), and thus,

(2) is determined.

Fromthe matchscore between two cases, we assess case similarity inthree steps:

First step Compute the whole matchscores.

Second step Compute the partialmatch scores.

Third step Contrast the whole match scores and the partial one, and assess with hy-

p othetical cases.

where the whole match scores mean the degree of the similarities for the whole of

cases, the partial matchscores mean that for the part of cases.

In rst step, we compute the whole match scores to search the precedent which has

a high match score. It narrows a numb er of cases to several candidates for the most

similar precedent. Next, in Second step, cases are divided into several parts. And then,

wecompute thepartialmatchscores. Finally,inthird step,wecontrasttwomatchscores

and assess withhypothetical cases. The resulting dierenceshaveimportantinformation

to assess case similarity.

This strategy is called divide and assessment.

We illustrate the system with precedentsabout complicityin crime. In criminal law,

thereare threesituations ofcriminality as: preparation,attempt andconsummation. We

regard these classications as State and examine how they are related temp orally with

aairs that construct the legal case.

Weassumenew legalsituation, andassess similaritiesagainst21precedents. Inmany

legalreasoningsystems,itcreateshypotheticalsbymodifyingnewcase,andtheyareused

by mo difying precedents. As a result, we can nd the knowledge that does not exist in

the precedent database. This kindof knowledgemightbe aissue of legal.

We compared this system with HYPO that is legal reasoning system. HYPO has a

set of \dimensions" representing factors that can aect the relative strength of cases. It

corresponds totemporalrelations inthis system. Besides this, thereare severalthings in

common. BecauseHYPOisoneofthe mostsophisticatedCBRsystem,itissupp orted by

thesecorrespondencethattheapproachofthissystemiseective. Moreover,wepointout

the dierences between twosystems, and we can clear the merit of similarity assessment

in this system.

This system isimplemented with QUIXOTE(adeductiveobject-oriented database lan-

guage) that has abductiveinference mechanism tocomplement lacking information. Be-

causemost ofthe legaldata andknowledgeareincomplete,this mechanismisveryuseful

in legalreasoning.