Full text of "A Pragmatic Legal Expert System"

James Popple

1,1 ED

LEGAI

isiiKUCujyii

A PRAGMATIC LEGAL EXPERT SYSTEM

To my parents, Kevin and Marli,
for their unfailing love and support;
and to Paula: for whom, but for whom

A Pragmatic

Legal Expert System

JAMES POPPLE

Visiting Fellow

The Australian National University

Dartmouth

Aldershot • Brookfield USA • Singapore • Sydney

All rights reserved. No part of this publication may be reproduced, stored in a
retrieval system, or transmitted in any form or by any means, electronic, mechani-
cal, photocopying, recording, or otherwise withoutthepriorpermissionof Dartmouth
Publishing Company Limited.

Published by

Dartmouth Publishing Company Limited

Go wer House

Croft Road

Aldershot

Hants GUI 13HR

England

Dartmouth Publishing Company
Old Post Road
Brookfield
Vermont 05036

USA

British Library Cataloguing in Publication Data

Popple, James
A pragmatic legal expert system. - (Applied legal
philosophy)

1. Law 2. Expert systems (Computer science)
I. Title
340'.0285*533

Library of Congress Catalogiitg-in-Publication Data

Popple, James, 1964-

A pragmatic legal expert system / James Popple.

p. cm. -(Applied legal philosophy)
Includes bibliographical references and index.
ISBN 1-85521-739-2

1. SHYSTER. 2. Law-Methodology-Data processing. 3. Law-
-Interpretation and construction-Data processing. 4. Expert
systems (Computer science)-Australia. 5. Law-Australia-
-Methodology-Data processing. I. Tide. II. Series.
K212.P664 1996

340M-dc20 96-7058

CIP

ISBN 185521 739 2
Printed in Great Britain by the Ipswich Book Company, Suffolk

Contents

Figures ix

Series preface xi

Preface xiii

Chapter 1: Legal expert systems 1

1.1

Introduction

1.2

Jurisprudence

1.3

Jurimetrics and the behaviourists

V.i

1.4

Rule-based systems

1.5

Case-based systems

1.6

Hybrid systems

1.7

Conceptual models: deep and shallow

1.8

Conclusion

Notes

Chapter 2: A pragmatic approach to case law

2.1

Introduction

2.2

Design criteria

2.3

A model of legal reasoning

2.4

Experts, users, and expert users

2.5

Knowledge representation

2.6

A specification language

2.7

Weighting attributes

2.8

Detecting attribute dependence

2.9

Calculating distances

2.10

Nearest cases and nearest results

2.11

Writing a report

2.12

Safeguards

101

2.13

Testing and evaluation

103

2.14

Conclusion

111

Notes

118

vi A pragmatic legal expert system

Chapter 3: Implementing SHYSTER 125

3.1 Introduction 126

3.2 The Shyster module 128

3.3 The Statutes module 128

3.4 The Cases module 130

3.5 The Tokenizer module 131

3.6 The Parser module 132

3.7 The Dumper module 138

3.8 The Checker module 141

3.9 The Scales module 146

3.10 The Adjuster module 147

3.11 The Consultant module 147

3.12 The Odometer module 148

3.13 The Reporter module 151

3.14 Conclusion 159
Notes 159

Chapter 4: Case studies 163

4.1 Introduction 164

4.2 A simulation of FINDER 164

4.3 The authorization of copyright infringement 173

4.4 Employees and independent contractors 191

4.5 The implication of natural justice 207

4.6 Conclusion 230
Notes 234

Chapter 5: Conclusion 243

5.1 Introduction 244

5.2 Evaluating SHYSTER 244

5.3 Future research 250

5.4 The contribution of the SHYSTER project 252

Notes 253

Appendix A: Example reports 255

A.l Introduction 256

Notes 256

A.2 Report file for Parker v. BA 257

A.3 Report file for APRA v. Jain 263

A.4 Report file for Re Porter 270

A.5 Report files for Ainsworth v. CJC 278

Contents vii

Appendix B: A complete example 289

B.l Introduction 290

Notes 290

B.2 Log file for BWIU v. Odco 291

B.3 Dump file for the Employee specification 294

B.4 Probabilities file for the Employee specification 322

B.5 Weights file for the Employee specification 323

B.6 Distances file for BWIU v. Odco 323

B.7 Report file for BWIU v. Odco 331

Bibliography 341

Cases 367

Statutes 373

Index 375

Figures

2.1 Structure of an example specification 71

2.2 Forms of functional dependence 81

2.3 Two-way association table 85

2.4 Example distances 91

2.5 SHYSTER's algorithm for choosing cases 94

2.6 SHYSTER's algorithm for using cases: part A 96

2.7 SHYSTER's algorithm for using cases: part B 97

2.8 SHYSTER's algorithm for using cases: part C 98

3.1 SHYSTER's command line switches 129

3.2 Keywords in SHYSTER's specification language 131

3.3 EBNF definition of SHYSTER's specification language 133

3.4 Attribute values for the Employee area 139

3.5 Example attribute values 142

3.6 Probabilities for example attributes 143

3.7 Extract from probabilities for the Employee area 146

3.8 Similarity measures 149

3.9 Distances for BWIU v. Odco 150

3.10 Extract from the log file for BWIU v. Odco 157

4.1 Probabilities for the Finder area 170

4.2 Weights for the Finder area 171

4.3 Distances for Parker v. BA 174

4.4 Probabilities for the Authorization area 180

4.5 Weights for the Authorization area 181

4.6 Distances for CBS v. Ames 184

4.7 Distances for Narich v. CPT 198

4.8 Relationship between areas in the Natural specification 210

4.9 Weights for the Natural specification 215

4.10 Probabilities for the Natural specification 216

4.11 Distances for Twist v. Randwick in the Natural area 219

4.12 Summary of test cases 231

4.13 Summary of generated testing 233

Series preface

The objective of the Dartmouth Series in Applied Legal Philo-
sophy is to publish work which adopts a theoretical approach to
the study of particular areas or aspects of law or deals with general
theories of law in a way which focuses on issues of practical moral
and political concern in specific legal contexts.

In recent years there has been an encouraging tendency for legal
philosophers to utilize detailed knowledge of the substance and
practicalities of law and a noteworthy development in the theor-
etical sophistication of much legal research. The series seeks to
encourage these trends and to make available studies in law which
are both genuinely philosophical in approach and at the same time
based on appropriate legal knowledge and directed towards issues
in the criticism and reform of actual laws and legal systems.

The series will include studies of all the main areas of law.
presented in a manner which relates to the concerns of specialist
legal academics and practitioners. Each book makes an original
contribution to an area of legal study while being comprehens-
ible to those engaged in a wide variety of disciplines. Their legal
content is principally Anglo-American, but a wide-ranging com-
parative approach is encouraged and authors are drawn from a
variety of jurisdictions.

Tom D. Campbell

Series Editor

Faculty of Law

The Australian National University

Preface

The history of the development of legal expert systems has, for
the most part, been characterized by the development and imple-
mentation of complex models of legal reasoning.

This book aims to show that a legal expert system need not
be based upon a complex model of legal reasoning in order to
produce useful advice. It advocates a pragmatic approach to the
design of legal expert systems — an approach based on the way in
which lawyers deal with the law on a day-to-da}' basis.

It argues that a system based upon a simple model of legal
reasoning can still produce good advice, where that advice is eval-
uated by reference to the accuracy of the system's predictions and
to the quality of its arguments. Furthermore, such a system, with
its simpler knowledge representation structure, makes simpler the
process of knowledge acquisition.

These arguments are made theoretically, and then by example.
I have developed a legal expert system which is based on a prag-
matic approach to the law. The development and testing of that
system, called SHYSTER, is described in this book.

SHYSTER is of a general design, so that it can operate in different
legal domains. It was designed to provide advice in areas of case
law that have been specified by a legal expert using a specially
developed specification language.

SHYSTER is a case-based legal expert system. Its knowledge of
the law is acquired, and represented, as information about cases.
It produces its advice by examining, and arguing about, the sim-
ilarities and differences between cases. By contrast, a rule-based
expert system represents the law using rules. A hybrid system
uses both rule-based and case-based techniques. SHYSTER has
been designed so that it can be linked with a rule-based system to
form a hybrid legal expert system.

xin

xiv A pragmatic legal expert system

Although SHYSTER attempts to model the way in which lawyers
argue with cases, it does not attempt to model the way in which
lawyers decide which cases to use in those arguments. It uses
statistical techniques to quantify the similarity between cases, and
chooses cases on the basis of that similarity measure. Its repres-
entation structure was designed to be as simple as possible while
complex enough to allow SHYSTER to produce good advice.

The body of this book is divided into five chapters:

• Chapter 1 discusses previous work of relevance to the devel-
opment of legal expert systems. The value of jurisprudence
to the development of such systems is also discussed, and the
adoption of a pragmatic approach (as opposed to a jurispru-
dentially pure approach) is recommended.

• A pragmatic approach is proposed in chapter 2. It incor-
porates a simple model of legal reasoning, and uses a simple
knowledge representation structure. Comparisons are made
between this approach and others. This approach is adopted
for the development of SHYSTER. Specific design criteria for
SHYSTER are detailed, and methods of testing and evaluating
the system are discussed.

• The implementation of SHYSTER is explained in chapter 3.
The twelve modules that comprise SHYSTER are described,
and their operation is demonstrated using examples.

• Four different specifications have been written for SHYSTER,
and these are used as the basis of case studies in chapter 4.
Each specification represents a different area of case law. Sev-
eral different methods are employed to test SHYSTER and
these specifications.

• In chapter 5, conclusions are drawn about SHYSTER and its
approach to case law. Some enhancements to SHYSTER are
suggested, avenues of future research are identified, and the
nature of the contribution made by the SHYSTER project is
discussed.

There are two appendices to this book. Appendix A provides
example reports relating to each of the four specifications. Each of
these reports is SHYSTER's advice on one of the test cases used in
chapter 4. A complete example of SHYSTER's output for another
test case is given in appendix B.

Preface xv

I have endeavoured to explain SHYSTER and its approach to case
law as clearly as possible. Occasionally — especially in chapters 2
and 3 — this explanation involves recourse to mathematical detail.
Readers for whom the sight of a formula provokes either great
fear or profound indifference, and who don't feel a need to under-
stand the intricacies of SHYSTER'S inner workings, are advised to
disregard any offending mathematics.

For those who would like to experiment with the system. SHYSTER
and the case law specifications discussed in this book are available
on the worldwide web at http://cs.anu.edu.au/software/shyster.

This book is based upon my doctoral thesis which I wrote while a
research student at The Australian National University, and com-
pleted in April 1993. I thank the members of my supervisory
panel — -Robin Stanton. Roger Clarke. Peter Drahos and Malcolm
Newey — for their counsel and encouragement, and Robin Creyke
and Phillipa Weeks for giving so freely of their time and legal ex-
pertise. I have also been provided with valuable advice on this
work by my thesis examiners — Kevin Ashley, Bob Moles and Alan
Tyree — and by Richard Susskind.

Notwithstanding this assistance, for which I am most grateful,
responsibility for this book remains mine.

James Popple
April 1995

shy-ster \'shTsto(r)\ n -s [prob. after Scheuster //1840 Am. attorney fre-
quently rebuked in a New York court for pettifoggery] : one who is profes-
sionally unscrupulous esp. in the practice of law or politics . . .

Webster's Third New International Dictionary (1961)

shyster ('Jaisto(r)) . . . [Of obscure origin.

It might be f. SHY a. (sense 7, disreputable) 4- -ster: but this sense of the
adj. is app. not current in the U.S.]

. . . ; A lawyer who practises in an unprofessional or tricky manner; especially,
one who haunts the prisons and lower courts to prey on petty criminals; hence,
any one who conducts his business in a tricky manner' (Funk's Stand. Diet.
1895). Also attrib. or adj. Orig. and chiefly U.S. slang ...

The Oxford. English Dictionary (second edition, 1989)

shyster. An unscrupulous lawyer (note that the definition presumes the
existence of scrupulous ones) . . .

The term does not come from — as suggested in various dictionaries — the
surname Scheuster, supposedly a lawyer noted for shyster-like practices; from
the name of the Shakespearean character, Shylock; ... or from any of the
various meanings of shy (e.g., to be shy of money). Rather . . . shyster evolved
from the underworld use of shiser, a worthless fellow, which derived in turn
from the German scheisse, excrement, via scheisser, an incompetent person
(specifically, one who cannot control his bodily functions) . . .

Hugh Rawson (1991)
A Dictionary of Invective

xvn

Legal expert systems

In the yahoo world of computer public relations, it is often the loudest mouths
which lead the unsuspecting computer user into that lonely canyon of empty
pockets and broken promises. It now seems . . . that parts of the academic
world are fast approaching that decibel level so far achieved only by computer
salesmen. Thus ... it seems that space can only be booked on the band-wagon
if one is prepared to make more outrageous claims than the next man. And
such claims are being made for the application of AI to the law.

Philip Leith (1986) 1

The computer scientists, encouraged by the modern positivists, fail to recog-
nize . . . that law, positive morality and ethics are inseparably connected parts
of a vast organic whole. Judgments are involved at every stage of the legal
process and machines cannot make judgments. In stating that legal rules can
be applied without further judgment; that they apply in an all or nothing
fashion; that legal decision making follows the form of the syllogism or that
it is a pattern-matching routine, the modern positivists, joined now by the
computer scientists take us along a dangerous road.

Robert N. Moles (1987)
Definition and Rule in Legal Theory:
A Reassessment of H. L. A. Hart and the Positivist Tradition

Aussi, lorsqu'un homme se rend
plus absolu, songe-t-il d'abord a
simplifier les lois. On commence,
dans cet Etat, a etre plus frappe
des inconvenients particuliers, que
de la liberte des sujets dont on ne
se soucie point du tout.

Thus when a man takes on absolute
power, he first thinks of simplifying the
law. In such a state one begins to be
more affected by technicalities than by
the freedom of the people, about which
one no longer cares at all.

Montesquieu (1748)
De I'Esprit des lois 3

A pragmatic legal expert system §1-1

1.1 Introduction

This chapter reviews previous work of relevance to the design
of legal expert systems. Jurisprudence and its value to the de-
velopment of such systems is discussed (§1-2) and the field of
"jurimetrics" and some behaviouristic research is examined (§1.3).
The development of legal expert systems is discussed under three
headings: rule-based, case-based, and hybrid systems. 4

Rule-based systems are examined first (§1.4), and special at-
tention is paid to the work of four projects. The problems of
knowledge acquisition and representation, and fact representation
are also discussed (§1.4.5 and §1.4.6). Rule-based systems have
been used to represent statutes and cases. However, as explained
in §1.4.7, rule-based systems are fundamentally inadequate for rep-
resenting case law.

Case-based systems are examined (§1.5) and three projects are
outlined. Systems that use rules in order to represent case law are
considered to be rule-based systems; only systems which adopt
case-based reasoning methods are considered to be case-based.
Applying this distinction is usually straightforward, though not
always. 5 Attempts have been made to use semantic networks to
represent case law but, as explained in §1.5.4, they are not suited
to the task.

Hybrid systems (§1-6) employ both rule-based and case-based
methods.

The search for conceptual models of legal reasoning is discussed
(§1.7). Finally, conclusions are drawn from the literature as to the
best approach to the design of legal expert systems (§1.8).

But first it is necessary to explain what is meant by a legal expert
system, and to outline some key concepts.

1.1.1 Legal expert systems

The range of computer applications in the law is wide. It extends
from general applications, of use to lawyers, to applications de-
signed specifically for the law. This book is concerned only with
a subset of those systems that make use of artificial intelligence
(Al) techniques to solve legal problems.

Legal AI systems can usefully be divided into two categories:
legal retrieval systems and legal analysis systems. Legal retrieval

§1.1 Legal expert systems

systems allow lawyers to search through databases containing de-
tails of statutes and decided cases. AI techniques may be employed
to simplify this task: e.g. by searching for keywords which have
not been input by the user but are deduced to be equivalent to, or
sufficiently related to, the input keywords.' Legal analysis systems
take information about a set of facts and determine the ramifica-
tions of those facts in a given area of law.

Mehl claims that there is no fundamental difference between
these two categories — that the difference is one of degree only. 8
Shannon and Golshani suggest that the difference between s} r stems
based on a "conceptual model of legal analysis" and text retrieval
systems is that the latter do not "understand" any area of the
law. 9 Similarly. Susskind says that "knowledge-based" systems,
as opposed to "database" systems are capable of "applying their
knowledge of the law to the problem data presented to them." 10

A better distinction — one which avoids the vexed question of
whether any AI system can really be said to know or understand
anything — can be made by reference to the output of the system.
The output from a legal analysis system is such that, if it had been
produced by a human, that human would be said to have legal
expertise. By contrast, the output from a legal retrieval system
could be produced by a human possessed of no legal expertise;
such output is used in, and is not the product of. legal analysis. 11

This book is concerned only with legal analysis systems.

Legal analysis systems can be divided into two categories: judg-
ment machines and legal expert systems.

A judgment machine is a machine designed to replace a human
judge. Such machines were first proposed forty years ago, though
no such proposals have been made for nearly tw r o decades. Writ-
ings on judgment machines are discussed below because they are
of historical interest, and because the idea of a judgment machine
raises some issues which are also relevant to the second category
of legal analysis systems: legal expert systems.

A legal expert system, as the term is used in this book, is a
system capable of performing at a level expected of a lawyer. AI
systems which merely assist a lawyer in coming to legal conclusions
or preparing legal argument are not here considered to be legal
expert systems; a legal expert S3*stem must exhibit some legal
expertise itself. This definition does not exclude systems that can
only be used by legal experts; several systems — including SHY-
STER — have been developed for use exclusively by lawyers. 12

A pragmatic legal expert system §1-1

1.1.2 So urces of law

The law in so-called "common law countries'' — e.g. Australia. Bri-
tain, Canada and the USA- — is derived from legislation and from
case law.

Legislation, or statute law, consists of statutes and delegated
legislation. In Australia, statutes are made by the Federal Parlia-
ment, or a State Parliament, and enacted when given royal assent
by the Governor-General, or a State Governor. Delegated legis-
lation (rules, regulations, ordinances, by-laws, etc.) is made by a
person or body to whom legislative power has been delegated by
statute.

Case law, or the common law, is judge-made law: judicial resolu-
tions of specific disputes. The sources of case law are the published
reports of the cases as heard before various courts.

Legislation takes precedence over case law; 13 parliaments can
override judge-made law by legislative enactment. However, judges
have the task of determining the meaning of legislation.

Any legal expert system which seeks to deal with the law in a
common law country must account for both statute law and case
law.

1.1.3 The doctrine of precedent

The development of case law is based upon the principle of stare
decisis which holds that courts should apply the doctrine of pre-
cedent. Morris et al. summarize the general rules of the doctrine
of precedent as follows:

• each court is bound by decisions of courts higher in its hierarchy;

• a decision of a court in a different hierarchy may be of consid-
erable weight, but will not be binding;

• only the ratio decidendi of a case is binding;

• any relevant decisions, although not binding, may be considered
and followed; and

• precedents are not necessarily abrogated by lapse of time. 14

Courts have a tendency to follow similar cases, even when not
strictly bound to do so. According to Cross:

It is a basic principle of the administration of justice that like
cases should be decided alike. This is enough to account for the
fact that, in almost every jurisdiction, a judge tends to decide a

§1.2 Legal expert systems

case in the same way as that in which a similar case has been
decided by another judge. 15

There is considerable argument about the extent to which judges
actually apply the doctrine of precedent. 10 Some legal theorists
contend that the doctrine is simply part of the public discourse
that judges use to justify their decisions. Stone claims that "the
degree of certainty and stability in the law secured by the doctrine
of stare decisis is far less than it appears to be," and that one of
the important social functions of the doctrine is "to maintain at a
maximum the feeling and appearance of certainty and stability." 17
Yet, judges of the highest court in Australia have been known to
follow previous cases which they believed to be wrongly decided. 18

What is clear is that judges use previously decided cases to
justify their decisions — if not to reach them — and that lawyers
use previously decided cases in legal argument. Any legal expert
system which deals with case law assumes the application of the
doctrine of precedent, at least to this extent.

1.1.4 Terminology

<->.

For consistency, the following three terms are used in this book
for a variety of terms that are used throughout the literature. An
attribute is a legally important fact; some authors use the words
"descriptor," "dimension." "factor." "feature" or "variable." The
instant case is the situation about which the expert system is in-
terrogated; this is sometimes called the "actual case." the "current
fact situation," the "hypothetical case" or the "new case." Lead-
ing cases make up the case base of a legal expert system; some
call these the "reference set" or the "training set."

1.2 Jurisprudence

1.2.1 The importance of jurisprudence

For centuries, lawyers and philosophers have written on, and ar-
gued about, the nature of human laws. On the face of it. it would
seem sensible for the developer of a legal expert system to have re-
gard to this jurisprudential theory before designing her/his system.
However, as Susskind complained in 1987, most of the published
research on various legal expert systems makes no use of jurispru-
dential resources. 19

A pragmatic legal expert system §1-2

Some expert system researchers have been doubtful as to the
value of jurisprudence to their research. Niblett, for example,
claims that "a successful expert system is likely to contribute more
to jurisprudence than the other way round". 20 Susskind disagrees:

It is beyond argument . . . that all expert systems must conform
to some jurisprudential theorj' because all expert systems in law
necessarily make assumptions about the nature of law and legal
reasoning. To be more specific, all expert systems must embody
theories of legal knowledge, legal science, the structure of rules,
the individuation of laws, legal systems and sub-systems, legal
reasoning, and of logic and the law (as well perhaps as elements
of a semantic theory, a sociology, and a psychology of law), the-
ories that must all themselves rest on more basic philosophical
foundations. If this is so, it would seem prudent that the general
theory of law implicit in expert systems should be explicitly artic-
ulated using (where appropriate) the relevant works of seasoned
theoreticians of law. Perhaps the reason that there is, as yet,
no overwhelmingly successful system is that the vast corpus of
apposite jurisprudential material has not yet been tapped in the
construction process. 21

Although it is true that all legal expert systems necessarily make
assumptions about the nature of law and legal reasoning, it does
not follow that they must conform to some jurisprudential theory.
A lawyer must have a model of the law (maybe unarticulated)
which includes assumptions about the nature of law and legal
reasoning, but that model need not rest on basic philosophical
foundations. It may be a pragmatic model, developed through
experience within the legal system. Many lawyers perform their
work with little or no jurisprudential knowledge, 22 and there is no
evidence to suggest that they are worse, or better, at their jobs
than lawyers well-versed in jurisprudence.

Susskind concedes that it is possible to build a legal expert
system "without jurisprudential insight 11 , but suggests that such a
system would be of very poor quality:

Because successful legal knowledge engineering presupposes so
profound a familiarity with the nature of law and legal reason-
ing, it is scarcely imaginable that such a mastery could be gained
other than through immersion in jurisprudence. 23

Harris, himself a jurisprudent, takes a different view:

People acquire those technical skills of legal reasoning and legal
argumentation which make up the concept of 'good lawyer' by

§1.2 Legal expert systems

immersing themselves in substantive legal subjects. Jurisprudence
has to do, not with the lawyer's role as a technician, but with any
need he may feel to give a good account of his life's work— either
to fellow citizens, or to himself, or to any gods there be. 24

If Harris is right- — and the existence of good (though jurispruden-
tial!}' illiterate) lawyers suggests that he is — then the importance
of jurisprudence to legal expert system design is questionable. A
legal expert system need only operate at the same level of ab-
straction as does a lawyer, rather than at the philosophical level
of a jurisprudent. The fact that many lawyers have mastered the
process of legal reasoning, without having been immersed in juris-
prudence, suggests that it may indeed be possible to develop legal
expert systems of good quality without jurisprudential insight.

This does not mean that legal expert systems designers should
completely ignore jurisprudential literature. However, as a legal
expert system need not conform to any jurisprudential theory, a
pragmatic approach to expert system design may be preferable to a
jurisprudential!}* pure one. Susskind, as Niblett complains, "gives
the role of jurisprudence in the design of legal expert systems a
greater significance than it deserves." 25

It is well beyond the scope of this book to cover ''the vast corpus
of apposite jurisprudential material". 20 however several areas of
jurisprudence of relevance to expert system design are discussed
below .

1.2.2 Scientific and mechanical jurisprudence
The dictionary defines jurisprudence as:

The science which treats of human laws (written and unwritten)
in general; the philosophy of law.

Pound refers to the "scientific character" of the law:

Sir Frederick Pollock gives us the clew when he defines the reasons
that compel law to take on this scientific character as three: the
demand for full justice, that is for solutions that go to the root of
controversies; the demand for equal justice, that is a like adjust-
ment of like relations under like conditions; and the demand for
exact justice, that is for a justice whose operations, within reason-
able limits, may be predicted in advance of action. In other words,
the marks of a scientific law are, conformity to reason, uniformity,
and certainty. 28

8 A pragmatic legal expert system §1-2

This approach to the law has been much criticized. Frank contends
that uncertainty is inherent in the legal process, and that seeking
certainty in general legal principles is simply "an expression of in-
fantile emotional attitudes which have persisted into adulthood." 29

In his 1908 paper. Mechanical Jurisprudence, Pound distin-
guishes scientific jurisprudence from mechanical jurisprudence:

Roman law in its decadence furnishes a striking example [of mech-
anical jurisprudence]. The Valentinian "law of citations" made a
selection of jurisconsults of the past and allowed their writings
only to be cited. It declared them, with the exception of Pap-
inian, equal in authority. It confined the judge, when questions of
law were in issue, to the purely mechanical task of counting and of
determining the numerical preponderance of authority. Principles
were no longer resorted to in order to make rules fit cases. The
rules were at hand in a fixed and final form, and cases were to be
fitted to the rules. 30

By contrast, Pound says, scientific law is "a reasoned body of
principles for the administration of justice". 31

Loevinger also argues that there is a role for science in law,
but sees little science in jurisprudence. In 1949 he proposed an
alternative to jurisprudence which he termed jurimetrics:

The next step forward in the long path of man's progress must be
from jurisprudence (which is mere speculation about law) to juri-
metrics — which is the scientific investigation of legal problems . . .
The inescapable fact is that jurisprudence bears the same relation
to a modern science of jurimetrics as astrology does to astronomy,
alchemy to chemistry, or phrenology to psychology. It is based
upon speculation, supposition and superstition; it is concerned
with meaningless questions; and, after more than two thousand
years, jurisprudence has not yet offered a useful answer to any
question or a workable technique for attacking any problem. 32

This proposal spawned a new field of stud}*, some work from which
is discussed below. 33

Loevinger calls those who fear the dangers of mechanized juris-
prudence "quixotic and uncomprehending" 34 — though he was not
a mechanical jurisprudent himself. 35 Some, however, saw the de-
velopment of computers as an opportunity to develop a judgment
machine: a machine that could replace a judge.

§1.2 Legal expert systems 9

1.2.3 Judgment machines

In 1955, Lasswell predicted that:

When machines are more perfect [sic] a bench of judicial robots
. . . can be constructed. The machine would apply a system of
"weights" to allegations of fact made bj' parties to a controversy,
and also to the justifications advanced in support of the claims put
forward by participants. Litigation can proceed by counsel for the
plaintiff and the defendant pressing buttons that translate their
cases into the physical signs built into the machine. Many results
would be "no decision." However, the machine could be designed
to settle a controversy of this kind with a "random" operation (b} r
lot). 30

He was not altogether serious:

It is a challenging task for legal historians to assist in constructing
a robot whose weights would give substantially the same result as
those produced by the [US Supreme] Court at various periods. The
task would not be too difficult for some justices on some issues.
But a robot facsimile of the less repetitive members of the Court
would provide a genuine challenge to the engineers. 37

However, some have taken the idea of a judgment machine very
seriously indeed. In 1949, Frank wrote that a "logic machine 11
might "disclose all possible available alternative legal rules," al-
though judges would still have to exercise the "the sovereign
prerogative of choice 11 between the rules on the basis of the judges'
"conscious or unconscious notions of policy."' Similarly Mehl
wrote, in 1959, that a machine could perform some of the func-
tions of a judge, but that a role for humans would remain because
"the solution to a legal problem may depend upon extra-rational
factors, involving the whole of human experience". 39

As recently as 1977, it has been seriously suggested that a judg-
ment machine could replace a human judge. DAmato proposed
that a machine could take the relevant facts of a case as input and
produce a number in the range —1 to +1 (where a positive num-
ber indicates a victory for the plaintiff). Given the multiplicity of
factors, he claims, a result of zero would be extremely unlikely, 40
although it is not clear why this would be so. These facts would
be determined by a jury, but the law would be decided by the
machine. Somewhat grudgingly, he allows for a vestige of human

10 A pragmatic legal expert system §1-2

control: an appeal court could review all of the machine's determ-
inations in a certain numerical range (e.g. —0.05 to 0.05) within
which the cases would be so close that a re-examination might be
required. The review court's subsequent decision would then be
incorporated into the system. 41

The idea of human judges being replaced by machines has been
trenchantly criticized. According to Weizenbaum:

The very asking of the question, "What does a judge . . . know
that we cannot tell a computer?" is a monstrous obscenity. That
it has to be put into print at all, even for the purpose of exposing
its morbidity, is a sign of the madness of our times.

Computers can make judicial decisions . . . They can flip coins in
much more sophisticated ways than can the most patient human
being. The point is that they ought not be given such tasks. They
may even be able to arrive at "correct" decisions in some cases —
but always and necessarily on bases no human being should be
willing to accept. 42

But D'Amato sees advantages in replacing human judges by ma-
chines:

Would we lose a judge's "judgment," and how important would
such a loss be to our legal system? Surely computers do not make
"judgments" the way humans do, and so we would lose the "hu-
man" aspect of legal judgments. But what specifically do we lose
when we lose the humanness of judgments? Is human judgment
just a euphemism for arbitrariness, discretion, or bias? 43

By contrast. Stone stresses the importance of legal change spring-
ing from "deviance, tentativeness, and even indecision in judg-
ment."

It is these phenomena above all which promote a judge's sensit-
ivity to new ideas or to newly perceived social situations and stir
Hamlet-like introspection. On these there rest some of the main
foundations of the social good we call "justice." 44

Proponents of the idea of automated judges claim that such sys-
tems would reduce the cost of the legal system, find inconsistencies
in the law. and provide a level of certainty in the law which does
not exist at present. 45 D'Amato claims that a judgment machine
would allow people to live under the rule of law and not under the
"rule of persons." 40

§1.2 Legal expert systems 11

D'Amato's argument is based upon "two bold initial premises."
He assumes that, from a jurisprudential point of view, the law has
been made completely determinable. He also assumes that, if a
computer can be programmed to make judicial decisions, human
"discretion" will have been completely removed. 47

His views on judgment machines are extreme, and have not been
widely accepted. For the most part, designers of legal analysis
systems have focused on building systems which provide advice to
lawyers or to laypeople, rather than systems which usurp the role
of judges. However, many of these designers have made D'Amato's
first assumption — that the law T is completely determinable — often
tacitly. 48

1.2.4 Petrifaction of the law

The idea of a judgment machine removing uncertainty and ambi-
guity in the law raises the possibility of the petrifaction of the law.
This possibility is also relevant to legal expert system design.

D'Amato claims that, once programmed, the law would become
settled. The computer would stop "progress," although the le-
gislature could always step in if anomalous results were being
produced. 49 Similarly, Kayton believes that the use of proposi-
tional logic could identify legal ambiguity and provide guidelines
for its resolution. He asks:

Would not each concrete demonstration of ambiguity and its res-
olution tend to rigidify the law and eventually destroy the very
flexibility which has made the common law viable? . . . Despite the
dangers of generalization, it is submitted that . . . [this question]
should be answered in the negative. Information brought to bear
in a rational pursuit is alwa}^ better than ignorance or confusion.
The elimination of ambiguities, rather than ossify the law, would
produce an optimum condition for the purposeful, intelligent, ami
efficient development of the law. 50

Schubert, another early researcher in this field, is not convinced of
the possibility, or the importance, of certainty in law. He writes:

. . . the ideal of certainty in law is tolerable only in the context of
an empirical world in which forces inducing change are so manifold
that the attainment of the goal is never possible. 01

Tyree argues that a deterministic computer judge could "over-
rule himself" if cases are added as thev are decided. 52 But as

12 A pragmatic legal expert system §1-2

Pound pointed out almost ninety years ago, the problem is more
than mere ossification of the law:

The effect of all system is apt to be petrifaction of the subject
systematized. Perfection of scientific system and exposition tends
to cut off individual initiative in the future, to stifle independent
consideration of new problems and of new phases of old problems,
and to impose the ideas of one generation upon another. 03

This problem is inherent in expert system design — in the law
and in other domains. There are dangers, for example, in the use
of a legal expert system by judges. Stone considers the use of
predictive techniques in the law and warns that:

. . . reliance at the judgment seat on new predictive techniques
would seriously threaten the judge's central concern with justice.
For if the results thus predicted for him do affirmatively guide him
in present decisions, each judge will tend to vote somewhat more
consistently with his past record. The margin of deviation would
accordingly disappear in deference to predicted patterns, aided by
the human tendency to follow the less agonizing because already
trodden path . . . 54

Judges. Stone says, have a duty to strive wholeheartedly for justice
u at the moment of judgment." 05 Although an expert system may
be of some use to judges of lower courts, there is a critical bor-
der between a machine serving the legal order and the dangers of
subversion of that order.

For whenever an appellate judge faces the duty to do justice in
this sense, he must address himself to it with his own present
experience and insights. It would be a corruption of justice for
him to shortcut this duty by resting on predictions of his future
decisions based on his own past behavior — still more so if the basis
is the average past behavior of a group of judges. DG

So. as Stone points out, the "pioneers and experts of the new
techniques" must face "the limits of the contributions they can
make." 57

1.2.5 Clear rules and clear cases

H. L. A. Hart was the most famous of the positivist legal theorists.
His major work, The Concept of Law, was published in 1901. Hart
realizes that because the law is expressed in natural language, it

§1.2 Legal expert systems 13

is subject to considerable semantic indeterminacy. This he terms
the open texture of law. However, he claims that it is possible
to use rules deductively to solve "clear cases": that is "those in
which there is general agreement that they fall within the scope
of a rule." 58 He also contends that:

. . . the result of the English system of precedent has been to pro-
duce, bj' its use, a body of rules of which a vast number, of both
major and minor importance, are as determinate as any statutory
rule. They can now only be altered by statute, as the courts them-
selves often declare in cases where the 'merits 1 seem to run counter
to the requirements of the established precedents. 59

As Susskind writes, one ramification of Hart's analysis is that:

... all expert systems in law whose inference procedures are solely
deductive will function exclusive^ in the clear case domain, and
will be of no aid in the solving of "problems of the penumbra." 60

But what constitutes a "clear case" is itself far from clear. As
Hart concedes:

... it is a matter of some difficulty to give an exhaustive account
of what makes a 'clear case' clear or makes a general rule obvious
and uniquely applicable to a particular case. 61

Moles disputes the very existence of clear rules and clear cases.
He performs a detailed analysis of the application of an example
of an ostensibly clear rule in a British statute: a provision which
prescribes the circumstances in which an injunction should be is-
sued in domestic violence cases. 02 He summarizes the effect of
three cases as follows:

(1) Before B v. B 63 we are concerned with a statutory provision
which on its face appears to be clear and comprehensible (at least
to non-lawyers) and it would appear that an injunction should
issue in cases such as those we have looked at.

(2) After Cantliff, 64 the provision has been considered twice
by the [English] Court of Appeal within two weeks. All of the
six judges who considered the matter are in agreement that the
injunction should not issue and that the rule is clear.

(3) The matter is further considered by a specially constituted
Court of Appeal of five, 00 who decide that the rule is clear, but
different from that in (2). and that the injunction should issue. 00

14 A pragmatic legal expert system §1-2

Moles also analyses the judicial application of the rule of pre-
cedent^ — which, with this statutory rule, "must be regarded as
amongst the clearest available to us" ° —to demonstrate that "our
experience of the legal system bears little relationship to Hart's
account of it". 68

Leith developed a legal expert system which operated on "clear
rules" in the law. 09 He later recanted, saying that:

. . . the very idea of a clear rule is inherently confusing and is not
observable in the real operation of the judicial process . . . judicial
creativit}' is not an aberation [sic] of the legal process, but (as
Moles . . . suggests), the very heart of the law. 70

As is explained below, 71 many expert system designers have adop-
ted a view of the law that allows for clear rules and clear cases —
without considering the possibility that the law may not be like
that at all.

1.2.6 Legal realism and rule scepticism

In the 1920s and 1930s, a movement called legal realism developed
in America. 72 Realists rejected the importance that mechanical
jurisprudence placed on rules. However, the rule scepticism of the
American Realists is meek by comparison with the strong inde-
terminacy thesis. Drahos and Parker explain that:

According to Karl Llewellyn a characteristic feature of Realism
was the rejection of simple (by which he meant general) rules
and the substitution of more detailed classificatory schemes which
better captured the specific nature of judicial rule-making . . . On
Llewellyn's account, rule scepticism emerges as a set of doubts
about the veracity of legal actors' claims to be following the legal
rules they say the}' are. This is not rule scepticism in the strong
sense of denying the existence of rules, however. Legal actors
may simply be following some other rules. The rule scepticism of
American Realism could perhaps be more accurately described as
rule cynicism. 73

Kripke. by contrast, is a true rule sceptic. His argument is
an example of the Wittgensteinian Paradox.' 4 Consider the two
functions "plus" (+) and "quus" (©). The + function is the math-
ematical function, addition. The © function is defined as follows:

L 5, otherwise.

§1.2 Legal expert systems 15

Suppose that Kripke has used + in the past, but always with values
of x and y smaller than 57. He performs the computation 68 + 57
and gets a result of 125. Yet it could be said that when he thought
he was using the "plus" function in the past, he was in fact using
"quus." As Kripke explains:

... in this new instance, I should apply the very same function or
rule that I applied so many times in the past. But who is to say
what function this was? In the past I gave myself only a finite
number of examples instantiating this function. So perhaps in the
past I used 'plus* and '+' to denote . . . 'quus' . . . Who is to say
that ['®'] is not the function I previously meant by '-p? 75

There is no justification for Kripke answering 125 rather than 5.
There is no way of determining (from his past behaviour — even his
past thoughts) whether by "plus" Kripke meant + or ©. Rules are
derived from a finite number of examples; for any given rule, there
is always an alternative rule which also explains those examples.

Mathematicians could counter that the meaning of + is well
defined. But Kripke argues that:

. . . scepticism about arithmetic should not be taken to be in ques-
tion: we may assume, if we wish, that 68 + 57 is 125 ... I cannot
doubt coherently that 'plus*, as I now use it, denotes plus! Perhaps
I cannot . . . doubt this about my present usage. But I can doubt
that my past usage of 'plus* denoted plus . . .

. . . There is no objective fact— that we all mean addition by '+',
or even that a given individual does — -that explains our agreement
in particular cases. Rather our license to say of each other that
we mean addition by '+' is part of a 'language game* that sustains
itself only because of the brute fact that we generally agree. 76

Drahos and Parker write that:

The consequences of this argument, if valid, are shattering. Rules
turn out to be no more than leaps in the dark and the whole notion
of rule following seems illusory. 77

The implications of Kripke's argument have been realized by
legal theorists.' 8 As Drahos and Parker point out:

Saying that there are no rules to follow, only social practices,
means that propositions about the law are potentially open to
wild fluctuations. The argument also puts paid to any possibility
of a correspondence theory of truth in law. 79

16 A pragmatic legal expert system §1-2

They propose a solution to this problem:

The Kripkean argument does not prevent people from saying that
they are following rule X. Rather it stops them from being able
to justify the existence of that rule by reference to some objective
meaning. This still leaves the practice, as opposed to the justific-
ation, of successful rule following to be accounted for. 80

Their solution is to view the law as a set of rules and conven-
tions: "a type of rule used to fix or interpret the meaning of other
rules." 81 They claim there is a distinction between rule knowledge
and rule understanding, the latter being "a matter of absorbing
conventions relating to rule use." 82 A lawyer requires knowledge of
the rules, and understanding as to how to apply those rules prop-
erly. Drahos and Parker argue that the problems facing designers
of legal expert systems "flow from the difficulties of representing
rule understanding rather than rule knowledge."* Rules can be
used to represent rule knowledge; the problem is "how to rep-
resent with tolerable accuracy . . . conventions which confer rule
understanding." 84

But they concede that, as with other rules, conventions have to
"run the gauntlet of scepticism". 85

1.2.7 A jurisprudential consensus?

The only major examination of the role of jurisprudence in the
development of legal expert systems is Susskind's 1987 book Ex-
pert Systems in Law: A Jurisprudential Inquiry. 86 As discussed
above, 87 he argues that all expert systems must conform to some
jurisprudential theory.

He also sets out to find consensus in jurisprudential theory. As
there is far too much literature in the field for one person to cover
it all, he limits his choice of material: "the vast majority" of
sources that he uses are British analytical jurisprudential writ-
ings since the mid-1950s and early 1960s, "the impetus for which",
he concedes, "was derived very largely from the publications of
H.L.A. Hart." 88

Susskind concludes that "there are no theoretical obstacles, from
the point of view of jurisprudence, to the development of rule-
based expert systems in law of limited scope." 89 He also claims

§1.2 Legal expert systems 17

that the divergence of views within jurisprudence has been over-
stated because legal theorists tend to focus on the differences.
He claims that there is a jurisprudential consensus, "albeit of
mundane and limited application". 90

Susskind provided the theoretical justification for not only his
own development work, 91 but for much of the subsequent work of
other developers of legal expert systems. He could also be said
to have provided, retrospectively, legal theoretical justification for
most of the work on rule-based expert systems that preceded his
book.

However. Susskind's approach to jurisprudence is fundamentally
flawed. As Moles points out:

[Susskind] said that he would carry out a surve}' of the jurispru-
dential literature. He acknowledged, of course, that it would not
be possible to survey the whole of the jurisprudential literature.
In fact, he determined that law was a system of rules by "survey-
ing" only those whose avowed position was based on the fact that
the law was a system of rules . . .

I would venture to suggest that this is in fact a misuse of the
survej' technique . . . Susskind was perfectly familiar with the work
of Hart and his followers, and was well able, therefore, to find any
number of books and articles which supported the "law as rules"
view. He then developed his position ... on the basis of what this
purported consensus within jurisprudence had to say.

. . . Of course, Susskind was telling certain sections of the Al
and Law community what they wanted to hear, and hence their
enthusiasm for it. 92

Moles also notes Susskind's admission that "the most rigorous
of these writings constituted the source materials with greatest
potential given the overall purpose of the project." J Because
the first objective of Susskind's work was to design, develop and
implement an expert system in law, 94 Moles cites this admission
as evidence that Susskind prejudged his survey.

In the light of this criticism, it is ironic to note Susskind's own
caution that:

... a little jurisprudential knowledge can be a dangerous thing! It
is tempting for the jurisprudential neophyte to become an ardent
devotee of a particular school of thought within legal theory and
to go on from there to implement all and onty the teachings of
that school. This course of action should be avoided at all costs.

18 A pragmatic legal expert system §1-2

Familiarity with a wide range of works should be achieved prior
to commitment to any particular jurisprudential posture. 95

Clark, another critic of Susskind, says that:

The shortcomings of [his] book coincide with the limits of positivist
legal theory, and even then some doubt must remain as to whether
expert systems in law mark a revival of the kind of "mechanical
jurisprudence" which Hart opposed so vigorously from within the
confines of positivist legal thought. 90

But the strongest criticism of Susskind comes from Leith:

. . . [Susskind] believes that he can come to some sort of comprom-
ise with the various theoretical positions taken by the renowned
thinkers of the field and produce "a general theory" (an indication,
I might suggest, of [his] poor theoretical conceptions) . . .

. . . Susskind sees jurisprudence as providing a variety of theoret-
ical models which can be modelled mathematically and translated
into computer programs.

. . . [0]f necessity formal specification requires a formalisation
of law; there can be no "informal models 1 ' which are mysteriously
formalised into a computer model . . .

. . . Therefore, in order to use formal specifications, Susskind
must provide a formal specification of law which can then be in-
corporated into the rule- format of a computer program . . . And,
of course, formal specifications of law are renowned for their the-
oretical and practical inadequacy. Legal formalism can, surely,
hardly be the compromise he wishes might arise from the conflict-
ing positions of Kelsen, Hart, Dworkin et al. 97

. . . [I]f he really does believe that his informal theoretical models
can be transformed into formal theoretical models without loss of
their informal attributes, then I must suggest that he has really
little understanding of the discipline of computer science. 98

Susskind's claim to have found a consensus in jurisprudence-
even one of mundane and limited application — is absurd. As
shown above. 99 the views of just two jurisprudents (Hart and
Moles) are completely irreconcilable. Similarly, Hart's views are
totally at odds with those of Kripke. 100 Susskind states his theory,
and develops an expert system which conforms to that theory. His
mistake is to claim that his theory is definitive, in that it represents
a jurisprudential consensus.

§1.3 Legal expert systems 19

1.3 Jurimetrics and the behaviourists

As mentioned above. 101 the term "jurimetrics" was coined by Loev-
inger. He defines "jurimetrics" as "the study of law and legal
problems by scientific methods and concepts, the employment of
science in law to the extent that it is applicable or adaptable." 102
Loevinger is neither a mechanical jurisprudent nor a positivist:

To begin with we must be clear that science offers us neither ul-
timate nor certain answers to legal problems. The dream that
science might someday tell us which of several competing interests
is the more important is a vain one. Science essaj's no such an-
swers in any field. Science does not assign social or ethical values.
Science may, indeed, provide data from which social or ethical
judgments may be made; but the judgments will remain with man

There is no prospect of any process that will preclude considera-
tion of social desirability or wisdom. The opportunity will always
be available to argue that precedent should not be followed, and
that considerations of policy, or expediencj 1 ", require a different
rule or a special result . . .

. . . Science does not and will not offer us any law machines that
give automatic answers to specific questions put to them, whether
as to particular cases or as to ultimate legal issues such as the
relative importance of interests that may be in conflict. By the
same token, science will provide us with no formulae or calculus
that will give us certainty either of prediction, analysis or answers
to ultimate questions such as which interest is to be preferred or
which desire has greater social value. 103

Although these comments are eminently reasonable, some of Loev-
inger's turns of phrase are exasperating. In 1949 he claimed that
putting the law on a "rational basis" was the "indispensable con-
dition" of the survival of the human race. 104 In the light of such a
ridiculous statement, and after his trenchant criticism of the value
of jurisprudence (quoted above 105 ), it is not surprising that, as
Gardner says:

The attitudes Loevinger and his colleagues expressed were never
adopted by the legal profession generally . . . As a movement
within the legal profession, jurimetrics has not been much heard
from since the early 1970s. 100

Nevertheless, some of the work of his colleagues, applying various
statistical and logical techniques to legal analysis, is wortlry of
comment here. 107

20 A pragmatic legal expert system §1-3

1.3.1 Kort

Kort uses mathematical expressions to represent American judicial
decisions. 108 He identifies attributes which are of importance and
represents decided cases in the following form:

2=1

where n is the number of attributes, Aij is the value of the ith
attribute for the jth case (attribute values being or 1), Wi is
the weight of the rth attribute, and Vj is the number of votes of
judges favourable to the party seeking redress in the jth case. This
approach is justified on the basis that the decisions of split courts
do not necessarily constitute two opposite extremes, but represent
certain degrees of support for one party. 109

By solving these simultaneous equations, weights are obtained
for the attributes. The sum of the weights of those attributes
present in the instant case is. according to Kort, the number of
likely judicial votes in favour of the party seeking redress.

1.3.2 Lawlor

Lawlor uses logical expressions to represent previously decided
cases. 110 He analyses right-to-counsel cases heard before the US
Supreme Court over 30 years, and finds that in all of these cases
each justice behaved consistently with his own "personal stare
decisis." He identifies legally significant attributes and builds a
logical expression which represents the behaviour of each justice.
A program uses these logical expressions to predict the likely out-
come, given a composition of the Court specified by the user.

Lawlor's system successfully predicted the US Supreme Court's
overruling of Betts v. Brady 111 in Gideon v. Wainwright. 112 But it
predicted a 5:4 majority; the Supreme Court's decision was unan-
imous.

Applying "traditional" stare decisis (i.e. considering the court as
a whole, rather than the decisions of individual justices) Lawlor's
system did not predict the Supreme Court's change. Given the
same cases, Kort's approach did not predict the change either.

This failure draws criticism from those who disagree with such a
"logarithmic approach to justice." 113 Wiener complains that "ad-
vocates of the computer" rest their arguments on an assumption

§1.3 Legal expert systems 21

that courts will adhere to the doctrine of stare decisis, which does
not always hold. 114

In defence of the behaviourists, Kay ton refers to their failure to
predict the decision in Gideon v. Wainwright and writes:

Should we have expected otherwise? Of course not! A reversal is
by definition a logical inconsistency. That which by stare decisis
had been called black is now called white. 115

The developers of these prediction systems accept that they rely on
the doctrine of precedent. Hence, a reversal of previous authority
will always be beyond the predictive capacity of these systems — as
it is beyond the predictive capacity of most lawyers.

1.3.3 Nagel and Schubert

Nagel and Schubert 110 examine the personal attitudes of US Su-
preme Court justices towards various political and economic re-
lations, and claim that these "off-the-bench" attitudes (as Nagel
calls them) affect the justices 1 decision-making.

Correlations were . . . made between responses to specific items and
various decisional propensities. For example, there was a high and
statistically significant correlation between disagreeing with [one
questionnaire item] ("Our treatment of criminals is too harsh; we
should try to cure, not to punish them") and being above the
average of one's court with regard to the proportion of times one
voted for the prosecution in criminal cases. Off-the-bench judicial
attitudes thus do seem to correlate in a meaningful way with on-
the-bench judicial decisions. 117

1.3.4 Haar, Sawyer and Cummings

In the mid-1970s Haar. Sawyer and Cummings made use of re-
gression analysis to build a predictive model for zoning amend-
ment cases in Connecticut. 118 Regression analysis is a statistical
technique for analyzing the relationship of a set of independent
variables to a dependent variable. For Haar et al. the dependent
variable is the outcome of the cases, the independent variables are
attributes.

They identify 167 attributes which "appeared to be important"
to the courts, 40 of which were deemed significant using a x 2
( "chi-square" ) test for association. This is "too many to use in a
regression analysis", 119 so Haar et al. employ two different methods

22 A pragmatic legal expert system §1-3

to reduce this number further: grouping attributes on the basis of
"experience, knowledge, and intuition", 120 and factor analysis.

1.3.5 The role of prediction

The behaviourists' predictive research has been criticized because
they do not attempt to model legal reasoning. 121 But. in not
so doing, the behaviourists just reflect the influence upon them
of the American Realists. They adopt Holmes's analysis that
for any individual "the law is simply a prediction of the way in
which the public force possessed by the government will act upon
him." 122 Loevinger claims that some method of legal prediction is
indispensable"; 123 Lawlor says that the ultimate goal of all sci-
entific methods is reliable prediction of future events and t; [r]eliable
prediction is also one of the ultimate goals of law." 124 Susskind
(certainly no behaviourist) also stresses the importance of the pre-
diction of judicial decisions. 125
Tapper w r rites that:

Although the statistical techniques employed by some of these
workers have been criticized, there can be no real doubt that this
work provides a successful approach to the analysis of the decisions
which have been reached in the past, and at least as satisfactory
a method of predicting future decisions as can be arrived at by
native wit and unaided intuition. 126

However, he criticizes the conclusions that have been drawn using
these techniques:

Either the behaviourist is to content himself with observing the
objective phenomena, in which case he can conclude nothing as to
motivation, or he is to ascribe motivation to the phenomena, in
which case he ceases to be objective. It is precisely at the point
at which decisions of the [US] Supreme Court, for example, are
characterized as pro-civil liberties or anti-labour that doubts arise
as to the real objectivity of the studies.

The same general line of argument may be advanced against
the fact-oriented approach . . . Here the point at which the study
loses its objectivity is in the characterization of the facts present
in the case. 127

(Although this ma}* be a valid criticism of a behaviouristic study, it
is not an argument against the use of predictive techniques where
the characterization of facts has been performed — subjectively,
admittedly — by a legal expert, as is done for SHYSTER.)

§1.3 Legal expert systems 23

Gardner is also critical of the jurimetric programs:

. . . programs, done from a political science viewpoint rather than
a legal one, in which the data concern legally irrelevant matters
such as the ideology and social background of individual judges. 128

By "legally irrelevant matters" Gardner means those matters to
which judges do not — and/or should not — explicitly have regard
when coming to their decisions. But if a system is designed to pre-
dict decisions then any attribute which assists in prediction should
be included — regardless of whether regard ought to be had to that
attribute. These "legally irrelevant matters" are not necessarily
irrelevant to the prediction of judicial decisions.

Stone dismisses concern about "jurimetric" prediction: "the be-
havioralists [sic] may have no conscious designs on the integrity
of the decisional process. The judgment of justice may be of no
concern to them." 129

He defines the "judgment of justice" by reference to a situation
where a judge makes a decision which:

. . . does not merely declare the existing law but decides what
justice requires that the law should be. (. . . [W]e include tacitly
here the reversal of earlier decisions, that is, creative decisions
which unsettle and resettle law: this is a fortiori creative.) It
is this kind of creative judgment which we have here termed "a
judgment of justice." 130

The behaviourists are. Stone says:

. . . onty observers looking at what has already been done in judg-
ment. And when in the course of prediction they turn their
attention to future judgments, it is to ask not what the judge
should do to further justice but only what kind of decision he will
give if he acts consistently with values attributed to him on the
basis of his past decisions. 131

In other words, the behaviourists' work attempted to be predictive
not normative.

When a lawyer gives advice, she/he is expected to make some
prediction as to the likely result. This prediction need not be em-
phatic: it may be no more than a tentative statement about the
strength of a person's legal position. Hence, legal expert systems
should have some degree of predictive capacity. Like the beha-
viourists' research, a legal expert system's predictive power is a

24 A pragmatic legal expert system §l--&

projection based on past cases. There is — and can be — no allow-
ance for changing social mores. This is not a serious limitation
upon such a system's utility: predicting a judge-made change in
the law is beyond all but the very best lawyers.

However, as discussed below, 132 the justification that a legal
expert system provides for its prediction is also important.

1.4 Rule-based systems

There are many examples of rule-based legal expert systems, the
most important of which are discussed in detail below: the work
of McCarty; Bench-Capon. Kowalski and Sergot; Gardner; and
Susskind.

The first proposal for a rule-based legal expert system was made
by Buchanan and Headrick in 1970. They complained that inter-
disciplinary work between lawyers and computer scientists had
"floundered on the misconceptions that each has of the other's
discipline". 133 and suggested that the computer modelling of legal
reasoning would be a fruitful area for research. 134

Their proposal asserts that "[i]n the absence of any reason to
speculate on how they carry on their work, [lawyers] now ap-
ply complex sets of rules without being aware of the rules them-
selves." 13B They make no reference to jurisprudential writing, but
they do make it clear that their approach is based upon two as-
sumptions about human problem-solving in general: "(1) problems
can be broken down into a set of subproblems. and (2) the solution
to any subproblem requires a series of decisions that are governed
by decision rules." 130

The literature is replete with examples of projects in which the
assumption that lawyers work with rules is unstated and/ or un-
supported. Maggs and deBessonet, for example, tacitly make this
assumption. 137 The law, they claim, can be expressed as rules
using propositional calculus. A program which implements these
rules could answer questions from a user, and allow the checking
of statute law for redundancy and contradiction. (It is not clear
what role case law plays in their system, or indeed in their model
of the law . )

Popp and Schlink's JUDITH system 138 uses rules to represent
parts of the German Civil Code. The principles behind JUDITH
are "strikingh* similar" to those of the MYCIN system (an expert
system dealing with bacterial infections) 13& —so similar that, its

§1.4 Legal expert systems 25

developers claim, it would be possible to create a legal know-
ledge base for MYCIN and a medical knowledge base for JUDITH. 140
Michaelsen and Michie's TAXADVISOR system 141 is implemented
using MYCIN. 142 TAXADVISOR is designed to assist lawyers to
advise clients on taxation and estate planning. 143

Meldman's system uses two different kinds of rules: general rules
which define the elements of the claim, and specific rules extrac-
ted from cases. 144 Things and relations are used to represent the
"everyday world of human affairs", 145 and are classified hierarch-
ically into categories. A fact comprises two things and a relation
between them; facts are assembled into situations. These situ-
ations are compared with the situation of the instant case, and
the system determines the extent to which the instant case falls
within or near the law of intentional torts (e.g. assault and bat-
tery).

Waterman and Peterson's LDS system 140 is a rule-based system
for the field of product liability. It does not determine whether li-
ability exists, but is designed to assist legal experts in settling
product liability cases. LDS is a typical example of a system
which developed without any legal theoretical justification. After
admitting that there is no "deep model of the legal process 11 , 14 '
Waterman and his colleagues proceed on the unstated assumption
that any such model must involve rules:

One might expect that the large body of legal rules and regulations
that have been accumulated and formalized in the legal domain
would make expert system development easier. Unfortunately,
this is not the case. Instead, this characteristic of the domain,
having rules that ahead}'' exist, has led to trouble . . . First, the
formal rules that define and regulate legal activity are often am-
biguous, contradictory and incomplete. And second, there exists
a body of informal rules or procedures about how to access, in-
terpret and use the 'formal' rules. Without these informal rules
the formal rules can not be used in any efficient or cost-effective
way. 148

This body of rules, the}' write, "needs to be mapped into code" 149
before a legal expert system can be built.

Bing's SARA system 150 is designed to analyse discretionary de-
cisions. It uses rides to represent legal norms: some strict and
some discretionary. These discretionary norms are weighted us-
ing correlation techniques. SARA allows a lawyer "to back up his
qualitative legal reasoning by quantitative indications. 11151

26 A pragmatic legal expert system §l--&

Stamper's LEGOL language 1 ' 2 also uses rules to represent legal
norms. Pattison and Ciesielski use a rule-based system to review
contracts. 153 SoftLaw's STATUTE, a commercially successful sys-
tem, uses rules to represent statutes, regulations and departmental
guidelines in taxation, social security and veterans' affairs law. 154

1.4.1 McCarty

McCarty has been described as "the father of AI and law 11 . 15 " His
TAXMAN project 150 is concerned with the development of a com-
putational theory of legal reasoning, using corporate tax law as an
experimental problem domain.

He claims that a computer-based legal consultation system must
be able to represent the "facts," at some comfortable level of
abstraction, and the "law," which would consist of a system of
"concepts" and "rules." These concepts and rules are relatively
abstract (i.e. they subsume large classes of lower-level factual de-
scriptions), and they have normative implications (i.e. they specify
which actions are permitted and which are obligatory).

Legal analysis, in its simplest form, would then be a process of ap-
plying the "law" to the "facts". Put this way, the paradigm seems
to be an ideal candidate for an artificial intelligence approach:
the "facts" would be represented in a lower-level semantic net-
work, perhaps; the "law" would be represented in a higher-level
semantic description; and the process of legal analysis would be
represented by a pattern-matching routine. 157

However, McCarty' concedes, the representation of facts in such
a system is more difficult than in other expert systems, because
"the facts of a legal case typically involve all the complexities of
daily life: human actions, beliefs, intentions, motivations, etc." 158
Even if the facts can be represented, he writes, the rules will often
be problematic:

Some rules, usually those embodied in statutes, have a precise
logical structure, and this makes them amenable to the existing
artificial intelligence techniques. But it is a commonplace among
lawyers that the most important legal rules do not have this form
at all: instead they are said to have an "open texture"; their
boundaries are not fixed, but are "constructed" and "modified"
as they are applied to particular factual situations. A sophistic-
ated legal consultation system would not be able to ignore these
complexities, but would have to address them directly. 159

§1.4 Legal expert systems 21

McCarty also makes the startling claim that the "simplest 1 '
problems for first-year law students are the hardest for an AI sys-
tem because the student draws upon ordinary human experience:

Paradoxically, the cases that are most tractable for an artificial
intelligence system are those cases, usually involving commercial
and corporate matters, which a lawyer finds most complex. There
is a simple reason why this is so. A mature legal sj'stem in an
industrialized democracy is composed of many levels of legal ab-
stractions . . . Because of their technical complexity, the legal rules
at the top levels of this conceptual hierarchy are difficult for most
lawyers to comprehend, but this would be no obstacle for an arti-
ficial intelligence system. 160

McCarty chose the area of tax law for his system because "com-
mercial abstractions, in fact, are artificial and formal systems
themselves, drained of much of the content of the ordinary world"
and, by legal standards, well structured. 101 The field of corpor-
ate tax law, he says, is "very near the apex of the hierarchy of
commercial abstractions" 162 — "whatever that means", comments
Moles. 163

In TAXMAN I, McCarty's first prototype system, the basic "facts"
of a corporate case are captured in a relatively straightforward rep-
resentation (e.g. a corporation issues securities). Below this level
is an expanded representation of the meaning of various entities
(e.g. a security interest) in terms of their component rights and
obligations. Above this level — presumably above both levels, al-
though this is not made clear — is the "law" (statutory rules which
classify transactions as taxable or non-taxable etc.). 104

McCarty found that, although the rules are complex, the under-
lying representations are manageable. He concludes from his early
work that "the construction of an expert consultation system in
this area of the law is a feasible proposition." 165

McCarty sees the development of legal expert systems as an
opportunity to contribute to jurisprudence. Although the juris-
prudential literature includes "many illuminating examples and
many valuable insights about the structure and dynamics of legal
concepts", he complains that "taken as a whole" it is "notoriously
imprecise". 166

The TAXMAN S3 f stem adds a strong dose of precision and rigor to
these discussions of linguistic and conceptual problems. Its critical

28 A pragmatic legal expert system §l--&

task is to clarify the concepts of corporate reorganization law in
such a woy that they can be represented in computer programs. 167

Moles is very critical of rule-based expert systems designers, and
McCarty in particular. McCarty takes a positivist approach to the
law, and Moles complains that:

The computer scientists have taken the de-humanizing aspects of
modern positivism to their extreme. . . .

The computer scientists, encouraged b} r the modern positivists,
fail to recognize the point which Austin 108 correctly emphasized
throughout his work— that law, positive morality and ethics are
inseparably connected parts of a vast organic whole. 109

Most damning is Moles* s statement that:

The sad thing is that [McCarty] has not shown the slightest aware-
ness of the nature of the legal enterprise. Far from having emphas-
ized any difficulties, he shows that he simply does not understand
what thej 1- are. 170

McCarty does identify two major limitations to the approach
taken in TAXMAN I. Firstly, he concedes that the factual descrip-
tions, although manageable in the corporate domain, would be
too complex in (for example) the average contract or tort prob-
lem. Secondly*, the higher-level conceptual representations are not
adequate for all domains because judicially created concepts are
incurably open-textured, have a dynamic structure with the ca-
pacity to evolve and adapt to new situations, and the evolution
of these concepts is governed by a sense of purpose. 1 ' 1 Having
recognized these problems, McCarty set out to solve them with
TAXMAN II.

TAXMAN II 172 uses "prototypes and deformations" to represent a
legal concept by specifying the prerequisite conditions for that con-
cept, a set of cases (real and hypothetical) in which that concept
does or does not apply, and a set of transformations for getting
from one case to another. If a given case (representing a legal
concept) can be transformed into the instant case, while still sat-
isfying the prerequisite conditions for the concept, then it can be
used in argument as an example of that concept.

Ashley sees McCarty's work as a significant advance, but points
to several shortcomings of TAXMAN II:

The work is largely an exercise in knowledge representation. Mc-
Carty does not set forth a control or process model that clarifies

§1.4 Legal expert systems 29

how a program would actually generate a legal argument . . . The
reported research involves a hand simulation of the arguments in
one US Supreme Court case 173 and work done on hand simulations
of several subsequent cases.

. . . [TAXMAN II] has no mechanisms for comparing cases in
terms of how on point the}' are, for distinguishing cases, or select-
ing the best precedents . . .

McCarty's model assumes a much neater domain than exists
in law. He assumes that in reality, legal cases are consistently
allocated as positive and negative exemplars of concepts. They
are not. He assumes that there is a near match between concepts
and the features of a case that are relevant to the concept. There
is not. 174

McCarty's definition of legal primitives has also been criticized.
Moles writes:

McCarty appears not to appreciate that 'corporations', 'securities',
'property', 'dividends' and so on are not subsumed 'beneath the
law', but are each the products of complex legal analysis. The
question of whether certain transactions are taxable or not is in-
timately tied into that legal analysis. 170

Ashley makes a similar point:

The problem with such primitives, if they are taken seriously as
a means for defining concepts, is that they assume what is to be
shown. Far from being a primitive, that someone has a right or a
duty in a given fact situation is an arguable legal conclusion that
must be justified by citing authorities. 176

1.4.2 Bench-Capon, Kowalski and Sergot

Bench-Capon, Kowalski. Sergot and their colleagues use programs
written in the PROLOG language to model statutes. 177 Their ap-
proach to legislation is the most extreme of all expert system
developers, due to their attitude towards knowledge acquisition.
They write:

The formalisation of legislation by means of rules has almost all
the characteristics of an expert system. It differs, however, in one
important respect. In a classical expert system, before knowledge
can be formalized, it has to be elicited from the subconscious of
an expert. Eliciting this knowledge is generally regarded as the
main bottle-neck in the construction of expert sj'stems. It is en-
tirety absent, however, in the case of legislation which is already

30 A pragmatic legal expert system §l--&

formulated and written down. Thus the use of expert system tech-
niques for representing legislation has virtually all the advantages
of expert systems without the attendant disadvantages of eliciting
the knowledge. 178

This statement is nothing less than astounding. Even if it is ac-
cepted that statute law can be represented using PROLOG clauses.
it is a bold claim indeed to assert that constructing those clauses
requires no expertise. As Moles says. "[c]learly these researchers
do not distinguish between the writing (which is the legislation)
and the meaning of that writing." 1 ' 9

Bench-Capon et al. have worked with several statutes. Most
famous is their work with the British Nationality Act 1981 (UK). 180
Consistent with their approach to knowledge acquisition, their
PROLOG representation of the British Nationality Act was im-
plemented in two months by a student, "without any expert legal
assistance." 181

They also represented the Supplementary Benefits Act 1976
(UK) and its regulations using a similar method, and made a sim-
ilar claim about the importance of legal expertise:

For our project, the accuracy of the representation was not a crit-
ical consideration at this [early] stage. Our formalisation could
therefore be undertaken with no expert legal assistance ... In
general, accuracy of the formalisation is, of course, critical, par-
ticularly if one were constructing a representation to be used in
practice. 182

Predictably — and correctly — this approach has been strongly
criticized. Moles points out that:

This is to assume . . . that the problem with regard to "accurac}'"
is merely a matter of changing the detail of content. It fails to
appreciate that an expert may have a great many useful things to
saj' about how one goes about the process of interpretation. The
expert advice will therefore have implications for the method being
employed and the way in which the knowledge is structured. 183

Kowalski and Sergot make this assumption explicit:

Access to an expert adviser might well have changed the exact
form of the rules in our program, but it would not have changed
the method we used to formulate and compute with the rules. 184

But how can they possibly be sure?

§1.4 Legal expert systems 31

On the need for legal expertise in the development of legal expert
systems. Susskind quotes Hayes-Roth. Waterman and Lenat:

It is verj' easy to be deluded into thinking one knows a great deal
about the domain. Remember: the expert became one only after
years of training and experience. 185

The work of Bench-Capon et al. has been most vehemently cri-
ticized by Leith. He quotes them:

The formalisation of the British Nationality Act is an axiomatic
theory similar, for example, to an axiomatisation of Euclidean
Geometry. In principle, a,ny logical consequence of the axiomat-
isation can be generated and tested mechanically . . . 180

then comments:

Does this sound like a wild claim? I suspect so. No lawyer, with
whom I have so far discussed Kowalski's "legal" work, has been
impressed in the least: the common repty is, "Surely he doesn't
believe that?" And this is the rub. In Al research, hinders are
prepared to provide research funding for the most inane of ideas
without appealing to those with some expertise in the area of
"intelligence" being researched. 187

1.4.3 Gardner

The aim of Gardner's research 188 is not to develop a program that
"solves" legal problems:

Instead, the objective is to enable the program to recognize the
issues a problem raises and to distinguish between those it has
enough information to resolve and those on which competent hu-
man judgments might differ. Toward this end a heuristic distinc-
tion between hard and easy questions is proposed. 189

Her chosen domain is an aspect of the law of contract that is
dominated by case law: offer and acceptance. She claims that,
although an area based on statute might seem easier for an Al
program to handle, the reverse is true. Case law must be taken
into account in statutory areas too, and statutory interpretation
raises its own problems. Hence, she concludes, "[b]eginning from
statutes therefore seems likelier to add a layer of complication
than to remove one." 190 Susskind agrees: "Gardner ... is right in
recognising that it is unrealistic to focus on statute at the expense
of precedent. There is a lesson here for all workers in the field." 191

32 A pragmatic legal expert system §l--&

Gardner's system has four different levels, the last of which is
not implemented. The network level is an augmented transition
network which represents legal states and events. The rule level is
a set of rules which operate on objects at the first level. These rules
are "definitions of the major concepts" 192 in the law of offer and ac-
ceptance. The third level is a set of examples which explain those
predicates in the rule level which are undefined, but whose resolu-
tion is "clear" ; this includes (non-legal) common sense knowledge.
Gardner, citing the legal positivists. assumes that open texture
does not render all cases hard:

The first three levels of the program are devoted to identifying the
hard questions. From an opposite viewpoint, the first three levels
try to identify the eas}' questions and resolve them. 193

The fourth (unimplemented) level would deal with the hard ques-
tions.

Susskind comments that:

Ultimately, the human expertise that Gardner tries to encapsu-
late in heuristic form is the ability to assess, in advance (in a
sense), whether a case does indeed raise easy or hard questions.
This will indicate if a conclusion may be inferred without further
ado or whether human judgement is required. She makes a brave
attempt, but ... I was left wondering about the generalisability
of her ideas and their applicability in other, far more extensive,
branches of law. 194

Ashley, too, is critical of Gardner's approach:

... an attorney distinguishes hard from easy questions in terms
of comparing the strengths of the best argument he or she can
make with the best arguments an opponent can make. Gardner's
program provides no measure for evaluating the strengths of com-
peting arguments . . .

Even if Gardner's program evaluated case-based arguments, it
could not do so ver} r realistically given the way cases are repres-
ented. Distinguishing, for example, is not possible because there
is nothing to distinguish. 195

Unlike McCarty and Bench-Capon et al., Gardner does confront
the arguments of the (American) Realists:

If legal realism is right, it appears to make the AI paradigm of
rule-based expert systems inappropriate, at least with any simple
mapping from legal rules to knowledge-base rules. There are

§1.4 Legal expert systems 33

several directions one might go instead. One direction would em-
phasize the idea that it is individual decisions, not general rules,
that have authoritative status as law. With this emphasis, one
might look for a method of reasoning from the decisions in past
cases to a conclusion in a present case . . . 19G But this cannot be
the whole story. If it were, why would present-da}' law professors,
thoroughly aware of the insights of realism, continue to expect
their students to know rules? 197

This is a strange argument; it says more about the way in which
the law is taught than it does about whether a rule-based ap-
proach to the law is appropriate. (And it must be remembered
that the American Realists were not true rule sceptics, as dis-
cussed above. 198 )

Gardner identifies two other directions to take "if legal real-
ism is right" : emphasizing the behaviouristic side of legal realism
(e.g. the jurimetrics research discussed above 199 ), and a third direc-
tion which Gardner herself adopts — "to retain an important place
for legal rules but to reinterpret their significance." 200 She claims
that:

Rulelike sentences can be understood as useful cognitive constructs,
needed to find order in (or impose order upon) an unwieldy mass of
individual decisions. Once articulated, they can provide guidance
as to how future decisions can be kept in some rough conformance
with this order; or, if the articulated rule seems to be a bad rule,
it can suggest a way of saying how the course of decisions ought
to be changed. 201

This does not really represent a reinterpretation of the significance
of rules. Only the most ardent rule-follower would suggest that
rules were anything else.

Gardner develops a rule-based model for offer and acceptance
cases, but makes the significant concession that:

... a rule-based model of case law must be understood, like any
academic legal writing, as a secondary source. The official sources
are the decisions. There has never been agreement on what it
would mean for rulelike generalizations from decisions to be both
accurate and appropriate. Thus, the basis for even uncontroversial
rules remains undefined in legal theory. 202

This is an interesting, and quite pragmatic concession, coming as
it does from someone whose work has been characterized as "fairly
clearly purist in nature". 203

34 A pragmatic legal expert system §l--&

1.4.4 Susskind

Susskind's work on the importance of jurisprudence to the devel-
opment of legal expert systems is discussed in detail above. 204 He
chose the Scottish law of divorce as his first experimental legal
domain. 200 He has also been involved in the development of the
Latent Damage System: a legal expert system concerned with the
time periods within which claimants may start negligence proceed-
ings where they have suffered latent damage or loss. 200 Capper
and Susskind claimed in 1988 that this was the first legal expert
system built in the UK by lawyers for lawyers. 20 '

Susskind believes that statutes and some cases — the clear cases
—can and should be represented using rules. With clear cases,
he claims, it is possible to "draw legal conclusions on the basis of
literal interpretations of the formal legal sources." 208 However:

... it should not be taken for granted that the entire common law
system can be reduced to a collection of rules . . . Simpson has
forcefully contended that such a reductionist model misrepresents
the common law and is inconsistent with its development, content,
and scope . . , 209

Other methods, he says, would be required in order to represent
cases which are not clear: methods which reason with uncertainty
and draw "probabilistically phrased conclusions." 210 Yet, he con-
tends:

. . . although the common law may not be sufficiently represented
in terms of rules, it cannot be doubted that it is invariably possible,
desirable, and necessary to interpret individual cases in the form
of individuated rules. 211

I doubt that it is possible, dispute that it is desirable, and
demonstrate — with SHYSTER — that it is not necessary to repres-
ent cases using rules.

1.4.5 Knowledge acquisition and representation

Feigenbaum wrote in 1981 that ;i [t]here are many important prob-
lems of knowledge representation, utilisation, and acquisition that
must be solved, but the acquisition problem is the most critical
'bottleneck' problem." 212 As discussed above, 213 Bench-Capon
et al. claim that there is no such bottleneck for knowledge acquis-
ition from statute law because legislation is "already formulated

§1.4 Legal expert systems 35

and written down." 214 However, this demonstrates a serious mis-
understanding of the law. Knowledge acquisition is as much a
problem in the legal domain (statute and case law) as it is in any
other. I propose a solution to this problem, at least in regard to
case law. in chapter 2.

In order to avoid the knowledge acquisition bottleneck, some re-
searchers have examined the possibility of the machine-processing
of statutes. If the only input is the words of the statute itself then
this problem is one of understanding natural language — a research
field in which results have not fulfilled original hopes.

A more feasible approach is to convert the words of the statute
into a machine-readable form. Allen has developed one such form
and claims that if statutes were drafted using this form, not only
would the automatic logical analysis of their contents be possible.
but humans would be able to read and work with them more
easily. 215 Unless and until legislation is expressed in machine-
readable form — an extremely unlikely event, the desirability of
which is far from clear — builders of legal expert systems which deal
with statute law must use human expertise to extract knowledge
from statutes.

The process of acquiring knowledge from statutes is usually seen
as a process of writing rules. However. Shannon and Golshani
warn that doing this in an ad hoc fashion is unsatisfactory because
the rules cannot be checked for correctness, and such an approach
may lead to dissimilar rule formulations which do not work well
together. 216 Instead they recommend following precise and con-
sistent methods (like those of Allen) to formulate rules. This, they
say, would reduce the scope for dissimilar rule formulations. 217

Shannon and Golshani's belief that there is a "correct" inter-
pretation of a statute is shared by many researchers in this field.
In believing this, they go even further than the legal positivists on
whose theories much of their research is based. 218

1.4.6 Fact representation

The representation of legal facts in a rule-based system — indeed
any expert system — is difficult. As with the representation of
statutes, researchers have developed normalized forms for fact rep-
resentation. An example is deBessonet and Cross's atomically
normalized form (ANF). 219 Shannon and Golshani use a modi-
fied example, taken from some of deBessonet and Cross's work

36 A pragmatic legal expert system §l--&

modelling the Louisiana law of causality, to demonstrate how a
statement of fact can be represented using ANF.
The statement:

A lessor believes that the lessee caused a defect in the leased
premises which requires that the lessee fix the defect

is represented in ANF as:

(Lessor Believes

((Lessee Caused (Property Has Defect))
Causes

(Lessee Must-Fix Defect))).

This decomposes into clauses as follows:

"Lessor Believes A" where
A is: "Bl Causes B2, n and
Bl is: "Lessee Caused C," and

B2 is: "Lessee Must-Fix Defect," and
C is: "Property Has Defect." 220

In this ANF representation, and the decomposed clauses, the
lessor believes that the lessee caused a defect in the leased premises
and that (as a result) the lessee is required to fix the defect. There
is at least one other interpretation of the original statement:

"Lessor Believes Bl n and "A"

where A and Bl have the same meanings as above. This inter-
pretation may not be "better" than Shannon and Golshanis, but
it is certainly plausible. In demonstrating a method of fact rep-
resentation they (unwittingly) provide an excellent example of the
enormous difficulties that representing facts can present. 221

It makes no more sense to say that any given representation of
the facts is "correct 11 than it does to make the same claim of a
representation of the law.

1.4.7 The inadequacy of rules for case law

Although several developers have used rule-based systems to model
statute law and case law, rules are fundamentally inadequate for
representing cases.

§1.4 Legal expert systems 31

Tyree, Greenleaf and Mowbray claim that it is inappropriate
to model case law using a rule-based system — not because it is
theoretically impossible, but because "it is not the natural way in
which lawyers reason with cases." 222 As discussed above, 223 it may
be theoretically impossible to write such rules: the sceptical view
of rules applies to representing statutes and cases. But there is
a fundamental difference between these two sources of law which
means that, even if rules are appropriate for representing statutes,
they are not appropriate for representing cases.

Lawyers apply statutes in a rule-like fashion. This is under-
standable given the rule-like form in which they are written. How-
ever, lawyers reason with cases by arguing about their similarities
and differences. As Tyree et al. explain, it used to be thought that
each decided case stood for a rule of law. "It is now clear that any
interpretation of the legal significance of a case must be in the
larger context of the legal material in which it is embedded". 224 In
support of this assertion, they cite Stone who argues that:

. . . however much we trj' to conceal the truth by using singular
terms like "case", "precedent", "decision" or "holding", the truth is
that the ratio decidendi of a case has always to be sought in a body
of judicial discourse, that is, of communications by judges which
enter the legal materials as a more or less complex collocation of
words in a written report.

Perhaps, Tyree et al. suggest, this is why reasoning with case
law using the usual production rule formulation has had little
success. 220 (The way in which lawyers argue with statutes and
cases is examined further in the next chapter. 227 )

The fact that case law is embodied in cases makes knowledge
acquisition for a rule-based system extremely problematic. Ex-
tracting rules from case law is difficult for a legal expert, because
that is not the way in which legal experts view case law. Kow-
alski identifies this "profiling 11 of cases as the bottleneck in the
construction of case-based legal expert systems. 228

It is not possible to automate this knowledge acquisition process
using inductive methods because the number of decided cases in
any given area of law is usually so small that inductive inference al-
gorithms cannot be used. 229 Even if an area of law has a sufficiently
large number of cases, those cases are, in a sense, unrepresentat-
ive of that area: so-called "problems of the penumbra" 230 are more

38 A pragmatic legal expert system §1-%

likely to require determination by the courts than are straightfor-
ward cases. Once such a problem has been resolved, other cases
in which the same problem arises are less likely to be taken to
court — if they are taken to court they are less likely to be repor-
ted. (Reasons for the paucity of reported cases are discussed in
detail in chapter 2. 231 ) This further complicates the induction of
sensible rules. And. of course, it is not at all clear how any such
rules could be used in legal reasoning with cases even if they could
be induced.

Developers of case-based systems like FINDER and HYPO (and.
to some extent, the hybrid CABARET system) 232 have recognized
the inadequacy of rules for representing case law. However this
view has not been universally accepted. Notable amongst its
critics is Berman. His argument against case-based reasoning is
examined, and refuted, in the next chapter 233 —after SHYSTER's
pragmatic approach to case law has been fully explained.

1.5 Case-based systems

Bench-Capon and his colleagues chose the British Nationality Act
as an object of research because, as a fairly recently enacted stat-
ute, they claim it was "free of the complicating influence of case
law." 234

In Australian courts, even new statutes are interpreted in the
light of previously decided cases. This applies to decisions inter-
preting an expression in a similar statute, in the same or in a dif-
ferent jurisdiction, and (to a lesser extent) to the re-enactment of
a statutory provision after a judicial decision as to its meaning. 235

If Bench-Capon et al. had sought expert advice they would have
learnt that, similarly, in the United Kingdom the prior legal his-
tory of the language and concepts used in a statute are relevant
to its interpretation. 236

A statute — even a newly enacted statute — must always be in-
terpreted in the light of case law.

In 1988. according to Pearce and Geddes. approximately 50% of
recent reported Australian cases required the court to rule upon
the meaning of some legislative instrument. In a further 25% of
cases courts were required to apply legislation, "its meaning this
time not being in dispute." 237 Clearly, any useful legal expert
system must be able to take account of the legal effect of previously
decided cases.

§1.5 Legal expert systems 39

Ashley claims that the following operations are general to all case-
based reasoning:

. . . (1) ordering relevant cases and potentially relevant cases in
terms of how analogous they are to the problem situation, (2) se-
lecting the most analogous cases, (3) identifying configurations of
counterexamples, (4) hypothetically modifying the problem situ-
ation to explore contingencies, and (5) comparing case-based ana-
lyses of different problem situations to explain differences. 238

In order to have some basis upon which to determine whether
two cases are similar, information about certain attributes of those
cases must be gathered. As Lambert and Grunewald explain:

The first task of [case-based] reasoning is to pick, from the infinite
number of respects in which cases can be similar and dissimilar,
a manageable set of [attributes] that could support a conclusion
that one case is so similar to another that it will likety have the
same outcome. Without such constraint, one would be faced with
the commonsense impracticability, if not the jurisprudential im-
possibility, of defining the entire set of [attributes] that any case
in the domain can have, together with the full ranges of possible
values that those [attributes] could take.

Taking this arbitrarily restricted, but practically necessary, case
structure, one could in principle generate the complete set of cases
belonging to the domain and produce therebj' a case base con-
taining one case that would be exactly the same as any possible
test case in the domain. But the size of such casebases would
be intractable for case structures possessing more than an ex-
tremely small number of [attributes]. Therefore the next task of
the reasoner is to install in the case base a set of cases considered
typical of those one is likely to encounter in the domain. These
can be either real or hypothetical cases that an expert concludes
collectively capture the essence of the domain. 239

As with statute law, there is no "correct" answer to a question
of case law. So, as Ashley writes, a case-based legal expert system:

. . . does not "decide" a case; it makes arguments on behalf of the
respective parties but does not necessarily determine a winner.
The program would be useful as an attornej^s assistant, spotting
issues, strengths, weaknesses, and precedents that an attorney
representing a client in the [instant case] would want to take into
account, or as part of a legal tutoring system. 240

40 A pragmatic legal expert system §1-%

Nevertheless, as discussed above. 241 an ability to predict the
likely outcome of a case is a component of legal advice, both human
and computerized.

Three examples of case-based systems are discussed below: the
nearest neighbour analysis of Mackaay and Robillard (§1.5.1), and
the FINDER (§1.5.2) and HYPO (§1.5.3) systems. These examples
are compared with SHYSTER in the next chapter. 242

1.5.1 Nearest neighbour analysis

Mackaay and Robillard were the first to examine the use of nearest
neighbour analysis in predicting judicial decisions. 243 Nearest
neighbour analysis is a statistical technique first developed in the
early 1950s. 244 As Cover and Hart describe it:

The nearest neighbor decision rule assigns to an unclassified sample
point the classification of the nearest of a set of previously classi-
fied points . . .

If it is assumed that the classified samples . . . are independ-
entty identical^ distributed ... it is reasonable to assume that
observations which are close together (in some appropriate met-
ric) will have the same classification, or at least will have almost
the same posterior probability distributions on their respective
classifications. 245

Mackaay and Robillard chose as their domain Canadian cap-
ital gain cases decided in the ten years before 1908. These cases
were the subject of earlier research by Lawlor, using different
techniques. 240 They screened the set of "initial or standard" cases
and removed those in which the decision did not coincide with
the majority decision amongst its nearest neighbours. This, they
claim, "has the advantage of purifying the standard or reference
cases by eliminating those that appear to be erroneous in relation
to the other ones." 247

Mackaay and Robillard used, as their similarity metric, the num-
ber of different attributes. They weighted each attribute equally,
on the basis that such a method is more reliable than differen-
tial weighting via multiple regression. 248 Their results compare
favourably with those of Lawlor, using the same cases. 249

1.5.2 FINDER

Tyree, Greenleaf and Mowbray use nearest neighbour analysis in
their FINDER system. 200 FINDER is a case-based system which

§1.5 Legal expert systems 41

gives advice on the law of trover — the law concerning the rights of
the finders of lost chattels. This area of law is unusual in that it
is based entirely on cases.

FINDER has a database of leading trover cases, and a set of
attributes which were of legal significance in those cases: e.g. "Was
the chattel attached to the land or premises where it was found?"
For each of the leading cases. FINDER has a vector of attribute
values; each attribute value (yes or no) answers the corresponding
attribute's question for that case. Hence, each vector of attribute
values represents the relevant facts of that case. The user provides
FINDER with the relevant facts of the instant case by giving a yes
or no answer to each of the attribute questions.

FINDER assigns a weight to each attribute, equal to the inverse
of the variance of the values of that attribute across all the cases. 251
FINDER uses these weights to find the weighted Euclidean distance
between the instant case and each of the leading cases. It uses
the nearest case, and the nearest case with the opposite result,
to build an argument about the likely result in the instant case:
i.e. whether or not the finder should be allowed to keep the found
chattel. Several statistical techniques are employed to reduce the
possibility of giving bad advice.

FINDER is designed to provide a comprehensive report for the
user as to its opinion. As Tyree explains:

Usually explanation and justification are provided in the [non-legal
expert] system as a means of establishing the user's confidence in
the advice which is given by the system . . .

By contrast, the legal expert system provides the justification,
that is, the reasoning process, as one of its major products. The
user, particularly if a professional, may care little for the predic-
tion of the system, but the reasons provided for the prediction
could be useful even if the predictions were always wrong. If the
expert system can provide good arguments, then these are useful
as a product in themselves. 252

Tyree understates the role of prediction in a legal expert system.
Certainly, a legal expert system's argument is important. But so,
too, is its predictive capacity (as discussed above 253 ).

SHYSTER adopts and expands upon FINDER's approach to case
law. SHYSTER's approach is detailed in chapter 2. Further
details of FINDER are explained there, by comparison with
SHYSTER. 254 FINDER has been simulated using SHYSTER, as
described in chapter 4. 25u

42 A pragmatic legal expert system §l-$

1.5.3 HYPO

Ashley and RisslancTs HYPO system is a case-based legal expert
system which makes use of hypotheticals in building its argu-
ments. 250 Their aim was to build a working model of "making
reasonable arguments in law," which Ashley describes as a "messy
domain . . . that lacks a strong theoretical model that would sup-
port deductive reasoning techniques." 257

HYPO has four knowledge sources. Its case representation
language — the first knowledge source — has two tiers: "legal case
frames" are used to store basic information about cases (including
the instant case) and hypotheticals, and the factual objects and
relations that are important in those cases; "factual predicates"
are used to summarize the facts of a case represented by the legal
case frames. "They are generalized factual statements that con-
firm whether certain legally significant relationships are true in
the case". 258

HYPO'S case base — its second knowledge source — contains thirty
legal cases (including hypotheticals) concerning trade secrets law.

Its third knowledge source are its dimensions. These are con-
structs for representing factors: "stereotypical facts of legal cases
important for the strength of a plaintiff's position on a particular
kind of claim." 259 A claim can be thought of as a result that one
of the parties desires. 200

Each dimension has a list of prerequisite factual predicates:
information about those facts which make a case more or less ex-
treme along the dimension, and how a change in those facts makes
the case better or worse for the plaintiff along the dimension. A
dimension is applicable if all of its prerequisites are satisfied; a
dimension is a near miss if all of its prerequisites are satisfied
except those associated with facts which locate the instant case
somewhere along the dimension. These dimensions:

. . . are not definitional elements of a claim; they do not purport
to specify necessary and sufficient conditions for determining the
existence of a claim. Instead, the}' represent collections of facts
that tend to strengthen or weaken an assertion that the claim
applies to a fact situation. 261

HYPO's fourth knowledge source is a set of criteria for evaluating
the strength of an argument: e.g. if one case is more "on point"
than another case, the former case is better. 202

§1.5 Legal expert systems 43

Given the facts of an instant case, HYPO selects the relevant
cases (the cases which share at least one dimension with the instant
case) and generates an argument based on those cases. HYPO's
arguments are "three-ply" : it makes a point for one side ("drawing
the analogy between the problem and the precedent" 203 ), responds
with a point for the other side (attempting to distinguish the cited
case, and citing other cases as counterexamples), then makes a
final rebuttal (attempting to distinguish the counterexamples).

The best cases to cite in the first instance are those in which the
result favours the chosen side, and which share (with the instant
case) at least one applicable dimension favouring that side. HYPO
also uses "most-on-point near-miss cases" in its arguments. These
are cases that would be analogous to the fact situation if all of the
dimensions that were near misses applied.

Of course, as Ashley points out, there may be more than one
analogous case. Because the evaluation criteria are not well defined,
the choice of most analogous case may depend on which evaluation
technique is used. "Thus it may be useful for decision-making or
explanation to define "most analogous' less restrictively to yield a
larger set of alternatives." 204 The most analogous cases may lead
to conflicting results. However:

. . . comparing and contrasting the conflicting most analogous cases
in a symbolic way can help educate the decision maker. She or
he sees the alternative ways of answering a question and is bet-
ter prepared to make a wise decision. She or he also sees how
small changes in the problem could lead to different results. The
law's adversarial system institutionalized this approach to decision
making. 205

HYPO also generates hypothetical variants of the instant case-
variants that would strengthen/weaken the case for each side. 200

HYPO is a sophisticated case-based system. But, its complicated
structure for knowledge representation means that knowledge ac-
quisition is difficult. Although HYPO is based on a general theory
of law, it was developed and tested using only one domain: Amer-
ican trade secrets law. The applicability of HYPO's approach in
other domains has not been demonstrated.

1.5.4 The inadequacy of semantic networks

Semantic networks represent another potential approach to repres-
enting case law. Hafner's LRS system, for example, uses a semantic

44 A pragmatic legal expert system §1-6

network to represent statutes and cases of relevance to negotiable
instruments law. 207 Branting cites two reasons for using a semantic
network, rather than a "feature- vector" representation of cases, in
his GREBE system: 268

First, any particular set of case features can represent only a small
portion of the nearly limitless variety of event sequences that can
give rise to legal claims. A second and more fundamental reason is
that determining the legally relevant aspects of a new case is fre-
quently the most difficult step in legal reasoning. Systems limited
to case descriptions consisting of sets of legally relevant features
ignore this step or force it onto the user. The capacity to rep-
resent and create multiple, competing arguments about the legal
consequences of a set of facts depends on a representation that is
free of bias towards any particular analysis. Any such unbiased
representation must be of a finer granularity than legally relevant
features. 269

I contend that no representation can be unbiased. Furthermore,
representations with exceptionally fine granularity face the same
criticism as do deep conceptual models of legal reasoning (dis-
cussed below 2 ' ): viz. they operate at an inappropriate level of
abstraction for use in an expert system.

Using semantic networks raises other problems. As Branting
concedes:

Unfortunately, representing complex cases in a semantic network
formalism is extremely laborious and difficult. Knowledge repres-
entation is a sufficiently immature field that each new case may
raise representational issues that are more difficult than the legal
issues posed by the case. 271

Semantic networks are of dubious use for representing case law.
At the very least, their use is not consistent with a pragmatic
approach to case law which. I argue, is appropriate for expert
system design.

1.6 Hybrid systems

The law in Australia and other common-law countries is based on
both statutes and cases. For a legal expert system to be of use in
most legal domains, it must be able to take account of statute law
and case law.

§1-6 Legal expert systems 45

The first legal expert systems used rule-based methods to rep-
resent both statutes and cases. It was not until the late 1980s that
a few researchers examined the possibility of combining rule-based
and case-based methods to produce a hybrid legal expert system.
Risslancl and Skalak wrote in 1989 that ;i [t]his sort of hybrid ar-
chitecture . . . has not been much researched to date", 272 and only
a few projects have arisen since.

The most important of these hybrid systems is CABARET. 273
CABARET deals with a small area of US taxation law: home of-
fice deductions. It treats its rule-based and case-based systems as
co-reasoners, each capable of operating on its own. Some thirty
heuristics control how the two systems work together. For ex-
ample, some of its heuristics concern how to "broaden" a near
miss rule (i.e. one in which all but one conjunct can be estab-
lished) :

• Use CBR [case-based reasoning] to find cases where the rule did
not fire, but the consequent of the rule still held. (That is, show
that the missing conjunct is not necessary to fire the rule.)

• Use CBR to find cases where the rule did fire, and point out the
similarities between those cases and the present case. (Show
that effectively j'ou have the missing conjunct.)

• Use CBR to find similar cases where the rule did not fire, but the
ultimate disposition of the case was consistent with the user's
point of view. (Show that the rule firing is not necessarj' for the
ultimate result the user wants.) 274

As can be seen from these heuristics, CABARET treats cases as
examples of rules filing or not firing. CABARET'S hybrid structure
is a mixed one: it mixes rules and cases.

Branting's GREBE system 270 deals with Texan worker's compens-
ation law. GREBE is a hybrid system. Its rule base contains
statutory rules and common-law rules. It uses a semantic net-
work to represent case facts, 270 and utilizes precedent constituents:
"Each precedent constituent acts as a warrant connecting some
subset of the facts of a precedent to one of eight distinct legal
predicates." 277 These precedent constituents are. effectively, rules
which allow GREBE to use portions of precedents. Branting claims
that this improves the system's case matching capacity; it can
match portions of cases where the entire cases would not match.

46 A pragmatic legal expert system §1-7

Of course, all case-based systems should be capable of match-
ing cases which are not completely identical. GREBE's precedent
constituents simply constitute a different approach to the notion
of similarity between cases than that adopted by (for example)
FINDER and SHYSTER.

PROLEXS (in its earlier versions 278 ) was a hybrid system. If the
sum of the weights of a list of weighted factors in the instant
case exceeded a certain threshold, the instant case was said to
match the "case-prototype" and "other types of (mostly rule-
based) reasoning took over". 279 The weights and the threshold
were determined by the knowledge engineer.

However, the developers have since adopted a different approach.
PROLEXS now uses a neural network to model the Dutch law of
"suitable employment." 280 Its developers claim that neural net-
works have advantages over rule-based systems because a neural
network generalizes from the examples it is provided with, though
they concede that a neural network cannot explain its decision by
referring to explicit rules since its knowledge is not symbolic. 281

PROLEXS's developers comment that "the generalizing capacity
of the network is already evident: in a sense the network has
discovered a new rule." 282 Clearly PROLEXS's approach to the
law is, now, rule-based.

All of these, and other, 283 hybrid systems use rules and cases to
represent case law. A cleaner division between the rule base and
the case base — one which does not mix rules and cases— is pro-
posed in the next chapter. 284

1.7 Conceptual models: deep and shallow

As Susskind says, "purely rule-based systems can cope only with
problems for which they have explicitly represented and applicable
rules". 285 In response to this problem — "to fill the gaps in rule-
based legal knowledge bases" 280 —several researchers have sought
to develop conceptual models of the legal domain. 28. Foremost
amongst these is McCarty 288 who identifies "the construction of
a conceptual model of the relevant legal domain" as "the most
critical task in the development of an intelligent legal information
system". 289

McCarty advocates the development of what he calls "deep con-
ceptual models." as opposed to "shallow" ones. As Greenleaf,

§1.7 Legal expert systems 41

Mowbray and Tyree comment. "[t]he meaning to be given to 'deep'
and 'shallow' is not always clear." 290 It seems that by "deep," Mc-
Carty means a conceptual model that is detailed enough to express
the important facts about a particular legal world, yet abstract
enough to suppress the irrelevant detail. 291 According to Shan-
non and Golshani, truly deep conceptual models closely approach
human reasoning, because they model the meaning behind words.
not just the words themselves. 292

Greenleaf et al. claim that there are at least three levels at which
a legal expert system might be expected to model legal knowledge:

(i) the system could include only the heuristics of legal ex-
perts as to the outcomes which are likely in particular situations,
but without provision of any justification based on primary legal

sources;

(ii) the representation could include justification based on the
primary legal sources, but without any explicit model of those
sources; certain heuristics concerning the relationship between
these sources, e.g., principles of interpretation, may be implied
in the representation or the inference system;

(iii) the system could include an explicit causal model which
serves to define the relationships among the concepts employed
in the primary sources. Justification would then, presumably, be
based on the model. 293

A "deep" model, then, would be one of type (hi).

Susskind claims that the results obtained by researchers who
are trying to develop conceptual models "are not universalizable
in so far as they do not seem to be offering any coherent guidance
regarding the development of conceptual models in other legal
domains." 294

But the problem is more fundamental than this. Deep con-
ceptual models of legal reasoning, like jurisprudence, operate at
too high a level of abstraction to be of use in legal expert system
development.

Furthermore, developing a conceptual model of legal reason-
ing amounts to the writing of rneta-rules; 295 meta-rules have all
the limitations inherent in rule-based systems, with the single ex-
ception that they can cope with problems for which a rule-based
system does not have explicit rules. A conceptual model does not
meet the arguments put by rule sceptics. 290

48 A pragmatic legal expert system §1-7

I agree with Greenleaf et al.:

We believe that the absence of anything even resembling a "deep"
model of more than the smallest subset of the legal domain means
that expert systems of type (iii) are far in the future and will re-
quire very substantial resources to build. Whether the expenditure
of these resources is necessary or even justified, at least for the
purpose of building expert systems, will depend upon the per-
formance of level (ii) type systems. This is an empirical problem
which may only be resolved bj' building systems and evaluating
their performance. 297

SHYSTER is a system of type (ii).

As to the value of developing conceptual models of the law, it
is worth noting the words of Stone:

When excessive pretensions are avoided, it may be a worthwhile
intellectual activity to construct general concepts of law, or of par-
ticular notions found within legal orders, and to draw from these
logical implications concerning the conceivably possible arrange-
ments and contents of legal orders . . .

Such jurisprudential activity . . . does not, despite many mis-
conceptions, help to discover or create any actual law. Its raison
d'etre is basically to extend knowledge, and to order complex legal
materials for mnemonic purposes of legal study and legal reform.
Of course, like all efforts to extend knowledge, analytical juris-
prudence also serves to sharpen the mind. Also, since it exposes
premises from which existing legal rules may claim to have been
inferred, logical analysis may provide a basis for substantive cri-
ticism of law which includes the rationalising and the testing of
rationalisations offered for such rules. These are all legitimate
outcomes of logical analysis; but they must always be carefully dis-
tinguished from erroneous uses of these outcomes. Among these
erroneous uses is their tactical use in legal tasks, for instance to
persuade courts to a particular view of the law. 298

Of course it is just these legal tasks with which a legal expert
system should be designed to deal.

McCarty describes his ''language for legal discourse" as taking
"a first concrete step" towards the realization of a deep concep-
tual model. 299 Moles is disparaging: "McCarty is only taking his
first steps after some 15 years (although apparently he is going to
continue to use the same tools as previously). " 300

§1.8 Legal expert systems

1.8 Conclusion

Susskind argues that "jurisprudence can and ought to supply the
models of law and legal reasoning that are required for computer-
ized implementation in the process of building all expert systems
in law." 301 In fact, jurisprudence is of limited value to developers
of legal expert systems. For a legal expert system to be capable of
producing advice similar to that which one might get from a law-
yer, it needs to operate at the same pragmatic level of abstraction
as does a lawyer — not at the philosophical level of jurisprudence.
Many lawyers operate without jurisprudential insight; why not.
then, develop a legal expert system based upon a similar prag-
matic approach?

Jurisprudence is of greater value to developers of judgment ma-
chines, who are concerned more with the nature of law and justice
than with the nature of lawyers' arguments. But the social de-
sirability of judgment machines is questionable, and whether such
machines are possible is debatable. I doubt that anyone would
seriously advocate their development today.

The expectations of legal analysis systems have changed since
the proposals of the 1940s and 1950s. Interest now focuses not
upon judgment machines but upon legal expert systems: systems
which are designed not to pass judgment but to provide legal ad-
vice. Developers have, as Stone urged them to in 1964. faced the
limits of the contributions they can make. 302

Providing legal advice, in an adversarial legal system like that
in Australia, requires the construction and analysis of arguments
and counter-arguments. It also requires prediction of the likely
outcome, or at least some comment on the relative strengths of
the arguments. The jurimetrics researchers focused on prediction;
case-based systems like FINDER and HYPO focus on argument con-
struction, although they can identify one side's argument as being
stronger than another's.

Although a legal expert system should have a degree of predict-
ive power, it should not be normative. Its predictions must be
based on projections from the past. It need make no allowance
for new issues of policy, changing social mores, or other factors to
which a judge may, openly or otherwise, have regard when coming
to a decision. Without regard to these factors, a legal expert sys-
tem can still provide useful legal advice. Taking all of these factors
into account requires a level of predictive skill which is beyond all
expert systems — and beyond all but the most prescient of lawyers.

50 A pragmatic legal expert system

Susskind blames the lack of successful working expert systems on
the developers' failure to use jurisprudence. 30 Moles attributes
it to the developers' "uncritical acceptance of law as a system
of rules. " 3(M Certainly, most legal expert system designers have
embraced the work of Hart and the legal positivists, without con-
sidering the work of the legal realists or the rule sceptics.

Although a rule-based approach may be appropriate for repres-
enting statute law, it is not appropriate for representing case law.
and a case-based approach is clearly inappropriate for representing
statute law. A legal expert system should account for both stat-
ute law and case law. A hybrid approach, utilizing rule-based and
case-based techniques, addresses these representational problems.

Despite what Bench-Capon et al. say, 30 ' legal expertise is es-
sential in the development of a legal expert system. Knowledge
acquisition is as much a problem for the expert system designer
in the legal domain as it is in any other; legal expert systems are
subject to the knowledge acquisition bottleneck.

Deep conceptual models of legal reasoning are inappropriate
for legal expert systems. They contribute to the difficulty of
knowledge acquisition. They also operate at the same level of
abstraction as jurisprudence, so they have little relevance to the
pragmatic level of abstraction at which lawyers operate. A con-
ceptual model of legal reasoning attempts to model precisely a
process which is not fully understood. Hence, developers of these
deep models confuse precision with accuracy.

A legal expert system should operate at the same pragmatic
level of abstraction as does a lawyer. The approach taken in the
development of SHYSTER, detailed in the next chapter, is based
upon this conclusion.

Notes

1 Leith 1980b at 97.

2 Moles 1987 at 271.

3 Montesquieu 1973, liv. VI, ch. II at 84, footnote omitted. Translation by
Peter Brown, Lecturer, Department of Modern European Languages, Faculty
of Arts, The Australian National University.

4 The development of legal expert systems raises legal questions itself.
For example, who — If anyone — is liable for negligent advice provided by such
a system? These questions are beyond the scope of this book, but see Thomas
1979; Campbell 1984; Boden 1985; Willick 1985; Clarke 1988; Capper and
Susskind 1988 at 101-38.

Legal expert systems 51

5 Bain's JUDGE system (1980), although strictly speaking a rule-based
system, adopts an approach which is similar to that of a case-based system.
It analyses the similarities and differences between cases to give advice about
sentencing for certain crimes including murder, assault and manslaughter.
JUDGE starts with a rule about a crime (provided by the user) and an appro-
priate sentence taken from a case. When a new case is added, JUDGE finds
similar crimes and generates a rule about the new case on the basis of the
similarities and differences between the new case and those similar crimes.

6 McCarty 1980a identifies a third category of legal AI systems: "integ-
rated legal systems." He cites, as an example, computerized title registration
systems which make decisions about people's rights and obligations (at 2).
It is hard to see why such a system would not be better classified as a legal
analysis system, albeit with some of the features of a legal retrieval system.

7 On legal retrieval systems see Tapper 1963, 1984; Bing and Selmer 1980,
chs 8-15; Bing 1984a, 1984b, 1987, 1989: Croydon 1980; Dick 1991. For
an example of early work in the field see Allen, Brooks and James 1962.
Sprowl 1979 describes a document preparation system (see also Cook, Hafner,
McCarty, Meldman, Peterson, Sprowl, Sridharan and Waterman 1981). For
a summary of Australian legal retrieval systems (CLIRS, SCALE, etc.) see
Greenleaf, Mowbray and Lewis 1988.

8 Mehl 1959 at 759.

9 Shannon and Golshani 1988 at 306.

10 Susskind 1987a at 3.

11 This is, of course, a variation on the famous test for machine intelli-
gence proposed by Turing 1950.

12 For example, HYPO (§1.5.3) and the Latent Damage System (§1.4.4).
Apropos SHYSTER and legal expertise, see §2.4.

13 The doctrine of parliamentary supremacy over the courts developed in
England in the seventeenth century: see Morris, Cook, Creyke and Geddes
1992 at 8.

14 Morris et al. 1992 at 31.

15 Cross 1961 at 4.

16 See Stone 1964a, 1985: Harris 1980: Krygier 1986. See also the discus-
sion in §1.2.5.

17 Stone 1964a at 325.

18 In Queensland v. Commonwealth (1977) 139 CLR 585, Gibbs J (at 601)
and Stephen J (at 604) of the High Court of Australia felt bound by the prin-
ciple of stare decisis to follow Western Australia v. Commonwealth (1975) 134
CLR 201 — a decision in which they had been in the minority — despite still be-
lieving that that case had been wrongly decided. And in Western Australia v.
Commonwealth (1995) 183 CLR 373, Dawson J (at 493-4) felt bound to fol-
low Mabo v. Queensland (1988) 166 CLR 186 and Mabo v. Queensland [No. 2]
(1992) 175 CLR 1 — both decisions in which he had been in the minority —
though he expressly did not resile from his position in those cases.

19 Susskind 1987a at 19.

20 Niblett 1981 at 3.

21 Susskind 1987a at 20, emphasis in the original.

22 In 1987, only four of the twelve law schools in Australia had compulsory
courses in jurisprudence: Pearce, Campbell and Harding 1987 at 106-7.

52 A pragmatic legal expert system

23 Susskind 1987a at 26.

24 Harris 1980 at 2.

25 Niblett 1988 at 33.

26 Susskind 1987a at 20, as quoted above.

27 The Oxford English Dictionary, second edition, 1989.

28 Pound 1908 at 605.

29 Loevinger 1949 at 472, citing Frank 1930.

30 Pound 1908 at 607.

31 ibid, at 605.

32 Loevinger 1949 at 483, footnote omitted. See also Loevinger 1963.

33 §1.3.

34 Loevinger 1963 at 35.

35 As explained in §1.3.

36 Lasswell 1955 at 398. As Gardner 1987 comments, "[w]hy one robot
was not enough was not discussed; the idea may have been that hard cases
would be decided by a vote of the random number generators the robots were
to incorporate" (at 81-2).

37 Lasswell 1955 at 398.

38 Frank 1949 at 207.

39 Mehl 1959 at 758.

40 D*Amato 1977 at 1280.

41 ibid, at 1290-1.

42 Weizenbaum 1976 at 226-7.

43 D'Amato 1977 at 1281.

44 Stone 1964b at 555.

45 This increased level of certainty would be a result of freely available
(automated) judicial advisory opinions.

46 D'Amato 1977 at 1278.

47 ibid, at 1279.

48 §1.4.

49 D'Amato 1977 at 1290.

50 Kayton 1964 at 302.

51 Schubert 1968 at 63, emphasis omitted.

52 Tyree 1980.

53 Pound 1908 at 606.

54 Stone 1964b at 554. See also Spengler 1963: an earlier cautionary
article about judgment machines.

55 Stone 1964b at 558, emphasis omitted.

56 ibid, at 559.

57 ibid, at 560.

58 Hart 1983 at 106.

59 Hart 1961 at 132. As discussed in §1.2.6, a true rule sceptic would
agree with Hart that the rules embodied in cases are "as determinate as any
statutory rule" — that is, not at all.

60 Susskind 1986 at 189-90.

61 Hart 1983 at 106.

62 Section 13(2) of the Domestic Violence and Matrimonial Proceedings
Act 1976 (UK). See Moles 1987 at 142-57.

63 [19781 Fain 26.

Legal expert systems 53

64 Cantliff v. Jenkins [1978] Fain 47.

65 Davis v. Johnson [1979] AC 264.

66 Moles 1987 at 155, footnotes added. Compare Moles 's analysis with
that of MacCormick 1978 at 19-37. In order to demonstrate the possibility
of purely deductive legal justification he provides a detailed analysis of the
judgment of Lewis J in Daniels v. R. White and Sons Ltd [1938] 4 All ER 258.
He argues that that judgment is an example of deductive reasoning — a series
of applications of modus ponens: p D q\ p ; .'. q.

67 Moles 1987 at 172.

68 id.

69 Leith 's system is called ELI (see also §2.13.1).

70 Leith 1985b at 9. See also Leith 1985a, 1986a.

71 §1.4.

72 See Llewellyn 1931, 1951, 1962. See also Holmes 1897.

73 Drahos and Parker 1992 at 110, referring to Llewellyn 1931 at 1237.

74 Kripke 1982 bases his argument on Wittgenstein's Philosophical In-
vestigations, but Drahos and Parker 1992 cite devotees of Wittgenstein who
deny that he was the source of Kripke's argument or conclusion (at 111).

75 Kripke 1982 at 8-9.

76 ibid, at 13, 97.

77 Drahos and Parker 1992 at 112.

78 For example Yablon 1987 who agrees with Kripke, and Bjarup 1988
who does not.

79 Drahos and Parker 1992 at 113. It also puts paid to Susskind's claim
of a consensus in legal theory (§1.2.7): Hart's and Kripke's viewpoints are
totally irreconcilable.

80 Drahos and Parker 1992 at 112.

81 ibid, at 114.

82 id.

83 ibid, at 115.

84 id.

85 ibid, at 114.

86 Susskind 1987a.

87 §1.2.1.

88 Susskind 1987a at 27-8. See also Susskind 1987b at 2.

89 Susskind 1987a at vii.

90 ibid, at 27.

91 Susskind's expert system development work, as opposed to his juris-
prudential work, is discussed in §1.4.4.

92 Moles 1991 at 153-4.

93 Susskind 1987b at 2.

94 ibid, at 1.

95 ibid, at 2. Unaccountably, Moles 1991 does not make this point in his
criticism of Susskind.

96 Clark 1988 at 429.

97 Kelsen 1945, 1967; Hart 1961, 1983; Dworkin 1977, 1978.

98 Leith 1987 at 131-2, emphases omitted, footnote added.

99 §1.2.5.

100 51.2.6.

54 A pragmatic legal expert system

101 §1.2.2.

102 Loevinger 1961 at 259.

103 ibid, at 266, 269-70, 274.

104 Loevinger 1949 at 493.

105 §1.2.2.

100 Gardner 1987 at 69. Note, however, that the Jurimetrics Journal
of Law, Science and Technology (ISSN 0022-6793). formerly Modern Uses of
Logic in Law, is still published quarterly, and has been since 1959 — since 1966
under its current title.

107 In addition to the references made in the following sections, see Tan-
enhaus, Schick, Muraskin and Rosen 1963.

108 Kort 1963a, 1963b.

109 Kort 1963b at 145.

110 Lawlor 1963, 1968, 1981.

111 (1942) 316 US 455.

112 (1963) 372 US 335.

113 Wiener 1962 at 1023.

114 ibid, at 1024.

115 Kayton 1964 at 312, some emphasis omitted, some emphasis added.

116 Nagel 1960, 1963, 1986, 1989; Schubert 1963a, 1963b. See also Spaeth
1963; Aubert 1963.

117 Nagel 1963 at 40, emphases and footnotes omitted.

118 Haar, Sawyer and Cummings 1977.

119 ibid, at 758.

120 id.

121 See Gardner's comments quoted in §2.14. Note, however, that Gard-
ner is less critical of Haar et al. because "[bjy reflecting on their statistical
results in the light of a traditional legal analysis of the same cases, the authors
are able to provide recommendations for attorneys and litigants in zoning
cases and to raise questions about zoning law for the attention of both court
and legislature" (Gardner 1987 at 76, referring especially to Haar et al. 1977
at 742-50).

122 Loevinger 1949 at 461, citing Holmes 1897.

123 Loevinger 1961 at 269.

124 Lawlor 1963 at 339.

125 Susskind 1987a at 95-6.

126 Tapper 1973 at 249, footnote omitted.

127 ibicl. at 250.

128 Gardner 1987 at 196 (a note to p. 75). Gardner cites as examples
Schubert 1975 and Goldman and Sarat 1978, though her criticism clearly
applies to Nagel's work too.

129 Stone 1964b at 518. See also Stone 1966 at 688. Stone refers to
Schubert 1963b and Ulmer 1963 (see also Ulmer 1964).

130 Stone 1964b at 550, emphases in the original.

131 ibid, at 553, emphasis in the original.

132 §1.5 (especially §1.5.2).

133 Buchanan and Headrick 1970 at 40.

134 ibid, at 41.

135 ibid, at 45.

Legal expert systems 55

136 id.

137 Maggs and deBessonet 1972.

138 Popp and Schlink 1975.

139 Buchanan and Shortliffe 1984.

140 Popp and Schlink 1975 at 309.

141 Michaelsen and Michie 1983.

142 More precisely, EMYCIN: MYCIN with its domain knowledge re-
moved.

143 The domain of taxation law has been widely used in rule-based expert
systems research: apart from McCarty's TAXMAN project (§1.4.1), see Tor-
sun 1987; Sherman 1987, 1989. See also the hybrid CABARET system (§1.6).

144 Meldman 1977. See also Cook et al. 1981.

145 Cook et al. 1981 at 692.

146 Waterman and Peterson 1980. See also Cook et al. 1981; Waterman,
Paul and Peterson 1987.

147 Waterman et al. 1987 at 26.

148 ibid, at 27-8. These "informal rules' 1 are similar to the "conventions"
proposed by Drahos and Parker 1992, and discussed in §1.2.6.

149 Waterman et al. 1987 at 28.

150 Bing 1980; Borchgrevink and Hansen 1980.

151 Bing 1980 at 132.

152 Stamper 1980; Stamper, Tagg, Mason, Cook and Marks 1982. See
also Cook et al. 1981. Stamper later decided that LEGOL is inadequate, and
developed NORMA (1986).

153 Pattison and Ciesielski 1990. See also Ciesielski 1990.

154 Johnson and Mead 1991.

155 Susskind 1989a at 29.

156 McCarty 1977, 1980a, 1980b, 1980c.

157 McCarty 1980a at 3.

158 id.

159 id.

160 ibid, at 4.

161 id.

162 id.

163 Moles 1987 at 270.

164 McCarty 1980a at 4-5.

165 ibid, at 5.

166 McCarty 1977 at 839.

167 ibid, at 839-40.

168 Austin 1885.

169 Moles 1987 at 270-1, footnote added.

170 ibid, at 269.

171 McCarty 1980c at 29-30.

172 McCarty and Sridharan 1981.

173 Eisner v. Macomber (1920) 252 US 189 (see McCarty 1982).

174 Ashley 1990 at 224-5, footnote added, extraneous punctuation re-
moved. Ashley's (i.e. HYPO's) mechanism for determining how "on point"
are two cases is discussed in §2.9.3.

175 Moles 1987 at 270, referring to the structure of TAXMAN I.

56 A pragmatic legal expert system

176 Ashley 1990 at 220-7. Ashley criticizes Goldman, Dyer and Flowers
1987 for a similar use of primitives in their STARE system.

177 Kowalski and Sergot 1985, 1990; Sergot, Cory, Hammond, Kowalski,
Kriwaczek and Sadri 1986a, 1986b; Bench-Capon, Robinson, Routen and Ser-
got 1987; Bench-Capon and Sergot 1988; Kowalski 1989; Bench-Capon 1991.

178 Sergot et al. 1986a at 49. Note that this statement is qualified in an-
other paper by the same authors published in the same year: "The knowledge
elicitation problem is almost entirely absent in the formalization of legisla-
tion' 1 (Sergot et al. 1986b at 383, emphasis added).

179 Moles 1991 at 144.

180 Kowalski and Sergot 1985; Sergot et al. 1986a, 1986b; Bench-Capon
and Sergot 1988; Kowalski 1989.

181 Sergot et al. 1986a at 41.

182 Bench-Capon et al. 1987 at 192.

183 Moles 1991 at 159. He also refers to Bench-Capon 1991 (a collection
of articles by Bench-Capon, Sergot, and others) and notes that, "[ajlthough
the book is called Kno w ledge-B ased Systems and Legal Applications, it would
appear from [the biographical notes] that there was very little emphasis on
involving people with legal expertise" (at 146). Moles's article sparked a
debate: see Tyree 1992; Moles and Dayal 1992.

184 Kowalski and Sergot 1990 at 207.

185 Hayes-Roth, Waterman and Lenat 1983 at 165, quoted by Susskind
1986 at 176.

186 Sergot et al. 1986a at 46.

187 Leith 198Gb at 97, some extraneous punctuation removed. Leith's
remarks are particularly interesting in the light of Susskind 's criticism of Leith
for having built a legal expert system without legal expertise (§2.13.1).

188 Gardner 1983, 1984, 1985, 1987.

189 Gardner 1987 at 4.

190 ibid, at 6.

191 Susskind 1990 at 222.

192 Gardner 1987 at 120.

193 ibid, at 119.

194 Susskind 1990 at 225. Grunbaum 1988 is similarly critical of Gard-
ner's "courageous" 1 attempt (at 639). Jones 1988 also doubts "[t]he utility of
the system outside this domain" (at 32).

195 Ashley 1990 at 226.

196 See Levi 1949.

197 Gardner 1987 at 22, footnote added.

198 §1.2.6.

199 §1.3.

200 Gardner 1987 at 22.

201 id.

202 ibid, at 33.

203 Susskind 1989a at 30.

204 §1.2.1 and §1.2.7.

205 Gold and Susskind 1986; Susskind 1986, 1987a, 1987b.

206 Capper and Susskind 1988; Susskind 1989b.

207 Capper and Susskind 1988 at vii. See also Susskind 1989b at 24.

Legal expert systems

208 Susskind 1987a at 255.

209 ibid, at 87, citing Simpson 1986.

210 Susskind 1987a at 256.

211 ibid, at 87.

212 Feigenbaum 1981 at 226. See also Feigenbaum and McCorduck 1983
at 75.

213 §1.4.2.

214 Sergot et al. 1986a at 49.

215 Allen 1963, 1965, 1968, 1981, 1982; Allen and Caldwell 1963; Allen
and Saxon 1985, 1988. See also Bench-Capon 1987.

216 Shannon and Golshani 1988 at 308. They also warn that ad hoc rule
formulation means that subsequent designers/users cannot trace the evolution
of a set of rules from the words of the statute. However, I suggest that this
disadvantage could easily be obviated by the provision of good documentation.

217 ibid, at 308-9.

218 However, Dworkin 1978, for example, contends that legal problems
do have a correct answer.

219 deBessonet and Cross 1985, 1986, 1988.

220 Shannon and Golshani 1988 at 310, footnotes omitted, one footnote
added. An extraneous right parenthesis has been removed from the ANF
representation .

221 This criticism of Shannon and Golshani *s modified example does not
apply to the original example given by deBessonet and Cross 1985 who refer,
in much less ambiguous terms, to the lessee's belief about damage caused by
her /himself (at 208).

222 Tyree, Greenleaf and Mowbray 1988 at 232.

223 §1.2.6.

224 Tyree et al. 1988 at 232.

225 Stone 1985 at 56. See also Tyree 1989 at 136.

226 Tyree et al. 1988 at 232.

227 §2.3.

228 Kowalski 1991 at 30.

229 Tyree et al. 1988 at 232. Quinlan, Compton, Horn and Lazarus 1987
give two examples of the use of an inductive technique. The first involves
2647 cases, from which a training set of 2000 and a test set of 647 are extracted
(at 165-6). The second involves 3066 cases: a training set of 2300 and a test
set of 766 (at 168-70). These examples use an inductive inference tool called
C4: a descendent of ID3. See also Quinlan 1986, 1988.

230 Susskind 1986 at 190.

231 §2.13.3.

232 §1.5 and §1.0.

233 §2.14.2.

234 Sergot et al. 1986b at 370.

235 Pearce and Geddes 1988 at 56-62.

236 Bankowski and MacCormick 1991 at 377.

237 Pearce and Geddes 1988 at 1.

238 Ashley 1990 at 127.

239 Lambert and Grunewald 1991 at 194.

58 A pragmatic legal expert system

240 Ashley 1985 at 116. See Ashley and Aleven 1991 for a description of
their research into developing an intelligent tutoring system for teaching law
students to argue with cases.

241 §1.3.5.

242 §2.14.1.

243 Mackaay and Robillard 1974. They did not actually develop an expert
system, but their work is relevant to the development of legal expert systems
and, hence, is discussed here.

244 The first formulation of the nearest neighbour rule appears to have
been by Fix and Hodges in 1951.

245 Cover and Hart 1967 at 21.

246 Lawlor reported his results in a paper presented to the 1971 Annual
Meeting of the American Political Science Association. Excerpts from that
paper, published in 1972, omit details of his work with these capital gain
cases. Lawlor 's approach was to find a weight for each of the attributes such
that the sum of those weights, where the attribute is present, is less than 1 if
the case was decided against a given party, and greater than 1 otherwise.

247 Mackaay and Robillard 1974 at 308.

248 See Schmidt 1971; Claudy 1972.

249 Mackaay and Robillard 1974 at 322. They also examine the visual
representation of cases, scaling the cases onto one- and two-dimensional maps
using the same similarity metric as they use for their nearest neighbour ana-
lysis. These maps allow easy visual identification of the borderline cases.

250 Tyree 1985, 1986, 1989; Tyree, Greenleaf and Mowbray 1988, 1989.
FINDER's approach to case law is based upon a mathematical concept of the
similarity of cases described by Tyree 1977.

251 Treating each YES as a 1, and each NO as a 0.

252 Tyree 1989 at 12-13.

253 §1.3.5.

254 §2.14.1.

255 §4.2.

256 Ashley 1985, 1986, 1989a, 1989b, 1990; Rissland 1983, 1985, 1990;
Rissland, Valcarce and Ashley 1984; Ashley and Rissland 1987a, 1987b, 1988;
Rissland and Ashley 1987, 1989.

257 Ashley 1990 at 2-3.

258 ibid, at 37.

259 id.

260 For example, a finding that there has been trade secrets misappropri-
ation.

261 Ashley 1990 at 112-3.

262 ibid, at 279. "On pointness" in HYPO is discussed in §2.9.3.

263 ibid, at 20.

264 ibid, at 142.

265 id.

266 See Rissland 1989 for a discussion of argument with hypotheticals.

267 Hafner 1981. See also Cook et al. 1981. "Negotiable instruments'" are
cheques and promissory notes.

268 GREBE is described in §1.6.

269 Brantins 1989 at 109.

Legal expert systems 59

270 §1.7.

271 Branting 1989 at 109.

272 Rissland and Skalak 1989a at 48.

273 Rissland and Skalak 1989a, 1989b; Skalak 1989: Skalak and Rissland
1991, 1992. See also Rissland 1990.

274 Rissland and Skalak 1989b at 527.

275 Branting 1989, 1991.

270 See §1.5.4 for a discussion of the inadequacy of semantic networks for
representing case law.

277 Branting 1991 at 150.

278 Oskamp, Walker, Schrickx and van den Berg 1989; Walker, Zeinstra
and van den Berg 1989.

279 van Opdorp, Walker, Schrickx, Groendijk and van den Berg 1991
at 280.

280 Oskamp 1989; Oskamp et al. 1989; Walker et al. 1989; Wolstenholme
1989; van Opdorp et al. 1991.

281 van Opdorp et al. 1991 at 280.

282 ibid, at 285.

283 Berman and Hafner 1991; Vossos, Zeleznikow, Dillon and Vossos 1991.
See also Sanders 1991 (a hybrid planner).

284 See especially §2.2.7.

285 Susskind 1987a at 151.

286 ibid, at 153.

287 Hafner 1981, 1987; deBessonet and Cross 1985, 1986, 1988.

288 McCarty 1984, 1986, 1989, 1990, 1991.

289 McCarty 1984 at 126.

290 Greenleaf, Mowbray and Tyree 1987 at 11.

291 McCarty 1983 at 267-8. See also Bench-Capon 1989.

292 Shannon and Golshani 1988 at 311. They propose a "happy medium"
between shallow and deep conceptual models.

293 Greenleaf et al. 1987 at 11.

294 Susskind 1987a at 153.

295 Meta-rules are similar to the "conventions" proposed by Drahos and
Parker 1992, and discussed in §1.2.6.

296 §1.2.6.

297 Greenleaf et al. 1987 at 11.

298 Stone 1985 at 46, footnotes omitted, emphases in the original.

299 McCarty 1989 at 180.

300 Moles 1991 at 163.

301 Susskind 1987a at 20.

302 Stone 1964b at 560, as quoted in §1.2.4.

303 Susskind 1987a at 20.

304 Moles 1987 at 7.

305 §1.4.2.

A pragmatic approach
to case law

I know you Lawyers can, with ease,
Twist words and meanings as you please;
That language, by your skill made pliant,
Will bend to favour ev'ry client;
That 'tis the fee directs the sense
To make out either side's pretense.
When you peruse the clearest case,
You see it with a double face;
For scepticism's your profession;
You hold there's doubt in all expression.
Hence is the bar with fees supply d,
Hence eloquence takes either side . . .

John Gay (1732)
The Dog and the Fox

Then there is the doctrine of precedent, one of my favourite doctrines. I
have managed to apply it at least once a year since I've been on the Bench.
The doctrine is that whenever you are faced with a decision, you always follow
what the last person who was faced with the same decision did. It is a doctrine
eminently suitable for a nation overwhelmingly populated by sheep. As the
distinguished chemist, Cornford, said: "The doctrine is based on the theory
that nothing should ever be done for the first time."

Lionel Murphy (1979)
The Responsibility of Judges'

We do not use to judge of Cases by fractions.

John Finch (1637)
R. v. Hampden

til

)2 A pragmatic legal expert system §2-1

2.1 Introduction

The adoption of a pragmatic approach to legal expert system
design is advocated in the previous chapter. This chapter de-
scribes one such pragmatic approach to case law — the approach
which was adopted for the development of SHYSTER.

SHYSTER'S design criteria are set out in §2.2. SHYSTER is based
upon a model of legal reasoning which is described in §2.3. Users of
a system like SHYSTER require some legal expertise, as explained
in §2.4.

SHYSTER's knowledge representation structure is described in
§2.5. This structure was designed to facilitate specification of
different areas of case law using a specification language which
is described in §2.6. Areas of case law are specified in terms of
the cases and attributes of importance in those areas. SHYSTER
weights its attributes (as described in §2.7) and checks for depend-
ence between them (§2.8).

In order to choose cases upon which to construct its opinions,
SHYSTER calculates distances between cases (§2.9) and uses these
distances to determine which of the leading cases are nearest to
the instant case (§2.10). SHYSTER uses information about these
nearest cases to construct a report (§2.11). Several safeguards are
employed so as to warn the user when SHYSTER's opinion may
be suspect (§2.12). Methods of testing and evaluating SHYSTER's
performance are discussed in §2.13.

Conclusions are drawn in §2.14. SHYSTER's approach to case
law is compared with those of other systems, and an argument
that an approach like SHYSTER's is inappropriate for case law is
examined and refuted.

2.2 Design criteria

The following design criteria were used in the development of SHY-
STER. They are based upon the conclusions drawn in the previous
chapter, 4 and are presented here in the future tense.

2.2.1 Users and output

SHYSTER will be designed to be used by lawyers. Hence, SHYSTER
will attempt to imitate the manner in which lawyers write advice
for their clients, and for each other; its output will be constructed
so as to resemble legal advice produced by and for a lawyer.

§2.2 A pragmatic approach to case law

The user will be assumed to have legal expertise, though no
specific expertise in the area of law about which SHYSTER is in-
terrogated. This approach will not significantly restrict the utility
of the system, and has been adopted by other expert system de-
velopers. Susskind, for example, suggests that:

. . . the users of expert systems in law should be lawyers, or at least
those with considerable familiarity with the workings of the legal
and court systems . . . for a system to be used responsibb 1 ", the user
must be aware of the possible role in legal reasoning of 'principles'
... and of 'purpose 1 ... Moreover, he must be sensitive to the
drawbacks and implications of 'compartmentalizing* the law . . .
and capable too of recognizing those occasions when some legal
aid cannot help him with any problem at hand . . .°

A system designed to be used exclusively by lawyers is still a legal
expert system as defined in the previous chapter.

(The role of legal expertise in the development and use of a legal
expert system is further discussed in §2.4 below.)

2.2.2 A pragmatic model of legal reasoning

As SHYSTER's advice will be in a form that might be produced by
a lawyer, SHYSTER will be designed to operate at the same level
of abstraction as does a lawyer.

Lawyers operate on a day-to-day basis at a pragmatic level of
abstraction which is different to the philosophical level of a jur-
isprudent. A legal expert system should be built upon a model
of legal reasoning, but this model need not conform to any jur-
isprudential theory about the nature of law. The model of legal
reasoning adopted for SHYSTER will reflect the way in which law-
yers reason with statutes and cases in areas of private law\ 7

(The model of legal reasoning adopted for SHYSTER is explained
in §2.3 below.)

2.2.3 Knowledge representation

The representational structure used for SHYSTER will be as simple
as possible while complex enough to allow SHYSTER to produce
good advice.

64 A pragmatic legal expert system §2-2

Simple knowledge representation is consistent with the choice
of a pragmatic model of legal reasoning over a more complex,
or deeper, model. Furthermore, complex knowledge representa-
tion requires commensurately complex knowledge acquisition. A
simpler representation — one which makes no attempt to model ac-
curately the way lawyers represent legal problems- — will obviate,
to some extent, the knowledge acquisition problem.

The knowledge representation structure will be similar to that
used for the FINDER system: cases will be represented as points
in space, the dimensionality of which is the number of relevant at-
tributes. Like FINDER, SHYSTER will allow attributes to have yes
or no values. In addition, SHYSTER will allow an attribute's value
to be unknown. The smaller the distance between two points in
this space, the more similar they will be considered to be.

Using such a structure, SHYSTER will choose cases to use in
argument on the basis of similarity. This structure will be less
sophisticated than, for example, HYPO. However, this simpli-
city can be justified on the grounds that it will greatly simplify
the knowledge acquisition process— avoiding, to some degree, the
knowledge acquisition bottleneck. Despite this simpler structure,
SHYSTER will still be capable of arguing with hypotheticals like
HYPO, though not necessarily with the same sophistication.

(SHYSTER's knowledge representation structure is detailed in
§2.5 below. Differences between SHYSTER and FINDER are ex-
plained in §2.14.1.)

2.2.4 Generality of application

SHYSTER will be of general design, so that it can operate in more
than one legal domain. The only restriction upon applicable do-
mains will be that they conform to the model of legal reasoning
adopted for the system.

Generalizability has been claimed by several legal expert system
developers, 8 but to date none has been demonstrably generalizable
(or even demonstrably widely applicable). SHYSTER will be de-
signed so that once information about different areas of case law
has been specified for it, it can be interrogated as to the ramific-
ations of any or all of those areas of law for a given situation.

(SHYSTER's case law specification mechanism — a specification
language — is introduced in §2.6, and its use is illustrated by ex-
ample in the next chapter. 9 )

§2.2 A pragmatic approach to case law 65

2.2.5 Prediction and argument

SHYSTER will be designed to make a prediction about the likely
result in a case. This prediction will be based upon previously
decided cases, assuming (as must any legal expert system which
takes account of case law) the application of the doctrine of preced-
ent. SHYSTER will also produce legal argument supporting, and
opposing, the predicted outcome. The calculations that SHYSTER
uses to reach its conclusions and to construct its legal arguments
will not form part of those arguments, although those calculations
will be accessible to the user.

A legal expert system's predictive ability and its ability to con-
struct legal argument are both important: prediction is a valuable
component of legal advice, but the nature of the adversarial sys-
tem requires that a lawyer be able to argue a case, and be prepared
to respond to counter-arguments.

Of course, SHYSTER will not be normative. A prediction will
merely be a statement about the likely outcome — a statement
about the relative strengths of the arguments that are constructed.

(SHYSTER's method of constructing arguments is explained in
§2.11.)

2.2.6 Evaluation of advice

SHYSTER'S advice will be evaluated by reference to the accuracy
of its predictions and the quality of its arguments. These are two
of the three main criteria by which a lawyer's advice is evaluated;
the third criterion is the cost of the advice which, for the purposes
of this project, is ignored.

As a principal method of evaluating its advice. SHYSTER will
be given details on real cases and its output compared with the
actual results and legal arguments in those cases. (This and other
methods of testing and evaluation are discussed in detail in §2.13.)

2.2.7 A hybrid structure

SHYSTER will be designed so that it can be linked with a rule-
based system to form a hybrid system.

A case-based s^'stem is inappropriate for representing statute
law. 10 Some researchers have recommended the use of both rules
and cases to represent case law. As discussed in the previous
chapter, 11 Susskind advocates the use of rules to represent statutes

66 A pragmatic legal expert system §2-3

and clear cases, and hints that other methods of representation
could be employed for the hard cases. In effect he proposes divid-
ing a rule-based system from a case-based system at the boundary
between clear and hard cases. The major drawback with this ap-
proach is that there is considerable doubt as to whether there is
such a thing as a clear case — and even if there is, no-one has de-
vised a method of identifying one. 12

CABARET uses rules and cases to represent case law. but its
structure is a mixed one. Rules are used to guide case-based ar-
guments. Although it is not clear exactly where the two S3*stems
meet, the division falls somewhere in case law. The problem with a
mixed approach is that it complicates the model of legal reasoning
and, consequently, the system's knowledge representation.

Deciding exactly how to divide the law into rule-based and case-
based sources is an arbitrary process. For SHYSTER, a clean and
intuitive approach will be adopted. Rule-based techniques will be
used only for the representation of statute law; case-based tech-
niques will be employed for case law. This division can be clearly
defined because the sources of law (statutes and cases) are clearly
defined. It also simplifies the knowledge acquisition process.

2.3 A model of legal reasoning

A legal expert system must be based upon a (possibly implicit)
model of legal reasoning. For the development of SHYSTER, a
pragmatic model of legal reasoning was adopted. That model is
explained and set out here.

2.3.1 Private and public law

Traditionally*, a distinction has been drawn between private and
public law. 13 Private law concerns relationships between citizens;
public law concerns relationships between citizens and the state.
Private law is characterized by commercial law; public law is char-
acterized by constitutional law or international law.

This distinction has been rejected by some jurisprudents. 14
However, it is certainly true that some areas of law are more
overtly concerned with matters of policy than are others. In these
areas — public law areas — precedent is given less weight than it is
in private law areas. In areas of private law, where predictability is
crucial, the doctrine of precedent is given greater weight (at least
ostensibly 15 ).

§2.3 A pragmatic approach to case law

This distinction is not sharp: precedent still applies in public
law, and matters of policy underlie much of private law. The
difference is a matter of emphasis. However, this difference in em-
phasis affects the way in which lawyers reason in different areas
of law. SHYSTER's model of legal reasoning assumes the applica-
tion of the doctrine of precedent, and has no regard to matters of
policy. This model reflects the way in which lawyers reason with
statutes and cases in areas of private law.

2.3.2 The functions of legal reasoning

Susskind identifies three functions of legal reasoning: justification,
prediction and persuasion. Judges, he says, need to provide "at
least ostensible reasons" 10 to justify their decisions. Lawyers try
to predict judicial or official behaviour, and try to persuade the
courts. 17 These three functions, Susskind points out, are "not fun-
damentally incompatible with one another". 18 Indeed, the designer
of a legal expert system does not have to choose one of these three
functions:

For there is an underlying, more restricted, and yet fundamental,
model of legal reasoning, common to all three accounts of function
just noted, that should be at the core of all current systems. 19

Susskind, it must be remembered, claims to have found a con-
sensus in jurisprudential theory. 20 This explains his reference to
a model. There is no jurisprudential consensus, and there is no
single model of legal reasoning. However. Susskind is correct in
that, if a model of legal reasoning is to be used for the development
of an expert system, it should be "common to all three accounts
of function" that he identifies.

2.3.3 Adopting a model of legal reasoning

For the purposes of the development of SHYSTER, the following
model of the process of reasoning with statutes and case law was
adopted. 21

A lawyer examines the facts of the case in question — the instant
case — and determines which area of law, and which statutes (if
any) apply. These statutes are applied to the facts of the instant
case. The meaning of a concept in a statute may be open-textured,
and may determine the result of the application of that statute to
the instant case.

68 A pragmatic legal expert system §2-3

A lawyer argues about the meaning of an open-textured concept
by reference to the facts of the instant case and those of previously
decided cases. The results of some cases are desirable in that they
ascribe a meaning to an open-textured concept which (when the
statute is applied) leads to a desired result in the instant case.
No two cases can be completely identical, given the plethora of
facts associated with any given case. Some of these differences
may be insignificant, and much of a lawyer's reasoning by analogy
concerns the legal significance of these differences.

A lawyer argues with cases in the following fashion:

• if the result of a previously decided case is desirable, she/he
argues that there are no legally significant differences between
the previous case and the instant case, so the previous case
should be followed;

• if the result of a previously decided case is undesirable, she/he
argues that there is some legally significant difference between
the previous case and the instant case upon which the previ-
ous case should be distinguished.

However, as Hart points out, "the class of such differences can
never be exhaustively determined." 22

This model allows that some concepts in a statute may be
open-textured, but assumes that these concepts are amenable to
full definition — at some level of abstraction — by reference to case
law. That is, open-textured statutory concepts can be defined
by arguing with cases, a process which (in turn) may involve
arguing about other open-textured concepts. These case-based
open-textured concepts are also assumed to be amenable to defin-
ition by further reference to case law. The model does not allow
that a case-based open-textured concept may be defined by refer-
ence to statutes. This assumption simplifies the model and is not
a significant restriction upon its application.

The appropriate level of abstraction below which a concept is
considered to be fully defined (i.e. answ T erable by the user) depends
upon the legal expertise of the user, as explained below. 23

That part of this model that deals with cases is consistent with
what Ashley describes as the standard model of analogical legal
reasoning. 24 This model has three steps: identifying a proper pre-
cedent; analyzing the facts and comparing and contrasting the

§2.4 A pragmatic approach to case law 69

precedent with the instant case; and determining whether the fac-
tual similarities or the differences are more important under the
circumstances (i.e. deciding whether to follow or distinguish the
precedent).

As Ashley points out, this model provides no guidance as to
which similarities and differences are more important, or for decid-
ing between competing analogies. He argues that "HYPO's model
of analogical legal reasoning meets both of these criticisms", 25 and
he is right in the sense that HYPO has a well-defined algorithm for
choosing important similarities and differences and for choosing
between the most analogous precedents. So too has SHYSTER, as
is explained below. 20

2.4 Experts, users, and expert users

As already discussed. 2 ' legal expertise is essential to the develop-
ment of a legal expert system. I am both a lawyer and a computer
scientist. I developed SHYSTER and tested it with the assistance
of three lawyers expert in different areas of law. 28

In the model of legal reasoning described above, the level of
abstraction at which open-textured concepts are considered to be
fully defined depends on the level of legal expertise possessed by
the user of the system.

Take, for example. Shannon and Golshani's discussion of the
development of a rule-based expert system. They identify one of
the major problems of expert system design as "the overwhelming
problem of open texture, illustrated by the swelling volume of
case law." 29 In fact, they don't see the problem as being literally
overwhelming, for they propose a solution. If a predicate is open-
textured, they say, the designer has two options: to rely on the
user's assessment of whether the predicate is true, or to produce
new rules which define that open-textured concept.

As explained in the previous chapter, 30 the second option is
inappropriate. The first option represents a very high level of
abstraction at which open-textured statutory concepts are deemed
to be fully defined. It is, of course, not a very practical level of
abstraction to choose: the only people qualified to use such a
system would have no need of it.

The approach adopted in SHYSTER is to assume that the user
has legal expertise, though no specific expertise in the area of
law that is represented. 31 And SHYSTER'S ability to argue with

A pragmatic legal expert system $2.-5

instantiations of the facts of the instant case means that a user
who is unable to answer any of SHYSTER's questions can force
SHYSTER to consider all the possibilities (as explained below 32 ).

Legal expertise is required at the first step of the model of legal
reasoning described above: i.e. determining which statute applies.
Unless a single expert system can cover the whole field of the
law — an unlikely prospect, discussed below 33 —then legal expertise
is required at the top level to identify the relevant area of law and
to choose an appropriate expert system.

For the remainder of this book, the term the legal expert is used
to refer to the person who specifies areas of law for SHYSTER.
The user of the system — also a legal expert, though with different
expertise — is called the user.

2.5 Knowledge representation

SHYSTER adopts and expands upon the approach to case law ad-
opted by FINDER. However, where FINDER gives advice only on
the law of trover. SHYSTER gives advice in an area of case law
specified by a legal expert. SHYSTER's method of representing
knowledge about case law was designed so as to be complex enough
to allow the production of good advice, yet simple enough to facil-
itate knowledge acquisition and avoid the knowledge acquisition
bottleneck.

A program written in SHYSTER's case law specification language
is called a specification. A specification may contain any number
of areas. Each area is specified in terms of attributes, leading
cases, and hypothetical (ideal) cases, 34 and relationships between
these entities.

The legal expert may also specify a hierarchy of courts which
applies in all areas of the specification. This hierarchy allows SHY-
STER to take account of the relative ranking of leading cases when
constructing its opinion.

2.5.1 Areas

For SHYSTER, an area of law represents an open-textured concept.
Each area has at least two results. A result is a possible resolution
of the open-textured concept that the area represents. 30

If the attributes within an area are also open-textured, they
may be defined by reference to other areas in the same specific-
ation. For example, an area may be specified in order to define

§2.5

A pragmatic approach to case law

root node

Figure 2.1: The structure of an example specification, repres-
ented as a directed acyclical graph. The circles are parent nodes,
representing open-textured concepts; the squares are leaf nodes,
representing concepts which are considered to be fully defined.

a statutory open-textured concept. That area may have several
open-textured attributes which are linked to other areas, which in
turn may also include open-textured attributes. This linkage is
achieved by binding the result returned from one area to a partic-
ular value for the appropriate op en- textured attribute.

There is no theoretical limit to the number of levels of areas that
may be used, but every open-textured attribute must be defined,
at some level, in terms of areas which have no open-textured at-
tributes; circular definitions are not allowed. The same area may
be linked to more than one attribute.

This structure can be thought of as a directed acyclical graph.
an example of which is given in figure 2.1. 3G The circles are par-
ent nodes, representing open-textured concepts; the squares are
leaf nodes, representing concepts which are considered to be fully
defined (i.e. answerable by the user). The top-level parent node is
called the root node.

12 A pragmatic legal expert system $2.-5

Each area has a parent node, with a child node for each of its
attributes. The root node is the top level — usually statutory—
open-textured concept. The user need only provide values for the
leaf nodes; SHYSTER determines a value for each parent node-
each open-textured concept — until a value is obtained for the root
node — the top-level open-textured concept.

In 1959, Mehl suggested that "a machine covering the whole
field of law would be simpler and less cumbersome than a series of
machines handling separate legal sectors."' 7 Susskind disagrees:

. . . the practical problems faced in engineering a system to function
even in a limited legal domain of application are so numerous
that it is likely that the only way 'a vast field of law' [Mehl's

words 38 ] could be catered for is through the networking of small

systems. 39

The approach to case law adopted for SHYSTER lies somewhere
between these two approaches. Theoretically, a single specification
could represent a vast field of law using a large number of areas.

2.5.2 Attributes

For SHYSTER, there are two types of attribute. An open-textured
attribute is called an external attribute because it is defined by
reference to another area of case law: an external area. Its value
is determined by the result in that external area. The value of a
local attribute must be input by the user. In figure 2.1, the parent
nodes (with the exception of the root node) represent external
attributes; the leaf nodes represent local attributes.

Each external attribute has an association between each of the
possible results in its external area and an attribute value. When
the result of the external area is determined, the appropriate value
is given to the attribute. Each local attribute has an associated
question. The user gives the attribute its value by answering that
question.

An area's attributes should represent all of (what the legal ex-
pert deems to be) the relevant similarities and differences between
cases in that area. These may be questions of fact or — because
an attribute may represent an open-textured concept— questions
of law. A question of fact is represented by a local attribute; a
question of law can be represented by a local or an external attrib-
ute, depending on the extent of its open-texture and the assumed
legal expertise of the user. A point of law which is below the

§2.5 A pragmatic approach to case law

appropriate level of abstraction is considered to be fully defined
(i.e. answerable by the user) and can be represented by a local
attribute despite its open texture.

In allowing attributes to be questions of fact and of law. this ap-
proach differs from those discussed in chapter 1 where an attribute
is considered to be a "legally important fact." However, this ap-
proach is a natural consequence of the model of legal reasoning
adopted for SHYSTER. 40

An attribute need not have been judicially enunciated for the
legal expert to include it in the specification. 41 An attribute may
be a political attribute: i.e. one which Gardner would characterize
as "legally irrelevant." 42 The fact that an attribute ought not to
be relevant is no reason to exclude it; SHYSTER is predictive, not
normative.

Attribute direction allows any value for any attribute (local or
external) to be "directed" towards a result (or results). Attribute
direction indicates that the occurrence of that value for that at-
tribute suggests that result (or results). Such an occurrence is not
conclusive, merely suggestive. SHYSTER uses attribute direction
as a safeguard, as explained below. 43

2.5.3 Leading cases and attribute values

Each of SHYSTER's areas includes details of the important cases
decided in that area of the law. These are the leading cases. Not
all cases in the area should be specified, only the important ones:
the best cases for SHYSTER to use in legal argument.

By letting the legal expert decide which are the important cases,
SHYSTER avoids the need to "screen" or "purify" the leading cases,
as Mackaay and Robillard do. The legal expert's choice of cases
is assumed to be a good one, even if some cases "appear to be
erroneous in relation to the other ones." 44

For each leading case, the legal expert specifies various items of
information including a fact vector. The fact vector is a vector of
attribute values: one value for each attribute specified in the area.
Each attribute value is either yes, no. or unknown. The fact
vector represents the relevant facts of the case — although some of
these "facts" may actually be questions of law. 40

Every attribute has a value in each case's fact vector (wherever
possible, a known value: yes or no) even if that attribute was
not the subject of legal argument and/ or was not judicially con-
sidered in that case. Lawyers in their arguments, and judges in

74 A pragmatic legal expert system §2-6

their judgments, do not always address all of the matters of legal
significance in a case. 40

Furthermore, if a case is decided before the case in which an
attribute is first judicially enunciated, that attribute is assumed
to have always been important, even though not mentioned until
the later case. This assumption is consistent with the declaratory
theory of law. 47 That theory has been strongly criticized, 48 but
this assumption can be justified on the basis that the previously
decided cases used in a specification are assumed to be consistent
with each other and (taken together) to represent the law as it is,
not as it used to be. 49 This is not to deny that judges make law.
and that case law evolves over time. It is simply to assume that
the cases which a lawyer would use in argument in a given area of
law form a consistent whole.

A result must be associated with each case. This corresponds
to the decision reached by the court.

2.5.4 Ideal points

The legal expert may also specify ideal points. An ideal point 50
represents the best case for a given result. A fact vector is spe-
cified, representing the ideal combination of attribute values for
that result. For an ideal point, an attribute value of unknown in-
dicates that the value of that attribute does not matter. Only one
ideal point may be specified for each result. 01 SHYSTER uses ideal
points to provide a safeguard against giving erroneous advice. 52

2.6 A specification language

I designed a language in which areas of case law can be spe-
cified for use by SHYSTER. Programs written in this language-
specifications — reflect the structure of SHYSTER's knowledge rep-
resentation (described above 53 ).

Each specification can contain any number of areas. Each area
contains information on results, attributes, cases and ideal points.
Results are represented as strings of characters: statements in
English as to the effect of that outcome. Attributes are represen-
ted by a collection of strings explaining the effect of each possible
attribute value. Local attributes also have a string which is the
question to ask the user when determining the value of that at-
tribute for the instant case.

§2. 7 A pragmatic approach to case law 75

Leading cases are represented by citation information (names.
dates, etc.) and a vector of attribute values: one value for each of
the attributes specified in the area. A vector is used to represent
each ideal point.

Identifiers are used to link these concepts. So. for example, each
result has an identifier and each leading case is linked to a result by
use of one of those identifiers. Areas also have identifiers, so they
can be accessed by name by a rule-based system and/or linked
to external attributes in other areas. The result from an area is
indicated to the rule-based system (or the external attribute in
the case-based system) that invoked it by use of a result identifier.

Each specification can include a hierarchy of courts with a list of
strings: each describing a court and each with an identifier. Any
case in any area in the specification can be linked to a court in
that hierarchy using its identifier.

In this fashion, identifiers are used to link entities within, and
between, areas in the same specification.

The use of this language to specify areas of case law for SHY-
STER is illustrated by example in the next chapter. 54 A formal
definition of the syntax of SHYSTER'S specification language is
given in figure 3.3.

2.7 Weighting attributes

In order to construct its opinion. SHYSTER quantifies the "dis-
tance" between the instant case and each of the leading cases.
These calculations, explained below. 55 use weighted attributes.
The question of how — indeed, whether— to weight attributes to
account for their relative importance is controversial.

One approach is to ask the legal expert to quantify the weight
to be given to each attribute. For example, the developers of
LESTER 00 assign weights that, they say, "reflect the associations
and relative significance an expert would attach to particular case
features, with some adjustments we have arrived at through trial
and error." 57 But lawyers are not used to thinking in this fashion.
Ashley and Rissland write:

In the legal domain, attornej's do know what [attributes] are im-
portant in a particular legal claim. Although they may be willing
to say in the abstract that a certain [attribute] is more important
than other [attributes], they almost never will venture numerical
weights to distinguish the [attributes'] importance. 58

A pragmatic legal expert system §2.7

Lawyers are unwilling, or unable, to give numerical weights to
attributes. If attributes are to be weighted then their weights
should be determined without assistance from the legal expert.

Aldenderfer and Blashfield warn:

While the concept of weighting is simple, its practice is difficult,
and very few guidelines exist. Williams 59 describes five types of
weighting, the most common being the a priori manipulation of
[attributes]. Sneath and Sokal 00 . . . argue strongly against a pri-
ori weighting, and suggest that the appropriate way to measure
similarity is to give all [attributes] equal weight. 01

However, Everitt points out that:

. . . the consequences of choosing one or other of the plethora of
similarity indices are in many cases equivalent to the adoption of
different schemes of [attribute] weighting, and so the concept of
"equal weighting" is not as simple as it seems at first sight. It
should also be remembered that [attributes] not included in the
analysis are effectively given a zero weighting compared with those
included. 02

SHYSTER adopts the approach to attribute weighting used by
FINDER. Each attribute value of yes is assigned a value of 1;
each no is assigned a value of 0. (These values are completely
arbitrary — although they must be different — as they form the basis
of calculations to determine relative weights of attributes.) Each
attribute is assigned a weight equal to the inverse of the variance of
the numerical values of that attribute across all the leading cases.

Unknown values are ignored for this purpose. It would be in-
appropriate to give unknown a numerical value of (say) 0.5 on
the basis that it is neither yes nor no. unknown is not a halfway
point between yes and no; it simply indicates that a value is not
known.

Using the inverse of the variance is diametrically opposite to the
standard approach adopted in statistical classification problems
which deems high-variance variables to be the most important. 03
As Tyree explains:

It is not that low-variance facts are of themselves important, but
that low-variance relevant facts are more important than high-
variance relevant facts. The}' are the facts which have been in-
cluded by the expert in spite of the fact that they do not appear
to assist greatly in the separation of the cases into two classes. 64

§2. 7 A pragmatic approach to case law

The aim is to quantify the importance that the law attaches to an
attribute, not that attribute's efficiency in discriminating between
the leading cases. 60

The variance a^ of the numerical values of an attribute Aj across
n cases is defined as follows: 00

^ 2 = -E(^-^) 2

where A^ is the value of the rth attribute for the jth case, and A;
is the mean of all attribute values for the ?th attribute. Because
Aij is either or 1. a/ 2 ranges from to 0.25. Consequently an
attribute's weight (the inverse of a/ 2 ) ranges from 4 to infinity.

An attribute with all known values the same has zero variance
and is assigned infinite weight. ' This seems paradoxical, but is
actually appropriate. Consider an area of law in which a value of
yes for an attribute Ax is enormously suggestive of a result R. 6S
In an instant case in which A x = yes it may be clear that, on
the strength of A x alone, the likely result is R. Such a case may
never reach a court; a result of R may be so likely that no-one
seeks judicial determination of the instant case. So, it is possible
that A x = no in all of the leading cases in this area. If SHYSTER
is asked for its opinion on an instant case where A x = yes, it is
appropriate that SHYSTER treats that attribute value as being of
considerable importance.

HYPO does not assign weights to attributes because, according to
Ashley, the concept of an attribute's weight "though intuitively
attractive, is. on closer view, highly problematic."* ' He gives five
reasons for not giving attributes numerical weights. Three of these
can be easily countered in SHYSTER's case.

• Domain experts may not reason in terms of weighting schemes,
especially numerical ones. Attorneys generally concede that [at-
tributes] are useful in analyzing legal problems, but they rarely
are willing to apply weights or probabilities to those [attributes].

For this reason, as discussed above, SHYSTER does not require the
legal expert to weight attributes.

78 A pragmatic legal expert system §2-7

• Weighting [attributes] is not justified by any authoritative means.
Even if attorneys did assign weights to [attributes,] they would
disagree on what those weights should be. In addition, attornej's
would not actually be able to cite [attribute] weights in their
arguments to a court because weighting is not an accepted kind
of argument.

Bing is not convinced by Ashley's reference to "authoritative
means", which he sees as criticizing a weighting scheme "for not
being an objective method, or a method which is related to the
legal argument."' SHYSTER'S method is objective, though, as
Bing points out, "its relation to legal argument is less obvious." 71

But, as explained below, 72 SHYSTER's advice is evaluated by
reference to the accuracy of its predictions and the quality of its
arguments. SHYSTER does not use attribute weights in its ar-
guments, only in choosing the cases to use in those arguments.
Hence, the extent to which its weighting method relates to legal
argument is irrelevant.

Furthermore, the fact that lawyers would disagree on weights is
no reason not to weight attributes. It means that no weighting
method can be definitive; it does not mean that no weighting
method can be effective.

• Reduction to numerical weights obliterates information needed
for symbolic comparison of cases. The business of attorneys
is arguing about the competing [attributes] in the light of the
precedents. If the [attributes] are collapsed into a number, there
is nothing left to argue about.

Although this may be true given HYPO's approach to argument
(discussed in the previous chapter 73 and below) it is not true of
SHYSTER. SHYSTER does not collapse attribute information into
a number, but uses numbers to decide which cases to use in ar-
gument. The choice of cases determines the manner in which
attributes are used in argument.

Two of Ashley's reasons for not weighting attributes (both
closely related) do apply to SHYSTER:

• [An attribute's] weight is highly contextual and depends on indi-
vidual problem situations. Although an attorney may consider
one factor generally to be more important than another, she or
he is always mindful of peculiar cases where the opposite is true

§2.8 A pragmatic approach to case law

• Premature commitment to a weighting scheme may cut off fruit-
ful lines of inquiry. A rigid scheme may cause an attorney to
overlook a factor that, although not generally important, is cru-
cial in a particular situation. 74

To address these concerns, Ashley and Rissland advocate a "sym-
bolic least commitment approach" to attribute weighting, in which
weighting is postponed for as long as feasible. 70 This approach is
adopted in HYPO. As Ashley explains:

HYPO clusters the applicable [attributes] according to how they
appear in the most-on-point cases, interprets the effect of the
clustered [attributes] in the light of the most-on-point cases, and
criticizes and tests the interpretations in the light of the salient
differences among the most-on-point cases by distinguishing pre-
cedents, citing counterexamples, and posing hypothetical that
change magnitudes and combinations of [attributes] in the prob-
lem . . .

Although by the end of [this process] HYPO has not actu-
ally assigned weights to the competing [attributes], it has dealt
symbolically with the problem of weighting. It has generated
precedent-citing arguments in favor of alternative interpretations
of the weights of the [attributes] within the context of the prob-
lem. 76

SHYSTER's simple knowledge representation means that HYPO's
approach to weighting is inappropriate for SHYSTER. Neverthe-
less, there can be no doubt that, in the legal domain, some at-
tributes are of greater importance than others. For this reason,
and despite the problems identified by Ashley, SHYSTER weights
its attributes. However, it is important not to lose sight of what
Lambert and Grunewald call "the necessary arbitrariness of any
weighting function." 77

2.8 Detecting attribute dependence

A functional dependence exists between two attributes when there
is a function which maps the values of one of the attributes dir-
ectly to the values of the other. A stochastic dependence exists
when the occurrence of a given value for one attribute affects the
probability of the occurrence of a given value for another. The ex-
istence of functional or stochastic dependence between attributes
may indicate shortcomings in the legal expert's specification.

80 A pragmatic legal expert system §2.1

Consider this (extreme) example. The legal expert chooses
two attributes which, although worded differently, are identical:
i.e. they ask precisely the same question, in two different ways.
The values of these two attributes across the leading cases are
identical, and each attribute is assigned the same weight. By ef-
fectively choosing the same attribute twice, the legal expert has
given that attribute twice its appropriate weight.

In this example, there is both a functional dependence and a
stochastic dependence between the two attributes. 78 Alternatively,
if two attributes differ only very slightly in their values across the
cases then there is a stochastic dependence (though no functional
dependence) between them: i.e. the occurrence of a given value
for one attribute affects the probability of the occurrence of a
given value for the other. This stochastic dependence may indicate
that the legal expert has inadvertently chosen two very similar
attributes: attributes that ask nearly the same question. Similarly,
there is a dependence between two attributes if their values are
completely, or nearly completely, different.

For every pair of attributes there is a pair of attribute values
corresponding to each of the leading cases. SHYSTER detects at-
tribute dependence by examining the known pairs (pairs where
both the attribute values are known) for each pair of attributes in
the area. If either, or both, of a pair of attribute values is unknown
then both values are ignored. (An unknown value is not a value
that could form part of an attribute dependence; it is an absence
of known values.)

Attribute dependence does not necessarily mean that the spe-
cification must be re-written. Even if two attributes have exactly
the same values across the leading cases, the legal expert may de-
cide that they do not ask the same question: i.e. that a case can
be imagined where the values of those two attributes are differ-
ent, despite the fact that they do not differ in any of the leading
cases. 79 SHYSTER warns the legal expert of attribute dependence
so that she/he can reconsider her/his choice of attributes/

Detecting functional dependence between attributes is straightfor-
ward. There are only four functions which map an attribute A x
to an attribute A Y • These functions — the four forms of functional
dependence — are described in figure 2.2. The equivalence function
produces the same value in A Y as in A x ] the inverse function pro-
duces the opposite value; the YES function sets Ay to yes, and
the NO function sets Ay to no, regardless of the value of Ax-

§2.9 A pragmatic approach to case law 81

Equivalence function: yes •— » yes,

NO •— » NO

Inverse function: yes *-* no,

NO h- » YES

yes function: yes «-» yes,

NO i— » YES

no function: yes *-* no,

NO •— » NO

Figure 2.2: The four forms of functional dependence.

SHYSTER checks all pairs of attributes, and warns if the equi-
valence function, or the inverse function, maps the values of one
attribute to those of the other, yes functions and no functions are
ignored; each is a constant function (i.e. each produces the same
result regardless of the value it is given), hence neither amounts
to a relationship between a pair of attributes. 81

It is not possible to prove that two attributes are stochastic-
ally independent, hence it is not possible to detect stochastic
dependence with complete certainty. However, SHYSTER can de-
tect evidence of stochastic dependence, and warns the legal expert
when such evidence exists. How this is done is detailed in the next
chapter. 82

2.9 Calculating distances

SHYSTER's choice of the cases with which to construct its argu-
ments is based upon a notion of similarity between cases. SHY-
STER quantifies this similarity using distance measures: the smal-
ler the distance between two cases, the more similar they are.

SHYSTER calculates two different types of distance: the known
distance is defined as the sum of the weights of every attribute for
which those two cases have different known values; the unknown
distance is defined as the sum of the weights of every attribute for
which either of the two cases has an unknown value.

A known distance of zero indicates that the two cases are
identical — at least as far as the known attributes are concerned. A
large unknown distance indicates that the values of some import-
ant (i.e. heavily weighted) attributes are unknown, casting some
doubt on the reliability' of the known distance calculation for those
two cases. The unknown distance can be thought of as a meas-
urement of possible error: it is the maximum distance that could
be added to the known distance if all of the unknown attribute
values were known.

82 A pragmatic legal expert system §2.9

Known and unknown distances are calculated between the in-
stant case and each of the leading cases. These distance measure-
ments are treated as characteristics of each leading case. So, for
example, a statement that case j has a smaller known distance
than case k means that the known distance between the instant
case and case j is less than the known distance between the instant
case and case A;.

.Statisticians make use of many different similarity measures. How-
ever, two aspects of SHYSTER/s approach to case law mean that
each of the commonly used similarity measures reduces to one of
three measures — six, allowing for attribute weighting. This is be-
cause SHYSTER's known values are binary, and because its choice
of cases is based on the relative distance/ similarity between cases:
i.e. it has regard to the fact that case k is further from the instant
case than is case j, but not how much further.

The decision to use known and unknown distance as SHYSTER'S
similarity measure is explained below. 83 after a brief survey of the
different techniques of measuring similarity.

2. 9. 1 Similarity measures

Cluster analysis is the separation of data into groups on the basis
of similarity. Entities are grouped so that two entities in the
same group are more similar than two entities in different groups.
Similarity measures are used to quantify the similarity between
every entity and every other entity. SHYSTER does not perform
cluster analysis, but it does quantify the similarity between the
instant case and each of the leading cases. 84

Many different types of similarity measure have been developed.
The most widely used measures are called metrics. Aldenderfer
and Blashfield explain:

The quantitative estimation of similarity has been dominated by
the concept of metrics; this approach to similarity represents cases
as points in a coordinate space such that the observed similarities
and dissimilarities of the points correspond to metric distances
between them . . . The dimensionality of the space is determined
by the number of [attributes] used to describe the cases. 85

§2.9 A pragmatic approach to case law 83

For a similarity measure to be a metric it must satisfy the fol-
lowing four criteria: 80

• given two cases j and /c, dj k = d k j > 0, where dj k is the
distance between case j and case k\

• given three cases j. A; and I, dj k < d^ + d ke (this is called the
triangle inequality or the metric inequality):

• given two cases j and A;, if dj k =£ then j is not identical to k;
and

• given two identical cases j and j\ d^> = 0.

Many researchers have argued against the use of similarity meas-
ures which do not meet these criteria. Aldenderfer and Blashfield
point out that:

[Measures] that are not metrics may not be jointly monotonic;
that is, the values of different [measures] used with the same data
will not necessarily vary conjointly, raising the disturbing issue
that these [measures] could suggest quite different relationships
among the entities. 87

However, it is not essential that a similarity measure be a metric.
For example, the Pearson product-moment correlation coefficient
( u a popular similarity measure". 88 discussed below) is not a metric.

There are four different kinds of similarity measure: distance
measures, association coefficients, correlation coefficients, and
probabilistic similarity coefficients.

Strictly speaking, probabilistic similarity coefficients do not ac-
tually calculate the similarity between two cases. They take into
account the distribution of the frequencies of the attribute val-
ues over all the cases, and are calculated during the formation
of clusters. 89 Probabilistic similarity coefficients are inappropriate
for use with SHYSTER because it does not perform cluster analysis.

The three kinds of similarity measure appropriate for SHYSTER
are distance measures, association coefficients, and correlation
coefficients.

84 A pragmatic legal expert system §2-9

Distance measures

Two popular distance measures are Euclidean distance and Man-
hattan distance. The Euclidean distance between two cases j and
k is defined as

\ i=l

where Aij is the value of the ith attribute for the jth case, 90 and n
is the number of attributes. 91 The Manhattan distance is defined

■Aij Am

Both Euclidean distance and Manhattan distance are specific
examples of the class of metric distance functions known as Min-
kowski metrics. The Minkowski metric is defined as

where K is some constant: for Euclidean distance, K = 2; for
Manhattan distance, K = 1. In the binary case (e.g. SHYSTER,
where each attribute value of yes is assigned a value of 1. and each
no is assigned a value of 0),

I . . \k f 0. if A iA = A 4

Hence, for binary attribute values, the Minkowski metric reduces
to

&jk

where A^* is the number of differences in the corresponding at-
tribute values of case j and case k.

For increasing values of A jft , this metric always increases regard-
less of the value of K . As SHYSTER is concerned only with the
relative distance between cases, it does not matter which value of
K is used. Choosing K = 1 gives a distance measure dj^'-

djk = Ajk-

This distance measure is a metric as it satisfies all four metric
criteria. It ranges from to ?i: the smaller the value of djk, the
nearer case j is to case k.

§2.9 A pragmatic approach to case law 85

case k

YES NO

YES

a b

c d

a is the number of yes/yes pairs;
b is the number of yes/no pairs; etc.

Figure 2.3: A two-way association table defining a, 6, c and d.

Association coefficients

Association coefficients are used to describe similarity between
cases with binary attributes. These coefficients are usually ex-
pressed in terms of a, b, c and d. The two-way association table
in figure 2.3 defines these variables.

Many different association coefficients have been proposed—
according to Aldenderfer and Blashfield. there are more than
thirty. 92 Everitt says that the reason for this proliferation of coef-
ficients is:

. . . uncertainty over how to incorporate negative matches [d] into
the coefficients, and also whether or not matched pairs of [attrib-
utes] are equally weighted, or carry twice the weight of unmatched
pairs, or unmatched pairs carry twice the weight of matched
pairs. 93

Everitt gives several examples of association coefficients. Jac-
card's coefficient.

a + b+ c'

is one measure of the association between two cases j and k. It
ignores d — the number of no/no pairs. This coefficient was de-
veloped for use in applications where it would be inappropriate to
treat two cases as being similar for lacking the same features as
well as sharing the same features. 94 Similarly,

and

2a + b + c a + 2(b+c)

ignore negative matches — and give double weight to matched pairs
and unmatched pairs, respectively. And

a+b+ c+ d

i^ivcs no positive significance to negative matches.

A pragmatic legal expert system §2.9

Coefficients such as these are inappropriate for use by SHYSTER
because the occurrence of no/no pairs is as important as the oc-
currence of yes/yes pairs. An attribute value of no does not
indicate the lack of a feature; it means that the answer to the at-
tribute's question is "no." Simply rephrasing SHYSTER'S attribute
questions could turn all yes/yes pairs into no/no pairs.

The other Wo example coefficients that Everitt gives are

a+d 2(a+d)

and

a + b+c+ cl 2(a + d) + b + c

The first, called the simple matching coefficient, is the number of
matching pairs as a fraction of the total number of pairs. The
second gives matching pairs twice the weight of non-matching
pairs. Coefficients such as these are appropriate for use by SHY-
STER because they do not distinguish between yes/yes pairs and
no/no pairs.

As Aj k = b + c and the number of attributes n = a + b+ c+ d,
coefficients of this kind can be generalized to

K(n-A jk )
K(n-A jk ) + A jk

where A' is some constant. For increasing values of Aj*, this coef-
ficient always decreases regardless of the value of K. Because
SHYSTER is concerned only with relative measures of similarity, it
does not matter which value of K is used. The simplest is K = 1,
giving

n-A jh
n

which ranges from to 1. Subtracting it from 1 yields an in-
versely proportional association coefficient Sj k which satisfies all
four metric criteria:

Sjk —

\?*

This association coefficient also ranges from to 1: the smaller
the value of S^.. the nearer case j is to case k. 95

§2.9 A pragmatic approach to case law 87

Correlation coefficients

The most popular correlation coefficient is Pearson's product-
moment correlation coefficient. 90 This coefficient is defined as

E 4-^ )[A ik -A

where n is the number of attributes. Aij is the value of the ith
attribute for the jth case, and Aj is the mean of all attribute
values for the jth case.

In general, this correlation coefficient has several limitations. 9 '
It is not a true metric. 98 Furthermore, as Aldenderfer and Blash-
field explain:

. . . the use of [this] method to calculate the correlation of cases
does not make statistical sense, because one must obtain the mean
value across different [attribute] types rather than across cases, as
in the standard use of the method. The meaning of the "mean"
across these [attributes] is far from clear. 99

Although this is true in general, in SHYSTER all attributes are of
the same type. Hence it does make sense to calculate the mean
across the attributes.

Despite its drawbacks, this coefficient has been widely used. It
ranges from —1 to +1: 100 the larger the value of r^., the nearer
case j is to case k.

2.9.2 Weighted similarity measures

Each of the three similarity' measures discussed above can be
weighted to take account of the importance of each attribute.
The weighted distance measure ct- k is

<-* = EK- A J

X W;

i=l

where Wi is the weight of the zth attribute. It ranges from
to ^22-i w * : the smaller the value of c£- k , the nearer case j is to
case k.

88 A pragmatic legal expert system §2-9

The weighted association coefficient S'- k is

n I j

E \Aij - A ik \ x Wi

2=1

It ranges from to 1: the smaller the value of Sj ki the nearer case
j is to case k.

The weighted correlation coefficient r r - k is

where A!- is the weighted mean of all attribute values for the jth
case. It ranges from —1 to +1: the larger the value of r'- k , the
nearer case j is to case k.

2.9.3 Choosing a similarity measure

For SHYSTER — and allowing the weighting of attributes — each of
the commonly used similarity measures reduces to one of six meas-
ures: dj k , d r . k i Sjk, Sj k , Tj k or r'- A . Mackaay and Robillard use dj k
as their similarity measure; FINDER uses d'- k .

According to Ashley, HYPO uses four comparison metrics. 101
The basic measure is on pointness: the degree of overlap of di-
mensions shared by the instant case and a given case, relative
to that of other cases. HYPO also has regard to the outcome of a
leading case (because "depending on the procedural context of the
case, some outcomes are more determinative than others" 102 ), the
magnitude of shared dimensions, and a case's potential relevance
as a near miss (i.e. whether a small hypothetical change would
make that near miss case more on point).

Strictly speaking. HYPO uses only one similarity measure: on-
pointness. The outcome of a case and the magnitude of shared
dimensions are weighting considerations; 103 making hypothetical
changes to a case is an argument technique. None of these three
is a similarity measure in the sense that that term is used in this
book.

§2.10 A pragmatic approach to case law

When choosing an appropriate similarity measure for a particular
set of data, it is important to have regard to the sort of data,
and to what is to be measured. Because SHYSTER treats cases as
points in n-dimensional space, where n is the number of attrib-
utes, a metric should be an appropriate similarity measure. As
discussed above, 104 attribute weighting is also appropriate. SHY-
STER'S known and unknown distance measures are variations on
the weighted distance measure d- k which is a metric.

However, it is important to note Sneath and Sokal's warning
that "when all is said and done, the validation of a similarity
measure ... in a given field has so far been primarily empirical". 105
The testing of SHYSTER is described in chapter 4. In order to
compare empirically the various similarity measures, SHYSTER
calculates (in addition to the known and unknown distance) values
of djk, Sjk, S r - k , Tju and r'- h . These extra measures are also used
as safeguards. 100

2.9.4 Infinite distance

As explained above, 10 ' each attribute's weight ranges from 4 to
infinity. In determining a distance — known or unknown — SHY-
STER may have to deal with one or more attributes with infinite
weight.

To handle such possibilities. SHYSTER's distances have an infin-
ite and a finite component. The infinite component is the number
of infinitely weighted attributes that differ between the two cases.
A distance of "2oo + x" is considered to be greater than "oo + y" ,
regardless of the values of x and y (the finite components). The
infinite component of this first distance should not be thought of
as being the sum of two infinities. Rather, it represents the effect,
upon the distance between two cases, of an infinite weight in each
of two dimensions in n-dimensional space.

2.10 Nearest cases and nearest results

Having determined the distance between the instant case and each
of the leading cases, SHYSTER can decide which of the leading
cases is nearest to the instant case. Having chosen a nearest case,
SHYSTER can decide upon the most likely result: the result in the
nearest case.

A pragmatic legal expert system §2.10

In fact, SHYSTER deals with several nearest cases and can also
deal with two or more cases which are equidistant from the instant

case.

2.10.1 Nearest cases

The nearest known neighbour is the case with the smallest known
distance, and with no unknown distance (i.e. with an unknown
distance of zero). The nearest unknown neighbour is the case
which has the smallest sum of known and unknown distances, and
non-zero unknown distance. The nearest of these — the case with
the smallest sum of known and unknown distances — is called the
nearest neighbour.

Consider the example distances in figure 2.4. taken from one
of the tests performed in chapter 4. 108 There are fourteen cases:
C\ . . . C14. Each case has a known and an unknown distance. Zero
distances are indicated by "-". The nearest known neighbour (and
the nearest neighbour) is C 5 ; the nearest unknown neighbour is C 3 .

2.10.2 Nearest results

The result of the nearest neighbour is termed the nearest result
Applying the doctrine of precedent, SHYSTER assumes that the
decision in the instant case will be the same as that in the nearest
neighbour. SHYSTER also finds the nearest known and nearest
unknown neighbours for every other result — i.e. from amongst the
cases in which another result was reached. These are termed the
nearest known other and the nearest unknown other. The nearest
of these cases is called the nearest other: there is a nearest other
for each other result.

The example cases whose distances are given in figure 2.4 are
grouped by result: the first seven cases have one result, the second
seven have another. 109 The nearest result is the result of the
nearest neighbour: C 5 . The nearest known other (and the nearest
other) is C 12 ; the nearest unknown other is C 13 .

2.10.3 Equidistance

In each comparison made during these classifications of cases and
results it is possible that the distances being compared are equal.
If tw T o distances are the same then their cases (or results, as ap-
plicable) are said to be equidistant, because they are equidistant
from the instant case.

§2.10

A pragmatic approach to case law

Case

Known

distance

Unknown
distance

C 2

c 3

Cr,

c 6
c 7

43.57
36.45
17.76
42.85
17.09
14.92
59.32

5.63
4.02

62.65

C 8

c 9

Cio
Cn

^- / 12
Cl8

C14

54.48
71.79
55.74
55.21
46.13
28.54
49.58

54.97

Figure 2.4: Distances for fourteen example cases. The cases
are grouped by result: the first seven have one result, the second
seven have another. The nearest known neighbour (and the
nearest neighbour) is C$; the nearest unknown neighbour is C3.
The nearest known other (and the nearest other) is C12; the
nearest unknown other is C13.

SHYSTER allows any number of equidistant cases in all of the
categorizations explained above. Equidistant cases are used in
argument (one after another, in order of importance) wherever
a single case would be used. SHYSTER also allows equidistant
results — results whose nearest neighbours are equidistant — but
only for other results; it allows only one nearest result so that
the open-textured concept that the area represents is defined and
an answer is returned by the case-based system for use in other
areas of case law. or by a rule-based system.

Mackaay and Robillard consider the problem of equidistance and
propose two solutions. 110 The nearest result can be that which has
the greatest number of equidistant cases; if there is no majority,
then no prediction can be made. Alternatively, the set of nearest
neighbours can be repeatedly extended to include the next nearest
neighbour(s) until a majority is achieved. This approach is not

92 A pragmatic legal expert system §2.1 1

adopted for use with SHYSTER. If there is a great distance between
the nearest neighbours and the next nearest neighbours, it is in-
appropriate to use those next nearest neighbours to resolve the
problem.

SHYSTER chooses between equidistant results by reference to the
relative importance of the courts in which the equidistant cases
were decided and (if further resolution is required) the year in
which those cases were decided (more recently decided cases are
considered to be more important). If the equidistance remains
unresolved. SHYSTER refuses to venture an opinion, writes an er-
ror message and stops. I had originally intended to add further
comparisons, resolving equidistance by reference (for example) to
the nearest ideal points, or to attribute direction. However, test-
ing of SHYSTER indicates that the two stage approach adopted is
adequate. The utility of going further is doubtful; to make further
distinctions would be to split a very fine hair.

2.11 Writing a report

SHYSTER constructs a report — a legal opinion — about the instant
case. This opinion includes a statement as to the likely result in the
instant case, and justification of that statement. This justification
is crucially important. As Tyree, Greenleaf and Mowbray point
out:

. . . the justification in a legal system is the main product, for
the justification is no more and no less than the legal arguments
which support the suggested outcome. It is in the nature of legal
reasoning that these arguments must also address the support for
the opposite outcome. It is these arguments which, if the matter
goes to court, must be presented for adjudication. 111

SHYSTER's approach to report generation is based on that of
FINDER, but SHYSTER takes account of unknown distance and
equidistance, and argues about the effect upon its argument of
hypothetical changes to the attribute values of the instant case.
This last feature allows SHYSTER to adopt, to some extent, aspects
of HYPO's approach to arguing with hypotheticals. HYPO:

Summarizes the cases that can be cited in favor of a position,
characterizes how strong^ they support the position, focuses the
attorney's attention on the most significant cases and hypothet-
icals, . . . and facilitates comparing arguments between cases and
hypotheticals . . . U2

§2.11 A pragmatic approach to case law 93

So too does SHYSTER, although the sophistication of its reporting
is limited by what is SHYSTER'S main advantage: the simplicity
of its knowledge representation.

2.11.1 Arguing with the instant case

SHYSTER opens its report with some introductory comments
about the area of law. then boldly declares its opinion 113 that the
result in the instant case will be the nearest result. It then uses the
nearest neighbour and the nearest others to justify that statement,
before closing with some concluding remarks about the area. The
opening and closing remarks are general comments provided by
the legal expert.

How SHYSTER chooses the cases to use in argument, and how
those cases are used, is described below.

Choosing cases

Figure 2.5 gives a pseudo-code description of SHYSTER's algorithm
for choosing leading cases to use in argument, and the order in
which to use those cases. SHYSTER will always use the nearest
known neighbour in argument. It will use the nearest unknown
neighbour in two circumstances: if it is the nearest neighbour, or
if it would be the nearest neighbour but for its unknown distance.
As explained above, 114 the unknown distance can be thought of
as a measurement of error: the maximum distance that could be
added to the known distance if all of the unknown attribute values
were known. Whichever of the nearest known neighbour and the
nearest unknown neighbour is the nearer is used in argument first.
A slightly different approach is used with the nearest others. For
every other result SHYSTER uses the nearest known other in argu-
ment. The nearest unknown other is used in two circumstances:
if it is the nearest other, or if it would be nearer the instant case
than the nearest neighbour — not just the nearest other — were it
not for unknown distance. The nearest unknown other is used
in this second circumstance because, if all of the unknown attrib-
ute values w r ere known, the nearest unknown other could be the
nearest neighbour and SHYSTER's opinion as to the likely result
would be different. The nearest unknown other is only used in
argument in either of these two circumstances; if it is used at all,
it is used before the nearest known other.

A pragmatic legal expert system

§2.11

FOR the nearest result DO

IF the nearest known neighbour is
the nearest neighbour THEN
use the nearest known neighbour;
IF (were it not for unknown distance) the nearest
unknown neighbour would be nearer the
instant case than the nearest neighbour THEN
use the nearest unknown neighbour;
END
ELSE

use the nearest unknown neighbour;
use the nearest known neighbour;
END
END

FOR every other result DO

IF the nearest unknown other is the nearest other OR
(were it not for unknown distance) the nearest
unknown other would be nearer the instant
case than the nearest neighbour THEN
use the nearest unknown other;
END

use the nearest known other;
END

Figure 2.5: SHYSTER'S algorithm for choosing the cases upon
which to base its opinion.

The nearest known neighbour is the case with the smallest
known distance, and with no unknown distance; the nearest
unknown neighbour is the case which has the smallest sum of
known and unknown distances, and non-zero unknown distance.
The nearest neighbour is the case with the smallest sum of
known and unknown distances. The nearest result is the res-
ult of the nearest neighbour. Similarly, for every other result
there is a nearest known other and a nearest unknown other;
the nearest of these is the nearest other.

The steps marked [Tl are refined in figure 2.6; [b] in figure 2.7;
|~cj in figure 2.8. For simplicity, this description and the descrip-
tions in these other figures assume no equidistance.

§2.11 A pragmatic approach to case law 95

This description is simplified in one respect: it assumes that
there is only one nearest known case and one nearest unknown case
for each result. If there are two or more equidistant cases. SHY-
STER uses each case — one after another, in order of importance.

Using cases

How each leading case is used in argument varies depending on
several factors. Figures 2.6. 2.7 and 2.8 give pseudo-code descrip-
tions of SHYSTER's algorithm for using cases. Each description
is a refinement of some steps in the algorithm for choosing cases
described in figure 2.5. Words within quotation marks are para-
phrasings of words that SHYSTER uses in its opinion.

Each case is summarized, then the similarities and differences
between the case and the instant case are explained. If the case
has unknown attribute values, those attributes are also detailed.

For the nearest result, it is argued that (because of the simil-
arities, and despite the differences) the result in the instant case
should be the same. For every other result, it is argued that (be-
cause of the differences, and despite the similarities) the result in
the instant case should be different.

If SHYSTER's opinion is a desirable result for the user, she/he
can use SHYSTER'S discussion of the nearest case, and the differ-
ences between the nearest case and the instant case, as the basis
for a legal argument. Alternatively, if SHYSTER's opinion is not
a desirable result for the user, she/he can base a legal argument
upon SHYSTER's discussion about the nearest others.

SHYSTER's reports make no reference to "nearest cases," "nearest
others," etc. No weights or distance measures are included or
discussed. The calculations which SHYSTER uses to reach its con-
clusions are not part of its reports, although they are written to
various intermediate files. Each report refers only to the simil-
arities and differences between the instant case and the leading
cases.

A pragmatic legal expert system §2.1 1

IF (were it not for unknown distance) the nearest
unknown neighbour would be nearer the
instant case than the nearest neighbour THEN
predict that the result will be the nearest result,
citing the nearest known neighbour and
the nearest unknown neighbour:
ELSE

predict that the result will be the nearest result,
citing just the nearest known neighbour;
END

FOR the nearest known case DO
summarize the case;
IF there is no distance between the

case and the instant case THEN
"the two cases are identical";
ELSE

list the similarities between the two cases;
list the differences between the two cases;
"nevertheless, the case should still be followed":
END
END

IF (were it not for unknown distance) the nearest
unknown neighbour would be nearer the
instant case than the nearest neighbour THEN
cite the nearest unknown neighbour as another

case in which the nearest result was reached;
FOR the nearest unknown neighbour DO
summarize the case;

IF there is some known distance between
the case and the instant case THEN
list the similarities between the two cases;
list the differences between the two cases:
ELSE

"the two cases may be identical, and . . .";
END
"I would have suggested, that this case be followed

instead of the nearest neighbour except that . . .";
list the unknown attributes:
END
END

Figure 2.6: Part of SHYSTER/s algorithm for using cases in
argument: a refinement of the steps marked [a~[ in the algorithm
described in figure 2.5.

§2.11 A pragmatic approach to case law

predict that the result will be the nearest result, citing the nearest
unknown neighbour and the nearest known neighbour;

FOR the nearest unknown neighbour DO

summarize the case;

list the similarities between the case and the instant case;

list the differences between the two cases;

list the unknown attributes;

"nevertheless, the case should still be followed";
END

FOR the nearest known neighbour DO

summarize the case;

list the similarities between the case and the instant case;

list the differences between the two cases;

"nevertheless, the case should still be followed";
END

Figure 2.7: Part of SHYSTER'S algorithm for using cases in

argument: a refinement of the steps marked B in the algorithm
described in figure 2.5.

Both FINDER and HYPO report on the similarities and differ-
ences in this fashion. However, as Ashley concedes:

We expect more from law students' explanations . . .

(and, presumably, from lawyers* explanations)

... of why a precedent should or should not be followed than a
discussion of the superficial, factual similarities and differences as-
sociated with factors. We expect their explanations to invoke some
principled analysis of why the similarities and differences matter
and to structure their explanations to reflect the relevant statutes
and court-made rules. Thus, at first glance, HYPO's approach
to identifying important similarities and differences may not seem
philosophically satisfying . . . 115

The same criticism applies to SHYSTER, but is countered on the
pragmatic grounds that the aim is to produce a working expert
system whose simple model of legal reasoning obviates the problem

A pragmatic legal expert system §2.1 1

IF the nearest unknown other is the nearest other OR (were it not
for unknown distance) the nearest unknown other would be
nearer the instant case than the nearest neighbour THEN
"if the nearest unknown other and the nearest known other

are followed then the result will be this (other) result";
IF (were it not for unknown distance) the nearest
unknown other would be nearer the instant
case than the nearest neighbour THEN
FOR the nearest unknown other DO
summarize the case;

IF there is some known distance between
the case and the instant case THEN
list the similarities between the two cases;
list the differences between the two cases;
ELSE

"the two cases may be identical, and . . ." ;
END
"I would have suggested, that this case be followed

instead of the nearest neighbour except that . . .";
list the unknown attributes;

"nothing in the case warrants changing the prediction";
END
ELSE

FOR the nearest unknown other DO
summarize the case;

list the similarities between the case and the instant case;
list the differences between the two cases;
list the unknown attributes;

"nothing in the case warrants changing the prediction";
END
END
ELSE

"if the nearest known other is followed then
the result will be this (other) result";
END

FOR the nearest known other DO

summarize the case;

list the similarities between the case and the instant case;

list the differences between the two cases;

"nothing in the case warrants changing the prediction";
END

Figure 2.8: Part of SHYSTER'S algorithm for using cases in
argument: a refinement of the steps marked fcl in the algorithm
described in figure 2.5.

§2.11 A pragmatic approach to case law

of knowledge acquisition. Similarly, Ashley makes a pragmatic
argument in defence of HYPO:

Undoubtedly fundamental legal principles play a role in legal ana-
logical reasoning . . . But one cannot hope to model that kind
of adversarial reasoning until one understands the simpler, more
factual analogical comparisons among precedents . . . 11G

2.11.2 Arguing with instantiations

As well as using unknown distance, SHYSTER has a second method
of taking unknown attribute values into account.

SHYSTER instantiates the unknown attribute values in the in-
stant case to create instantiations of the instant case in which all
the attribute values are known. SHYSTER treats each instantiation
as if it were a new instant case, and determines the nearest cases
and nearest results.

Because there are only two known attribute values (yes and
no) there are 2" different instantiations, where n is the number of
unknowns in the instant case. To avoid writing unnecessarily long
reports, SHYSTER only reports on an instantiation if its nearest
result is different to that of the instant case.

Instantiation is a useful feature. If the user is unable to answer
an attribute question, she/he simply answers unknown. If hav-
ing known values for the instant case's unknown attribute values
would make a difference to the result, the relevant instantiations
are reported on in full. 117 This feature can also be used where the
user knows the answer to an attribute question, but wants to test
the effect of providing a different answer. 118

Instantiation is used as a safeguard against giving erroneous
advice. 119 It is also used in chapter 4 to perform generated tests
(which are described below 120 ).

2.11.3 Arguing with hypotheticals

HYPO generates hypotheticals: hypothetical variations of the in-
stant case that are stronger or weaker for a particular side. There
are five heuristics for modifying the instant case: make a near-miss
dimension apply; strengthen or weaken a case along an applicable
dimension; move a case along a related dimension: make a case

100 A pragmatic legal expert system §2.1 1

extreme along a dimension; and make a case into a near-win given
a target. 121 Ashley identifies several uses for hypotheticals in legal
argument: to factor a complex situation into component parts
(e.g. by exaggerating strengths, weaknesses, or by hypothetically
eliminating features); to create a test case that puts an issue or
pits competing attributes against each other; to present, support,
and attack positions in an argument (e.g. by testing consequences
of a tentative conclusion); etc. 122

Ashley describes hypothetical reasoning as the key to "exploring
the dialectics between cases and principles or between cases and
rules." 123 Whether or not it is appropriate to think of cases as
rule exemplars (as Ashley seems to). SHYSTER's simplified repres-
entation of case law does not allow it to reason with hypotheticals
to the same extent as does HYPO. However, SHYSTER does exam-
ine hypothetical variations in order to alert the user to the effect
of such variations. Showing how a case can be strengthened and
weakened can be particularly useful where there is some uncer-
tainty surrounding one or more "known" attribute values.

For SHYSTER, the number of possible hypothetical variations
upon the instant case is 2" — 1. where n is the number of known
attribute values. 124 Generating all the possible hypothetical vari-
ations is of little use, as well as being computationally intensive.
Instead, SHYSTER examines all possible variations which can be
achieved by making no more than a certain number of changes
to the known attribute values in the instant case; that number is
specified by the user. 125

As with instantiations, SHYSTER treats each hypothetical as if
it were a new instant case, and determines the nearest cases and
nearest results. A hypothetical is considered eligible to be reported
on if its nearest result is different to that of the instant case, or if
it has the same nearest result but its nearest neighbour is nearer
the instant case than is that of the instant case. Of these eligible
hypotheticals, only the nearest are chosen to be reported on — up
to a certain (user-specified) number for each result.

The hypothetical reports give the user information about how
the argument about the instant case can be strengthened and
weakened by changing only a specified number of attributes. If
the user wishes to examine the effect of varying a specific attrib-
ute or attributes, she/he can use SHYSTER's instantiation feature.

§2.12 A pragmatic approach to case law 101

2.12 Safeguards

SHYSTER employs several safeguards so that it can warn the user
in situations where its advice may be suspect. In certain circum-
stances, the results of employing these safeguards are logged, for
the user, in a log file. Only in some of these circumstances is a
warning issued.

SHYSTER's safeguards are based on the extra similarity meas-
ures, ideal points specified by the legal expert, centroids (described
below 120 ), attribute direction, equidistance and instantiations.

2.12.1 Extra similarity measures

As well as calculating known and unknown distance, SHYSTER
also determines the similarity between the instant case and each
of the leading cases using the extra similarity measures discussed
above: 12, viz. djk, Sjk, S'- k , Tju and r'- k . If any of these measures
suggest different nearest neighbours or nearest others then those
differences are logged. In this way, cases which might be useful to
the user's argument, but are not mentioned in SHYSTER's report,
are brought to the user's attention.

A warning is issued if either of the extra weighted measures
(the weighted association coefficient or the weighted correlation
coefficient) suggests that a case with a different result to that
of the nearest neighbour ought to be the nearest neighbour. This
may be cause for concern about SHYSTER's opinion as to the likely
result.

A different result suggested by one of the unweighted measures
is not considered important enough to warn the user about, on the
basis that attribute weighting is essential in the legal domain. 128

2.12.2 Ideal points

As discussed above, 129 the legal expert may specify ideal points.
SHYSTER treats each ideal point as if it were one of the leading
cases, and determines the distance between it and the instant case.
It is reasonable to expect that the result in the nearest ideal
point will be the same as that of the nearest neighbour; after
all. the nearest ideal point represents the ideal combination of

102 A pragmatic legal expert system §2-12

attribute values for its result. If the nearest ideal point has a
different result to that of the nearest neighbour, this fact is logged.
More serious would be a situation where an ideal point with
a different result is at least as near to the instant case as is the
nearest neighbour. If this occurs, a warning is issued.

2.12.3 Centroids

For each result. SHYSTER calculates the mean of each attribute
for the cases with that result and creates a vector of the average
attribute values for that result. 130 These vectors of attribute values
are called centroids. SHYSTER treats each centroid as if it were
one of the leading cases, and determines the distance between it
and the instant case.

As with ideal points, it is reasonable to expect that the result
in the nearest centroid will be the same as that of the nearest
neighbour. The nearest centroid represents the average combine
tion of attribute values for its result. If the nearest centroid has
a different result to that of the nearest neighbour then this fact
is logged because, as Tyree et al. explain, the instant case "is, in
some sense, near the common boundary of the [result] groups and
so must be considered as a 'difficult 1 case." 131 However, a warning
is not issued unless a centroid with a different result is at least as
near to the instant case as is the nearest neighbour.

2.12.4 Attribute direction

As mentioned above, 132 SHYSTER allows the legal expert to spe-
cify attribute directions. These indicate that the occurrence of a
certain value for a certain attribute suggests a certain result or
results.

For each result, SHYSTER sums the weights of each attribute
for which the value of that attribute in the instant case is direc-
ted towards that result. 133 This sum is termed a direction. The
larger — the stronger — a direction, the greater the extent to which
the attribute values in the instant case "direct" SHYSTER towards
that result.

There are three types of attribute direction for each result. The
specified direction is calculated using the legal expert's attribute
direction in the case law specification. The ideal point direction
is calculated using ideal points; if an ideal point is the only ideal
point in the area with a given value for an attribute then that value

§2.13 A pragmatic approach to case law 103

for that attribute is considered to be directed towards the ideal
point's result. By an analogous method, the centroid direction is
calculated using each result's centroid.

For each type of direction, the result with the strongest dir-
ection is said to be suggested by that direction. If any of these
three directions suggest a result different to the nearest result, that
fact is logged. Only if the specified direction suggests a different
result is a warning issued. This distinction between directions is
made because the specified direction, based as it is on information
provided by the legal expert, is less likely than the other directions
to suggest a different result anomalously.

2.12.5 Equidistance

As explained above, 134 SHYSTER chooses between equidistant res-
ults by reference to the relative importance of the equidistant cases
and the year in which those cases were decided. If equidistance
has to be resolved in such a fashion. SHYSTER issues a warning;
equidistant results cast doubt on SHYSTER'S choice of nearest res-
ult.

2.12.6 Instantiations

As explained above, 135 SHYSTER creates instantiations of the in-
stant case in which all attribute values are known, and treats each
as if it were a new instant case.

SHYSTER reports on any of these instantiations of the instant
case which have a different result to that of the uninstantiated
instant case. In such an event, doubt is cast on SHYSTER's con-
clusion; there is a combination of known attribute values which
is consistent with the instant case but which leads SHYSTER to a
different result, so a warning is issued.

2.13 Testing and evaluation

SHYSTER's general structure allows the testing of SHYSTER's ap-
proach to case law in different areas of law. Testing, using four
different case law specifications, is described in chapter 4. The
choice of test domains — deciding which areas of law to specify for
testing purposes — is discussed below.

The principal testing method involves specifying an area of law
for SHYSTER, then giving it information about a case which was

104 A pragmatic legal expert system §2-13

actually decided in that area but which is not one of the leading
cases in the specification. Unfortunately there is a paucity of such
test cases. 130 Even if the specification is written so as to represent
the law as it was a few years ago, the number of cases decided
since then (all potential test cases) is small.

So other testing methods are also employed. Generated testing
uses SHYSTER'S ability to generate instantiations to create many
different imaginary cases. Reflexive testing involves removing a
leading case and testing it using the case base from which it was
removed. (Cases used with the principal testing method are re-
ferred to as "test cases," to distinguish them from those used in
generated tests or reflexive tests.)

For simplicity, the tests described in chapter 4 are often referred
to as tests of their specification. Of course, those tests are actually
testing SHYSTER, using that specification. The quality of SHY-
STER'S advice is a function of both its approach to case law and
of the quality of the specification. 137 Methods of evaluating the
quality of SHYSTER's advice are examined below. 138

2.13.1 Choosing a test domain

Susskind suggests that (insofar as any area of the law is self-
contained) an area of law chosen as a test domain should be
relatively autonomous: its sources should be limited in number
and reasonably well defined. It should be small enough to al-
low extensive coverage and its problems should not require the
use of "a great deal of 'common-sense' knowledge": 139 not be-
cause reasoning in the law r doesn't require common sense — on
the contrary — but because artificial intelligence techniques cannot
cope satisfactorily with common sense. Given Susskind's consen-
sual approach to jurisprudence, it is not surprising that he also
suggests that there should be agreement amongst experts as to
the scope and content of the test domain. 140

Furthermore, he says, the domain must be one in which problem
solving requires expertise. He is critical of Leith's ELI system, 141
which deals with British welfare rights — a domain which was:

. . . chosen because of its simplicity which allowed Leith (not him-
self trained in law) to "become 'expert' in it." The system was
not constructed, therefore, with the assistance of a legal expert,
there could not have been any inclusion of experts' heuristics, and
this factor might incline us to doubt whether the designation "ex-
pert system" is appropriate. Moreover, if a non-lawyer could, in

§2.13 A pragmatic approach to case law 105

a fairly short period, develop expertise in an area of law, then
we might justifiably query whether that chosen area is indeed a
suitable domain of application. For the chosen legal domain ought
to be one whose problems do indeed require expertise (normally
acquired over many years), and not relatively brief research, for
their resolution. 142

As Clark comments, SusskintTs criteria of suitability "would seem
to narrow the scope of expert systems in law quite considerably." 143

The four specifications used to test SHYSTER are discussed in
chapter 4. 144 Each is quite different from the others. The Finder
specification is a simulation of the FINDER system, and deals
with a completely case-based area of law. The Authorization
specification deals with the definition of a specific open-textured
statutory concept, and is an example of an area of law in which
there are more than two possible results. The Employee specifica-
tion defines an open-textured concept which is common to several
different statutes and areas of case law.

Gardner justifies her choice of an area of the law on the grounds
that it is "relatively well developed and stable." 145 Although that
is true of the first three SHYSTER specifications, it is not true of the
fourth: the Natural specification deals with a recently developed
(and still developing) area of Australian administrative law.

These four specifications satisfy most of Susskind's criteria.
They are all specified in terms of a limited number of well-defined
cases. Each domain is fairly small, yet sufficiently complex that
solving problems requires legal expertise. However, no claim is
made that any of these specifications embodies a legal consensus
as to what the law is. Apart from the Finder specification, I de-
veloped each specification myself, with the assistance of an expert
in the relevant field. Accordingly, each specification represents one
interpretation of its field.

2.13.2 Evaluating SHYSTER 's opinion

There are a number of levels at which SHYSTER'S opinion can be
evaluated. It states its prediction as to the likely outcome, on
the assumption that the result will be the same as it was in the
case which it deems most similar to the instant case. SHYSTER's
prediction of the likely result is considered "good" if it is the same
as the result in the actual case, and "bad" otherwise. 146

106 A pragmatic legal expert system §2.13

In their written judgments, judges will often explicitly follow, or
refer favourably, to certain cases before coming to their conclusion.
Sometimes they will explicitly refuse to follow a case on the basis
that it is distinguishable on its facts. Some reported judgments
also include precis of the arguments put to the judges by counsel
for the parties involved.

SHYSTER's opinion is only as good as the cases it chooses upon
which to base its arguments. If a case which SHYSTER chooses is
referred to by a judge, or cited in argument before the court, then
it is considered a "good" case for SHYSTER to have chosen. This is
true for those cases upon which SHYSTER bases its arguments (its
nearest neighbours) and its counterarguments (its nearest others).
SHYSTER is proposing these cases as the best arguments for each
of the possible results.

For example, a judge may explicitly follow a case in coming
to her/his conclusion. If SHYSTER chooses that same case as the
basis of its counterargument then it will come to a bad conclusion,
by following another case. However, it has identified a good case
upon which to base an argument for one party; its conclusion is
bad, but at least one of its chosen cases is good.

Generally, SHYSTER's choice of a case is considered "bad" if
that case was neither cited in forensic argument nor in judgment.
In some circumstances, however, I argue that SHYSTER's choice
of a case is good even though it was not used by counsel or by the
court. All such divergences from the general rule are discussed in
detail in chapter 4.

Sometimes the case which is used as a test was decided before
one of the leading cases that SHYSTER chooses. If this happens.
SHYSTER's choice is only characterized as "good" or "bad" on the
advice of the legal expert.

As discussed above, 147 SHYSTER employs several safeguards to
protect against giving bad advice. If SHYSTER issues a warning
and SHYSTER's prediction as to the likely result is bad. that warn-
ing is characterized as "good" ; if the predicted result is good, the
absence of a warning is also considered to be "good." If SHYSTER
issues a warning even though SHYSTER's prediction is good, the
warning is "bad" ; similarly if the prediction is bad and SHYSTER
issues no warning, the absence of a warning is deemed "bad."

§2.13 A pragmatic approach to case law 101

When lawyers use a case in argument, it may be that only one
aspect of that case is considered. This is true of the test cases,
and of the leading cases that make up the test specifications. For
example. Salemi v. MacKellar [No. 2] l4S is used as a test of the
Natural specification because the High Court had to determine
whether Salemi had a right to be heard before the Minister ordered
his deportation. In that case, the Court also considered the issue
of whether news releases were "instruments under the hand of the
Minister." But that issue is ignored in the discussion of Salemi v.
MacKellar in chapter 4. It is not relevant to the specification
that the case is used to test, and is not taken into account when
evaluating SHYSTER's opinion.

2.13.3 The paucity of test cases

One of the major obstacles to developing 149 and testing a case-
based legal expert system like SHYSTER is the paucity of reported
cases. This can be attributed to the filtering effect of various stages
of the legal system.

Consider a client who seeks legal advice as to her/his legal op-
tions in some matter. Good legal advice will filter out a hopeless
case. 150 If, for example, a lawyer recognizes the matter as being
identical to a previously decided case which suggests that the cli-
ent will lose, her/his advice will probably be not to proceed. So,
for any given leading case in a SHYSTER specification, there may
be many substantially identical cases which proceeded no further
than a lawyer's office.

Even if the client's legal case is a strong one, there are many
reasons why it might never reach court. Most people find the cost
of legal redress prohibitively expensive. Further, the time delay
involved may well dissuade a person from taking legal action, or
defending an action. For any number of non-legal reasons, pro-
spective parties to a case may choose not to proceed, and may
u settle out of court."

Finally, not all cases that are decided in court are reported.
Only those cases which the court reporters deem significant are
included in the law reports. The legal domain is quite different
from other areas of case-based expert system development where
the developer may have access to a large number of cases which
are relatively unfiltered.

108 A pragmatic legal expert system §2-13

In the face of this lack of test cases, Gardner used examination
problems to test her system on the basis that, although they may
not be as complex as real cases, they are reasonably difficult and
the sorts of questions that lawyers are expected to be able to
handle. 151 This is certainly true— and SHYSTER is tested with
some hypothetical cases in this way — but I contend that actual
cases make better tests. Although the legal system greatly restricts
the number of cases that are reported, it ensures that those cases
that are reported make good — that is, difficult- — tests. If a case
had been straightforward, it is highly unlikely that it would have
proceeded through the various filters that the legal system provides
to emerge as a reported judgment.

The four specifications described in chapter 4 are tested with a
total of seventeen previously decided cases which are not amongst
the specified leading cases. Although a small number, it compares
favourably with the testing of other case-based systems: Tyree,
Greenleaf and Mowbray tested FINDER with just one case; 152 Ash-
ley evaluated HYPO's performance using only four cases — all of
which were taken from the case base. 153

In addition to the real cases used to test SHYSTER, there are four
hypothetical test cases. These were proposed by the legal experts
as interesting tests that have not yet been before the courts. The
seven ideal points specified in three of the four specifications are
also used as test cases.

2.13.4 Generated tests

Generated tests are performed using SHYSTER's capacity to argue
with instantiations. 154 SHYSTER can be used to generate, and give
advice on, all possible cases in an area. However, there is no point
in generating the entire search space in this fashion. In an area of
reasonable size, the number of generated cases would be so large
as to make evaluation of SHYSTER's advice impractical. 155

Instead, the legal expert is asked to specify a number of attrib-
ute values whose presence in a case means that a certain result, or
results, follows (i.e. that result, or results, would be reached if that
case were to be heard by a court). A fact vector is constructed
containing these known attribute values, with the remaining at-
tribute values set to unknown, and SHYSTER is made to generate
all instantiations of that fact vector.

§2.13 A pragmatic approach to case law 109

Although it is not possible to evaluate SHYSTER'S choice of cases
in all of these instantiations — -these generated cases — it is to be
hoped that SHYSTER will choose a good result (i.e. one of the res-
ults specified by the legal expert) in most, if not all. of them. The
number of good choices of result as a fraction of the number of gen-
erated cases is SHYSTER's success rate for a generated test. Some
of these generated cases may represent paradoxes: combinations
of attribute values which are impossible. Paradoxical generated
cases are ignored when determining SHYSTER's success rate. 150

SHYSTER's ideal point warnings 157 are of particular importance
to the generated tests. Many of the generated cases may be ex-
tremely unusual: i.e. combinations of attribute values that are
very unlikely to occur, though not impossible. Assuming that its
leading cases are carefully selected by the legal expert, SHYSTER
is more likely to choose its nearest cases badly when given an ex-
tremely unlikely and unusual instant case than when it is given a
realistic combination of attribute values as its instant case. Ideal
point warnings are designed to detect these extremes, and are
taken into account when evaluating the results of generated tests.

2.13.5 Reflexive tests

A reflexive test is performed in the same fashion as a test per-
formed using the principal testing method, except that one of the
leading cases in the current specification is used as the test case.

Of course, if SHYSTER is presented with a fact vector which is
identical to that of one of the leading cases in its specification, it
will simply follow that leading case. Such a test would demonstrate
nothing. So, before a reflexive test is performed, the leading case
that is to be used as a test case is removed from the specification.
SHYSTER is effectively asked "if this case were decided now, in
the light of all the leading cases except itself, how would it be
decided?" (The term "reflexive" is coined to describe such a test
because the case is being applied to the specification whence it
came.)

Mackaay and Robillard criticize Lawlor for adopting a reflexive
approach to testing. They point out that:

. . . when all cases are used to determine optimal [attribute] weights
. . . there can be no surprise to find that on the basis of those
weights each case is correctly classified; the results would look
unduly promising. 108

110 A pragmatic legal expert system §2-13

This criticism does not apply to reflexive testing of SHYSTER
because SHYSTER recalculates attribute weights (without the re-
moved leading case) for each of the reflexive tests.

The specification that is used for a reflexive test is diminished:
it excludes the test case. It must be assumed that the case that
has been removed belongs in the specification, otherwise the legal
expert would not have included it. Hence the diminished specific-
ation no longer represents the area of law that it was written to
represent.

This means that the result of a reflexive test does not assist in
evaluating SHYSTER's approach to case law: SHYSTER's opinion
and the cases it chooses cannot sensibly be compared with the
judgment in the actual case. However, the results of reflexive
tests do provide information about the specification itself.

It is verj' unlikely that a specification of reasonable size would
have a leading case for every possible combination of attribute
values. Yet it is highly desirable that a specification be capable of
handling new cases which are not identical to any of the leading
cases.

Consider a specification in which conducting a reflexive test for
every leading case in that specification yields a good result. All of
the leading cases contribute to the extent that each case's result
can be determined from the other cases. Such a set of results would
indicate that the specification is successful at handling new com-
binations of attribute values. Conversely, consider a specification
in which every reflexive test yields a bad result. That specification
is clearly very poor at handling new fact situations.

In reality, a given specification is unlikely to reach good conclu-
sions in all of its reflexive tests, or in none of them. The number
of good conclusions will probably lie somewhere in between. If a
large proportion of a specification's reflexive tests yield a good res-
ult, that is indicative of that specification's suitability to handle
new cases.

It is important to note the following points about reflexive testing.
Even if SHYSTER's opinion in a reflexive test is very good — i.e. it
comes to the same conclusion as did the court in that case, and
for the same reasons — that is no reason to remove that case from
the specification permanently. It does not follow that, because
the diminished specification is all SHYSTER needs to reach a good
conclusion, the excluded case adds nothing to the specification. A

§2.14 A pragmatic approach to case law 111

new case could occur in which the best opinion would be to follow
the excluded case, and not the cases that SHYSTER chooses in the
reflexive test. This will almost certainly be true for a new case
with the same fact vector as the excluded case. The diminished
specification may reach a good conclusion, but not necessarily for
good reasons. 159

Removing a case from a specification can introduce attribute
dependencies which were not there before. It is also quite likely
that some of the cases which SHYSTER chooses upon which to
base its opinion in a reflexive test will have been decided after the
test case. If that happens, it is impossible to evaluate SHYSTER's
choice of cases.

The weight given to the decisions of a given court may vary
between jurisdictions. For example, decisions of the English House
of Lords are binding on lower courts in England, but only strongly
persuasive in Australian courts. Three of the four specifications
discussed in chapter 4 use Australian and English cases, 100 and
each specification aims to represent the Australian law. If an Eng-
lish case is used for a reflexive test, SHYSTER may choose a case
which was not referred to in the test case but which might have
been a good case to choose had the test case been heard in an
Australian court.

Reflexive tests are not tests of SHYSTER's approach to case law.
though they do provide information about a particular specifica-
tion. A reflexive test has been performed for every leading case in
each of the four specifications discussed in chapter 4. The results
of these tests are discussed in that chapter. 161

2.14 Conclusion

SHYSTER treats cases as points in space, the dimensionality of
which is the number of attributes. The instant case is placed at its
appropriate point in this space, and the nearest leading cases are
determined. These nearest cases are used to produce an argument
(based on similarities and differences between the cases) about
the likely outcome in the instant case. That argument relies on
the doctrine of precedent; it assumes that the instant case will be
decided the same way as was the nearest case.

SHYSTER can also instantiate unknown attribute values, thus
testing all possible configurations of the instant case. A limited
number (specified by the user) of hypothetical variations of the

112 A pragmatic legal expert system §2.14

instant case can also be tested, to see whether the case can be
strengthened toward some result. SHYSTER also applies several
safeguards and the user is warned if SHYSTER has some doubt
about the veracity of its advice.

The report that SHYSTER generates makes a prediction and
justifies that prediction by reference only to cases and their sim-
ilarities and differences: the calculations that SHYSTER performs
in coming to its opinion do not appear in that opinion.

SHYSTER models legal knowledge at the second of the three
levels that Greenleaf, Mowbray and Tyree identify: 102 its ap-
proach includes justification based on the primary legal sources.
but without any explicit model of those sources; and principles of
interpretation are implied in that approach.

By taking a pragmatic approach to representing case law. SHY-
STER follows one of the approaches that Gardner recommends tak-
ing "if legal realism is right" : emphasizing the behaviouristic side
of legal realism. 103 Gardner herself takes a different approach, 104
and says of the behaviourists' programs:

These are not AI programs . . . The programs are concerned with
predicting judicial decisions, or more generally with analyzing ju-
dicial behavior, from a data base in which legal rules have no
role. Traditional modes of reasoning are replaced by mathemat-
ical methods . . , 105

Gardner does not explain why she thinks that these "prediction
programs" are not AI programs. If a program can successfully
predict judicial decisions — even without adopting a "traditional
mode of reasoning" — it is at least arguable that it is artificially
intelligent.

Areas are used in SHYSTER to represent open-textured concepts:
statutory or case-based. This structure facilitates the linking of
a rule-based system with SHYSTER's case-based system to form a
hybrid system, capable of dealing with statutes and case law. 100

A hybrid approach has a number of benefits. It has the advant-
age of approximating the approach which a lawyer would take
when given a legal problem. The rules, derived from a statute,
are applied until the meaning of some open-textured concept is
required. Faced with this problem, a lawyer would turn to the
common law in order to further clarify the meaning of the statute.
So, too, does SHYSTER: the lawyer's two-stage approach is clearly
modelled.

§2.14 A pragmatic approach to case law 113

A consequence of SHYSTER's approach is that it returns a res-
ult from each area of case law, for use in another area, or by a
statutory rule base. It is certainly true, as Ashley points out. that
u the goal of a theory of analogical legal argument should not be
to explain what the right answer is." 167 But in returning a result
from an area. SHYSTER does not pretend to be giving the "right
answer"; it is merely attempting to predict the "likely answer."
applying the principle of stare decisis. SHYSTER's opinion is pre-
dictive, not normative, and a good legal expert system should have
predictive power. 108

Gardner cites various writers who identify the following as diffi-
culties associated with this approach to case law: separating find-
ings of fact from legal conclusions; determining what the judges
in the leading cases believed the facts to be; categorizing facts
appropriately; and deciding what aspects of the facts should be
included. 109 But these difficulties are not peculiar to legal expert
systems which take an approach to case law like SHYSTER's. All of
these difficulties are inherent in the problem of dealing with case
law. SHYSTER expects the legal expert to have regard to these
difficulties when specifying an area of case law. This is not an
unreasonable expectation: knowing how to overcome these diffi-
culties is one of the characteristics of legal expertise.

2.14.1 Comparisons with other approaches

SHYSTER adopts and expands upon the nearest neighbour ap-
proach to case law taken by FINDER. (Mackaay and Robillard
apply nearest neighbour techniques to cases, too, although they
did not actually develop an expert system.) SHYSTER also adopts
aspects of HYPO's approach to reasoning with hypotheticals.

One difference between SHYSTER and both FINDER and HYPO
is SHYSTER's generality: while FINDER deals only with the law of
trover and HYPO deals with trade secrets law, SHYSTER allows a
legal expert to define an arbitrary number of areas of case law in
each specification.

SHYSTER's knowledge representation is more complex than
FINDER'S, but simpler than HYPO's. 170 Unlike SHYSTER, FINDER
does not allow the specification of unknown attribute values: un-
known values are entered as nos. Tyree suggests that an altern-
ative approach, u possibly better." would be to select a value at
random: 171 SHYSTER's use of unknowns is better still.

114 A pragmatic legal expert system §2.14

SHYSTER allows any number of possible results in an area. Both
HYPO and FINDER treat all cases (including the instant case) as
having one of two possible results. Ashley claims that:

. . . subject to some qualification, there are only two possibilities:
either the plaintiff won the case or did not. The qualification
is that depending on the procedural context of the case, some
outcomes are more determinative than others. 172

This qualification relates not to the number of possibilities but to
the importance of the leading cases (which is captured by SHY-
STER using its hierarchy of courts). Although there may be only
two possible results in the areas of law covered by FINDER and
HYPO, this is not true in other areas. 173

SHYSTER's approach to weighting attributes assumes attrib-
ute independence, so it checks all areas for attribute dependence.
SHYSTER uses the same method to weight its attributes as does
FINDER; hence, FINDER also assumes attribute dependence, al-
though its developers do not address the problem. Fortunately.
for FINDER, there is neither functional dependence, nor evidence
of stochastic dependence, between its attributes. 1 ' 4 SHYSTER.
like FINDER but unlike HYPO, assumes that each attribute has
the same weight in all cases.

SHYSTER'S distance metric is similar to FINDER's, except that
it takes account of unknown distance and infinitely weighted
attributes. 175 FINDER does not use extra similarity measures, spe-
cified directions or instantiations as safeguards, as does SHYSTER.
FINDER does use centroids as a safeguard. The use of ideal points
as a safeguard is suggested by Tyree et al. but is not implemented
in FINDER. 176

Apart from a few cosmetic additions, the structure of SHY-
STER'S reports is similar to those of FINDER. However, SHYSTER'S
algorithm for choosing which cases to use in argument is more
complicated because of the need to account for unknown distance.
And, unlike FINDER, SHYSTER handles equidistant cases and ar-
gues with instantiations and hypotheticals.

Two of SHYSTER's features — specified direction and hypothet-
icals— are designed to incorporate some of HYPO's functionality,
without the need for HYPO's more complex knowledge represent-
ation.

HYPO represents the attributes that favour each side, and treats
a problem as a collection of possibly competing attributes. All at-
tributes are required to favour one side or the other. 1 ' ' Similarly.

§2.14 A pragmatic approach to case law 115

SHYSTER's specified direction feature allows the legal expert to
specify a result favoured by a certain value for a certain attribute.
However, SHYSTER does not require that all attributes (or attrib-
ute values) be directed, for the simple reason that not all attribute
values can be directed towards a result or results. 178

SHYSTER's reasoning with hypotheticals is not as sophisticated
as HYPO's, due to SHYSTER's simpler representation of case law.
However, SHYSTER is able to examine limited hypothetical vari-
ations, and inform the user if the effect of these variations is to
strengthen or weaken its argument about the instant case.

2.14.2 A Sisyphean journey?

Having explained SHYSTER's approach to case law r . I can now
mount a detailed refutation (foreshadowed in the previous chap-
ter 179 ) of Berman's argument that case-based reasoning techniques
are inappropriate for modelling case law. Berman is a proponent
of deep rule-based models. He believes that "lawyers make their
decisions on the basis of their judgment as to whether a par-
ticular rule will be applied to the facts of a specific case." 1SG I
argue in chapter l 181 that developers of deep conceptual models
of legal reasoning confuse precision with accuracy. The refutation
that follows deals only with Berman's argument against case-based
reasoning, with particular reference to SHYSTER.

Berman claims that the choice for the developer in the legal do-
main is a "Sisyphean journey" with case-based reasoning or "down
hill with rules". 182 He refers to work in case-based legal reason-
ing employing "frame-based structures, transition nets, semantic
networks, discrimination trees, connectionist models, etc., 183 and
claims that:

To legal scholars well versed in the subtleties of legal reason-
ing these particular representations of legal cases, though sem-
inal works of considerable scientific importance, constitute a mere
simulacrum of legal thought.

First, the models do not contain choice of law rules to account
for those legal cases that implicate the law of more than one jur-
isdiction . . .

This could be achieved using a hybrid system. "Choice of law"
rules could be implemented in the rule base (or a meta-rule base)

116 A pragmatic legal expert system §2.14

to ensure that the appropriate areas of case law were applied. Al-
ternatively, assuming that the user is an expert, the user could
make this choice her/himself (as discussed above. 184 such an as-
sumption is reasonable).

Second, the models do not account for the fact that some preced-
ents are weakened by divided courts . . .

SHYSTER allows the legal expert to specify a hierarchy of courts.
Several "courts" in this hierarchy could be different compositions
of the same court. For example, there is no reason why she/he
could not include in this hierarchy a "court" of four out of seven
justices of the High Court of Australia, and rank it appropri-
ately. The legal expert could even include a minority judgment
as a separate case in the specification if it was deemed sufficiently
important. SHYSTER only uses its hierarchy for resolving equidis-
tance, but there is no reason why a case-based system could not
make more use of such a hierarchy in its operation.

Third, the models do not take into consideration that judicial
opinions cany varying precedential values . . .

SHYSTER takes this into consideration in two ways: by use of a
hierarchy, and by allowing a legal expert to choose the important
cases and, by implication, to reject the less important cases.

Fourth, the models do not account for the fact that the preceden-
tial weight may turn on when the case was decided . . .

As with the relative importance of a case, SHYSTER takes the
year in which a case was decided into account when resolving
equidistance — there is no reason why a case-based reasoning sys-
tem could not give the time since a case was decided more im-
portance than does SHYSTER. Further, the legal expert should be
expected not to include, in a specification, cases which are so old
as to be of little precedential weight.

Fifth, the models do not account for sub silentio overruling — the
disregard of precedents which have been so often distinguished or
ignored that they lack precedential value.

SHYSTER assumes that all of the cases chosen by the legal expert
are u good law." Disregarded precedents should be disregarded.

§2.14 A pragmatic approach to case law 111

Sixth, except for the works of McCarty . . . the models lack mech-
anisms for resolving tensions between conflicting lines of authority.

But tensions between conflicting lines of authority are unresolv-
able. The appropriate mechanism for dealing with conflicting lines
of authority is to present all the arguments to assist the user in
constructing her/his argument.

Seventh, the model does not account for judicial decisions motiv-
ated by political considerations unarticulated in opinions . . .

Political considerations, if the legal expert deems them important.
can be included as attributes in SHYSTER's case law specifica-
tions. 185

Eighth, these models have not provided for the computational
representation of legal fictions where the concepts of contracts,
easements and notice become spurious easements, quasi contracts,
and constructive notice.

Why these examples could not be represented in a case-based sys-
tem is not clear.

Ninth, these models do not consider that the precedential value
of a case may turn on the prestige of the judge who wrote the
opinion.

As discussed above, under Berman's second point, this presents
no difficulty for SHYSTER.

Tenth, and most importantly, the model does not represent ac-
curately the procedural posture of a case so that the resulting
arguments fail to distinguish cases in which courts have ruled on
matters of law from cases where appellate courts have merely af-
firmed findings of fact . . . 180

If a case involves merely an affirmation of a finding of fact, and
no ruling on a matter of law, then the legal expert is unlikely to
choose it as a leading case.

Berman confuses the role of a case-based system with the role
of a legal expert in the knowledge acquisition process. With the
possible exceptions of Bench-Capon et al., no-one would deny that
legal expertise is required in the development of a legal expert
system. Hence, it is reasonable to use the fact of legal expert
specification in SHYSTER to refute several of Berman's points.

118 A pragmatic legal expert system

Berman concedes that rules u fall far short of fully representing
legal knowledge", but:

For developers, as contrasted to researchers, the issue is not
whether the resulting base is "complete" or even "accurate" or
"self-modifying" — but whether the resulting rule base is sufficient-
ly complete and accurate to be "useful". 187

I agree that usefulness should be the principal criterion in legal
expert system design, but I argue that, in a case law domain, a
case-based system can better satisfy that criterion.

Notes

1 Gay 1967, vol.11 at 1-2.

2 Murphy 1980 at 5. Opening address delivered at the First National
Conference of Labor Lawyers, Adelaide, 29 June 1979.

3 (1637) 3 How St Tr 825 at 969.

4 §1.8.

5 Susskind 1987a at 12.

6 §1.1.

7 The difference between private and public law is explained in §2.3.1.

8 For example, Susskind 1987a at 255; Ashley 1990 at 249-50.

9 §3.5 and §3.6.

10 §1.4.7.

11 §1.4.4.

12 §1.2.5.

13 See Stone 1964a at 120; Kelman 1987; Bottomley, Gunningham and
Parker 1991.

14 For example, Kelsen 1945, 1967, and the authors referred to in the
previous note.

15 §1.1.3.

16 Susskind 1987a at 41.

17 id. Susskind also claims that law reformers use legal reasoning to try to
persuade the legislature to change the law. This is debatable. In a sense, law
reformers operate outside the realm of legal reasoning; their concern is that
the application of legal reasoning to the existing law produces undesirable
results.

18 id.

19 ibid, at 42.

20 §1.2.7.

21 See Morris, Cook, Creyke and Geddes 1992, especially chs 3-4, 8.

22 Hart 1961 at 131.

23 §2.4.

24 Ashley 1990 at 229, citing Burton 1985 at 40.

25 Ashley 1990 at 230, capitalization added.

26 §2.7, §2.9 and §2.10.

27 §1.4.2 and the beginning of §1.5.

A pragmatic approach to case law 119

28 Chapter 4.

29 Shannon and Golshani 1988 at 315.

30 §1.4.7.

31 See the discussion on attribute A§ in the Authorization area of the
Authorization specification (§4.3.2).

32 §2.11.2.

33 §2.5.1.

34 The word "hypothetical" is used here in its general sense. These hy-
pothetical ideal cases {ideal points: §2.5.4) are different to the hypotheticals
that SHYSTER uses in argument (§2.11.3).

35 For example, the Employee area of the EMPLOYEE specification (§4.4.2)
represents the open-textured concept of a worker's employment status. There
are two possible results: the worker is an employee, or the worker is an inde-
pendent contractor.

36 The structure represented in figure 2.1 is that of the NATURAL spe-
cification (§4.5.2). See also figure 4.8.

37 Mehl 1959 at 768.

38 id.

39 Susskind 1987a at 54, emphasis omitted. He continues: "But that is
not a job for the near future."

40 Berman and Hafner 1991 discuss a formal model for distinguishing
between questions of fact, questions of law, and mixed questions.

41 See, for example, the development of the NATURAL specification from
the work of McMillan 1991 (discussed in §4.5.1 and §4.5.2).

42 Gardner 1987 at 196 (a note to p. 75). See §1.3.5.

43 §2.12.4.

44 Mackaay and Robillard 1974 at 308 (see §1.5.1).

45 As discussed in §2.5.2.

46 An example of this can be found in the EMPLOYEE specification. The
legal expert identified the question of whether the worker's employer deducted
"pay as you earn" (PAYE) tax instalments from the worker's pay as being an
attribute (^15) in the Employee area. In Zuijs v. Wirth Brothers Pty Ltd
(1955) 93 CLR 561, Wirth Brothers deducted PAYE tax instalments from
Zuijs's pay. Dixon CJ, Williams, Webb and Taylor JJ of the High Court
mention this fact (at 567), but do not use it in coming to their decision.

47 Blackstone 1773 (especially vol. 1 at 69-70); Krygier 1986.

48 For example, Stone 1964a, 1985.

49 The Natural area of the NATURAL specification affords an example of
this. The attribute A3 concerns the question whether an administrative power
is of a nature that would suggest that procedural fairness would be applied in
its exercise. When Bread Manufacturers of New South Wales v. Evans (1981)
180 CLR 404 was decided, the answer in that case would have been YES. The
law has developed to the extent that, if that case were to be heard now, the
answer would be NO.

50 Ideal points are so-named because, as explained in §2.9.1, SHYSTER
treats cases as points in space, the dimensionality of which is the number of
attributes.

51 Apropos specifying more than one ideal point per result, see §5.2.5.

52 52.12.2.

120 A pragmatic legal expert system

53 §2.5.

54 §3.5 and §3.0.

55 §2.9.

56 Lambert and Grunewald 1989. LESTER deals with unjust discharge
from employment under collective bargaining agreements.

57 ibid, at 91.

58 Ashley and Rissland 1988 at 239.

59 Williams 1971.

60 Sneath and Sokal 1973.

61 Aldenderfer and Blashfield 1984 at 21. emphases added, footnotes ad-
ded.

62 Everitt 1974 at 50.

63 Tyree 1989 at 149 n. 16.

64 ibid, at 141, emphases added. Tyree refers to separating the cases into
two classes because FINDER allows only two possible results: the finder wins,
or the finder loses. SHYSTER allows an arbitrary number of results in each
specified area of law.

65 Tyree, Greenleaf and Mowbray 1988 at 245 n. 8.

G6 Neter, Wasserman and Whitmore 1982 at 185. This formula defines
the variance of a finite number of values — a finite population — and is different
from the formula that defines the variance given a sample of values.

67 FINDER does not deal with infinite weight as none of its attributes has
zero variance.

68 Real examples are ^5 in the Authorization area of the AUTHORIZATION
specification, and A4 in the Expectation area of the NATURAL specification.

69 Ashley 1990 at 175.

70 Bing 1992 at 104.

71 id. Bing's reference is to FINDER's method, but his comment applies
equally to SHYSTER's method.

72 §2.13.2.

73 §1.5.3.

74 Ashley 1990 at 175-6.

75 Ashley and Rissland 1988.

76 Ashley 1990 at 176-8, capitalization added.

77 Lambert and Grunewald 1991 at 194.

78 Actually, there is only a stochastic dependence in this example if there
is some variation of values within the attributes. In the extreme circumstance
where all of the attribute values are the same, the two attributes are (strictly)
stochastically independent.

79 An example of this is the functional dependence between A4 and A$
in the Employee area of the EMPLOYEE specification (see the discussion of
attribute dependence in §4.4.2).

80 See, for example, the discussion in §4.4.2 on the three attributes re-
moved from the first draft of the EMPLOYEE specification and replaced with
a single attribute (.A5).

81 However, if an attribute's known values are all YESs or all NOs (i.e. there
is a YES function, or a NO function, mapping the values of every other attribute
to those of that attribute), the legal expert is warned that the attribute has
been given infinite weight (§3.9.1).

A pragmatic approach to case law 121

82 §3.8.1.

83 §2.9.3.

84 This process takes O(n) time, where n is the number of leading cases,
whereas cluster analysis takes 0(n ) time.

85 Aldenderfer and Blashfield 1984 at 18, emphasis omitted.
80 id. Everitt 1974 at 56 lists all but the third criterion.

87 Aldenderfer and Blashfield 1984 at 18-19.

88 ibid, at 19.

89 Sneath and Sokal 1973 at 140-5; Aldenderfer and Blashfield 1984 at 33.

90 It is often necessary to standardize attribute values before calculating
Euclidean distance. Standardization is not necessary in SHYSTER as all of
the attribute values are of the same type.

91 In SHYSTER, n is the number of known pairs of attribute values. At-
tribute pairs with one or two UNKNOWNS are ignored for these purposes; they
are taken into account in the calculation of unknown distance.

92 Aldenderfer and Blashfield 1984 at 28.

93 Everitt 1974 at 51.

94 Gower's coefficient, another measure of similarity, is identical to Jac-
card's coefficient in the binary case: Aldenderfer and Blashfield 1984 at 31.

95 Note that, in SHYSTER, n may vary within the same area of law
because n is the number of known pairs. Hence, Sjk is not proportional to

djk-

96 Aldenderfer and Blashfield 1984 at 22.

97 See ibid, at 23-4; Everitt 1974 at 53-4.

98 Identical cases do not have a coefficient of zero, and the coefficient
often fails to satisfy the triangle inequality.

99 Aldenderfer and Blashfield 1984 at 23.

100 A correlation coefficient of —1 indicates a negative linear association;
a value of -hi indicates a positive linear association. A value of zero indicates
no linear association.

101 Ashley 1990 at 127.

102 ibid, at 128.

103 As explained in §2.7, HYPO adopts a "symbolic least commitment
approach" to attribute weighting.

104 §2.7.

105 Sneath and Sokal 1973 at 146.

106 §2.12.1.

107 §2.7.

108 These distances are for Narich Pty Ltd v. Commissioner of Pay-roll
Tax (1983) 50 ALR 417: one of the cases used to test the EMPLOYEE specific-
ation. Narich v. CPT is described in §4.4.3, and full details of its distances
are given in figure 4.7.

109 The Employee and Contractor results, respectively (§4.4.2).

110 Mackaay and Robillard 1974 at 309-10. They term this the problem
of "tied distances amongst neighbours."

111 Tyree, Greenleaf and Mowbray 1989 at 47.

112 Ashley 1990 at 39.

113 Insofar as a program can have an opinion.

114 $2.9.

122 A pragmatic legal expert system

115 Ashley 1990 at 230-1, capitalization added.
110 ibid, at 232.

117 Note that the use of unknown distance as a measurement of error
(discussed in §2.9) does not account for the possibility of an instantiation
producing a different result. An unknown attribute value in the instant case
has the effect of adding the weight of that attribute to the unknown distances
of all the leading cases: i.e. the leading cases are all pushed away from the
instant case by the same amount.

118 Used in this fashion, instantiation is similar to SHYSTER/s use of
hypotheticals (§2.11.3), only with a higher degree of user control.

119 §2.12.0.

120 §2.13.4.

121 Ashley 1990 at 85.

122 ibid, at 233. See also Rissland 1989.

123 Ashley 1990 at 232.

124 Unknown values are dealt with by instantiation (§2.11.2).

125 The number of hypotheticals h with exactly j differences from the
instant case, given n attributes, is (™) = n(n — l) ■ ■ ■ {n — j 4- l)/if So, the
number of hypotheticals with no more than k differences is (?) 4- (?) + • ■ •
4- (?) . The Employee area of the EMPLOYEE specification, for example, has 18
attributes. Allowing no more than 2 differences {k = 2) restricts SHYSTER
to 171 hypotheticals. If k = 3, h = 987; and if k = 4, h = 4047— still only
a small fraction of the number of possible hypothetical variations upon the
instant case: 2 — 1 = 202143 hypotheticals.

120 §2.12.3.

127 §2.9.1-§2.9.3.

128 §2.7.

129 §2.5.4.

130 Each YES is assigned a value of 1, and each NO is assigned a value
of 0. For the purposes of constructing the centroid, each mean (a number n in
the range to 1) is rounded to the nearest attribute value (0 < n < 0.5: NO;
0.5 < n < 1: YES). However, the unrounded means are used when calculating
the correlation coefficients r and r for those centroids.

131 Tyree et al. 1988 at 241.

132 §2.5.2.

133 SHYSTER uses the attribute's result weight (explained in §3.9.1)
when calculating the strength of directions.

134 §2.10.3.

135 §2.11.2.
130 §2.13.3.

137 This is true for test cases and generated tests. For reasons discussed
in §2.13.5, reflexive testing does not test SHYSTERs approach although it
does provide some information about the specification.

138 §2.13.2.

139 Susskind 1987a at 55.

140 ibid, at 53-5.

141 Leith 1985a, 1986a.

142 Susskind 1986 at 175-6. (It is probably unnecessary to point out that
Susskind wrote this article for a legal audience.) He continues: "This is not

A pragmatic approach to case law 123

to belittle Leith's achievements, for he was clearly working with limited re-
sources" (at 176). Leith bitterly disputes Susskind's comments, including his
definition of what constitutes a legal expert system. "If there is no agreement
[over suitable terminology in the field], why should computer scientists accept
the definition proposed by a lawyer? Will Susskind accept the definition of
'divorce' proposed by a computer scientist?" (1987 at 130). Susskind's early
experimental work was with the Scottish law of divorce.

143 Clark 1988 at 428.

144 §4.2-§4.5.

145 Gardner 1987 at 33. Gardner's chosen domain is one aspect of the
law of contract (§1.4.3).

140 There are two exceptions to this rule- — two tests where SHYSTER's
choice of nearest result is deemed "good" despite disagreeing with the ac-
tual cases: Twist v. Randwick Municipal Council (1976) 136 CLR 106 and
Salemi v. MacKellar [No. 2] (1977) 137 CLR 396. These exceptions are ex-
plained and justified in §4.5.3.

147 §2.12.

148 (1977) 137 CLR 396 (see §4.5.3).

149 As discussed in §1.4.7. one of the reasons why case-based knowledge
acquisition cannot be automated is that the number of decided cases in any
given area of law is usually so small that inductive inference algorithms cannot
be used.

150 And bad legal advice may filter out a hopeful case.

151 Gardner 1987 at 6.

152 Tyree 1986, 1989; Tyree et al. 1988, 1989. Tyree et al. 1989 also
describe (at 49-50) a reflexive test using Bridges v. Hawkesworth (1851) 21
LJQB 75.

153 Ashley 1990 at 183-93: Crown Industries Inc. v. Kawneer Co. (1971)
335 F Supp 749; Structural Dynamics Research Corporation v. Engineering
Mechanics Research Corporation (1975) 401 F Supp 1102: USM Corporation v.
Marson Fastener Corporation (1979) 379 Mass 90: 393 NE 2d 895; Amoco
Production Co. v. Lindley (1980) 609 P 2d 733.

154 §2.11.2.

155 For example, the Employee area of the EMPLOYEE specification has
18 attributes and, hence, a search space of 2 = 262 144 different cases.

156 Some of the cases generated for the EMPLOYEE specification represent
paradoxes. These impossible cases are discussed in §4.4.4.

157 SHYSTER issues a warning if the instant case is at least as near to
the ideal point for another result as it is to the nearest neighbour (§2.12.2).

158 Mackaay and Robillard 1974 at 312. Lawlor 1968 terms this reflexive
approach "cyclical sampling" (at 110-11).

159 Compare this with the approach taken by Mackaay and Robillard
1974. They say that cases which are "incorrectly' predicted in a reflexive test
are "suspect" because "[s]uch cases are difficult to explain on the basis of the
remaining ones" (at 312). Mackaay and Robillard were dealing with a large
set of cases, not a smaller set of expertly chosen cases as in SHYSTER.

160 The FINDER specification (§4.2.2) uses only English cases.

161 §4.6.3.

162 Greenleaf, Mowbray and Tyree 1987 at 11, as quoted in §1.7.

124 A pragmatic legal expert system

163 Gardner 1987 at 22.

164 §1.4.3.

165 Gardner 1987 at 74-5. Gardner refers to several of the researchers
discussed in chapter 1, including Kort 1963a; Lawlor 1963, 1972: Mackaay
and Robillard 1974; Haar, Sawyer and Cummings 1977: Borchgrevink and
Hansen 1980; Tyree 1981.

166 Green leaf, Mowbray and Tyree 1991 propose a similar mechanism for
their DATALEX project (at 223), though the extent to which it has been
implemented is not clear. See also Tyree et al. 1989 at 49.

167 Ashley 1990 at 254.

168 §1.3.5.

169 Gardner 1987 at 75. The representation of questions of fact and ques-
tions of law is discussed in §2.5.2.

170 HYPO's knowledge representation is explained in §1.5.3.

171 Tyree 1977 at 414 n. 40. There are two unknown values coded as NO
in FINDER (see the discussion on leading cases in §4.2.2).

172 Ashley 1990 at 128.

173 For example, the Authorization area of the AUTHORIZATION specific-
ation has three results.

174 §4.2.2.

175 Because FINDER does not allow unknown values, it has no need of
unknown distance. None of its attributes is infinitely weighted.

176 Tyree et al. 1988 at 241.

177 Ashley 1990 at 249.

178 An example can be found in the Authorization area of the AUTHORIZ-
ATION specification. An answer of YES to A3 ( tk Did the accused sell or hire the
infringer the means of infringing?") can be directed towards the result Auth,
but an answer of NO or UNKNOWN for that attribute cannot be directed to-
wards either of the other results: Not-Auth or Liable. The fact that the accused
did not sell or hire the infringer the means of infringing (A% = NO) suggests
neither that the accused did not authorize the infringement (Not-Auth) nor
that the accused is directly or vicariously liable for the infringement (Liable);
similarly for A3 = UNKNOWN. Another example is Aq in the Employee area of
the EMPLOYEE specification.

179 §1.4.7.

180 Berman 1991 at 308.

181 §1.8.

182 Berman 1991.

183 Including Gardner 1987; Goldman, Dyer and Flowers 1987; Hafner
1987; Rissland and Ashley 1987; B ranting 1989; Ashley and Aleven 1991.

184 §2.4.

185 For example, several attributes in areas in the NATURAL specification
embody political considerations.

186 Berman 1991 at 307-8.

187 ibid, at 308.

Implementing
SHYSTER

Hector Frome: "Justice is a machine that, when someone has once given it the
starting push, rolls on of itself."

John Galsworthy (1910)
Justice

He is no lawyer who cannot take two sides.

Charles Lamb (1833) 2

Thelma Todd: "I didn't know you were a lawyer.
You're awfully shy for a lawyer.''
Groucho Marx: "You bet I'm shy. I'm a shyster lawyer."

Monkey Business (1931) 3
125

126 A pragmatic legal expert system §3.1

3.1 Introduction

SHYSTER'S approach to case law is described in the previous
chapter. In this chapter, the implementation of that approach
is described, and illustrated using examples. SHYSTER was imple-
mented in a UNIX environment, and consists of a dozen modules
written in ISO C. 4 The structure of this chapter mirrors that of
SHYSTER, with the description of the implementation divided into
descriptions of each module.

The Shyster module (§3.2) is the top-level module for the whole
system. The Statutes module (§3.3) is the top-level module
for a rule-based system, presently unimplemented. The Cases
module (§3.4) is the top-level module for the case-based system.
The Tokenizer and Parser modules (§3.5 and §3.6) tokenize and
parse a program written in SHYSTER's case law specification lan-
guage. The Dumper module (§3.7) displays the information that
has been parsed. The Checker module (§3.8) checks for evidence
of dependence between the attributes. The Scales module (§3.9)
determines the weight of each attribute. The Adjuster mod-
ule (§3.10) allows the legal expert to adjust the weights of the
attributes. The Consultant module (§3.11) interrogates the user
as to the attribute values in the instant case. The Odometer mod-
ule (§3.12) determines the distances between the leading cases and
the instant case. The Reporter module (§3.13) writes SHYSTER's
legal opinion.

Except where otherwise indicated, the examples used in this
chapter are taken from the Employee specification, which is ex-
plained in chapter 4° and is used as the basis of the complete
example in appendix B.

3.1.1 Knowledge representation

SHYSTER uses records to represent entities (courts, areas, results.
attributes, cases, etc.) and makes multiply linked lists of these
entities to reflect the relationships between them.

For example, each area record has a pointer to the head of a
linked list of result records. Each result record has a pointer to
the head of a linked list of case records. Each case record has a
pointer to the head of a linked list of attribute value records. Each
attribute value record is an element in two lists, linked by case and
by attribute, forming part of a matrix of attribute values.

§3. 1 Implementing SHYSTER 1 2 1

Using linked lists allows the manipulation of these entities and
relationships without requiring the imposition of any limits upon
their numbers. The only limits that affect the user of SHYSTER
or the writer of specifications are on the maximum length of a
filename, and the maximum length of an identifier.

The information contained in each record, and the links between
those records, are built up by each of SHYSTER's modules until,
by the time that the Reporter module is invoked, the structure
is complete.

3.1.2 Output files

Each time SHYSTER/s case-based system is invoked it writes the
following files:

• a log file which summarizes SH YSTER's operation and includes
any warnings that have been issued;

• a dump file which is a dump of SHYSTER's internal represent-
ation of the case law specification;

• a probabilities file which gives dependence probability figures
for each attribute pair in each area;

• a weights file which gives details of the weights that SHYSTER
gives to each attribute in each area;

• a distances file for each invoked area which includes distances
and other similarity measures for the instant case, instanti-
ations and hypotheticals; and

• a report file for each invoked area which is SHYSTER's legal
opinion — its argument about the likely result in that area for
the instant case.

Only the first and the last of these are intended for the user. SHY-
STER's report files are completely self-contained; the user need
have no knowledge of SHYSTER and its operation in order to un-
derstand its reports. The log file should be comprehensible to
anyone with only passing acquaintance with SHYSTER's opera-
tion. A user who finds the information contained in a log file
arcane may safely ignore the file: of its contents, only the warn-
ings are crucial, and these are also written to the standard error
stream.'

128 A pragmatic legal expert system §3-2

All other files are intermediate files that SHYSTER produces
on its way to its report. They provide details about SHYSTER'S
internal workings for the information of the knowledge engineer
and the legal expert.

All of these output files are plain text files. With the exception
of the log file, they are all in IATgX format: i.e. they are suitable for
processing by the I£TgX document processor. 8 This contributes to
SHYSTER's portability, as IATjrX is widely available on many plat-
forms. Using I^TgX simplifies the footnoting of text, allows some
data to be displayed in a clear and economical tabular format,
and ensures the aesthetic quality of the output. All of the ex-
amples of SHYSTER output in this book — the example reports in
appendix A. the files in the complete example in appendix B, and
extracts in various figures in this chapter and the next — appear
exactly as produced by SHYSTER after processing by IAT^X. 9

3.2 The Shyster module

The Shyster module is the top-level module for SHYSTER. It
extracts the options and arguments from the UNIX command line,
initializes the rule-based system and the case-based system, then
invokes the rule-based system.

Figure 3.1 lists the command line switches and arguments that
SHYSTER recognizes. Only the -c switch must be used; all other
switches are optional. So. for example, if the -d switch is not used.
no dump file is written. However, if the -1 switch is not used then
the information that would have been written to the log file is
written to the standard output stream. 10

3.3 The Statutes module

The Statutes module is the top-level module for a rule-based
system. This module provides a skeletal structure within which a
rule-based system could be developed, and linked with the case-
based system to form a hybrid system.

§3.3

Implementing SHYSTER 1 2 9

-a

Enable weight adjustment (see §3.10)

-c specification

Read the case law specification from

"specification, els"

-d distances

Write each distances file to "distances- area .tex"

-D dump

Write the dump file to ^ dump. tex"

-e

Enable echo mode (see §3.11)

-h r c

Hypothesize, reporting on r hypothetical per

result with a limit of c changes

-i

Write LMgX code that can be included in another

LMgX document (i.e. not stand-alone code)

-1 /op

Write the log file to "/op. log"

-p probabilities

Write the probabilities file to "probabilities .tex'

-q

Enable quiet mode (don't summarize cases, etc.)

-r report

Write each report file to "report-area.tex"

-w weights

Write the weights file to "weights. tex"

Figure 3.1: The UNIX command line switches and arguments

recognized b}>"

SHYSTER. Apart from the -c switch, all switches

are optional.

The module has two functions.

The Initialize_Statutes function presently returns a pointer to
a dummy structure. If implemented, it would initialize the rule-
based system, by reading a statute law specification, and return a
pointer to SHYSTER's internal representation of that specification.

The Statute_Law function would invoke the rule-based system
proper. Each statutory concept which was not open-textured
would be resolved using rules. Each statutory concept which was
open-textured would be associated with an identifier correspond-
ing to an area of case law, specified for the case-based system.
When the rule-based system needed to resolve an open-textured
concept, the case-based system would be invoked using that iden-
tifier. The case-based system returns an identifier corresponding
to the result from the appropriate area. That result would be
bound in the rule-base to a value for the statutory open-textured
concept, thus resolving the concept.

130 A pragmatic legal expert system §3.4

At present, the Statute_Law function prompts the user for an
identifier, then invokes the case-based system seeking advice in the
area corresponding to that identifier. The result returned by the
case-based system is written to the log file.

3.4 The CASES module

The Cases module is the top-level module for the case-based sys-
tem. Its two major functions correspond to the two functions in
the Statutes module.

The Initialize JJ 'as es function calls the Tokenizer and Parser
to read the case law specification and build an internal represent-
ation of that specification. The Dumper is invoked to dump that
internal representation to the dump file. The Checker is used
to check for attribute dependence. Finally, the Scales module is
called to assign weights to all the attributes. The function returns
a pointer to SHYSTER'S internal representation of the specification
with all attributes weighted.

The Case_Law function takes a pointer to that internal rep-
resentation, and an identifier corresponding to one of the areas
in that specification. If weight adjustment is enabled, it calls the
Adjuster. It then invokes the Consultant to interrogate the user
as to the attribute values in the instant case. It is the Consultant
which recursively invokes the Case_Law function, if required, to
resolve open-textured attributes: i.e. external attributes.

Having ascertained the attribute values, the Case_Law function
calls the Odometer to calculate the distances between the instant
case and the leading cases, and to determine the nearest cases and
results. Then the Reporter is invoked to write a report about
the instant case.

The Cases module instantiates any unknown attribute values
in the instant case and repeatedly invokes the Odometer and the
Reporter to recalculate the distances and argue using each in-
stantiation. If the user has requested it, the Cases module also
makes hypothetical variations to the instant case and. again, re-
peatedly invokes the Odometer and the Reporter to recalculate
distances and argue with each hypothetical.

Output from these invocations of the Odometer and the Re-
porter is written to the distances file and report file for the current
area.

§3.5

Implementing SHYSTER 1 3 1

AREA

CLOSING

HELP

OPENING

SUMMARY

ATTRIBUTE

COURT

HIERARCHY

QUESTION

UNKNOWN

CASE

EXTERNAL

IDEAL

RESULT

YEAR

CITATION

FACTS

RESULTS

YES

Figure 3.2: Th

e keyword

3 in SHYSTER'S case law

specification

language.

3.5 The TOKENIZER module

The Tokenizer module reads the case law specification file, and
breaks it into tokens. The Tokenizer is invoked repeatedly by
the Parser; each time it is invoked, it returns the next token in
the specification.

3.5.1 Tokens

There are seven different valid types of token.

An identifier is any sequence of alphabetic characters, numeric
characters, and the - character (starting with an alphabetic char-
acter). If an identifier is more than 16 characters long it is trun-
cated to its first 10 characters and a warning is issued. An identi-
fier's case is significant.

A keyword is an identifier that has a special meaning to SHY-
STER. There are twenty keywords, and they are listed in figure 3.2.
Keywords are reserved: i.e. they cannot be used as identifiers.

A string is a sequence of characters, enclosed between a pair
of " characters. 11 There must be at least one character between
the two quotation marks, but SHYSTER imposes no upper limit
on the length of a string. SHYSTER converts a pair of consecutive
" characters within a string into a single " character. 12

A year is a positive integer of up to four digits.

An attribute vector is a sequence of Y, N and U characters within
parentheses. 13

The remaining two tokens are the = character and a token which
indicates that the end of the specification file has been reached.

132 A pragmatic legal expert system §3.6

3.5.2 Comments and whitespace

When the Tokenizer reads a */, character in the specification (ex-
cept in a string), it skips over the rest of that line: any characters
between the % and the end of the line are ignored. This allows the
legal expert to put comments in the specification file.

Whitespace 14 is required between adjacent identifiers/keywords,
and between adjacent strings. The Tokenizer treats each occur-
rence of whitespace in the specification as a single space. Hence,
extra whitespace can be freely added between tokens making the
specification easier to read without changing the way that it is
tokenized. This also applies inside strings; if a string is too long
to fit on a single line it can be split over several lines using white-
space, without changing the string as it is read by the Tokenizer.

3.6 The Parser module

The Parser module parses 15 the case law specification using the
tokens provided by the Tokenizer. A formal definition of the
syntax of the specification language is given, in Extended Backus-
Naur Form (EBXF), 16 in figure 3.3. The specification of case law
using this language is best illustrated by example.

3.6.1 Hierarchy

A specification starts with an (optional) hierarchy. This binds
court identifiers to strings which describe the courts in which the
specification's leading cases were decided. For example:

HIERARCHY

HC-5 'five justices of the High Court of Australia"
HC-4 'four justices of the High Court of Australia"
HC-3 'three justices of the High Court of Australia"
HC 'a single justice of the High Court of Australia"
FCA-3 'three judges of the Federal Court of Australia"
PC 'the Judicial Committee of the Privy Council"
CA 'the English Court of Appeal"
KB 'the King's Bench Division of the English High

Court"
= QB "the Queen's Bench Division of the English High

Court"

§3.6

Implementing SHYSTER 133

specification

[hierarchy ]
area { area }.

hierarchy

hierarchy-header hierarchy-block.

hierarchy-header

"HIERARCHY".

hierarchy-block

court-identifier string

{["="] court-identifier string}.

court-identifier

identifier.

area

area-header area-block.

area-header

"AREA" area-identifier.

area-block

[opening] [closing]

results

attribute { attribute }

case { case }

{ideal-point }.

area-identifier

identifier.

opening

"OPENING" string.

closing

"CLOSING" string.

results

results-header results-block.

results-header

"RESULTS".

results-block

result-identifier string
result-identifier string
{result-identifier string}.

result-identifier

identifier.

attribute

attribute-header attribute-block.

attribute-header

"ATTRIBUTE".

attribute-block

local-attribute | external-attribute.

local-attribute

"QUESTION" string
["YES" string {result-identifier}]
["NO" string {result-identifier}]
["UNKNOWN" string {result-identifier}]
["HELP" string].

Figure 3.3: A formal definition, in Extended Backus- Naur

Form (EBNF), of the

syntax of SHYSTER'S case law specific-

at ion language.

CONTINUED NEXT PAGE

The courts are listed in descending order of seniority: the earlier
in the list that a court appears, the better the authority of its
cases. An = character separates courts of equivalent rank. It is
not necessary to specify every court in the jurisdiction, because
courts are ranked relatively: the above example tells SHYSTER

134 A pragmatic legal expert system

§3.6

external-attribute

"AREA" area-identifier
["YES" string {result-identifier}
["EXTERNAL" result-identifier
{ result-identifier } ] ]
["NO" string {result-identifier}
["EXTERNAL" result-identifier
{ result-identifier } ] ]
["UNKNOWN" string {result-identifier}
["EXTERNAL" result-identifier
{ result-identifier } ] ] .

case

case-header case-block.

case- header

"CASE" string [string].

case-block

"CITATION" string
"YEAR" year

["COURT" court-identifier]
"FACTS" attribute-vector
"RESULT" result-identifier
["SUMMARY" string].

ideal-point

ideal-point-header ideal-point-block.

ideal-point-header

"IDEAL".

ideal-point-block

"FACTS" attribute-vector
"RESULT" result-identifier.

attribute- vector

11 (" attribute- value { attribute- value } ")".

attrib ute- value

"Y" | "N" | "U".

string

character {character}

identifier

letter { letter | digit | "-" }.

year

digit [digit] [digit] [digit].

Figure 3.3 (continued).

that FCA-3 is more important than PC — but not how much more
important. Each court identifier in the hierarchy must be unique.
Note that there are three different sorts of identifier: court
identifiers, area identifiers and result identifiers. SHYSTER knows
which sort of identifier to expect in various places in the specifica-
tion. Hence, the same identifier can be used (for example) to refer
to a result and to a court in the same area. Court identifiers apply
in all areas in the specification. The scope of a result identifier is
the area in which it is specified: i.e. the same identifier can be
used for two different results in two different areas. 17

§3. 6 Implementing SHYSTER 135

3.6.2 Areas

A specification can contain any number of areas. An area com-
mences with an area header:

AREA Employee

which binds an identifier to the area that is about to be specified.
So, if the case-based system is invoked with this identifier then
this is the area that it will use — this is the area that defines the
appropriate open-textured concept.

An opening string may be specified. This should be a brief
introduction appropriate to any opinion given in this area. It
is written at the beginning of SHYSTER's report for this area.
Similarly a closing string may be specified, containing concluding
remarks.

At least two results must be specified; this involves binding iden-
tifiers to strings. For example:

RESULTS

Employee "the worker is an employee"

Contractor "the worker is an independent contractor"

Each string is a statement which holds when the relevant result
occurs. The statement must be cast so that is makes sense when
prefixed with the following: "If case is followed then . . ."

3.6.3 Attributes

Any number of attributes may be specified. Each attribute is
either local or external.

Local attributes

For a local attribute, the first keyword after ATTRIBUTE is QUESTION.
For example: 18

ATTRIBUTE '/. worker paid by time

QUESTION "Did the employer pay the worker by time"
YES "the employer paid the worker by time"

Employee
NO "the employer did not pay the worker by time"

Contractor
UNKNOWN "it is not known whether the employer paid
the worker by time"

136 A pragmatic legal expert system §3.6

HELP "The employer could pay the worker by time

(e.g. by the hour, or by the week) or by
results . "

The QUESTION string is the question that the user will be asked if
the value of the attribute needs to be determined. The YES, NO
and UNKNOWN strings are optional, but the user will only be allowed
to answer the question with a value for which there is a string:
e.g. if no UNKNOWN string is defined, the user will not be allowed
to answer unknown. The HELP string is displayed at the user's
request, and should provide further information to assist her/him
in answering the question.

The result identifiers after the strings specify attribute direction:
a value of yes for this attribute is directed towards the Employee
result; a value of no is directed towards Contractor.

External attributes

For an external attribute, the first keyword after ATTRIBUTE is AREA.
This example is taken from another specification: 19

ATTRIBUTE '/. legitimate expectation

AREA Expectation

YES "the applicant had a legitimate expectation

which was affected by the decision"
Affected EXTERNAL Expectation
NO "the applicant did not have a legitimate

expectation which was affected by the

decision"
EXTERNAL No-Expectation
UNKNOWN "it is not known whether the applicant had a

legitimate expectation which was affected by

the decision"

The value of this attribute is to be resolved by reference to the
Expectation area. 20 The identifiers after the EXTERNAL keywords are
external result identifiers: result identifiers in the external area.
These results are associated with values for this attribute. So. if
SHYSTER returns a result of Expectation for the Expectation area,
the value of this attribute is set to yes; if it returns a result of
No-Expectation, the value is set to NO. 21 If it returns any other
identifier, SHYSTER exits with an error. 22

A value of yes for this attribute is directed towards the Affected
result in this area.

§3.6 Implementing SHYSTER 131

•3. 6. 4 Leading cases

Any number of leading cases may be specified as follows: 23

CASE "Queensland Stations Pty Ltd v. Federal

Commissioner of Taxation"
"Queensland Stations v. FCT"

CITATION

"(1945) 70 CLR 539"

YEAR

1945

COURT

HC-3

FACTS

(NYNYNNYYYNNNNNNNYN)

RESULT

Contractor

SUMMARY

"agreements were ent

Queensland Stations and some drovers . The
agreements stated that the drovers would . . ."

The first CASE string is the full name of the case; the second is a
shorter version of the name, and is optional. The CITATION string
is the case citation, and the YEAR is the year of the decision — not
necessarily the same as the year in the citation. 24 The (optional)
COURT identifier links the case to the court in which it was decided.
The FACTS attribute vector has one attribute value for each at-
tribute defined in the area. The order of these values corresponds
to the order in which the attributes have been defined. The Y,
N and U characters represent the values yes, no and unknown.
respectively. The RESULT of this case is Contractor.

The SUMMARY string is the legal expert's summary of the case.
The string may contain many sentences, but must be cast so as
to make sense when prefixed with the following: "In case y a de-
cision of court, . . ." This string may (but need not) include I^TgX
commands which will be processed when the case is listed in the
dump file, and if the case is used in a report file: e.g. \f ootnote.

A warning is issued if a result has been specified with no leading
cases, or (worse still) with neither cases nor an ideal point. 20 And
if two cases in an area have the same attribute values, or the same
attribute values except for unknown values, a warning is issued.
If these two cases have different results, this is also mentioned in
the warning: if two or more leading cases have identical attribute
values but different results then at least one of those cases was
wrongly decided, or the area needs another attribute to distinguish
between those cases. 20

138 A pragmatic legal expert system §3.7

3. 6. 5 Ideal points

Ideal points are specified as follows:

IDEAL

FACTS (YNYNYYYNNYYYNYYYYU)
RESULT Employee

This represents the ideal combination of attribute values for the
specified result.

3.7 The Dumper module

The Dumper module writes the dump file: a formatted version
of the case law specification that has just been tokenized and
parsed. This file is easier to read (once processed by IATgX) than
its corresponding specification, simplifying the development and
amendment of a specification. It also reflects SHYSTER's internal
representation of the specification. A complete dump file, for the
Employee specification, is given in §B.3.

The Dumper begins the dump file by displaying the hierarchy
of courts in a tabular format. The courts are numbered; courts of
the same rank share the same number.

For each area in the specification, the Dumper starts by dis-
playing the cases and their attribute values in a matrix format.
An example is given in figure 3.4. The attribute values are rep-
resented by a • symbol for yes. a X symbol for no. and a blank
space for unknown. 27

The attributes are named A t . . . A ls in the order of their ap-
pearance in the specification. The cases have been grouped by
their result, and named C x . . . C 14 in the order of their importance
within their group. (The more important of two cases is considered
to be the one decided by the more important court in the hierarchy
or. if their courts are equally important, the case that is more re-
cent.) The ideal points are also represented: /Employee and /contractor-
The rank of the court in which each case was decided is indicated
in the column labelled "c"; these numbers correspond to those in
the hierarchy display. This matrix allows the user easily to com-
pare the attribute values for all the leading cases and ideal points.

The Dumper writes out the opening and closing strings (if they
were specified). It then writes the result identifiers and their
strings.

§3.7

Implementing SHYSTER 139

■—

<u
>.

4-*

■ — i

in t-

t- 00

ro rt

iT> <0

t- 00

l-H

X •

X X

X •

•

X •

•

l-H

X X

•

X •

X X

• X

^~~^

l-H

X X

•

X X

•

1-4

l-H

•

X X

•

X X

• X

CD
CD

-r

l-H

X X

•

X X

•

X X

•

LU
CD

rJ=!

l-H

•

X •

•

X X

•

4-3

^-1

°9 — 1

<4-l

•

X •

• •

•

X X

•

• X

# x

•O

■*■ o

£ J7

X X

X •

•

X X

X •

•

X X

• •

•

X •

•

X X

X •

•

X X

•

la
>

t^
^

•

• •

•

X •

X X

J-3

•

• X

X •

•

X X

•

X X

J-3

•

X •

• X

•

X X

•

• X

4^
03

• X

X •

• ♦

X •

•

X X

• ♦

•

X •

•

X X

i-H

•

• •

X •

•

X X

•

X X

• 1— <

«)

l-H

Cfl

-*■ «o

to IV .2

oo a

O i-t

2 S e

o o

f-H i — 1

—

i-H (— 1 ,,

140 A pragmatic legal expert system §3.7

3.7.1 Attributes

The Dumper writes details on all attributes: local and external.

Local attributes

Local attributes are displayed in the following format: 28

A 12 : Did the employer pay the worker by time?

YES: the employer paid the worker by time.

=^ Employee
NO: the employer did not pay the worker by time.

^ Contractor

UNKNOWN: it is not known whether the employer paid the

worker by time.
The employer could pay the worker by time (e.g. by the
hour, or bj' the week) or by results.

Specified direction is indicated using a => symbol: a value of yes
for this attribute is directed towards the Employee result. 29 The
last paragraph is the help string.

External attributes

External attributes are displayed in the following format: 30

A4: -^ Expectation area

YES: the applicant had a legitimate expectation which was

affected by the decision.

^= Expectation

=> Affected
NO: the applicant did not have a legitimate expectation

which was affected by the decision.

*£= No-Expectation
UNKNOWN: it is not known whether the applicant had a

legitimate expectation which was affected by the de-
cision.

The link between this attribute and the Expectation area is indic-
ated using a •£> symbol. The association of results from the ex-
ternal area with values of this attribute is indicated using a <= sym-
bol: if the Expectation area returns a result of Expectation the value
of this attribute is set to yes. As with local attributes, specified
direction is indicated using a => symbol.

§3.8 Implementing SHYSTER 141

3. 7. 2 Leading cases

Leading cases appear grouped under the appropriate result head-
ing: e.g. "Cases in which the worker is an independent contractor."
They are displayed in the following format:

Cg: Queensland Stations Pty Ltd v. Federal Commissioner of Tax-
ation (1945) 70 CLR 539 ("Queensland Stations v. FCT")

A\\ the employer did not direct the manner in which the
work was to be done.

A 2 '. the worker was allowed to use her/his own discretion
in doing an aspect of the work that was not specified
beforehand.

Ai$: the employer and the worker did not express any in-
tention that the relationship would be one of principal
and independent contractor.

In Queensland Stations Pty Ltd v. Federal Commissioner
of Taxation 31 a 1945 decision of three justices of the High
Court of Australia, agreements were entered into between
Queensland Stations and some drovers. The agreements
stated that the drovers would . . .

The short case name appears in parentheses after the full citation.
The attribute values are described in full. The summary is format-
ted as it would be if the case were used in a report, complete with
footnotes.

3.7.3 Ideal points

Ideal points are displayed using a similar format to that used for
leading cases. Details of the ideal point's attribute values are given
under an appropriate heading. For example:

-^Employee (the ideal case in which the worker is an employee):

3.8 The Checker module

The Checker module examines every pair of attributes in each
area, checking for functional or stochastic dependence between

142 A pragmatic legal expert system §3A

Ax-

NO NO NO NO NO YES YES YES YES YES NO NO YES

Ay.

NO YES NO NO NO NO YES YES YES YES YES NO YES

Figure

3.5: Example attribute values for two attributes A x

and Ay

those attributes. Detecting functional dependence is straight-
forward. 32 And, although it is not possible to prove that two
attributes are stochastically independent, it is possible to detect
evidence of stochastic dependence.

3.8.1 Detecting evidence of stochastic dependence

Two events E x and E 2 are said to be stochastically independent if
the probability of their both occurring is equal to the product of
the probabilities of each of them occurring:

P(E 1 AE 2 ) = P(E 1 )P(E 2 ).

There are four types of known attribute pairs: yes/ yes. yes/no,
no/ yes, and no/no. Let TV be the number of known pairs in the
attributes A x and A Y . Let x be the number of yess in A x , and let
y be the number of yess in Ay. Let n be the number of yes/ yes
pairs for Ax and Ay. Because known attribute values are binary,
the number of any type of known pair is linked to the numbers of
all other types, and can be expressed in terms of N, x. y and n:

number of yes/ yes pairs = n (by definition),
number of yes/no pairs = x — n,
number of no/yes pairs = y — n,
number of no/no pairs = N — x — y + n.

So, in order to detect evidence of stochastic dependence it is ne-
cessary to examine the expectation and actual occurrence of only
one of the four possible pairs. The Checker examines only the
yes/yes pairs.

§3.8

Implementing SHYSTER 143

P(n = i)

P(n < i)

P(n > i)

0.0006

1.0000

0.0245

0.0251

0.9994

0.1836

0.2087

0.9749

0.4079

0.6166

0.7913

0.3059

0.9225

0.3834

0.0734

0.9959

0.0775

0.0041

1.0000

0.0041

0.0000

1.0000

0.0000

1.0000

0.0000

Figure 3.6

The probabilities for the

example attributes Ax

and Ay defined in figure 3.5. N = 13, a

= 6, y = 7.

Let P(n) be the probability of there being exactly n yes/ yes
pairs given N, x and y, and assuming random data:

P(n) =

number of configurations with exactly n YES/YES pairs

total number of configurations

N-y

x — n

Because the variables N, x, y, and n specify a combination of
attribute values, P(n) is the probability of that combination oc-
curring.

The manner in which the Checker uses this formula for P{n) to
check for evidence of stochastic dependence between attributes is
best illustrated by example.

Consider two attributes Ax and Ay, with known attribute val-
ues as shown in figure 3.5. 33 There are thirteen known pairs:
N = 13. There are six yess in Ax', x = 6. There are seven yess
in A Y \ y = 7. The formula for P(n) can be used to calculate
the probability that n = i for i = . . . N. Figure 3.6 lists these
probabilities in the column labelled "P(ii = i)".

144 A pragmatic legal expert system §3.1

There are five yes/ yes pairs in Ax and Ay: n = 5. Figure 3.6
shows that the probability of there being exactly five yes/yes
pairs (given the distribution of yess and nos in Ax and Ay, and
assuming random data) is 0.0734.

But. the probability of there being exactly five yes/yes pairs
is not of much use on its own. What is important is whether the
number of yes/ yes pairs is unusually high, or unusually low.

The "P(n < ?)" column gives the probability of there being i
yes/yes pairs or fewer: i.e. the cumulative total of the "P{n = i) n
column. The "P(?i > i)" column gives the probability of there
being i yes/yes pairs or more: i.e. the cumulative total of the
'^(71 = i)" column, summing backwards from the bottom. The
probability of there being five yes/ yes pairs or fewer is 0.9959. The
probability of there being five yes/yes pairs or more is 0.0775.

The method that the Checker uses to detect stochastic de-
pendence is analogous to using the x 2 ("chi-square" ) test for
independence, 34 except that that test assumes a normal distribu-
tion and. hence, a large sample whereas the Checker calculates
probabilities exactly and does not require a large sample. (SHY-
STER'S method is an example of what is called Fisher's exact
test. 35 )

It is necessary to choose a threshold of likelihood, below which
a given number of yes/yes pairs will be considered unusually high
or unusually low. When using the x 2 test, it is common practice to
use a threshold of 0.05. Using this threshold, the occurrence of five
yes/yes pairs in Ax and Ay is neither unusually high nor unusu-
ally low. This means that the combination of attribute values in
those attributes is not unusual: there is no evidence of stochastic
dependence between Ax and Ay.

Had there been six yes/yes pairs, that would have been unusu-
ally high because P(n > 6) = 0.0041. One or no yes/ yes pairs
would have been unusually low, because P(n < 1) = 0.0251.

For every pair of attributes, the Checker counts the actual num-
ber of yes/yes pairs and calculates the probability of there being
that number of yes/ yes pairs or fewer, and the probability of there
being that number of yes/yes pairs or more.

The Checker does not need to build a complete probability
table, like that in figure 3.6. in order to calculate these two prob-
abilities for each attribute pair. The probability of i yes/yes pairs
or fewer is P(n < i) = J2' n=0 P(n), and the probability of?' yes/yes

§3.8 Implementing SHYSTER 145

pairs or greater is P(n > i) = 1 — Zl^lo P{ n )- So. -P(^) needs only
to be calculated for n = . . . ?'.

The Checker does not use the formula for P(n) directly when
calculating probabilities. Because n > max(0, x + y — JV), the first
non-zero probability is either -P(O) or P(x+y—N). By substitution
into the formula for P(n).

and

P( x + y -N)= X ' y ' x + y>N.

N)(x + y- N)\

The Checker starts with the first non-zero probability and suc-
cessively multiplies it by P(i+ 1)/P(i): by substitution.

P(t+1) (i-x)(i-y)

P(i) (i+l)(N-x-y+i + l)'

This is a more efficient method of calculating each probability than
applying the formula for P(n) directly.

P(i+ 1)/P(i) becomes zero when i = x or i = y. so all values of
P(i) for i > min(x. y) are zero. This limit on i is never exceeded
because n < min(x,£/).

3. 8. 2 Writing matrices of probabilities

Probability figures are written, in matrix form, to SHYSTER's
probabilities file. Each cell in this matrix has two figures: 30

1.00
0.08

probabilit}' of the actual number of yes/yes pairs or fewer;
probability of the actual number of yes/yes pairs or more.

If a probability figure is not above the threshold of 0.05 it is
marked with a • symbol and a message is written to the log file to
warn the legal expert that there is evidence of stochastic depend-
ence. If there is an equivalence function or an inverse function
mapping one attribute to the other, the first probability figure is
marked with a ■ symbol and a warning message is written.

An extract from the probabilities matrix for the Employee area is
given in figure 3.7; the complete probabilities file is in §B.4. There
is functional dependence between attributes A 4 and A- . There is
evidence of stochastic dependence between attributes A 7 and A 9 .

146 A pragmatic legal expert system

§3.9

A 3

A 4

-4a

A 7

A 9

Aio

0.05
LOO

1.00
0.15

0.30

0.95

0.95
0.30

1.00
0.00

0.96

0.28

0.09
0.80

0.41
0.91

0.62
0.85

0.45

1.00

0.90
0.50

0.50
0.90

0.27
0.97

0.09
1.00

1.00
0.58

1.00
0.19

1.00
0.45

0.50
0.90

0.90
0.50

1.00

0.23

0.77
0.77

0.58
0.89

0.56
0.88

1.00
0.45

A 3

O.OOh
1.00

0.50
0.87

0.50
0.88

0.88
0.56

1.00
0.08

1.00
0.10

A A

0.87
0.50

0.88
0.50

0.56
0.88

0.08
1.00

0.10
1.00

0.99
0.12

0.80
0.71

0.22
0.98

0.73
0.77

0.58
0.88

0.02*
1.00

0.50

0.91

1.00
0.16

0.89
0.58

0.88
0.56

A s

Figure 3.7: An extract from the probabilities matrix for the
Employee area (SHYSTER output). The area has 18 attributes;
only 45 of the 153 cells in the complete matrix are shown here.

3.9 The Scales module

The Scales module assigns weights to attributes and writes, to
the weights file, a table of weights for each area in the specification.

•3. 9. 1 Calculating weights

For each attribute, the Scales module calculates a weight and
some result weights (one for each result).

The weight is the inverse of the variance 37 and it ranges from 4
to infinity. 38 A warning is issued if an attribute has infinite weight.

A result weight is the inverse of the variance of the attribute
values for that result. Result weights are used to calculate the
strength of attribute directions because the attribute directions
represent the extent to which the attribute values of the instant
case suggest a given result. 39

§3. 1 1 Implementing SHYSTER 141

3.9.2 Writing" tables of weights

The Scales module writes to the weights file a table of weights for
each area in the specification. The weights file for the Employee
specification is given in §B.5. Each table has columns for the
mean ft, variance a 2 and weight w of each attribute for each result,
and for the attribute as a whole. The mean column for each result
is that result's centroid. 40

3.10 The Adjuster module

The Adjuster module allows any of the attribute weights (in-
cluding result weights) in the current area to be set to any desired
value. This feature is intended for use by the legal expert — not
the user, who need not be aware of the manner in which SHYSTER
weights its attributes. Using the Adjuster, the legal expert can
test the effect of changing weights during the development of a
specification.

The Adjuster is not invoked unless weight adjustment is en-
abled using the -a switch on the command line. If a weight is
changed, an adjusted weights file is written for the appropriate
area.

3.11 The Consultant module

The Consultant module determines the attribute values for the
instant case in the current area. If an attribute is external, the
Consultant invokes the Cases module again and assigns the at-
tribute a value on the basis of the result identifier that is returned.
If an attribute is local, the user is interrogated as to the attribute
value.

At present, the Consultant uses a simple scrolling prompted
dialogue. If the user has enabled echo mode (using the -e switch on
the command line) then the Consultant echoes the user's input
by writing the appropriate attribute string to the terminal. This
ensures that the user understands the meaning of the attribute
value that she/he has entered. SHYSTER's modular design is such
that any desired interface could be employed by changing just this
module.

148 A pragmatic legal expert system $3.12

3.12 The Odometer module

The Odometer module performs all of SHYSTER's distance cal-
culations, and writes tables of distances to the distances file.

3. 12. 1 Calculating distances

As explained in the previous chapter, 41 for SHYSTER each of the
commonly used similarity measures reduces to one of six measures,
and a variant of the weighted distance metric d'- k was chosen. All
six similarity measures are summarized in figure 3.8.

The Odometer calculates the distances (known and unknown)
between the instant case and the leading cases, the ideal points,
and the centroids. It also calculates values for the extra similarity
measures, and the strength of the attribute directions. Using the
known and unknown distances it determines which are the nearest
cases and the nearest result.

If necessary, it resolves equidistant results by reference to the
rank of the courts which decided the nearest cases, and the recent-
ness of those cases. A warning is issued if equidistance has to be
resolved in this fashion.

SHYSTER uses a precision threshold of two decimal places in all
of its comparisons. 42 This allows for the possibility of rounding
errors having been introduced in SHYSTER's arithmetic, and re-
cognizes the danger of relying too much on precise quantification
of abstract notions.

The Odometer is invoked for the instant case, and for each
instantiation and each hypothetical. 43 It treats each instantiation
or hypothetical as if it were the instant case for the purposes of
performing its calculations.

3.12.2 Writing tables of distances

The Odometer writes, to the distances file, a table of distances
for the instant case, each instantiation, and each of the chosen
hypothetical. 44

A distances file for the Employee area is given in full in §B.6. An
extract from that file — the table of distances for the uninstantiated
and unhypothesized instant case — is given in figure 3.9. 4D The
instant case is Building Workers' Industrial Union of Australia v.
Odco Pty Ltd, 46 which is discussed in detail in the next chapter. 4 '

§3.12 Implementing SHYSTER 149

distance

measure:

djk

= Aj-a-

weighted

distance

= H| ^ ~ A ^\ X W *

measure:

2=1

association
coefficient:

Sjk

weighted
association

s jk

n 1 1

J] \Aij -Aik\ x «-,-

2=1 ' '

coefficient:

2=1

correlation
coefficient:

r i* =

weighted ^ ^J

2=1

£ (i4ij x u-i - A'A (Aik x tt-i - A',)

correlation

coefficient: / ^ ( x ^ _ A tf ^ (^ x „, . _ ^ ) '■

V 2=1 * = 1

Figure 3.8: Measures of similarity between a case j and a
case k (discussed in §2.9.1 and §2.9.2).

Aj-/j is the number of differences in the corresponding attribute
values of the two cases; n is the number of attributes; Aij is the

value of the ith attribute for the jth case; Wi is the weight of
the ith attrib
jth case; and
the jth case.

the ith attribute; Aj is the mean of all attribute values for the
jth case; and A - is the weighted mean of all attribute values for

The table in figure 3.9 includes an attribute value matrix, similar
to the one in figure 3.4. with the addition of the instant case C Instant
and a centroid for each result: /^Employee an d /^Contractor- As before,
the rank of the court that decided each case is in the column

150 A pragmatic legal expert system

§3.12

o "d

-"'".

CO CD

z
CO

—

■-

CO
CO

- 1

+
8

-11

CO
CO

5
LU

-II

*ft

a
u

-J\

-S3

^ .2
o

.—I

c-, co

CO CD

o CI

«3

■H

O CI

—

CI o

—

C» CI

—

-T

«o

r£2 O

d d

d ci

;

■'-

00 CI

1^ CI

«o

CO CO

«o

—

■go

O CI

—

—t CI

Ct CI

—

■^

»o

a .

o o

1 1

43 5 s

G5 00

o o

e»

- 1

CO >o

t-l *-H

CI -V

-J. ~f

--.

—

C< CI

!—■

—

o3 ^.

;

d d

;

d d

;■

-r co

o ~v

—

rH iH

•o

•v

—

S cq

-T »0

:-".

■* ■*

—

CO CO

■*

;

d d

;■

d d

1- CI

■* N

-"•".

io >o

•v

."-".■

ai co

.-:

CO CO

.-"

■H

~~ m

—

ct ■-<

:■

■-I i-H

i-i

—

3 .9

;

d d

—

;

» d

d d

t °

o ™

"-•

iH iH

— '

—

CD — 1

—

i-H i-H

■"•

—

I-H

— ,

*3 i-H

^ rH

" 1

— 1 fO

;■

—

O) CO

00 CO

:•'.

co o

-'.

I-H

■<*

d d

so ci

- 1

•p co

—

CO »o

-":

— T

fr-

.'-:

—

CI CI

—

1-1

CI T

«o

1- GO

-".

CO >o

■i;

•
X

X
X

X X
X X

X
X

•

X X
X

•

■
■

-■

X X
• X

•

•
X

•

•
X

•

CO M

X X

•

•-

■

X X

•

8 6^

•

X X

■

•

•-

■-■

X •

o-s

. M

» Q>

■i;

X X

•

■-■

X X

to^

^ o

■i;

X •

X X

•

•i*

•

X X

•

■-■

X X

•

■",

i-> ?3

1 "^

O CL>

•

X •

•

• •

•

■

X X

•

^ PI

k> °

QJ Q>

? *

X X

X •

•

■

■-■

X X

•

O ^

1o o

• X

X X

■

•

• •

•

■

X X

■-■

• X

•

•r-< JJ

T3 O

•

X •

•

■-■

• X

«4-l *-

■t;

•

X •

•

■■.

X •

•

■

X X

•

<D O

3 u

■^

•

• •

•

• X

•

■'■

X X

•

td »

J-3 "-•

■t;

•

X X

•

X •

•

• *

•

<l> ii

■t;

•

• •

•

• •

•

■-■

X •

•

•t;

•

• X

•t;

• •

•

X •

•

■-■

X X

•

Oi 40

CO CD

i-i CO

* *- -s

I E

Q) £ •

H CL> ^

-_ - t

£ 6"

o 1

■~

*■•

-"-

E5.3

§3. 1 3 Implementing SHYSTER 1 5 1

The known and unknown distances are in columns labelled "r/[<"
and "du"- Unweighted distance measures are in the column la-
belled "A" — not "d", to avoid confusion with the known and un-
known distances. Values for the other similarity measures (5, S',
r and r') are labelled appropriately. 48

The strength of each non-zero attribute direction is displayed
in the "Result" column. A => symbol indicates specified direction;
a => symbol indicates ideal point direction; a => symbol indicates
centroid direction. 49 Each number represents the strength of the
direction towards that result.

3.13 The Reporter module

The principal output from SHYSTER is contained in its report
file, written by the Reporter module. This module also imple-
ments SHYSTER's safeguards and writes information about them,
together with a general summary of its report, to the log file.

In each area, the Reporter is invoked once for the instant case,
and once for each instantiation and for each chosen hypothetical. 50
Only if an instantiation has a different result to that of the instant
case does the Reporter write a report on that instantiation.

A report file for the Employee area is given in full in §B.7. Again,
the instant case is BWIU v. Odco. Except where otherwise indic-
ated, the examples in this section are taken from that report file.
Other example report files are given in appendix A.

3.13.1 Arguing with the instant case

The Reporter argues that the result in the instant case will be
the nearest result, then builds a counterargument for each of the
nearest other results. This same process is also used to argue with
instantiations and with hypotheticals (as discussed below).

Arguing for the nearest result

For the instant case, the Reporter starts by writing the opening
string to the report file. It then states the facts of the instant case
and declares its opinion as to the likely result — the nearest result:

In my opinion — following Humberstone v. Northern Timber Mills—
the worker is an independent contractor.

152 A pragmatic legal expert system §3-13

Humberstone v. NTM (C 8 in figure 3.9) is the nearest neighbour.
It is also the nearest unknown neighbour. There are no nearest
known cases; all of the leading cases have some unknown distance
because of the two unknowns in the instant case.

The Reporter then summarizes the nearest neighbour: 01

In Humberstone v. Northern Timber Mills., 52 a 1949 decision of
three justices of the High Court of Australia, Humberstone carried
goods for NTM. He had originally held himself out as a carrier,

If the case is identical to the instant case^ — if it has the same at-
tribute values — then the Reporter announces that the two cases
are "on all fours." ; If there is no known distance between the
cases but there is some unknown distance, it declares that the Wo
cases "may be on all fours" and explains its reservation by refer-
ence to the unknown attributes. In this example there are some
known differences. First, the Reporter lists all of the similarities:

Three or more similarities are characterized as "significant"; two
similarities are "very significant"; a solitary similarity is called
"extremely significant." as it must be for such a dissimilar case to
be the nearest neighbour.

Then the Reporter lists all of the differences: 54

However, the instant case is not on all fours with Humberstone v.
NTM. In that case the worker was not in business on her/his own
account; the worker was allowed to employ others to assist with
her/his work; the worker was not required to work at specified
times; the employer did not pay the worker by time; and the
employer and the worker did not express any intention that the
relationship would be one of principal and independent contractor.

And if there are any unknowns in the leading case — there are none
in this example — then they are explained too.

Despite the differences, SHYSTER stands by its original state-
ment:

Nevertheless, I believe that Humberstone v. NTM should be fol-

1< iwed.

§3. 1 3 Implementing SHYSTER 1 5 3

If there are equidistant nearest neighbours they are all handled
in this fashion, with a linking paragraph written between each
pair of cases. This paragraph explains which case is the more
important, and why. For example, the report on the second hypo-
thetical for BWIU v. Odco includes the following paragraph which
introduces a discussion of Ready Mixed v. Minister, having just
discussed Australian Mutual Provident Society v. Chaplin: 55

In 1967, Ready Mixed Concrete (South East) Ltd v. Minister of
Pensions and National Insurance 5 * 5 was decided by the Queen's
Bench Division of the English High Court. (A case decided by
the Queen's Bench Division of the English High Court is not as
good authority as a case decided by the Judicial Committee of the
Privy Council — like AMP v. Chaplin; furthermore Ready Mixed v.
Minister is 11 years older than AMP v. Chaplin.)

Arguing for other results

For every other result, the Reporter discusses the nearest other.
It explains the effect of following the nearest other instead of the
nearest neighbour, then summarizes the nearest other:

If Ferguson v. John Dawson and Partners (Contractors) Ltd is
followed then the worker is an employee.

In Ferguson v. John Dawson and Partners (Contractors) Ltd 57
a 1976 decision of the English Court of Appeal, Ferguson fell off a
roof while removing some scaffolding boards. He claimed damages
against Dawson (the building contractors) for breach of statutory
duty relying on ...

Once again all of the similarities are listed, but this time the dif-
ferences are decisive:

However, there are several significant differences between the in-
stant case and Ferguson v. Dawson. In that case the employer
directed the manner in which the work was to be done; the worker
was not allowed to use her/his own discretion in doing an aspect
of the work that was not specified beforehand; the worker was an
integral part of the employer's business; . . .

(Like the similarities in the nearest neighbours, the differences in
the nearest others are characterized as "significant," "very signi-
ficant" or "extremely significant," depending on their number.)

154 A pragmatic legal expert system §3.13

The Reporter also compares the importance of this case with
that of the nearest neighbour. 58 before reiterating its opinion:

Note also that Ferguson v. Dawson is only a decision of the English
Court of Appeal and not as good authority as a case decided by
three justices of the High Court of Australia — like Humberstone v.
NTM.

Consequently, there is nothing in Ferguson v. Dawson to war-
rant any change in my conclusion.

The Reporter is not swayed from its conclusion by the import-
ance of the nearest other. For example, in the report on the first
hypothetical for BWIU v. Odco it writes:

Despite the fact that Cam v. Sargent is a decision of four justices
of the High Court of Australia (and better authority than a case
decided by three justices of the High Court of Australia — like
Humberstone v. NTM), there is nothing in Cam v. Sargent to
warrant any change in my conclusion.

Equidistant nearest others are handled in a similar fashion to
that used for equidistant nearest neighbours. The report on the
instant case concludes with the closing string.

Variations

Although the Reporter relies heavily on the strings supplied by
the legal expert in the specification, each report is more than a
mere regurgitation of those strings. The structure of each report
varies depending on the circumstances. For example, where a case
would have been the nearest case if not for its unknown distance,
the Reporter makes this clear — without any mention of unknown
distance. This example is taken from another report file: 59

I would have suggested that Stevenson v. Macdonald (1) be fol-
lowed (instead of Massey v. Crown Life) except that it is not
known whether the employer supervised or inspected the work; it
is not known whether the employer paid the worker by time; it
is not known whether the money that the employer paid to the
worker was stated to be a "fee"; . . .

Different situations are handled in different ways at several stages
in the process of preparing the report. For this reason, there are
many different possible reports, even ignoring the difference in case
names, summaries, and other strings. This means that SHYSTER'S

§3. 1 3 Implementing SHYSTER 1 5 5

reports read quite well. Although their style is a little stilted, it
is just possible that they could be mistaken for the work of a
lawyer — -although that was not one of the aims of the SHYSTER
project.

3.13.2 Arguing with instantiations

For the purposes of writing reports on instantiations, the Report-
er treats each instantiation as if it were the instant case and
follows the steps for arguing with the instant case outlined above. 00
The only differences are in the introductory comments, and in the
fact that the instant case is referred to as the "instantiated case."
For example, a report on an instantiation from another test case
commences as follows: 01

It may be that the following is true of the instant case: the em-
ployer would not make a profit/loss if the work performed by the
worker cost less/more than expected; and the employer neither
supervised nor inspected the work.

If that is so then in my opinion — following Ready Mixed Con-
crete (South East) Ltd v. Minister of Pensions and National In-
surance — the worker is an independent contractor.

3.13.3 Arguing with hypotheticals

As with instantiations, to write a report on a hypothetical the
Reporter treats that hypothetical as if it were the instant case
and follows the steps for arguing with the instant case outlined
above. 02 Again, the introductory comments are different, and the
instant case is referred to as the "hypothetical case."

The report on the first hypothetical in the report file for BWIU v.
Odco starts with these words: 03

Consider the instant case changed so that the following is true: the
worker was allowed to employ others to assist with her/his work;
and the employer and the worker did not express any intention
that the relationship would be one of principal and independent
contractor.

If that were so then I would be more strongly of the opin-
ion that — following Humberstone v. Northern Timber Mills— the
worker is an independent contractor.

156 A pragmatic legal expert system §3.13

If the hypothetical variations lead SHYSTER to a different conclu-
sion, as in the second hypothetical, the words are different again:

Consider the instant case changed so that the following is true:
the worker was not allowed to use her/his own discretion in doing
an aspect of the work that was not specified beforehand; and the
worker was an integral part of the employer's business.

If that were so then my opinion would be that — following
Ferguson v. John Dawson and Partners (Contractors) Ltd — the
worker is an employee.

3. 13.4 Safeguards

As well as constructing its reports, the Reporter implements
the safeguards explained in the previous chapter. 04 It determines
which would be the nearest neighbours applying, in turn, each of
the extra similarity measures: d^, Sj k , S'- k , r^ k and r'. h . If these
measures suggest different nearest neighbours from those chosen
by SHYSTER using its known and unknown distance, that fact is
noted in the log file.

Figure 3.10 is extracted from the log file for BWIU v. Odco: the
complete log file is in §B.2. The fact vector for the instant case
is given, then the short case names of the nearest neighbour and
the nearest other. All of the extra similarity measures suggest the
same nearest neighbour; there are no safeguard warnings. The
nearest result is declared to be Contractor (this is the identifier
that is returned by this invocation of the case-based system).

The fact vector for the first instantiation is given, and the dif-
ferences between it and the instant case are marked with carets.
In this instantiation, the nearest neighbour and nearest other are
the same as in the uninstantiated instant case. However, some
of the extra similarity measures now disagree with the choice of
nearest neighbour. These disagreements are set out under the
"Safeguards" heading. GD Extra cases which the extra measures
suggest as the nearest neighbour are marked with a +; if an extra
case has a different result to that of the instantiation, it is marked
with a * and its result appears in parentheses after the case name.
If an extra measure does not suggest one of the nearest neighbours.
the case name is marked with a -.

§3. 1 3 Implementing SHYSTER 1 5 1

Fact vector is (NYNYNNNUNNYUNNNNNY) .

Nearest neighbours :

Contractor :

C8 Humberstone v. NTM

Nearest others :

Employee :

C5 Ferguson v. Dawson

Nearest result for the instant case is Contractor.

Instantiation 1 is (NYNYNNNYNNYYNNNNNY) .

A A
Nearest neighbours :

Contractor :

C8 Humberstone v. NTM

Nearest others :

Employee :

C5 Ferguson v. Dawson

Safeguards :

Distance measures :

* C6 Stevenson v. Macdonald (2) (Employee)

+ Cll AMP v. Chaplin

+ C13 Stevenson v. Macdonald (1)

Association coefficients :
+ Cll AMP v. Chaplin

Correlation coefficients :

- C8 Humberstone v. NTM
+ Cll AMP v. Chaplin

Weighted correlation coefficients:

- C8 Humberstone v. NTM
+ Cll AMP v. Chaplin

Nearest result for instantiation 1 is Contractor.

Figure 3.10: An extract from the log file for BWIU v. Odco
(SHYSTER output).

158 A pragmatic legal expert system §3.13

All of these differences are logged, but only some of them will
cause a warning to be issued. If one or both of the weighted safe-
guard measures suggest that a case with a different result should
be the nearest neighbour then a warning is issued. In this example,
only the (unweighted) distance measures suggest a different result,
so no warning is issued.

The Reporter also writes to the log file if the ideal points,
centroids or any of the three attribute directions (specified, ideal
point or centroid directions) suggest a different result. However, a
warning is not issued unless an ideal point or a centroid with a dif-
ferent result is at least as near to the instant case (or instantiation
or hypothetical) as is the nearest neighbour, or if the specified
directions suggest a different result. Warnings are written to the
log file and to the standard error stream.

In summary, SHYSTER will issue a warning in each of the fol-
lowing circumstances:

• the weighted association coefficients suggest that a case, with
a different result to that of the nearest neighbour, ought to
be the nearest neighbour;

• the weighted correlation coefficients suggest that a case, with
a different result to that of the nearest neighbour, ought to
be the nearest neighbour;

• an ideal point suggesting a different result is at least as near
to the instant case as is the nearest neighbour;

• a centroid suggesting a different result is at least as near to
the instant case as is the nearest neighbour; or

• the specified directions suggest a different result or results.

As explained in the previous chapter, 00 SHYSTER will also issue a
warning if:

• there are two or more equidistant results.

And. as also explained in the previous chapter, ' a warning is
issued if:

• an instantiation of the instant case has a different result to
that of the uninstantiated instant case.

§3. 14 Implementing SHYSTER 159

3.14 Conclusion

SHYSTER is a working implementation of the pragmatic approach
to case law adopted, and justified, in the previous chapter. It
has been designed to allow a legal expert to specify areas of case
law using a simple specification language. SHYSTER'S case-based
system has been constructed so as to facilitate its linking with a
rule-based system to form a hybrid legal expert system.

Within the limitations imposed by its internal representation of
case law. SHYSTER is capable of producing sophisticated reports
on the likely outcome in user-specified instant cases. In the process
of constructing a report, SHYSTER produces several intermediate
files containing information of use to the knowledge engineer and
the legal expert.

The next chapter discusses the specification of areas of law and
the use of those specifications, and various testing methods, to
evaluate SHYSTER and its approach to case law.

Notes

1 Galsworthy 1941 at 36.

2 Lucas 1935 at 394. Letter to Samuel Rogers.

3 Paramount. Screenplay by Arthur Sheekman.

4 International standard ISO/IEC 9899: 1990; Australian standard AS
3955-1991. Kernighan and Ritchie 1988 describe ANSI C which is the same
as ISO C. Popple 1993b lists the code for each module in full.

5 §4.4.

6 250 characters and 16 characters, respectively.

7 UNIX's stderr stream.

8 Lamport 1986 describes L^TgX which is a set of macros for Knuth's
TgX system (1984). This book was prepared using J^TgX. SHYSTER/s I^TgX
output is suitable for processing by L^TgX version 2.09 (25 March 1992) and
TgX version 3.141. It can also be processed, in "compatibility mode," by
L^TgX 2g which is described by Lamport 1994.

9 The SHYSTER output that appears in the appendices to this book was
produced using the -i switch which causes SHYSTER to write L^TgX code
suitable for inclusion in another I^TgX document.

10 UNIX's stdout stream.

11 A string, as defined in figure 3.3, includes the enclosing quotation
marks. The TOKENIZER discards these quotation marks when the specific-
ation is read. Hence, the word string is used in two different ways in this
description of SHYSTER: before being tokenized, a string includes the en-
closing quotation marks; once internally represented by SHYSTER, a string is
only those characters which appeared between the enclosing quotation marks
in the specification file.

160 A pragmatic legal expert system

12 A string may contain commands which will be processed by LMgX
if that string is included in one of SHYSTER's files. LMgX commands are
prefixed with a \ character. The $, & and % characters have special meanings
in IATgX; if one of these characters appears in a string, SHYSTER converts it
into the command that produces the appropriate character: \$. \& or \L.

13 As with strings (see the note before last), attribute vectors are defined
in figure 3.3 to include the enclosing parentheses, but the TOKENIZER discards
these parentheses when the specification is read.

14 Whitespace is any sequence of spaces, tabs, vertical tabs, carriage re-
turns, or form feeds.

15 SHYSTER's is an LL(1) parser.

16 EBNF is described by Wirth 1977.

17 Note that the specification of an external attribute requires the use of
result identifiers from another area, as explained in §3.6.3.

18 This is A\i from the Employee area of the EMPLOYEE specification.

19 This is A 4 from the Affected area of the NATURAL specification.

20 The external area need not appear in the specification before it is linked
to an external attribute; forward references are allowed, and are not checked
until invocation. Recursive references (circular definitions) are not allowed
but SHYSTER does not detect them, unless the reference is to the area in
which the attribute is specified.

21 The specification of an UNKNOWN string in this example is not strictly
necessary; no result identifier returned by the Expectation area can produce a
value of UNKNOWN in this attribute. However, if it is not specified and one of
the leading cases or ideal points has a value of UNKNOWN for this attribute,
the DUMPER module will issue a warning.

22 Result identifiers are not checked for validity, so the legal expert should
ensure that these identifiers match result identifiers in the appropriate external
area. If the same result identifier is specified for two or all of the possible
attribute values, it is linked to that first specified (the order being YES, NO
then UNKNOWN).

23 The order in which cases are specified is unimportant. SHYSTER
groups cases by their result, and orders the cases in each group according
to the importance of the court in which each case was decided.

24 See, for example, Moffatt v. Kazana [1969] 2 QB 152 in the Finder area
of the FINDER specification, and Ready Mixed Concrete (South East) Ltd v.
Minister of Pensions and National Insurance [1968] 2 QB 497 in the Employee
area of the EMPLOYEE specification. Both cases were decided in 1967.

25 This warning is actually issued by the CASES module, not the PARSER,
but it is convenient to mention it here.

26 This happens in the NATURAL specification (§4.5.2), but note the dis-
cussion there as to why that specification was not changed.

27 I found that this sort of graphical representation conveys its informa-
tion more clearly than does one using Ys, Ns and Us, or ones and zeros.

28 As with the example specification of a local attribute in §3.6.3, this is
A12 from the Employee area of the EMPLOYEE specification.

29 If an attribute value is directed towards more than one result, result
identifiers are displayed separated by a V symbol.

Implementing SHYSTER 1 6 1

30 As with the example specification of an external attribute in §3.0.3,
this is A4 from the Affected area of the NATURAL specification.

31 (1945) 70 CLR 539.

32 §2.8.

33 These attributes are A4 and A$, respectively, from the Employee area
of the EMPLOYEE specification (see figure 3.4).

34 Conover 1971 at 154-8.

35 Fisher 1970.

36 This cell contains the probabilities for the example attribute values in
figure 3.5.

37 Because all of SHYSTER's known attribute values are assigned numer-
ical values of or 1, the variance <r/ (defined in §2.7) simplifies to Ai — Ai .
SHYSTER uses this simplified formula to calculate the variance.

38 If an attribute has no known attribute values (i.e. each of its values is
UNKNOWN) it is given no weight and a warning message is issued. A further
warning is issued (by the ODOMETER) if the instant case or an ideal point
has a known attribute value for the weightless attribute.

39 §2.12.4.

40 §2.12.3.

41 §2.9.3.

42 In addition to the precision threshold that it uses in its comparisons,
SHYSTER has a second precision threshold within which it considers two cases
to be equidistant. Both of these thresholds are the same — all comparisons
are precise to two decimal places — but they need not be; by changing one
constant, SHYSTER can be rebuilt with a threshold for deciding when two
cases are equidistant that is different to the threshold that it uses in its other
comparisons.

43 The ODOMETER is invoked twice for some hypotheticals. The first time
it performs its calculations without writing anything to the distances file. For
those hypotheticals that are chosen to be reported upon, the ODOMETER is
called a second time to perform the calculations again and to write details to
the distances file. Only a fraction of the hypotheticals are chosen: performing
the calculations twice for the chosen hypotheticals avoids having to store the
results of the calculations for all hypotheticals.

44 How hypotheticals are chosen is explained in §2.11.3.

45 Tables of distances extracted from other distances files can be found
in figures 4.3. 4.6, 4.7 and 4.11.

46 (1991) 29 FCR 104.

47 §4.4.3.

48 If one of the attributes is infinitely weighted, the values obtained for S'
become meaningless because both its numerator and its denominator are in-
finite; in this event, values for S' are not written to the dump file, and S' is
not used as a safeguard. Calculation of the weighted correlation coefficient r
is complicated when one of the attributes is infinitely weighted; if this hap-
pens, a very large weight is used in place of an infinite one. If either of two
cases has all attribute values the same, the correlation coefficients r and r'
are meaningless and are ignored for safeguard purposes.

49 If all three directions for a given result are of the same strength,
a =$> symbol and a single number appear in their place.

162 A pragmatic legal expert system

50 How hypotheticals are chosen to be reported on is explained in §2.11.3.

51 Cases are not summarized more than once in the same report — and
not summarized at all if the user has enabled quiet mode (using the -q switch
on the command line). In quiet mode, opening and closing strings are not
written either.

52 (1949) 79 CLR 389.

53 The rather quaint expression "on all fours" is used by FINDER (see,
for example, Tyree 1989 at 160). Although this expression may seem twee,
it is preferable to "the same,'' "identical," etc., because two cases are never
completely identical. Its meaning is clear from the context, and its oddness
draws the reader's attention to the fact that something less than sameness is
meant.

54 This extract illustrates one of SHYSTER's limitations: its inability to
distinguish between differences and distinctions, discussed in §5.2.4.

Ijlj (1978) 18 ALR 385. This paragraph actually links the discussion of
two equidistant nearest others, but paragraphs of the same structure are used
to link equidistant nearest neighbours.

56 [1968] 2 QB 497.

57 [1976] 1 WLR 1213.

58 If there is more than one nearest neighbour, the comparison is made
with the most important nearest neighbour.

59 This quotation is taken from the report file for Narich Pty Ltd v. Com-
missioner of Pay-roll Tax (1983) 50 ALR 417 which is used as a test case
in §4.4.3.

60 The opening and closing strings are not written again when arguing
with instantiations.

61 This quotation is taken from the report on the fourth instantiation
in the report file for Re Porter; Re Transport Workers Union of Australia
(1989) 34 IR 179 which is given in full in §A.4. Re Porter is used as a test
case in §4.4.3.

62 As with instantiations, the opening and closing strings are not written
again when arguing with hypotheticals.

63 See §5.2.4 for an explanation of how this extract illustrates SHYSTER's
inability to distinguish between differences and distinctions.

64 §2.12.

65 "Weighted association coefficients'" does not appear as a subheading
in this example because those coefficients agree with the choice of nearest
neighbour.

66 §2.12.5.

67 82.12.6.

Case studies

. . . the notion that a computer can predict the course of judicial decision rests
on assumptions that are demonstrably untenable, does violence to the very
nature of law, and is moreover certain to blunt the professional techniques of
any lawyer who relies on machines rather than on his own powers of reasoning
and advocacy.

Frederick Bernays Wiener (1962)
Decision Prediction by Computers: Nonsense Cubed — and Worse

The Law is the true embodiment
Of everything that's excellent.
It has no kind of fault or flaw,

And I, my Lords, embody the Law. William Gilbert (1882)

Iolanthe; or The Peer and the Peri

Great cases, like hard cases, make bad law.

Oliver Wendell Holmes (1904)
Northern Securities Co. v. United States

163

164 A pragmatic legal expert system §-&■!

4.1 Introduction

SHYSTER is unusual amongst legal expert systems in that it has
not been developed specifically for a given area of the law. Instead,
its case-based system has been designed to allow the specification
of different areas of law.

In order to test SHYSTER, and its approach to case law, four
specifications were written, each representing an area of Australian
law: an aspect of the law of trover (the Finder specification); the
meaning of "authorization" in copyright law (the Authorization
specification); the categorization of employment contracts (the
Employee specification); and the implication of natural justice
in administrative decision-making (the Natural specification).

These specifications are discussed in |j 4.2 §4.5. 4 For each spe-
cification, the law is briefly stated, and the specification and its
testing are described. (The methods used to test each specifica-
tion, and SHYSTER, are explained in chapter 2. 5 )

The facts of cases — cases which are not part of the specification-
are given to SHYSTER, and SHYSTER's opinions are compared with
the judgments in those cases. Each of these tests is discussed,
individually, below. If a specification includes ideal points, those
ideal points are also used as test cases. The results of these tests
are summarized in figure 4.12 near the end of this chapter, and
conclusions are drawn from them in §4.6.1.

Generated tests (explained in chapter 2°) are performed for each
specification for which legal expertise was available: i.e. all but the
Finder specification. Each generated test is also discussed below,
and summarized in figure 4.13 at the end of this chapter. Conclu-
sions are drawn from the results of these generated tests in §4.6.2,
and from the reflexive testing (also explained in chapter 2 7 ) of all
four specifications in §4.6.3.

Conclusions are drawn from all of this testing in the next chapter.

4.2 A simulation of FINDER

FINDER is a case-based legal expert system. 8 Because SHYSTER's
approach to case law expands upon that of FINDER, a specification
can be written which causes SHYSTER to simulate FINDER.

The simulation does not make use of all of SHYSTER'S features.
However, SHYSTER'S safeguard mechanisms, and the opportunity
to perform reflexive tests, mean that FINDER \s approach to case
law can be tested more extensively using SHYSTER than is possible
using FINDER itself.

§4.2 Case studies 165

4.2.1 The law

The law of trover — the law concerning the rights of the finders of
lost chattels — -is unusual in that it is based entirely on cases. The
1722 case of Armory v. Delamirie 9 established that the finder of
any article which has been lost has a general right to that article as
against all the world except the true owner. 10 Only a few reported
trover cases have been decided in the centuries since. Although
they are few, these cases have qualified considerably the general
rule stated in Armory v. Delamirie, and the law of trover has
become complicated.

In 1948. Lord Goddard CJ of the King's Bench Division of
the English High Court referred to several trover cases (Bridges
v. Hawkesworth, 11 Elwes v. Brigg Gas Co.. 12 South Staffordshire
Water Co. v. Sharman 13 and Hannah v. Peel 14 ) and said:

These cases, or rather the first three of them, have long been the
delight of professors and text writers, whose task it often is to
attempt to reconcile the irreconcilable. It is, however, right to say
that in recent years both the Corpus Professor of Jurisprudence
at Oxford and the Professor Emeritus of English Law at Cam-
bridge have expressed the opinion that Bridges v. Hawkesworth
was wrongly decided. If it was, the difficult}'' largety disappears.
But that much-battered case has lately been re-invigorated by
Birkett J's decision in Hannah v. Peel, and I am glad to think
that ... it is still for wiser heads than mine to end a controversy
which will no doubt continue to form an appropriate subject for
moots till the House of Lords la3 f s it to rest for all time. 15

Although wiser heads than Lord Goddard's have not yet laid
it to rest for all time. Donaldson LJ made what has been called
u an unusually clear statement of the principles to be applied" IG in
the 1981 case of Parker v. British Airways Board. 1 ' He listed the
following rights and obligations of the finder:

1. The finder of a chattel acquires no rights over it unless (a) it
has been abandoned or lost and (b) he takes it into his care and
control.

2. The finder of a chattel acquires very limited rights over it if
he takes it into his care and control with dishonest intent or in
the course of trespassing.

3. Subject to the foregoing and to point 4 below, a finder of a
chattel, whilst not acquiring any absolute property or ownership
in the chattel, acquires a right to keep it against all but the true

166 A pragmatic legal expert system §-&-2

owner or those in a position to claim through the true owner or
one who can assert a prior right to keep the chattel which was
subsisting at the time when the finder took the chattel into his
care and control.

4. Unless otherwise agreed, any servant or agent who finds a
chattel in the course of his employment or agency and not wholly
incidentally or collaterally thereto and who takes it into his care
and control does so on behalf of his employer or principal who
acquires a finder's rights to the exclusion of those of the actual
finder.

5. A person having a finder's rights has an obligation to take
such measures as in all the circumstances are reasonable to ac-
quaint the true owner of the finding and present whereabouts of
the chattel and to care for it meanwhile. 18

His Lordship then listed the following rights and obligations of the
occupier of the premises where the chattel is found:

1. An occupier of land has rights superior to those of a finder
over chattels in or attached to that land and an occupier of a
building has similar rights in respect of chattels attached to that
building, whether in either case the occupier is aware of the pres-
ence of the chattel.

2. An occupier of a building has rights superior to those of a
finder over chattels upon or in, but not attached to, that building
if, but only if, before the chattel is found, he has manifested an
intention to exercise control over the building and the things which
may 1"' upon it or in it.

3. An occupier who manifests an intention to exercise control
over a building and the things which may be upon or in it so as to
acquire rights superior to those of a finder is under an obligation
to take such measures as in all the circumstances are reasonable
to ensure that lost chattels are found and, upon their being found,
whether by him or by a third party, to acquaint the true owner of
the finding and to care for the chattels meanwhile. The manifesta-
tion of intention may be express or implied from the circumstances
including, in particular, the circumstance that the occupier mani-
festly accepts or is obliged by law to accept liability for chattels
lost on his "premises," e.g. an innkeeper or carrier's liability.

4. An "occupier" of a chattel, e.g. a ship, motor car, caravan or
aircraft, is to be treated as if he were the occupier of a building
for the purposes of the foregoing rules. 19

It is clear that the question of who owns a found chattel remains,
in Lord Godclards words, "a really difficult area of law". 20

§4.2 Case studies 16'

4.2.2 The Finder specification

The Finder specification contains a single area: the Finder area.
This area does not make use of all of SHYSTER's features — it has
no opening string, no closing string, no help strings, no unknown
attribute values, no ideal points and no attribute direction—
because FINDER does not have these features.

Only two features of this specification are not part of FINDER.
The first is a hierarchy of courts. This is not strictly necessary;
SHYSTER does not require that a hierarchy be specified. Specify-
ing a hierarchy allows SHYSTER to mention the court in which
a case was heard when it discusses that case. There are only
three courts in this hierarchy, representing two of the three divi-
sions of the English High Court: the Chancery division, and the
Kings/Queen's Bench division. Because all three courts are of
equal importance, the inclusion of the hierarchy does not affect
which cases SHYSTER chooses to justify its opinions.

The second feature of this specification that is not part of
FINDER is the inclusion of strings for the unknown values in all
attributes. This allows the user to answer unknown to any of the
attribute questions, and force SHYSTER to examine instantiations
of the instant case.

Results

The Finder area has two results: Win and Lose. These correspond
to FINDER'S two possible outcomes: "the finder wins" and "the
finder loses." The area — like FINDER — actually represents only
part of the law of trover: the resolution of conflict between the
finder of a chattel and another person who is not the true owner
of the chattel.

Attributes

There are ten attributes, whose questions are taken (slightly re-
worded) from the questions asked by FINDER:

Ax'. Was the finder the occupier of the premises where the chattel
was found?

A2'. Was the chattel attached to the land or premises where it was
found?

168 A pragmatic legal expert system §-&-2

A3: Was the other claimant (the non-finder) the owner of the
premises where the chattel was found?

A 4 : Was the other claimant the true owner of the chattel or did
she/he claim through the rights of the true owner?

A^: Did the finder hand over the chattel to the other claimant
after the finding?

Aq: Did one of the parties rely on the terms of an agreement made
with the other which purported to give her/him the right
to the chattel?

Aj: Was the finder a servant of the other claimant?

A&: Was the chattel hidden or in a position so as to be difficult
to find?

A 9 : Was an attempt made to find the true owner of the chattel
or, alternatively, was the chattel clearly abandoned?

Ai Q : Did either of the parties know of the existence of the chattel
prior to the finding?

Answering these questions should be fairly straightforward. Only
the use of the word "servant" in A 7 could cause difficulty. It
would seem to refer to an employee as opposed to an independent
contractor — a distinction which is the basis of the Employee spe-
cification described later in this chapter. 21 Indeed, Tyree suggests
that "servant" can be read as "employee" in FINDER. 22 However,
in one (and perhaps both) of the leading cases in which the value
of A 7 is yes, the finder was an independent contractor of the other
claimant. 23 Clearly "servant" has a broad meaning in FINDER; it
would probably be better to ask "Was the finder an employee of,
or independent contractor to, the other claimant?" However, in
the interests of an accurate simulation, the question for A 7 has
not been rephrased for the Finder specification.

Leading cases

The same cases used hy FINDER are used as leading cases in
the Finder area. As well as the 273-year-old case of Armory v.
Delamirie, 24 there are three nineteenth century cases — Bridges v.

§4.2 Case studies 169

Hawkesworth, 25 Elwes v. Brigg 26 and South Staffordshire v. Shar-
man 21 — and three twentieth century cases — Hannah v. Peel 28 City
of London Corporation v. Appleyard 29 and Moffatt v. Kazana. 30
All of these cases are English. Nevertheless, they represent the
law as it applies in Australia.

London v. Appleyard is used twice. In that case some work-
ers (including Appleyard) were employed by a firm which was
engaged by another firm (Yorkwin Investments Ltd) to perform
some construction work on premises leased by Yorkwin from the
ow r ner (the City of London). 31 The workers found some banknotes
during construction. These banknotes were claimed by Appleyard
and his colleagues, by Yorkwin, and by the City of London. Before
McNair J of the Queen's Bench Division of the English High Court
the case was dealt with as two separate conflicts: Appleyard (as
finder) against Yorkwin, and Yorkwin (as finder) against the City
of London. Hence the case appears in the specification as Lon-
don v. Appleyard (1) and as London v. Appleyard (2). 32 SHYSTER
treats each as a separate case; the facts are different, although the
result happens to be the same in both cases.

Unlike SHYSTER, FINDER does not allow the specification of un-
known attribute values: unknowns are coded as nos in FINDER. 33
Two of the no values in FINDER are actually unknown. 3 " 1 and two
other attribute values in FINDER are incorrect. 35 Nevertheless, the
fact vectors in the specification are those used in FINDER; in the
interests of accurate simulation, no unknowns are included and
the errors are not rectified. As it happens, including unknown val-
ues and rectifying errors changes neither SHYSTER's opinion nor
the cases SHYSTER chooses to justify its opinion in the testing
described below.

The summary strings for the leading cases are taken (slightly
reworded) from FINDER.

Attribute dependence

SHYSTER detects no functional dependence, and no evidence of
stochastic dependence, between the attributes in this specification.
The probabilities matrix for the Finder area is given in figure 4.1.

170 A pragmatic legal expert system

§-1-2

A 2

A 3

A 4

A G

A 7

A 8

A Q

^io

0.93
0.50

0.82
0.71

1.00
0.38

0.71
0.82

1.00
0.11

0.36
1.00

1.00
0.36

1.00
0.63

0.89
0.64

0.93
0.50

0.50
1.00

0.50
0.93

1.00
0.21

1.00

0.21

1.00
0.50

0.21
1.00

A 2

0.38
1.00

0.82
0.71

1.00
0.30

0.64
0.89

0.89
0.64

1.00
0.38

0.11
1.00

A 3

0.63
1.00

0.75
1.00

1.00

0.75

1.00
0.88

1.00
0.25

A 4

robabi

0.89
0.64

0.36
1.00

0.11
1.00

0.38
1.00

0.89
0.64

lities r

0.54
1.00

1.00

0.54

1.00
0.75

0.54
1.00

A G

natrix

1.00

0.54

1.00
0.75

0.54
1.00

for th

1.00
0.25

0.46
0.96

A s

e Finde

0.25
1.00

A 9

^igurc

TER

• 4.1:

DUtpilt

The p
)■

r area

(SHY

Weights

The table of weights for the Finder area, as extracted from the
weights file, is given in figure 4.2. The rightmost column indicates
that two attributes are of equal greatest importance: 30

A 4 : Was the other claimant the true owner of the chattel or did
she/he claim through the rights of the true owner?

Aq: Was an attempt made to find the true owner of the chattel
or, alternatively, was the chattel clearly abandoned?

That SHYSTER deems A 4 to be important is pleasing. The finder
has a general right to the found article as against all the world
except the true owner. If the other claimant is the true owner
then the finder should lose (as happened in Moffatt v. Kazana 37 ).
Although Aq is important, it probably does not deserve to be
considered as important as A+.

§4-2

Case studies 171

Attr.

Win
a 2

Lose
a 2

a 2

0.00

0.60

0.24

4.17

0.38

0.23

4.27

A 2

0.00

0.80

0.16

6.25

0.50

0.25

4.00

A 3

0.67

0.22

4.50

0.60

0.24

4.17

0.63

0.23

4.27

A 4

0.00

0.20

0.16

6.25

0.13

0.11

9.14

0.67

0.22

4.50

0.20

0.16

6.25

0.38

0.23

4.27

A 6

0.00

0.40

0.24

4.17

0.25

0.19

5.33

A 7

0.00

0.40

0.24

4.17

0.25

0.19

5.33

0.33

0.22

4.50

1.00

0.00

0.75

0.19

5.33

A 9

0.67

0.22

4.50

1.00

0.00

0.88

0.11

9.14

■4 10

0.33

0.22

4.50

0.20

0.16

6.25

0.25

0.19

5.33

Figure 4.2: The table of weights for the Finder area (SHYSTER
output).

4.2.3 Test case

A major drawback of the Finder area as a case study for SHYSTER
is the dearth of finder cases. Only one such case has been decided
since the most recent of FINDER's leading cases.

Parker v. BA

1981

In Parker v. British Airways Board, 3 * Parker (an airline passenger)
found a gold bracelet in the British Airways international execut-
ive lounge at Heathrow Airport, London. He handed the bracelet
to a British Airways employee, and left his name and address, ask-
ing that the bracelet be returned to him if the true owner could
not be found. 39

The owner never claimed the bracelet, and British Airways sold
it and kept the proceeds: i"850. When Parker discovered this.
he sued British Airways in the Brentford County Court and was
awarded ^850 damages plus -£50 interest. British Airways ap-
pealed to the English Court of Appeal claiming that, as occupier
of the premises, it had rights superior to Parker's.

Parker relied heavily on Bridges v. Hawkesworth. British Air-
ways submitted that that case could be distinguished in favour of
South Staffordshire v. Sharman. 40

172 A pragmatic legal expert system §-&-2

The three judges who heard Parker v. BA in the Court of Appeal
unanimously found for Parker. British Airways was held not to
have sufficiently manifested an intention to exercise control over
lost property before it was found in the executive lounge to acquire
rights in the bracelet superior to Parker's.

The English Court of Appeal is the highest court to have heard a
case in this area. Although not strictly bound by the previously de-
cided cases, the court took those cases into account. Donaldson LJ,
who delivered the principal judgment, exhaustively discussed the
authorities before stating the rights and obligations of the finder
and of the occupier in such cases (as quoted above 41 ). He gave
considerable emphasis to Bridges v. Hawkesworth. 42 Eveleigh LJ
and Sir David Cairns agreed, the latter declaring that Bridges v.
Hawkesworth "is the closest case on its facts to the present case." 43

SHYSTER's complete report file for Parker v. BA is given as an
example in § A.2. SHYSTER'S fact vector is (NNNNYNMYN); to quote
from the report file:

. . . the finder was not the occupier of the premises where the chat-
tel was found; the chattel was not attached; the other claimant
was not the owner of the premises where the chattel was found; the
other claimant was not the true owner of the chattel and was not
claiming through the rights of the true owner; the finder handed
over the chattel to the other claimant after the finding; neither
party relied on the terms of an agreement regarding the right to
the chattel; the finder was not a servant of the other claimant;
the chattel was not hidden and was not in a position so as to be
difficult to find; an attempt was made to find the true owner of
the chattel or, alternatively, the chattel was clearly abandoned;
and neither party knew of the existence of the chattel prior to the
finding.

Of these attribute values, only the statement that an attempt
was made to find the true owner of the chattel (^9 = yes) is
contentious. This contention could be important; A 9 is one of the
most important attributes, according to SHYSTER.

The British Airways employee to whom Parker gave the brace-
let handed it to the company's lost property' department. The
true owner never claimed it. It is not clear from the report of the
case whether Parker or British Airways made any attempt to find
the true owner beyond leaving the bracelet in the lost property

§-1.3 Case studies 173

department for some time (e.g. they may have advertised the fact
that the bracelet had been found, in the executive lounge). It is at
least arguable that the correct value of A 9 is unknown. But. as it
happens, changing the value of A Q from YES to unknown makes no
difference despite that attribute's weight; SHYSTER comes to the
same conclusion on the basis of the same cases, and both instan-
tiations have the same nearest result as does the uninstantiated
instant case.

SHYSTER concludes that Parker should win. SHYSTER'S table of
distances, as extracted from the distances file, is given in figure 4.3.
The nearest neighbour is Bridges v. Hawkesworth (Co); the nearest
other is London v. Appleyard (1) (C 5 ). SHYSTER issues no warn-
ings.

SHYSTER comes to a good conclusion, and its choice of nearest
neighbour is a good one. Its choice of nearest other is justifi-
able: the court in London v. Appleyard explicitly followed South
Staffordshire v. Sharman — the case upon which British Airways
relied. 44 As can be seen in figure 4.3, South Staffordshire v. Shar-
man (C 7 ) is SHYSTER's (equal) second-nearest other.

SHYSTER reports that Parker v. BA is not on all fours with
Bridges v. Hawkesworth because, in that case, the other claimant
was the owner of the premises where the chattel was found. Never-
theless, SHYSTER states its opinion that Bridges v. Hawkesworth
should be followed. SHYSTER'S report on this case is substantially
identical to that of FINDER given the same facts, 45 although this
is hardly surprising.

4.2.4 Conclusion: the Finder specification

SHYSTER is able to simulate FINDER. However due to an almost
total lack of test cases, the Finder specification is not of much use
in the testing of SHYSTER.

4.3 The authorization of copyright infringement

The Authorization specification deals with the meaning of "au-
thorization" in the Australian Copyright Act. This specification
is designed to be used with a rule-based system representing the

174 A pragmatic legal expert system

§4.3

rH '

3 CO

s "*.

r^ a,

E +

l/>

L".

.-H O

t~-

£ s

g =

Ol i— t

\r.

■^

O O

CD .2

to 00

b- CO

i —

1 —

I— <

, J-=>

o o

-^v o

J-3

L-

oo to

1— 1

|— 1

O CM

o o

o
o

O O

O tD

.— < CM

I— <

•&

O O

i-H CM

I— 1

-**

en .„

1 1

*t3

1 1

s — ' w

I -

b- 00

-f

co ii

CM ^

7 1

■**

I— <

5, u

— __

^t 1

i— i

i— <

I-*

s «

|— |

i— t i— i

■ — i

1 — 1

r—t

t £

e S

CL (L»

X •

•

O CD

C4-. rJ= j

•

X X

s °

X X

•

X X

•

^ -

J- »o

o ^?

£ <*

•

• •

•

^ «

ju o

^r
^

X X

•

03 to

+3 CO

•

• X

•

CD =S

ft 1

X X

•

^ CU

*. ^^ s

—
£
*-

l-H

Cfl CO c

-r

„

2^^

o o |

Figu

is 57
yard

§-1.3 Case studies 175

4.3.1 The law

The Copyright Act 1968 (Cth) is the statutory component of the
law of copyright as it applies in Australia. Sections 13(2). 36(1)
and 101(1) forbid the authorization of copyright infringement:

13. (2) For the purposes of this Act, the exclusive right to
do an act in relation to a work, an adaptation of a work or any
other subject-matter includes the exclusive right to authorize a
person to do that act in relation to that work, adaptation or other
subject-matter.

36. (1) Subject to this Act, the copyright in a literary, dra-
matic, musical or artistic work is infringed by a person who, not
being the owner of the copyright, and without the licence of the
owner of the copyright, does in Australia, or authorizes the doing
in Australia of, any act comprised in the copyright.

101. (1) Subject to this Act, a copyright subsisting by virtue
of this Part [copyright in subject-matter other than works] is in-
fringed b}' a person who, not being the owner of the copyright,
and without the licence of the owner of the copyright, does in
Australia, or authorizes the doing in Australia of, an}' act com-
prised in the copyright.

Nowhere in the Act is there any other reference to authorization;
the meaning of the word is left to the courts to interpret.

From 1912. until the 1968 Act came into force (in 1969), the
Copyright Act 1911 (UK) was in force in Australia by virtue of
the Copyright Act 1912 (Cth). Section 1(2) of the UK Act also
forbids authorization of copyright infringement:

1. (2) For the purposes of this Act, "copyright" means the sole
right to produce or reproduce the work or any substantial part
thereof in any material form whatsoever, to perform, or in the case
of a lecture to deliver, the work or any substantial part thereof in
public; if the work is unpublished, to publish the work or any
substantial part thereof; and shall include the sole right,—

(a) to produce, reproduce, perform, or publish any translation
of the work;

(b) in the case of a dramatic work, to convert it into a novel or
other non-dramatic work;

176 A pragmatic legal expert system §-1-3

(c) in the case of a novel or other non-dramatic work, or of an
artistic work, to convert it into a dramatic work, b}' way of
performance in public or otherwise;

(d) in the case of a literary, dramatic, or musical work, to make
any record, perforated roll, cinematograph film, or other
contrivance b}' means of which the work may be mechanic-
ally performed or delivered,

and to authorise any such acts as aforesaid.
As with the 1968 Act. "authorise" is not defined in the UK Act.

Since early this century there have been several cases in which
courts have had to decide whether copyright infringement had
been authorized under one of these Acts. Because of the simil-
arities between the UK Act and the 1968 Act. Australian courts
have taken into account cases concerning the former when inter-
preting the latter. 46

In one of the earliest of these authorization cases. Performing
Right Society Ltd v. Ciryl Theatrical Syndicate Ltd, Scrutton LJ
expressed his view that the words "to authorize any such acts as
aforesaid 11 at the end of s. 1(2) of the UK Act "are superfluous
and add nothing to the definition." 47 He considered them to be
a reference to acting through an agent. However, only two years
later Bankes LJ in Falcon v. Famous Players Film Co. decided
that "authorize 11 should be given its ordinary dictionary meaning
of "sanction, approve, countenance 11 . 48 Bankes LJ's interpretation
was adopted by Isaacs J of the High Court of Australia in 1928, 49
and has been used ever since. As Ricketson points out, this means
that:

. . . the notion of "authorisation" extends beyond the authority
given to an agent. Thus, if an infringing act is committed by an
agent acting within his real or ostensible authority, his principal
will be direi fcly liable on the ordinary rules governing the relation-
ship of principal and agent. The same, of course, applies where
the infringing act is committed by an employee acting within the
course of his employment. 50

Gibbs J discussed the meaning of authorization in some detail
in University of New South Wales v. Moorhouse. 51 He pointed out
that "authorize" can also mean "permit," though this is tempered
by the principle that a person cannot be said to authorize an

§-1.3 Case studies 11

infringement of copyright unless she/he has the power to prevent
it. 52 Furthermore:

Express or formal permission or sanction, or active conduct indic-
ating approval, is not essential to constitute an authorization . . .
However, the word "authorize" connotes a mental element and it
could not be inferred that a person had, by mere inactivity, au-
thorized something to be done if he neither knew nor had reason
to suspect that the act might be done. 53

Having cited several authorities, Gibbs J summarized the law as
follows:

It seems to me to follow from these statements of principle that
a person who has under his control the means by which an in-
fringement of copyright may be committed . . . and who makes
it available to other persons, knowing, or having reason to sus-
pect, that it is likely to be used for the purpose of committing an
infringement, and omitting to take reasonable steps to limit its
use to legitimate purposes, would authorize any infringement that
resulted from its use. 54

However, he warned that:

. . . the question of whether one person authorizes another to com-
mit an infringement depends upon all the facts of the case so that
a decision on a particular set of circumstances may be of no as-
sistance in other cases. 55

4.3.2 The Authorization specification

Having consulted a legal textbook. 50 the published judgments in
the leading cases to which that textbook refers, and a legal expert
in the field, 57 I wrote a SHYSTER specification of the meaning
of "authorization" in the Copyright Act This specification was
written so as to represent the law as it was in 1983- — -the textbook
states the law as it was then 58 — so that important cases which
have been decided since that time can be used as test cases.

The Authorization specification contains a single area: the
Authorization area. This area makes use of several of the features
not used in the Finder specification. There is a hierarchy which
ranks seven courts at six distinct levels, there is an opening string
(although there is no closing string), attributes have help strings,
some of the attribute values of the leading cases are unknown, and
ideal points and attribute direction are included.

178 A pragmatic legal expert system §-1-3

Results

The Authorization area has three results: Auth. Not-Auth and Liable.
These correspond to "the accused authorized the infringement."
u the accused did not authorize the infringement" and u the accused
is liable (directly or vicariously) for the infringement." 59 These are
three quite distinct results; the Authorization area demonstrates the
need for a case-based system to allow for more than two possible
outcomes. 00

Attributes

There are seven attributes:

A\: Was the infringer an employee of the accused?

A2'. Was the infringer an independent contractor to the accused?

.A3: Did the accused sell or hire the infringer the means of in-
fringing?

^4: Did the accused have the power to prevent the infringement?

j4 5 : Did the accused take reasonable steps to avoid the infringe-
ment?

A 6 : Did the accused know, or have reason to anticipate or suspect,
that the infringing act was to be, or was likely to be, done?

A?: Was the specific infringement causally related to an incite-
ment to infringe on the part of the accused?

Although there will be times when the answers to A x and A 2
are obvious, these questions are not always easy to answer. The
Employee specification described below is concerned solely with
deciding whether a worker is an employee or an independent con-
tractor. If the Employee area from that specification were included
in the Authorization specification then either or both of A x
and A 2 could be made external attributes: i.e. rather than asking
the user a single question, the attribute value would be determ-
ined by reference to the Employee area, which has 18 attributes. 01
The last specification described in this chapter (the Natural spe-
cification) illustrates this process: it has three areas, two of which
have external attributes. 62

§-1.3 Case studies 179

However, a simpler approach was adopted for the Authoriza-
tion specification. Both A\ and An have help strings which give
assistance to the user in answering the attribute question. For
example, the help string for A x is:

If the accused had control over the infringer's manner of doing
her/his work then the infringer was an employee of the accused.
If the infringer undertook to do something for the accused and
had discretion as to the manner in which it was to be done then
the infringer was an independent contractor to the accused, not
an employee.

(This is a greatly simplified statement of the law, as can be seen
by comparison with the Employee specification.)

Of the other attribute questions, only A$ should present any
difficulty. This attribute is very important. If the accused did
take reasonable steps to avoid the infringement then she/he will
not be said to have authorized the infringement, 03 although she/he
could still be directly or vicariously liable. In all of the cases that
make up the Authorization area the value of A 5 is no: i.e. in none
of the cases did the accused take reasonable steps. This is not
surprising; if reasonable steps had been taken, the cases would
probably never have been taken to court. The absence of yess for
A 5 means that SHYSTER assigns that attribute infinite weight.

The attribute has a fairly unhelpful help string: "Whether par-
ticular steps were reasonable depends on the facts of the case."
But the nebulous concept of "reasonableness" is one with which
lawyers very often have to deal. A lawyer would probably not
have much difficulty in answering this question, and she/he al-
ways has the option of answering unknown and forcing SHYSTER
to examine instantiations.

If Ar> is changed to unknown in all of the test cases in the testing
described below, SHYSTER comes to the same conclusion on the
basis of the same cases. This does not mean that A 5 is irrelevant;
its inclusion (and its infinite weighting — see below) means that if
the accused did take reasonable steps to avoid an infringement, the
specified directions and ideal point directions will strongly suggest
Not-Auth. 04

In theory, it should be possible to specify another area of case
law which defines this very open-textured concept of reasonable-
ness, however that might prove practically impossible. 05

180 A pragmatic legal expert system

§4.3

A 2

A 3

A 4

A G

-l7

0.75
1.00

0.50
1.00

1.00
0.63

1.00
1.00

1.00
0.88

1.00
0.50

0.21
1.00

1.00
0.30

1.00
1.00

0.25
1.00

0.79
0.79

A 2

0.40
0.95

1.00
1.00

1.00
0.56

0.64
0.83

A 3

1.00
1.00

0.67
1.00

0.95
0.40

A 4

1.00
1.00

1.00
0.44

A G

Figure 4.4: The probabilities matrix for the Authorization area
(SHYSTER output).

Leading cases

There are nine leading cases in the Authorization area. Three
of them arose under the Copyright Act 1911 (UK) or its suc-
cessors: PRS v. Ciryl 66 Falcon v. Famous Players 6 ' and A &M
Records Inc. v. Audio Magnetics Inc. (UK) Ltd. 68 Four arose un-
der the UK Act. in force in Australia by virtue of the Copyright
Act 1912 (Cth): Mellor v. Australian Broadcasting Commission 69
Winstone v. Wurlitzer Automatic Phonograph Co. of Australia Pty
Ltd, 70 Australasian Performing Right Association Ltd v. Miles 71
and Australasian Performing Right Association Ltd v. Canterbury-
Bankstown League Club Ltd. 72 And two arose under the Copyright
Act 1968 (Cth): UNSW v. Moorhouse 73 and RCA Corporation v.
John Fairfax and Sons Ltd. 74

SHYSTER warns that APRA v. C anterbury-B ankstown and Mel-
lor v. ABC have identical facts (except for unknown values), and
that APRA v. C anterbury-B ankstown and APRA v. Miles have
identical facts (except for unknown values) and different results.

The first draft of the specification included an early King's
Bench Division case: Performing Right Society Ltd v. Bradford
Corporation. 75 That case has an identical fact vector to that of
Mellor v. ABC. On reflection, the former case was omitted in
favour of the latter, later, and more important. Privy Council
decision." 5

§4.3

Case studies 181

Attr.

Auth

/i <T W

Not-Au

2
ft G-

Liable

2
ft a~

2
tr

0.00 0.00 oo

0.00

1.00 0.00

0.13

0.11

9.14

0.25 0.19 5.33

0.33

0.22

4.50

0.00 0.00

0.25

0.19

5.33

0.60 0.24 4.17

0.33

0.22

4.50

0.00 0.00

0.44

0.25

4.05

A 4

0.80 0.16 6.25

0.33

0.22

4.50

1.00 0.00

0.67

0.22

4.50

0.00 0.00 CO

0.00

0.00 0.00

0.00

1.00 0.00 oc

0.67

0.22

4.50

1.00 0.00

0.89

0.10

10.13

A-7

0.80 0.16 6.25

0.00

1.00 0.00

0.56

0.25

4.05

Figure 4.5: The

;able

weights for the Authorization

area

(SHYSTER output).

Attribute dependence

SHYSTER detects no functional dependence, and no evidence of
stochastic dependence, between the attributes in this specifica-
tion. The probabilities matrix for the Authorization area is given in
figure 4.4.

Weights

The table of weights for the Authorization area, as extracted from
the weights file, is given in figure 4.5. The rightmost column of
the table indicates that, as mentioned above, A fj is given infinite
weight.' 7 The next most important attributes are:

A G : Did the accused know, or have reason to anticipate or suspect,
that the infringing act was to be, or was likely to be, done?

A\\ Was the infringer an employee of the accused?

These attributes are indeed important, although A 4 ("Did the
accused have the power to prevent the infringement?") should
probably be more heavily weighted.

4.3.3 Test cases

The Authorization specification was written so as to represent
the law as it was in 1983. Three important authorization cases
have been decided since then, and they are used to test SHYSTER.
Another case was decided before 1983 but was not considered im-
portant enough to include in the specification. That case, CBS

182 A pragmatic legal expert system §-1-3

Inc. v. Ames Records and Tapes Ltd, tS is used as a test case too.
There are also two hypothetical test cases, chosen by the legal
expert.

The Authorization area has three results, and SHYSTER will al-
ways look for nearest cases for each result. So. SHYSTER argues
about the Liable result even in those test cases where the courts
did not explicitly consider the possibility that the accused was dir-
ectly liable. It is not inappropriate to argue about the Liable result
in such cases; the judges must have (at least tacitly) rejected that
conclusion themselves. It would, however, be cause for concern if
SHYSTER'S opinion were that the accused was directly liable in
such a case — but it does not make that mistake in any of these
tests.

CBS v. Ames 1981

In CBS Inc. v. Ames Records and Tapes Ltd,' 9 Ames operated a
record lending library. As well as hiring records. Ames sold blank
cassettes. CBS sold records to Ames, and held the copyright in
those sound recordings. CBS claimed that Ames knew that it
was likely that its customers would copy the records onto blank
cassettes at home, and that hence Ames was "authorizing" in-
fringement within the meaning of s. 1(2) of the Copvright Act 1956
(UK).

Before Whitford J of the Chancery Division of the English High
Court. CBS relied on Falcon v. Famous Players, Vigneux v. Cana-
dian Performing Right Society Ltd so (not one of the leading cases
in the Authorization specification), Winstone v. Wurlitzer and
UNSW v. Moorhouse.

Ames argued that, applying the dictionary meaning of author-
ization as did Bankes LJ in Falcon v. Famous Players, it had
not authorized any infringement. It also claimed that UNSW v.
Moorhouse was wrongly decided. 81 Ames relied on A &M v. Au-
dio Magnetics, pointing to the absence of evidence of a particular
case of infringement. CBS urged that A &M v. Audio Magnetics
be distinguished on the basis that "here the intention and desire
was to take advantage of home taping. 1 ' 82

Whitford J agreed with Ames, although he made no mention of
A &M v. Audio Magnetics in his judgment.

§-1.3 Case studies 183

SHYSTER'S fact vector is (NNYNNYN); to quote from its report file:

. . . the infringer was not an employee of the accused; the infringer
was not an independent contractor to the accused; the accused sold
or hired the infringer the means of infringing; the accused did not
have the power to prevent the infringement; the accused did not
take reasonable steps to avoid the infringement; the accused knew,
or had reason to anticipate or suspect, that the infringing act was
to be, or was likely to be, done; and the specific infringement was
not causally related to an incitement to infringe on the part of the
accused.

SHYSTER agrees with Whitford J that Ames did not authorize the
infringement. SHYSTER'S table of distances, as extracted from
the distances file, is given in figure 4.6. The nearest neighbour
is A & M v. Audio Magnetics (C s ): one of the cases upon which
Ames relied; the nearest others are Falcon v. Famous Players
(C 5 : Auth) — upon which CBS relied — and APRA v. Miles (C 9 : Li-
able). SHYSTER warns that the specified directions suggest the
Auth result, as can be seen in figure 4.6: the strength of the spe-
cified direction for Auth (3oo+4.17) is greater than that for the
other two results.

WEA v. Hanimex 1987

In WEA International Inc. v. Hanimex Corporation Ltd? 3 Hanimex
advertised one of its cassette tape products on radio. Humorous
advertisements implied that recordings of various artists 1 songs
on cassettes other than Hanimex's would melt in a hot car: "If
you don't want your favourite recordings ruined, use Fuji GTI Car
lapes.

WEA and other record companies owned the copyright in sound
recordings made by artists who were mentioned by name in the
advertisements. They claimed that Hanimex had authorized mem-
bers of the public to infringe their copyright, contrary to the
Copyright Act 1968 (Cth).

The case was heard in the Federal Court of Australia by Gum-
mow J. His judgment includes a long history of authorization
provisions in copyright legislation, and their judicial interpreta-
tion, since the turn of the century. Of the recent cases, he referred

184 A pragmatic legal expert system

§4.3

i o . •

-as ir

1 -

o o

-C

in in

+
8

i— •

i -

1
■H

8 8
-41 4

■fl

-H E-l W

- 0) jh

+= ^H O

o> o +^

- h-3

1*-

i— I

i -

? 1

2 S^

■

r~-

■

...

':■

-l-

A3 H o

t— 1

. — ' CJ

t -

<T)

r—t

"3 _ir S*

t*-

I— •

•^

-r

r -

(_)

T|*

Z-

■ — 1

-M +3

-I 1

-**

-f

t -

■*#

r--

— 1

1ft

l-H

t— 1

^tf

1 —

H .5 «>

^ +^> ^

« I 2

■ — i

Tt 1

,-H

■ — i

-**

I— 1

-f

-p

"t3

1 —

S ^ 05

r ■

i —

Tt*

a ^

1 -

5^ <D ^

-r.

"*F

-f

-**

I— 1

1 —

|— 1

■ — 1

r—i

co •£ •

■ — i

Tfl

r-:

^ ^ S

•

^ o ^

e tab
Audi
sPla

-t -^

•

l-H

•

a
a

■**

■£

I-

10 <

—

_•

£-2 fi«

•^

-i

Figu

bour
Falco
sugge

§-1.3 Case studies 185

to UNSW v. Moorhouse, CBS v. Ames and RCA v. Fairfax. 84 He
pointed out that RCA v. Fairfax was authority for the proposition
that authorization of copyright infringement requires an actual act
of infringement, and observed that:

... it has not even been shown that there has been any unauthor-
ised reproduction by any particular person of any of the sound
recordings in which [the record companies] hold copyright. 85

Additionally. Gummow J held that, in any event, the advertise-
ments were not "an invitation or incitement to or approval of"
copyright infringement. 80

SHYSTER'S fact vector is (NNYNNYN)— as it is in CBS v. Ames.
Hence, SHYSTER's distances and report files are identical to those
for that case. SHYSTER agrees that Hanimex did not authorize
the infringement, although it warns that the specified directions
suggest Auth.

The nearest neighbour is .4 &M v. Audio Magnetics. That this
case was not mentioned by Gummow J in his judgment is most
surprising. The facts of ^4 &M v. Audio Magnetics are very similar
to those of WE A v. Hanimex: they both concern advertisements
claiming high durability of blank cassettes and referring to specific
recording artists. A&M v. Audio Magnetics is a decision of the
Chancery Division of the English High Court, and not as good
authority in Australia as a decision of the XSW Supreme Court.
like RCA v. Fairfax. Nevertheless, it is amazing, given the length
of his discussion of the development of the law in Australia, Eng-
land and the US, that Gummow J did not refer to A & M v. Audio
Magnetics.

Both A &M v. Audio Magnetics and RCA v. Fairfax were decided
on the same point: that there must be an actual infringing act be-
fore someone can be said to have authorized an infringement. The
judge in RCA v. Fairfax explicitly followed A &M v. Audio Mag-
netics, amongst other cases, in coming to his decision. 87 Hence, I
claim that SHYSTER's choice of nearest neighbour is a good one,
despite the fact that it was not referred to in WEA v. Hanimex.

Gummow J made reference to only the first of SHYSTER's two
nearest others: Falcon v. Famous Players (Auth) and APRA v.
Miles (Liable).

186 A pragmatic legal expert system §-1-3

CBS v. Amstrad 1988

In CBS Songs Ltd v. Amstrad Consumer Electronics Pic 88 Am-
strad manufactured twin-deck tape recording machines. These
were designed so as to facilitate tape-to-tape copying, and were
advertised in a manner that was likely to encourage home tap-
ing and copying. However, the advertisements warned that some
copying could not be done without permission, and made it clear
that Amstrad had no authority to grant such permission.

CBS and other record companies sued Amstrad claiming that it
had authorized users of its twin-deck recorders to infringe copy-
right. Before the House of Lords. CBS cited many cases in ar-
gument, including Falcon v. Famous Players. Amstrad relied on
A &M v. Audio Magnetics, amongst others.

As in CBS v. Ames, A&M v. Audio Magnetics— although used
in argument — was not referred to in the judgments. Instead, Lord
Templeman (with whom the other four law lords agreed) found for
Amstrad on the basis of Bankes LJ's definition of authorization in
Falcon v. Famous Players, and following CBS v. Ames.

SHYSTER'S fact vector is (NNYNNYN)— as it is in CBS v. Ames and
WE A v. Hanimex.

SHYSTER agrees that Amstrad did not authorize the infringe-
ment. Its choice of cases is good: the nearest neighbour is A &M v.
Audio Magnetics; the nearest others are Falcon v. Famous Play-
ers (Auth) and APRA v. Miles (Liable). However, SHYSTER warns
that the specified directions suggest Auth.

APRA v. Jain 1990

In Australasian Performing Right Association Ltd v. Jain, 89 Jain
was the director and principal executive officer of a company (Val-
amo Pty Ltd) which was the proprietor of a tavern. Live bands
and a video music system performed works in which the APRA held
copyright. It was conceded that Valamo had infringed copyright.
but Jain claimed that he was not personally liable. The day-to-
day operations of the tavern were controlled by the licensee. Jain
claimed that the licensee was responsible for all the music at the
tavern, and that he (Jain) only took an interest if takings were
down, in which case he would advise the licensee to engage a dif-
ferent band.

§-1.3 Case studies 187

The Full Court of the Federal Court of Australia held that Jain
had authorized the infringement. Sheppard. Foster and Hill JJ
held that Jain had allowed:

... a situation to develop and to continue in which he must have
known that it was likely that the [APRA's] music would be played
without any licence from it. It was within his power to control
what was occurring [but] he did nothing at all. 90

In coming to their conclusion, their Honours mentioned many
cases. To the extent that they followed any particular case, they
followed UNSW v. Moorhouse. 91

The complete report file for APRA v. Jain is given as an example
in §A.3. SHYSTER'S fact vector is (NYNYNYN). 92 SHYSTER agrees
with the Federal Court that Jain authorized the infringement.
However, none of the cases chosen by SHYSTER was mentioned by
the Court. The nearest neighbour is Mellor v. ABC; the nearest
others are RCA v. Fairfax (Not-Auth) and APRA v. Miles (Liable).
There are no warnings.

Hypothetical case 1

The legal expert suggested a hypothetical case for testing the
Authorization specification: a fact situation about which he had
been asked for advice, but which has not come before a court for
consideration.

A residential college on a university campus provides personal
computer facilities for the use of its residents. It also makes avail-
able to the residents several items of software. Neither the college
nor the residents hold the copyright in this software, although the
software could easily be copied. The college is concerned as to
whether, if a resident were to copy an item of software, the college
could be said to have authorized that infringement.

The legal expert advises that the college could be said to have
authorized the infringement.

SHYSTER'S fact vector is (NMYNYN). SHYSTER'S conclusion is that
the college would have authorized the infringement. The nearest
neighbour is UNSW v. Moorhouse: the nearest others are RCA v.
Fairfax (Not-Auth) and APRA v. Miles (Liable). However, it warns
that the specified directions suggest Liable.

The legal expert advises that SHYSTER's choice of cases, nearest
neighbour and nearest others, is good.

188 A pragmatic legal expert system §-1-3

Hypothetical case 2

The legal expert suggested a second hypothetical authorization
case. A language school lends audio tapes to its students and
allows them to take those tapes home. One of the language school's
employees gives a student one of these tapes and tells him that
"lots of people copy them." Could the school be said to have
authorized an infringement if the student makes a copy of the
tape?

The legal expert advises that the language school would prob-
ably not be held to have authorized the infringement.

SHYSTER'S fact vector is (NNNNNYY). Its conclusion is that the
language school would not have authorized the student's infringe-
ment. The nearest neighbour is RCA v. Fairfax; the nearest others
are Falcon v. Famous Players (Auth) and APRA v. Miles (Liable).

However. SHYSTER warns that two of these cases are equidistant
from the hypothetical case; it chooses RCA v. Fairfax because it
is a decision of the New South Wales Supreme Court whereas Fal-
con v. Famous Players is merely a decision of the English Court
of Appeal. The weighted correlation coefficients suggest that Fal-
con v. Famous Players should be the nearest neighbour, and the
specified directions suggest Liable.

Again, the legal expert advises that SHYSTER's choice of cases
is good.

4.3.4 Generated tests

The legal expert provided three examples for generating tests using
the Authorization specification: generated tests 1, 2 and 3.

Generated test 1

The legal expert advises that, if the following is true, either the
accused authorized the infringement, or the accused is directly or
vicariously liable for the infringement:

. . . the infringer was an employee of the accused; the infringer was
not an independent contractor to the accused; and the accused
did not take reasonable steps to avoid the infringement.

§-1.3 Case studies 189

Given the fact vector (YNUUNUU). SHYSTER generates 16 instan-
tiations: one eighth of the total search space. In all but two of
these instantiations, SHYSTER's chosen results are good (i.e. Auth
or Liable).

In each of the two instantiations in which the result is Not- Auth.
the nearest ideal point is that of a good result but is further from
the instantiation than is the nearest neighbour, so no warning is
issued. In both of those instantiations, the accused did not have
the power to prevent the infringement (A 4 = no), and the specific
infringement was not causally related to an incitement to infringe
on the part of the accused (A- = no). The legal expert agrees
that these attribute values justify SHYSTER's choice of result in
both instantiations.

Generated test 2

The legal expert also advises that, if the following is true, the
accused did not authorize the infringement:

. . . the infringer was not an employee of the accused; the accused
did not have the power to prevent the infringement; and the ac-
cused took reasonable steps to avoid the infringement.

The fact vector is (NUUNYUU). SHYSTER generates 10 instanti-
ations, in all but four of which SHYSTER'S chosen result is good
(i.e. Not-Auth).

In each of those four instantiations, SHYSTER warns that an
ideal point suggesting the good result is at least as near to the
instantiation as is the nearest neighbour. The only attribute for
which these four instantiations have common values (apart from
the three attributes with known values in the above fact vector) is
the last: in each, the specific infringement was causally related to
an incitement to infringe on the part of the accused (A 7 = yes).
Although this attribute value does suggest Auth or Liable, 93 the
three known attribute values in the above fact vector are more
persuasive: SHYSTER's choice of result in these four instantiations
is bad.

190 A pragmatic legal expert system §-1-3

Generated test 3

The legal expert advises that, if the accused took reasonable steps
to avoid the infringement, she/he cannot have authorized the in-
fringement (although she/he may be directly or vicariously liable
for the infringement). SHYSTER's fact vector is (UUUUYUU). It
generates 04 instantiations: half the search space. In 13 of these
64 instantiations. SHYSTER's chosen result is bad (i.e. Auth).

In all 13. SHYSTER warns that an ideal point suggesting a good
result is at least as near to the instantiation as is the nearest
neighbour. The fact that, in each of these 13 instantiations, the
infringer was not an employee of the accused (A\ = no) does not
justify SHYSTER's choice of result.

4.3.5 Conclusion: the Authorization specification

In the test cases for the Authorization specification described
above, SHYSTER chooses a good result every time, and 83.33% of
its chosen cases are good. However, the small number of attributes
specified for the Authorization area means that SHYSTER has diffi-
cult}* distinguishing between cases: e.g. the first three test cases
share the same fact vector. This may indicate a deficiency in the
specification, or it may be indicative of a feature of the area of
law.

As shown in figure 4.13, SHYSTER's choice of result is good
in 80.21% of the instantiations in all three generated tests. If a
bad choice of result is considered good where an ideal point warn-
ing is issued, SHYSTER's success rate rises to 87.50%. 100.00%
and 100.00% for each test: 97.92% overall. (In none of the instan-
tiations in these three tests does SHYSTER issue an ideal point
warning when the choice of result is good: i.e. none of SHYSTER's
ideal point warnings is bad.)

The Authorization area is a good example of an area which could be
used to define an open-textured statutory concept in a rule base:
the meaning of "authorization" is undefined in sections 13(2).
36(1) and 101(1) of the Copyright Act. It is also a good example of
an area of law in which there are more than two possible results.
SHYSTER (when using the Authorization specification) pro-
duces good advice in the test cases, and chooses good results in
almost all of the generated tests.

§4.4 Case studies 191

4.4 Employees and independent contractors

The Employee specification deals with the legal categorization of
a worker as an employee or as an independent contractor.

4.4.1 The law

If a person works for another, one of two relationships exists:
either they are employer and employee — and theirs is a contract
of service — or they are principal and independent contractor — -and
theirs is a contract for services.

(The parties to a contract of service used to be called "master"
and "servant," but this terminology is no longer used. 94 Before it is
determined which sort of relationship exists, it is convenient to use
generic terms: "employer" and "worker" are used to mean "em-
ployer" and "employee" — in the case of a contract of service — or
"principal" and "independent contractor" — in the case of a con-
tract for services.)

This distinction is important; the law of employment is con-
cerned almost totally with contracts of service. 95 Furthermore,
an employer is vicariously liable for the actions of an employee if
the employee is acting within the course of her/his employment; a
principal would not normally be vicariously liable for the actions
of an independent contractor.

The distinction also affects the terms that will be implied into
the employment contract in the absence of an express agreement
between the parties, the applicability of industrial awards, the
applicability of statutes which may affect workers' compensation,
occupational health and safety, long-service leave, annual holidays,
unfair dismissal, payroll tax, fringe benefits tax, etc. 90

As Creighton, Ford and Mitchell point out:

It is most important therefore to be able to determine whether
any given worker is engaged under a contract of employment. Un-
fortunately neither courts nor parliaments have been consistent in
the attitude they have adopted to this process of categorization.
In particular the}' have contrived to make it virtually impossible
to draw a clear and consistent distinction between 'employees' and
'independent contractors.'

. . . Parliaments tend to adopt one of two approaches to the pro-
cess of categorization. The first, and simplest, is to ignore the

192 A pragmatic legal expert system §4.4

issue entirely, and to leave the matter to be determined by refer-
ence to common law criteria . . . The other approach is to include
some form of definition, usually in the interpretation section. In
some instances this definition is purely circular in form . . . whereas
in others it consists of either a broadening or a narrowing of the
common law concept (sometimes going as far as to include 'inde-
pendent contractors'). Whatever approach is adopted, the final
answer is to be found in the common law. 9 '

The most commonly applied criterion for distinguishing between
a contract of service and a contract for services is the "control
test." If an employer has a right of control over a worker's manner
of doing her/his work then the control test suggests that the con-
tract is of service and that the worker is an employee. Conversely,
if a worker has discretion as to the manner in which the work is to
be done then the test suggests that the contract is for services. 98

However, courts have explicitly had regard to many different
factors in deciding whether a worker is an employee or an in-
dependent contractor. In Price v. Grant Industries Pty Ltd, for
example, the Full Court of the Federal Court examined in detail
fifteen points which Price claimed were indicative of a contract of
service; Grant Industries relied on eleven different points which
it claimed were indicative of a contract for services. 99 This is an
area of the law in which many factors have to be balanced before
a conclusion can be reached.

4.4.2 The Employee specification

Having consulted a legal textbook. 100 Halsbury's, 101 the published
judgments in the leading cases to which they refer, and a legal
expert in the field, 102 I wrote a SHYSTER specification of the
distinction between an employee and an independent contractor.
This specification was written so as to represent the law as it was
in 1982 so that important cases which have been decided since
then can be used as test cases. 103

The Employee specification is used as the basis of the complete
example of SHYSTER's output files in appendix B. Its dump file
is given in full in §B.3.

The specification contains a single area: the Employee area.
There is a hierarchy which ranks nine courts appropriately, and
there is an opening string (although there is no closing string).

§4.4 Case studies 193

Results

The Employee area has two results: Employee and Contractor. These
correspond to "the worker is an employee" and "the worker is an
independent contractor."

Attributes

In keeping with the large number of important factors in this area,
there are eighteen attributes:

A\: Did the employer direct not only what work was to be done,
but also the manner in which it was to be done?

A 2 ' Was the worker allowed to use her/his own discretion in doing
an aspect of the work that was not specified beforehand?

.A3: Was the worker an integral part of the employer's business?

A4: Did the worker own the tools or provide the transport with
which she/he performed the work?

A$: Would the emplo} r er make a profit/loss if the work performed
by the worker cost less/more than expected?

A 6
A 7
A 8

A w
An

A\2

A 13

Was the work performed on the employer's premises?

Did the employer supervise or inspect the work?

Was the worker in business on her/his own account?

Was the worker allowed to employ others to assist with her/his
work?

Was the worker obliged to work only for the employer?

Was the worker required to work at specified times?

Did the employer pay the worker by time?

Was the money that the employer paid to the worker stated
to be a "fee"?

Ai^\ Was the money that the employer paid to the worker stated
to be "wages" or "salarj'"?

Ai$: Did the employer deduct PAYE 104 tax instalments from the
worker's pay?

A 16 : Did the employer pay the worker sick pay or holiday pa}'?

A 17: Did the employer and the worker express an intention that
the relationship would be one of employer and emploj^ee?

A\$: Did the employer and the worker express an intention that
the relationship would be one of principal and independent
contractor?

194 A pragmatic legal expert system §4.4

Answering these questions should be fairly straightforward.
Only A 3 could cause difficulty. The help string for that attrib-
ute suggests that:

If the worker was "part and parcel" of the employer 's business then
she/he was an integral part of the business, not merely accessory
to it.

And the user always has the option of answering unknown if un-
sure.

The first draft of the specification included three extra attrib-
utes:

• Did the employer have to pay the costs of performing the work?

• Did the employer stand to make a profit if the work performed
by the worker cost less than expected?

• Did the employer bear the risk of loss if the work performed bj'
the worker cost more than expected?

These attributes were chosen because courts have had regard to
these questions in various cases decided in this area. However,
SHYSTER detected functional dependence between each of these
attributes and the other two: every case in the specification had
the same attribute values for each of these three attributes. It
was decided that these three attributes were, in fact, three differ-
ent ways of asking the same question, so they were removed and
replaced with a single attribute:

^5: Would the emplo} r er make a profit/loss if the work performed
by the worker cost less/more than expected?

Leading cases

There are thirteen different leading cases in the Employee area.
Three of them concerned the applicability of an industrial award:
Cam and Sons Pty Ltd v. Sargent, 10 ^ Price v. Grant 106 and Aus-
tralian Timber Workers Union v. Monaro Sawmills Pty Ltd. 107
Three others concerned the payment of tax or payments to a
government contribution scheme: Federal Commissioner of Tax-
ation v. J. Walter Thompson (Australia) Pty Ltd, 10S Queensland
Stations Pty Ltd v. Federal Commissioner of Taxation 109 and
Ready Mixed Concrete (South East) Ltd v. Minister of Pensions

§4.4 Case studies 195

and National Insurance. 110 Two concerned workers' compens-
ation: Humberstone v. Northern Timber Mills 111 and Zuijs v.
Wirth Brothers Pty Ltd. 112 Two arose out of copyright claims:
Performing Right Society Ltd v. Mitchell and Booker (Palais de
Danse) Ltd 113 and Stevenson Jordan and Harrison Ltd v. Mac-
donald and Evans. 114 One was a claim for damages for breach
of statutory duty: Ferguson v. John Dawson and Partners (Con-
tractors) Ltd. llb Another concerned a worker's entitlement to long
service leave: Australian Mutual Provident Society v. Chaplin. 116
And one was a claim for compensation for unfair dismissal: Mas-
sey v. Crown Life Insurance Co. 117

Stevenson v. Macdonald is used twice. In that case Macdonald
and Evans claimed copyright in a book because its author had
written it while he was working for them. The book was divided
into sections, which were written under different circumstances.
The English Court of Appeal looked at each section separately, to
determine wdiether that section had been written by its author as
an employee of Macdonald and Evans. The case appears as Steven-
son v. Macdonald (1) — which deals with the first section — and
as Stevenson v. Macdonald (2) — which deals with the second. 118
SHYSTER treats each as a separate case; the facts and the result
are different in both.

Attribute dependence

The probabilities file for the Employee specification is given in
full in §B.4. SHYSTER detects functional dependence between A 4
and A§: in the leading cases, the inverse function maps the attrib-
ute values of either attribute to those of the other. Nevertheless,
the two attributes represent different questions: there is no reason
to believe that the worker not owning the tools or providing the
transport with which she/he performs the work (^4), and the em-
ployer standing to make a profit/loss if the work performed by
the worker costs less/more than expected (A$) are really the same
attribute. Hence, both attributes remain in the specification.

SHYSTER detects evidence of stochastic dependence between A 3
and A 17 , A 4 and A u , Ar and An, A 7 and A 9 , A g and An, and A u
and A l2 . Only in regard to two of these pairs — the first and the
last — -is there any danger that the attributes are actually asking
significantly similar (or significantly dissimilar) questions. After
careful consideration it was decided that all of these attributes ask

196 A pragmatic legal expert system §4.4

different questions, and all are important in distinguishing between
employees and independent contractors. So these attributes, too,
remain in the specification.

Weights

The weights file for the Employee specification is given in full
in §B.5. The rightmost column of the table of weights indicates
that three attributes are of equal greatest importance in the Em-
ployee area: 119

A\ 3 : Was the money that the employer paid to the worker stated
to be a "fee"?

Ai^\ Was the money that the employer paid to the worker stated
to be "wages" or "salar}'"?

.Aig: Did the employer pay the worker sick pay or holiday pay 1 ?

The fact that the courts have had regard to so many different
factors when deciding employee cases, makes it difficult to judge
SHYSTER's choice of most important attributes. However, A\
and A-2 should probably be more heavily weighted than they are.

4.4.3 Test cases

The Employee specification was written so as to represent the law
as it was in 1982. Four actual cases are used to test the Employee
specification — cases which were decided after that time. There are
also two hypothetical cases, chosen by the legal expert.

Narich v. CPT 1983

In Narich Pty Ltd v. Commissioner of Pay-roll Tax, 120 Narich held
the Australian franchise of Weight Watchers. Lecturers taught
classes, following a program detailed in the Weight Watchers hand-
book, and deducted their fees from the money that they collected
from class members. Clause 3 of the agreement between the lec-
turers and Narich stated that the lecturers were not employees of
Narich, but independent contractors.

§4.4 Case studies 191

The Commissioner claimed that the lecturers were actually em-
ployees of Narichj and that Narich was liable, under the Pay-roll
Tax Act 1971 (NSW), to pay payroll tax on the money kept by the
lecturers.

The Judicial Committee of the Privy Council cited AMP v.
Chaplin as authority for the principle that a statement in an
agreement that a worker is an "independent contractor" or an
"employee" is not decisive. Their Lordships held that, despite
clause 3, the effect of the agreement between the lecturers and
Narich was that the lecturers were employees of Narich.

SHYSTERS fact vector is (YNYNYNNNNNYYYNNNNY); to quote from
its report file:

. . . the employer directed the manner in which the work was to
be done; the worker was not allowed to use her/his own discretion
in doing an aspect of the work that was not specified beforehand;
the worker was an integral part of the employer's business; the
worker neither owned the tools nor provided the transport with
which she/he performed the work; the emploj^er would make a
profit/loss if the work performed by the worker cost less/more
than expected; the work was not performed on the employer's
premises; the employer neither supervised nor inspected the work;
the worker was not in business on her/his own account; the worker
was not allowed to emploj' others to assist with her/his work;
the worker was not obliged to work only for the employer; the
worker was required to work at specified times; the employer paid
the worker by time; the monej' that the employer paid to the
worker was stated to be a "fee"; the money that the emploj'er
paid to the worker was not stated to be "wages" or "salarj 1- "; the
employer did not deduct PAYE tax instalments from the worker's
pay; the employer paid the worker neither sick pay nor holiday
pay; the employer and the worker did not express any intention
that the relationship would be one of employer and emploj'ee;
and the employer and the worker expressed an intention that the
relationship would be one of principal and independent contractor.

SHYSTER agrees with the Privy Council that the lecturers were
employees, although it warns that the specified directions sug-
gest Contractor. Its table of distances, as extracted from the dis-
tances file, is given in figure 4.7. None of SHYSTER's chosen cases
was cited in the judgment: the nearest neighbour is Ferguson v.
Dawson (C 5 ); the nearest other is Massey v. Crown Life (Ci 2 ).

198 A pragmatic legal expert system

§4-4

<J1

00
CO

•*

-2 4

«o

c ^— -

E -

— -i

-ii

art

-|[

--JI

—,

o ■-<

- 1

«o

o *o

**■

■*r

O .-1

g e

.-1

» o

■v

—

: i

C-l -^

—

"O CO

—
■

o o
1

;

z
1

■

o o
i i

z
1

z
■

z
1

z
i

o o
1

ic o

■

l l

<D >

— t CM

«o

-"■"

—

CO CO

—

CO o

H O

-1

iO o

;

-i

•v

CO -o

—

■w

—

—t CM

—

o o

1 1

o o

1 1

CD SO

t- CM

i-t CM

QJ CD

rH CO

- 1

-T CO

■■*

—

"* "*.

O O

o o

T t-

CO CI

CO o

:■"

CM 1J

IO N

•o

—

«o

iO iO

O O

;

o o

;

o o

+r x

^T SO

e»

O CO

a a

— —

""■

—.

CO CM

-T

CD O

«o ~V

CD
CM

«o

ft *

CO o

«o

w c;

—

CM ~V

"*.

K so

■W

- 1

T **:

•o

«o

i-l IO

i^ ci

-■

" 1

•v —

«o

ci co

CO g

—

rH TJ.

—

•o

'-•'■

>o t-

•o

•V

—

-f -f

TC >o

CO CO

«o

K o

CO J3

•

X X

•

X X

•

• X

> ' +3

ft & o

■£

X X

•

X •

•

—

X X

•

X X

•

X X

■?

-: E 2

•

X X

•

X X

•

X X

^ LU ^

««S x o

o -^ u

X X

•

X X

CS ^ to

-7

•

• X

X X

• X

•

X X

•

X X

•

X X

c2 -B s

*- —

<P —V

•

• X

•

X X

•

X X

CO M *

o o3 *((•

X X

•

X X

•

X X

2 fl
c3 ., M

S^ 3

•

• •

•

• •

X •

•

• X

O o
MH CO QJ

o .. ^

•

• •

•

X •

X X

2 ^ T3

.£

•

• •

•

X X

•

c8 ' — ' a>

X X

•

X X

43 e ^S

-r

a O u

■■■

X •

•

• •

•

• •

r- 5C O

•

• •

•

X X

•

X •

^ .. s

■<

•

X X

•c

•

• •

•

X X

•

X X

^ +o

■

u
O

n f

1- -fl

° 1

2 a i

oil

-1

§4.4 Case studies

Stevens v. Brodribb 1986

In Stevens v. Brodribb Sawmilling Co. Pty Ltd, 121 Brodribb em-
ployed sniggers to fell trees, and truck drivers to carry the trees to
the sawmill. They used their own vehicles, set their own hours of
work, and were paid according to the volume of timber that they
delivered to the sawmill. Stevens, a truck driver, was injured due
to the negligence of a snigger. He claimed that the snigger was
an employee of Brodribb and. hence. Brodribb was vicariously li-
able for the snigger's negligence. He also claimed that he was an
employee of Brodribb and. hence. Brodribb was personally liable
to him for breaching the duty of care owed by an employer to an
employee.

Five justices of the High Court of Australia applied Humber-
stone v. NTM and AMP v. Chaplin, and decided that neither the
truck driver nor the snigger were Brodribb \s employees. 122

SHYSTERS fact vector is (NYYYNUNYYNMNMNM). SHYSTER agrees
with the High Court that Stevens was an independent contractor.
The nearest neighbour is AMP v. Chaplin: one of the cases that
the High Court applied. But none of their Honours referred to
SHYSTER'S nearest other: Cam v. Sargent. SHYSTER issues no
warnings.

Re Porter 1989

In Re Porter; Re Transport Workers Union of Australia, 123 five
truck drivers, who owned their own trucks, had nominated for
various offices in the Victorian branch of the TWU. The returning
officer rejected the drivers' nominations on the grounds that only
members who were employees were eligible for election, and that
the drivers were independent contractors. The drivers challenged
the returning officer's decision.

In the Federal Court, the TWU relied heavily on four Australian
cases (only one of which is part of the Employee specification)
in which owner-drivers of trucks had been held not to be em-
ployees of the company for which they worked: Humberstone v.
NTM, Wright v. Attorney-General for Tasmania, 12,1 Barro Group
Pty Ltd v. Fraser 125 and Stevens v. Brodribb. The union also re-
lied on the English case of Ready Mixed v. Minister. Gray J said

200 A pragmatic legal expert system §4.4

that, despite the consistency of these cases, "a balancing exercise
is always involved in the determination whether an employment
relationship exists." 12 ° On balance, he held, the five drivers were
employees.

The complete report file for Re Porter is given as an example
in §A.4. SHYSTER'S fact vector is (YNYYUNUNNYNYNNYWNY). 127
SHYSTER agrees with Gray J that the drivers were employees.
The nearest neighbour is Ferguson v. Dawson, which was not men-
tioned in the case. The nearest other is Ready Mixed v. Minister:
one of the cases cited by the TWU.

SHYSTER warns that one of the instantiations has a different
result to that of the instant case. If the employer would not make a
profit/loss when the work performed by the drivers cost less/more
than expected (A 5 = no), and the employer neither supervised
nor inspected the drivers' work (A 7 = no) — A 5 and A 7 are the
two attributes with unknown values in the instant case — then
SHYSTER'S opinion would be that the drivers were independent
contractors (following Ready Mixed v. Minister). The legal expert
confirms that this is a sensible distinction to make.

BWIU v. Odco 1991

In Building Workers' Industrial Union of Australia v. Odco Pty
Ltd, 12S Odco provided labour to builders in Victoria. 129 In the
agreement between Odco and the workers it was stated that the
workers were contractors, not employees. They undertook to work
for an agreed hourly rate, regardless of any industrial awards
which might apply. Odco encountered "hostility from within the
trade union movement, especially amongst officials of building
unions" 130 and "incidents" occurred which led Odco to bring an
action against the union for breach of s. 45D of the Trade Prac-
tices Act 1974 (Cth). The question of whether the workers were
employees or independent contractors of Odco was relevant to es-
tablishing a defence for the union under the Act.

The case was heard by the Federal Court. Wilcox, Burchett
and Ryan JJ applied Stevens v. Brodribb and held that the workers
were not employees of Odco. The fact that parties to an agreement
label workers as employees or as independent contractors is not
decisive — their Honours cited Ferguson v. Dawson^however they
held that, in this case, the "contractor" label was correct. 131

§4.4 Case studies 201

BWIU v. Odco is used as the test case in the complete example
in appendix B. The distances file for BWIU v. Odco is given in
full in §B.6; the report file is in §B.7. SHYSTER's fact vector
is (NYNYNNNUNNYUMNNNY). 132

SHYSTER's conclusion is that the workers were independent con-
tractors. Its choice of cases is good. The nearest neighbour is
Humberstone v. NTM (C 8 ). which was applied by the High Court
in Stevens v. Brodribb which was applied by the Federal Court in
BWIU v. Odco. The nearest other is Ferguson v. Dawson (C 5 ),
which their Honours cited in their judgment. No warnings are
issued.

Hypothetical case 3

Both hypothetical case 3 and case 4 are taken from a question set
in an examination for a university employment law course. 133 A
secretary. Amber, is employed in the Department of Legal Studies
in an Australian university. The written contract between the
university and Amber provides that salary, hours, allowances, and
leave shall be in accordance with the terms of a specified award.
Both the university and Amber contribute to a superannuation
scheme for Amber's benefit. Amber's appointment is to continue
until she is 65 years of age. after an initial probationary period.
There is a duty statement for Amber's position, and she performs
duties as required by the Head of the Department.

It is clear, and the legal expert confirms, that Amber is an
employee of the university.

SHYSTER's fact vector is (YNYNYYYNNUYNNYYYYN). It agrees that
Amber is an employee. The nearest neighbour is PRS v. Palais de
Danse; the nearest other is Massey v. Crown Life. Neither is a
good choice.

SHYSTER warns that the weighted correlation coefficients sug-
gest that Stevenson v. Macdonald (1), in which the worker was an
independent contractor, should be the nearest neighbour.

Hypothetical case 4

From the same examination question as hypothetical case 3 comes
another hypothetical case.

Bonny works as a secretary in the same department as Amber.
Bonny is registered with an employment agency. The Department

202 A pragmatic legal expert system §4-4

and the employment agency entered into a contract in which the
agency agreed to provide a competent secretary for a three month
period, and retained the right to provide a substitute at any time.
The university indemnified the employment agency for any liabil-
ity arising out of Bonny's work.

When Bonny registered with the employment agency she signed
a contract which stated that any work which she procured through
the agency she would perform as an independent contractor.
Bonny is paid a specified daily rate, is not eligible for any leave,
and is responsible for injury insurance and taxation liabilities.
Bonny has had her work period extended several times, and has
worked in the Department of Legal Studies for a year. Bonny and
Amber have similar duties and patterns of work. They both follow
guidelines issued by the Head of the Department.

The legal expert advises that, unlike Amber and on the basis of
BWIU v. Odco, Bonny is an independent contractor.

SHYSTERS fact vector is (YNYNYYYNNNYYNNNNNY). SHYSTER dis-
agrees with the expert, and concludes that Bonny is an employee
of the university. The nearest neighbour is Ferguson v. Dawson.
As in this hypothetical case, both the employer and the worker
in Ferguson v. Dawson agreed that the worker was an independ-
ent contractor. 134 In fact, the only difference that SHYSTER sees
between Ferguson v. Dawson and hypothetical case 4 is that, in
the hypothetical case, the work was performed on the employer's
premises. Nevertheless, SHYSTER's choice of nearest neighbour is
a bad one. The nearest other is Massey v. Crown Life: a good
choice. In Massey v. Crown Life, Massey was an employee for two
years, then performed the same duties for another two years as an
independent contractor. The differences between Massey as em-
ployee and Massey as independent contractor, are similar to those
between Amber and Bonny. SHYSTER issues no warnings.

The legal expert advises that one of the aims of this examination
question was to demonstrate the absurdity of the law in this area:
Amber and Bonny perform very similar jobs, yet the law sees
them as having quite different relationships with the university.
Hypothetical case 4 is an example of a tripartite employment re-
lationship: a relationship where one party pays an employment
agency which pays the worker. BWIU v. Odco was the first case in
which a court had to determine the employment status of workers
in such relationships. It was decided in 1991 and so is not part of

§4.4 Case studies 203

the Employee specification. SHYSTER's failure in this hypothet-
ical case illustrates the unsuitability of the Employee specification
for providing advice on such tripartite relationships.

4.4.4 Generated tests

The legal expert provided three examples for generating tests using
the Employee specification: generated tests 4, 5 and 6.

Generated test 4

The legal expert advises that, if the following is true, the worker
is an employee:

. . . the employer directed the manner in which the work was to be
done; the worker was an integral part of the employer's business;
the employer would make a profit/loss if the work performed by
the worker cost less/more than expected; the employer supervised
or inspected the work; the worker was not allowed to employ others
to assist with her/his work; the money that the employer paid to
the worker was stated to be "wages" or "salary"; the employer
deducted PAYE tax instalments from the worker's pay; and the
employer paid the worker sick pa}' or holiday pay.

SHYSTER'S fact vector is (YUYUYUYUNUUUUYYYUU). It generates
1024 instantiations: a mere l/256th of the search space. How-
ever, it is not possible that the money that the employer paid
to the worker was stated to be both a "fee" and "wages." This
paradox occurs whenever the values of both A i3 and A\ 4 are yes.
Because A I4 is always yes, changing A l3 from unknown to no
avoids these impossible cases and halves the size of the search
space. In 149 of the 512 instantiations in this reduced search space,
SHYSTER's choice of result is bad (i.e. Contractor): a success rate
of 70.90%.

In all 149 the nearest ideal point is that of the expected result;
and in all but 4 of those 149 : SHYSTER warns that that ideal
point is at least as near to the instantiation as is the nearest
neighbour. There is no (originally unknown) attribute for which
these 149 instantiations have common values. However, in 131
(or 87.92%) of them the worker was in business on her/his own

204 A pragmatic legal expert system §4.4

account (A& = yes). The legal expert advises that, although this
(together with the originally known attribute values) does not
amount to a paradox, it is an almost inconceivable combination of
attribute values.

Generated test 5

The legal expert also advises that, if the following is true, the
worker is an independent contractor:

. . . the employer would not make a profit/loss if the work per-
formed by the worker cost less/more than expected; the worker
was in business on her/his own account; the worker was allowed
to employ others to assist with her/his work; the employer did
not deduct PAYE tax instalments from the worker's pay; and the
employer paid the worker neither sick pa}' nor holiday pay.

Given the fact vector (UUUUNUUYYUUUUUNNUU), SHYSTER generates
8192 instantiations: l/32nd of the search space. Once again, a
paradox occurs whenever the values of both A\ 3 and A i4 are yes.
Removing these impossible cases reduces the size of the search
space by a quarter. In 1946 of the 6144 instantiations in this
reduced search space, SHYSTER'S choice of result is bad (i.e. Em-
ployee): a success rate of 68.33%.

Despite the fact that the nearest ideal point is that of the expec-
ted result in 1353 of those 1946, only 492 of them cause SHYSTER
to issue a warning. Furthermore, in 111 instantiations SHYSTER
issues an ideal point warning, even though the choice of result is
good.

Generated test 6

The legal expert further advises that, if the following is true, the
worker is an independent contractor:

. . . the worker was allowed to use her/his own discretion in do-
ing an aspect of the work that was not specified beforehand; the
worker owned the tools or provided the transport with which
she/he performed the work; the emploj^er would not make a profit/
loss if the work performed by the worker cost less/more than

§4.4 Case studies 205

expected; the worker was in business on her/his own account;
the money that the employer paid to the worker was stated to be
a "fee"; the employer did not deduct PAYE tax instalments from
the worker's pay; and the employer paid the worker neither sick
pay nor holiday pa}'.

SHYSTER'S fact vector is (UYUYNUUYUUUUYUNNUU). It generates
2048 instantiations. As with generated test 4, removing the para-
doxical cases halves the size of the search space. In 149 of the
1024 instantiations in this reduced search space, SHYSTER's choice
of result is bad (i.e. Employee): a success rate of 85.45%.

In all 149 the nearest ideal point is that of the expected res-
ult; and in all but 37 of these 149, SHYSTER warns that that ideal
point is at least as near to the instantiation as is the nearest neigh-
bour. The fact that, in 120 (or 80.54%) of these 149, the work was
performed on the employers premises (A 6 = yes) is not enough
to justify SHYSTER's choice of result in these instantiations.

4.4.5 Conclusion: the Employee specification

In all but one (i.e. 83.34%) of the test cases for the Employee
specification described above. SHYSTER reaches a good conclusion.
But only 41.67% of the cases that it chooses to use in argument
are good.

As shown in figure 4.13. SHYSTER's choice of result is good
in 70.78% of the instantiations in all three generated tests. (This
takes into account the reduction of the size of the search space
due to the removal of instantiations where the values of both .4x3
and A 14 are yes. Such instantiations do not represent plausible
cases.)

This performance is dominated by generated test 5: 135 the test
which produced, from SHYSTER, the worst performance of all the
generated tests described in this chapter. Alone amongst all these
generated tests, this test causes SHYSTER to issue ideal point
warnings for instantiations in which its choice of result is good.
If a bad choice of result is considered good where an ideal point
warning is issued and a good choice of result is considered bad
where an ideal point warning is issued. SHYSTER's success rate
for the three generated tests of the Employee specification rises
to 99.22%, 74.53% and 96.39%, respectively: 79.09% overall.

206 A pragmatic legal expert system §4.4

There are two tripartite employment cases amongst the test cases:
BWIU v. Odco and hypothetical case 4. In both of these tests, it is
important to be clear as to who is the "employer" for the purposes
of answering SHYSTER'S questions: for BWIU v. Odco the employer
was taken to be Odco (the employment agency); for hypothetical
case 4 it was taken to be the university. Although SHYSTER'S
choice of result and cases for BWIU v. Odco is good, the ability
of the Employee specification to handle tripartite employment
relationships is stretched by hypothetical case 4.

The Employee area also demonstrates one of SHYSTER'S minor
shortcomings. In its report, SHYSTER's summary of the facts in
a case can make difficult reading when, as here, there is a large
number of attributes. The wording of the strings for the last two
attributes means, for example, that the statement of facts in SHY-
STER'S report on Stevens v. Brodribb is a 259-word sentence that
concludes:

. . . the employer and the worker did not express any intention that
the relationship would be one of employer and employee; and the
employer and the worker did not express any intention that the
relationship would be one of principal and independent contractor.

It is to be hoped that no human legal expert would express the
facts so clumsily. However, this problem is inherent in SHYSTER's
method of writing reports; possible solutions are discussed in the
next chapter. 136

The Employee area could be linked to several statutory open-
textured concepts, as represented in a rule-based system. It could
also be linked to areas of case law (like the Authorization area) in
which the distinction between employees and independent con-
tractors is important.

SHYSTER, when using the Employee specification, chooses good
results in most of the generated tests. It is capable of providing
fairly good advice in cases involving bipartite employment rela-
tionships, as it was designed to do. Although it is not so well
suited to cases involving tripartite relationships, it performs well
in an area of law in which, according to Clark and Wedderburn.
the juridical concepts "approach anarchy." 137

§-1.5 Case studies 201

4.5 The implication of natural justice

The last specification discussed in this chapter concerns the im-
plication of natural justice principles in administrative decision-
making.

4.5.1 The law

The law in Australia has developed the notion of affording nat-
ural justice to a person who is affected by an administrative de-
cision. Natural justice principles might require, for example, that
the person affected be given the right to a hearing — to make
representations — before the decision is made. Or the decision-
maker might be obliged to give reasons for the decision, after it
has been made. These requirements are sometimes called require-
ments of "procedural fairness."

The development of this area of law has been convoluted:

The evolution of the doctrine of natural justice has been a journey
in fluctuation, typified by the regular pitch of basic principles from
one leaning to another. 138

In Kioa v. West, Mason J said:

The law has now developed to a point where it may be accep-
ted that there is a common law duty to act fairly, in the sense of
according procedural fairness, in the making of administrative de-
cisions which affect rights, interests and legitimate expectations,
subject onty to the clear manifestation of a contrary statutory
intention. 139

But McMillan asserts that:

The main residual difficulty concerning the implication of natural
justice has been the tendencj' to apply uncritically [this] superficial
test . . . the affect [sic] of a decision on a person's interest is only
one of man}' - factors that is relevant to deciding whether a natural
justice obligation attaches to the decision. 140

The notion of "legitimate expectations" expands the area con-
siderably. Legitimate expectations "are capable of including ex-
pectations which go beyond enforceable legal rights, provided they
have some reasonable basis". 141

208 A pragmatic legal expert system §4-5

McMillan discusses different ways of determining whether natural
justice principles apply to a given administrative procedure:

An option at one end of the spectrum has always been to treat
the obligation of procedural fairness as attaching universally to
all decisions, and to concentrate instead on the practical con-
tent of that obligation in any particular instance. This option
. . . was described by Deane J in Haoucher as "conceptually more

• c_ ■ » 142

satisfying .

However, that approach is too broad. As Wilcox J points out
in Minister for Arts, Heritage and Environment v. Peko-Wallsend
Ltd:

. . . the law has not yet reached the stage of applying the ob-
ligation of natural justice to every decision which disadvantages
individuals. 143

McMillan proposes an alternative:

... a preferable approach would be to determine whether natural
justice applied by examining in each case a number of different
factors. While the weight attached to some factors is clearly
greater, it is the overall balance that is important. This ap-
proach was propounded by the Privy Council in Durayappah v.
Fernando, 144 was initially applied in Australia in Salemi v. Mac-
Kellar [No. 2j 145 and FA I Insurances Ltd v. Winneke 146 but has
not been developed further. 147

His factors are:

• the nature of the property, right, interest, status, or legitimate
expectation;

• the effect or impact of the decision;

• the nature of the power being exercised;

• the statutory and factual criteria according to which the de-
cision was made;

• the nature of the officer making the decision:

• the statutory procedural framework under which the decision
was made; and

• the circumstances in which the decision was made. 148

§4.5 Case studies 209

McMillan discusses these factors in detail. I discuss next the law
relating to one of these factors because it significantly affects one
of the tests described below. 149

The statutory procedural framework under which the decision was
made may grant a person a right to appeal against a decision. The
existence of a statutory right to appeal used to suggest that natural
justice would not be implied. For example, in Twist v. Randwick
Municipal Council, Mason J stated:

Having regard to the subject matter of the section, . . . and more
particularly the comprehensive nature of the appeal to a District
Court judge, I am of opinion that s. 317b(5) [granting the right of
appeal] should be read as providing the exclusive remedy available
to an owner who wishes to challenge the validity or correctness of
an order made under s. 317b(1). 150

And, in the first Marine Hull and Liability Insurance Co. Ltd v.
Hurford, Wilcox J held that the Treasurer should have applied the
principles of natural justice before making the contested direction
—however, because the relevant Act provided for a review of any
such directions by the Administrative Appeals Tribunal:

. . . the legislature must be taken to have evinced an intention
that, in the event of the Treasurer failing to so act, the directions
are not to be regarded as being invalid in law. They are merely
susceptible of challenge before the Tribunal. 151

Recently, there has been a tendency to interpret the existence of
a statutory right to appeal as indicating that natural justice should
be implied. In the second Marine Hull v. Hurford, for example.
Davies J stated that:

The existence of a right to have a matter reconsidered and of
a right to have a matter reviewed by the Administrative Appeals
Tribunal may well affect the nature of the procedures which ought
to be adopted in complying with the rules of natural justice but,
ordinarily, it does not exclude them. 152

And in Bropho v. Western Australia, Rowland J held that the ex-
istence of a right of appeal to the Supreme Court indicated that
the decision-maker should be influenced by the same considera-
tions as would a judge of that Court; it indicated that questions
of natural justice should not be excluded. 153

210 A pragmatic legal expert system

§4.5

Natural area

Affected area

^^^

(a)

A 2

A 3

A 5

A 6

A 7

A 1

A 4

Expectation area

A 3

-4,

-4,;

Figure 4.8: The relationship between the attributes in the
three areas which make up the Natural specification.

Rowland J's approach is diametrically opposed to that of the
High Court in Twist v. Randwick, decided fourteen years earlier.
His decision was overturned on appeal by the Full Court of the
Supreme Court of Western Australia in Western Australia v.
Bropho} * though on a different point: 150 the Full Court did not
refer to the existence of a right of appeal. Although the law may
be evolving towards Rowland J's interpretation, it has not yet
reached that stage.

4.5.2 The Natural specification

Having consulted a legal expert in the field. 150 and the published
judgments in the leading cases, I wrote a SHYSTER specification
on the implication of the duty to observe natural justice. The
specification is based on the factors identified by McMillan. 157

The Natural specification contains three areas: the Natural
area, the Affected area and the Expectation area. (The relationship
between the attributes in these areas is illustrated in figure 4.8. lo8 )
There is a hierarchy which ranks nine courts, and the Natural and
Expectation areas both have opening strings.

§4.5 Case studies 211

Natural area

The Natural area has two results: Implied and Not-Implied. These
correspond to u a duty to observe natural justice is implied 11 and
u a duty to observe natural justice is not implied."

There are seven attributes in the Natural area, corresponding to
the seven factors identified by McMillan. The term applicant is
used to refer to the person who seeks the implication of natural
justice:

A\: Did the decision affect the property, right, interest, status, or
legitimate expectation of the applicant?

^2^ Is the decision apt to have a discrete impact on the interests
of the applicant?

^3: Is the power of a nature that would suggest that procedural
fairness would be applied?

A4: Did the statutory or factual criteria focus on matters which
were discrete to the interests of the applicant?

^ 5

A 7

Was the decision-maker a high-level policy-maker?

Is there a statutorj' right to appeal against the decision? 159

Were there circumstances which make an obligation to ob-
serve natural justice inappropriate?

Only A G is straightforward. But (with the exception of A x ) an-
swering the other questions, with the assistance of the help strings
which are included in the specification, should present no major
difficulties. For example, the help string for A 4 is:

The decisional criteria are of two kinds: the spectrum of consider-
ations to which the decision maker was authorized to have regard
(the statutory criteria), and the specific considerations to which
regard was had in fact (the factual criteria). Either set of criteria
can focus on matters which are discrete to the interests of the
applicant, or on matters of policy or public interest.

And the help string for A$ explains that:

Ministers, members of Cabinet, Governors and the Governor-
General are high-level policy-makers.

However, Ai is such a difficult question to answer that it is
defined as an external attribute, linked to the Affected area. The
user is never asked the A\ question, but is asked questions from
the Affected area instead.

212 A pragmatic legal expert system §4-5

Affected area

The Affected area has two results: Affected and Unaffected. These
correspond to "the decision affected the property, right, interest.
status, or legitimate expectation of the applicant" and "the de-
cision did not affect [those things]. 11 A result of Affected produces
a value of YES for A x in the Natural area; a result of Unaffected
produces a value of no.

There are four attributes in the Affected area:

A\\ Did the decision affect a financial, property or occupational
interest of the applicant?

Ao'. Did the decision affect the applicant's personal liberty?

.A3: Did the decision affect the applicant's reputation?

A4: Did the applicant have a legitimate expectation which was
affected by the decision?

Only A 4 is not a straightforward question. It, too, is defined as
an external attribute, and linked to the Expectation area.

Expectation area

The Expectation area also has two results: Expectation and No-
Expectation. These correspond to "the applicant had a legitimate
expectation which was affected by the decision 11 and "the applic-
ant did not have a legitimate expectation which was affected by
the decision." A result of Expectation produces a value of yes for
A4 in the Affected area; a result of No-Expectation produces a value
of NO.

There are six attributes in the Expectation area:

A\: Did the decision-maker break a promise or undertaking?

A 2 ' Did the decision-maker go against an established course of
practice?

A3: Did the decision involve a refusal to renew an existing in-
terest?

A 4 : Did the decision-maker or a statutory provision suggest that
an initial interest would be granted?

A$: Did the decision affect an established liberty or interest?

Aq: Was there a standard administrative procedure which the
decision-maker did not follow?

§-1.5 Case studies 213

There is a subtle distinction between A\, A 2 and A G which is best
illustrated by two examples.

In Haoucher v. Minister for Immigration and Ethnic Affairs, 160
the Minister had told Parliament that recommendations of the Ad-
ministrative Appeals Tribunal, on deportation matters, would be
overturned only in "exceptional circumstances 11 and when justified
by "strong evidence." The AAT recommended that the Minister's
order to deport Haoucher be revoked, but the Minister rejected
the recommendation, without giving Haoucher details of any "ex-
ceptional circumstances" or "strong evidence."

The value of A x is no: there may indeed have been "exceptional
circumstances" — the Minister did not promise to give details of
those circumstances. A 2 is yes: the Minister had an established
course of practice of following AAT recommendations. However,
Aq is no: the Minister went against practice, not procedure.

In Attorney-General of Hong Kong v. Ng Yuen Shiu, 161 an im-
migration official had announced that certain illegal immigrants
would be interviewed and, in deciding whether to allow them to
stay, each case would be "treated on its merits. 11 Ng (an illegal
immigrant) came forward the next day and was questioned, but
given no opportunity to put the merits of his case.

Although the decision-maker broke a promise (A t = yes), there
had not been time to establish a course of practice (A 2 = no), and
there was no standard procedure to be followed (A G = no).

Leading cases

There are fifteen leading cases in the Natural area. Seven of them
were decided by the High Court of Australia: Commissioner of
Police v. Tanos 162 Bread Manufacturers of New South Wales v.
Evans, 163 FAI v. Winneke 164 Kioa v. West, 165 South Australia v.
O'Shea 166 Haoucher v. Minister 167 and Annetts v. McCann. 168
Two were decided by the Federal Court of Australia: Marine
Hull v. Hurford 169 and Minister v. Peko-Wallsend. 17<} Two were
decided by the Supreme Court of New South Wales: Nashua Aus-
tralia Pty Ltd v. Channon 171 and Macrae v. Attorney-General for
New South Wales. 172 Two are decisions of the Judicial Commit-
tee of the Privy Council: Durayappah v. Fernando 173 and AG v.
Ng. 174 One was decided by the House of Lords: Council of Civil
Service Unions v. Minister for the Civil Service. 175 And one was a
decision of the Chancery Division of the English Court of Appeal:
Mclnnes v. Onslow-Fane. 176

214 A pragmatic legal expert system §4-5

Nine of these are also leading cases in the Affected area. And six
of them, plus two further cases — Heatley v. Tasmanian Racing and
Gaming Commission 177 and Cole v. Cunningham 178 — are used in
the Expectation area. For reasons discussed below, 179 the prom-
inent High Court cases of Twist v. Randwick 180 and Salemi v.
MacKellar 181 were not used in the Natural specification.

A peculiarity of the Natural specification is that, as SHYSTER
warns, there are three pairs of cases in the Natural area which
have identical facts and different results. 182 The existence of two
or more leading cases with identical facts but different results in-
dicates either that one of the cases was wrongly decided, or that
there is a need for more attributes in the area (to distinguish
between the cases). All of the leading cases in the Natural area
were correctly decided, and the attributes in that area are based
on the factors identified as being important by McMillan. Hence,
no change was made to the specification. If two or more of these
cases with identical facts but different results are the nearest cases,
SHYSTER warns the user and chooses the most important of them
as the nearest neighbour.

Attribute dependence

The probabilities matrices for the three areas in the Natural spe-
cification, as extracted from the probabilities file, are given in
figure 4.10. SHYSTER detects no functional dependence, and no
evidence of stochastic dependence, between any of the attributes
in any of the three areas.

Weights

The tables of weights for the three areas in the Natural specific-
ation, as extracted from the weights file, are given in figure 4.9.

In the Natural area, the rightmost column indicates that two
attributes are of equal greatest importance: 183

Ai'. Did the decision affect the property, right, interest, status, or
legitimate expectation of the applicant?

Aj\ Were there circumstances which make an obligation to ob-
serve natural justice inappropriate?

§4.5

Case studies 215

Attr.

Implied

/i <T~ U'

Not-Implied
ft a w

2
<7

1.00 0.00 oo

0.83 0.14 7.20

0.93

0.06

16.07

A 2

1.00 0.00 oo

0.67 0.22 4.50

0.87

0.12

8.65

0.67 0.22 4.50

0.33 0.22 4.50

0.53

0.25

4.02

A 4

0.63 0.23 4.27

0.17 0.14 7.20

0.43

0.24

4.08

A 5

0.78 0.17 5.79

0.67 0.22 4.50

0.73

0.20

5.11

0.22 0.17 5.79

0.00 0.00 oo

0.13

0.12

8.65

A 7

0.00 0.00 oo

0.17 0.14 7.20

0.07

0.06

16.07

Attr.

Affected

ft <T" W

Unaffected
ft a~ w

0.50 0.25 4.00

0.00 0.00 oo

0.44

0.25

4.05

A 2

0.38 0.23 4.27

0.00 0.00 oo

0.33

0.22

4.50

A 3

0.38 0.23 4.27

0.00 0.00 oo

0.33

0.22

4.50

A 4

0.50 0.25 4.00

0.00 0.00 oo

0.44

0.25

4.05

Attr.

Expectation
ft a~ w

No-Expectation
ft a w

2
<7

it'

0.33 0.22 4.50

0.00 0.00 oo

0.25

0.19

5.33

A 2

0.50 0.25 4.00

0.00 0.00 oo

0.38

0.23

4.27

A 3

0.17 0.14 7.20

0.00 0.00 oo

0.13

0.11

9.14

A 4

0.00 0.00 oo

0.00

A 5

0.83 0.14 7.20

0.50 0.25 4.00

0.75

0.19

5.33

A 6

0.17 0.14 7.20

0.00 0.00 oo

0.13

0.11

9.14

Fig

ure 4.

9: The tables c

f weights for th

e three areas in

the

Natural s

pecification: th

e Natural area, t

le Affected

area and

the Expect;

tion area (SHYS

TER output).

SHYSTER'S weighting of attributes in this area is fairly good. A x
and A 7 are very important attributes. Indeed if the value of A x
is no. then natural justice should never be implied. 184 However.
the legal expert advises that A 2 should be weighted more heavily
than it is.

All four attributes in the Affected area are weighted almost the
same. 185 This is appropriate — each of these attributes is equally
indicative of the applicant having been affected.

216 A pragmatic legal expert system

§4.5

A 2

A 3

-44

A 5

A G

A 7

1.00

0.13

1.00
0.47

1.00
0.57

0.73
1.00

1.00
0.87

1.00
0.93

1.00
0.20

1.00
0.31

0.94
0.48

1.00
0.74

1.00
0.87

A 2

0.95
0.30

0.34
0.95

1.00
0.27

0.47
1.00

A 3

0.17
0.99

0.84
0.69

0.57
1.00

A 4

0.48
0.94

1.00
0.73

0.87
1.00

A G

A 2

A 4

0.12
1.00

0.95
0.40

0.83
0.64

0.24
1.00

0.95
0.40

A 2

0.60
0.88

A 3

A 2

A 5

A 4

A 5

A 6

0.36
1.00

0.75
1.00

1.00
1.00

0.46
0.96

0.75
1.00

1.00

0.38

1.00
1.00

1.00
0.36

0.03
1.00

A 2

1.00
1.00

1.00
0.75

0.88
1.00

A 3

1.00
1.00

1.00

0.75

A 5

Figure 4.10: The probabilities matrices for the three areas in
the Natural specification: the Natural area, the Affected area
and the Expectation area (SHYSTER output).

§-1.5 Case studies 211

One of the attributes in the Expectation area has infinite weight:

A 4 : Did the decision-maker or a statutory provision suggest that
an initial interest would be granted?

This is an important attribute, but its infinite weighting probably
weights it too highly in relation to the other attributes in this

4.5.3 Test cases

Eight cases are used to test the Natural specification. Seven
of these were considered for the specification, but deemed not to
be leading cases. The eighth case, Ainsworth v. Criminal Justice
Commission, 18 ' was decided after the specification was written.

Twist v. Rand wick 1976

In Twist v. Randwick Municipal Council, 18 * Twist owned land in
Randwick. The Council resolved, under s. 317b(1) of the Local
Government Act 1919 (NSW), to have a building on his land de-
molished. Twist was given 60 days to demolish the building, and
then an extension of another three months.

Section 317b(5) gave Twist a right of appeal to the District
Court against the demolition order, but the Court's rules required
that such appeals be made within 60 days. Twist did not appeal
in time, and was not allowed an extension of time.

When the Council informed Twist that it would demolish the
building, he appealed to the New South Wales Supreme Court, and
then to the High Court. Twist claimed that the Council was bound
by the rules of natural justice to have given him an opportunity
to be heard before the demolition order was made. He relied on
Ridge v. Baldwin, 189 a decision of the House of Lords. The Council
relied on the nineteenth century English case of Vestry of St James
and St John, Clerkenwell v. Feary. 190 (Neither decision is part of
the Natural specification.)

Three justices of the High Court unanimously found against
Twist. Barwick CJ and Mason J held that the existence of a right
of appeal showed that Twist had no right to be heard before the
Council made its demolition order. 191 Jacobs J disagreed, but held
that the existence of a right of appeal did show that the Council's
failure to allow a hearing did not render its decision void. 192

218 A pragmatic legal expert system §4-5

SHYSTER'S fact vectors are (NNNNYN). (YNNN) and (YYYYNYN)— for
the Expectation. Affected and Natural areas, respectively. To quote
from its report files:

. . . the decision-maker did not break a promise or undertaking;
the decision-maker did not go against an established course of
practice; the decision did not involve a refusal to renew an exist-
ing interest; neither the decision-maker nor a statutory provision
suggested that an initial interest would be granted; the decision af-
fected an established liberty or interest; and there was no standard
administrative procedure which the decision-maker should have
followed.

. . . the decision affected a financial, property or occupational in-
terest of the applicant; the decision did not affect the applicant's
personal liberty; the decision did not affect the applicant's repu-
tation; and the applicant did not have a legitimate expectation
which was affected by the decision.

. . . the decision affected the property, right, interest, status, or
legitimate expectation of the applicant; the decision is apt to have
a discrete impact on the interests of the applicant; the power is
of a nature that would suggest that procedural fairness would
be applied; the statutory or factual criteria focused on matters
which were discrete to the interests of the applicant; the decision-
maker was not a high-level policy-maker; there is a statutory right
to appeal against the decision; and there were no circumstances
which would have made an obligation to observe natural justice
inappropriate.

In the Expectation area, SHYSTER concludes (as did the High
Court) that Twist did not have a legitimate expectation which was
affected by the decision. The nearest neighbour is SA v. O'Shea:
the nearest others are Haoucher v. Minister and Heatley v. TRGC.
SHYSTER warns that the specified directions suggest Expectation.

In the Affected area, SHYSTER concludes that the decision
affected Twist's property, right, interest, status, or legitimate
expectation. The nearest neighbour is Bread Manufacturers v.
Evans; the nearest other is Minister v. Peko-Wallsend.

SHYSTER's conclusion in the Natural area is that a duty to ob-
serve natural justice is implied. Its table of distances, as extracted
from the appropriate distances file, is given in figure 4.11. The
nearest neighbour is Commissioner v. Tanos (C 5 ), one of the cases
to which Mason J referred in coming to his decision; the nearest
other is Mclnnes v. Onslow-Fane (C15).

§4.5

Case studies 219

>^ to

T. < fl>

■c

■u

ffi ^

—

■*

OT S

+
8

1 -
on

-ft

a; -

M ^3

03 43

2 S

—

'J'

-t

^t 1

i-H

'~.

on
o

t-
o

u
i— i

r-
o

1--
C

■1
o

(TV

<T»

■

<£>

«J

(7>

IfJ

43 2 Si

5«j

■.1

I-

'~.

r-H

Z-

—

r-i

-r

—

i-H

—

•o

i—i

■fl*

u . ,e

C*l

—

1--

<£>

Randwi
sioner v
nslow-F

Z-

—

■-"

'J 1

I -

—

i-H

—

1— 1
o

c-i

TO
O

M
O

T- 1

< -2 ^

i-H

-•:

■q*

tfi

1—1

—

« S si

3 o 3

-~ s

f_. CO hS

tTl

■

fTt

r—

Tf*

S f

ur i
Mc

1 -

1£>

— ,

\ -

— .

1 -

Of)

—

1—1

If)

<D O OT

i-H

—

I -

i ~

94 H
C3 b0 o

^"S^

•

MH -S 43

o | ■

•

■

•

J ^

(6 C r-

X ^

•

■

•

<1> -d ^

*t ^

•

■

•

^^r.

•

• "D

_, 4> -5

•

4.1

utpi
Imp

—

'_"

—

—
—

—

■H

—

2 1

z
— ■

Figure

STERo
result is

220 A pragmatic legal expert system §4-5

This third conclusion is at odds with the decision of the High
Court. However, as discussed above, 193 the significance of the ex-
istence of a statutory right to appeal has changed over the years
since Twist v. Randwick was decided. I respectfully submit (and
the legal expert agrees) that Twist v. Randwick was wrongly de-
cided, at least insofar as it concerns the significance of a statutory
right to appeal. The case was not included in the Natural spe-
cification for this reason.

Apart from Commissioner v. Tanos, all of SHYSTER's chosen
cases were decided after Twist v. Randwick.

Salemi v. MacKellar 1977

In Salemi v. MacKellar [No. 2], 194 Salemi was an Italian citizen
who entered Australia under a temporary entry permit. He stayed
in the country after the permit expired, and became a "prohib-
ited immigrant" within the meaning of the Migration Act 1958
(Cth). The Minister of Immigration announced, in a press re-
lease, an amnesty under which prohibited immigrants who met
certain conditions would be granted resident status if they came
forward before a certain date. Salemi appeared to meet those con-
ditions, but the Minister refused his application for the benefit of
the amnesty and proposed to deport him under s. 18 of the Act.

Salemi cited Schmidt v. Secretary of State for Home Affairs 195
and claimed that the news release had given him a legitimate
expectation" that he would be entitled to remain in Australia.
Hence the Minister should have given him an opportunity to be
heard before deciding to issue the deportation order.

A majority of the High Court held that Salemi had no "legitim-
ate expectation." 19b and half of the Court held that the legislative
scheme indicated that Parliament did not intend that people af-
fected by deportation orders issued under s. 18 should have a right
to be heard. 197

SHYSTER'S fact vectors are (YUNYYN). (UYNY) and (YYYYYNN).

In the Expectation area, SHYSTER concludes that the applicant
did have a legitimate expectation which was affected by the de-
cision. The nearest neighbour is AG v. Ng: the nearest other is
SA v. O'Shea. SHYSTER warns that the weighted correlation coef-
ficients suggest that SA v. O'Shea should be nearest neighbour.

§4.5 Case studies 221

In the Affected area, SHYSTER concludes that the decision af-
fected the property, right, interest, status, or legitimate expecta-
tion of the applicant. Although not discussed in the case (apart
from the issue of legitimate expectation) this is clearly true. The
nearest neighbours are Haoucher v. Minister and Kioa v. West:
the nearest other is Minister v. Peko-Wallsend.

In the Natural area, SHYSTER disagrees with the High Court
and concludes that a duty to observe natural justice is implied.
The nearest neighbour is Kioa v. West: the nearest other is SA v.
O 'Shea. However, according to the legal expert, it is now generally
believed that Salemi v. MacKellar was wrongly decided. Salemi
did have a legitimate expectation, and natural justice should have
been implied. The case was not included in the Natural specific-
ation for this reason.

All of SHYSTER'S chosen cases were decided after Salemi v. Mac-
Kellar.

Heatiey v. TRGC 1977

Heatley v. Tasmanian Racing and Gaming Commission 198 is one
of the leading cases in the Expectation area. It is not a leading case
in either of the Affected or Natural areas, so it is used here to test
those two areas.

In Heatley v. TRGC. Heatley was warned off racecourses in Tas-
mania by the Commission, using its powers under s. 39(3) of the
Racing and Gaming Act 1952 (Tas). Heatley had been given no
notice that the Commission intended warning him off (and hence
was given no opportunity to make representations to the Commis-
sion) and was given no reasons.

A majority of the High Court held that the Commission was
bound by the rules of natural justice. 199 In the absence of an
emergency, the Commission should have given Heatley notice of its
intention to warn him off. and the grounds for taking such action.
Further, Heatley should have been given an opportunity to make
representations to the Commission before it made its decision.
Aickin J (with whom Stephen and Mason JJ agreed) held that
all members of the public — including Heatley — have a legitimate
expectation that they will be allowed onto racecourses. 200

222 A pragmatic legal expert system §4-5

SHYSTER'S fact vectors are (UNYY) and (YYYUNNN).

In the Affected area. SHYSTER agrees with the High Court that
the decision affected the property, right, interest, status, or le-
gitimate expectation of the applicant. The nearest neighbour is
FA I v. Winneke: the nearest other is Minister v. Peko-Wallsend.

In the Natural area, SHYSTER concludes, as did the High Court,
that a duty to observe natural justice is implied. The nearest
neighbour is Annetts v. McCann] the nearest other is Mclnnes v.
Onslow-Fane.

However. SHYSTER warns that these two cases are equidistant
from Heatley v. TRGC; it chooses Annetts v. McCann because it
is a decision of five justices of the High Court whereas Mclnnes v.
Onslow-Fane is merely a decision of the Chancer}' Division of the
English High Court. It also warns that both the weighted as-
sociation and weighted correlation coefficients prefer Mclnnes v.
Onslow-Fane as nearest neighbour.

All of SHYSTER's chosen cases were decided after Heatley v.
TRGC, however the legal expert points out that the cases it has
chosen in the Affected area both concerned licence interests; SHY-
STER'S choice would have been better if each chosen case had
concerned a mere interest.

Cole v. Cunningham 1983

Like Heatley v. TRGC, Cole v. Cunningham 201 is one of the leading
cases in the Expectation area. It is not a leading case in either of
the Affected or Natural areas, so it is used here to test those two
areas.

In Cole v. Cunningham, Cunningham had been encouraged to
resign from the public service because his superiors believed he
had been guilty of misconduct in the performance of his duties. He
had formed an attachment and begun to live with a Fijian woman
whose application for an extension of her temporary entry permit
he had processed. He was threatened with criminal prosecution
for harbouring an illegal immigrant, and told that "[i]f you resign
now it will be a normal resignation and you'll leave with a clean
record." 202 He resigned.

About eighteen months later. Cunningham sought reappoint-
ment to the public service and was told that he would be offered
a position subject to police and security clearances. The next day
he was told that he had been given an unsatisfactory report based
on the earlier events.

§4.5 Case studies 223

Bow en CJ, Sheppard and M or ling J J of the Federal Court of
Australia held that, in general, applicants for appointment or re-
appointment to the public service are not entitled to have natural
justice principles applied because those applicants have no legit-
imate expectation which can be affected by a refusal to appoint.
However. Cunningham did have a legitimate expectation that any
decision to reappoint him would not be made on the basis of his
past record. 203

SHYSTERS fact vectors are (YNNY) and (YYNYNNN).

In the Affected area. SHYSTER concludes that the decision af-
fected the property, right, interest, status, or legitimate expecta-
tion of the applicant — as did the Full Court of the Federal Court.
The nearest neighbour is CCSU v. Minister: the nearest other is
Minister v. Peko-Wallsend.

SHYSTER agrees with the Court that a duty to observe natural
justice is implied. The nearest neighbour is Annetts v. McCann:
the nearest other is Mclnnes v. Onslow-Fane.

As with Heatley v. TRGC, SHYSTER warns that these two cases
are equidistant from Cole v. Cunningham and chooses Annetts v.
McCann over Mclnnes v. Onslow-Fane because it was decided in
a higher court. It also warns that both the weighted coefficients
prefer Mclnnes v. Onslow-Fane.

Apart from Mclnnes v. Onslow-Fane, to which no reference was
made by the Court, all of SHYSTER'S chosen cases were decided
after Cole v. Cunningham. The legal expert approves of SHY-
STER'S choice of nearest neighbour in the Affected area, but not
its choice of nearest other: Minister v. Peko-Wallsend concerned a
property interest, where Cole v. Cunningham concerned a promis-
sory interest.

Ackroyd v. Whitehouse 1985

In Ackroyd v. Whitehouse (Director of National Parks and Wildlife
Service), 204 Ackroyd was a bird trapper. His licence was cancelled
without notice, hearing or reasons, by the Director under s. 134(1)
of the National Parks and Wildlife Act 1974 (Cth). Section 135
of the Act granted a person whose licence had been cancelled the
right to appeal to the Minister. Ackroyd claimed that the Director
ought to have applied natural justice principles before making his
decision. The Director cited Twist v. Randwick claiming that the
right of appeal to the Minister meant that natural justice ought not

224 A pragmatic legal expert system §4-5

to be implied. The New South Wales Court of Appeal disagreed,
and found for Ackroyd.

Kirby P held that the fact that Ackroyd had had his licence
renewed annually for several years "might reasonably give rise to
a legitimate expectation" that the licence would continue in oper-
ation unless some misconduct were proved. 203

SHYSTER'S fact vectors are (NUYNYN). (YNNY) and (YYYYNYN).

In the Expectation area. SHYSTER agrees with Kirby P that
Ackroyd had a legitimate expectation which was affected by the
decision. The nearest neighbour is FAI v. Winneke; the nearest
other is SA v. O'Shea. which was decided after Ackroyd v. White-
house.

In the Affected area, SHYSTER concludes (as did the Court) that
the decision affected the property, right, interest, status, or le-
gitimate expectation of the applicant. The nearest neighbour is
CCSU v. Minister; the nearest other is Minister v. Peko-Wallsend,
which was decided after Ackroyd v. Whitehouse.

Finally, in the Natural area. SHYSTER agrees with the Court of
Appeal that a duty to observe natural justice is implied. The
nearest neighbour is Commissioner v. Tanos; the nearest other is
Mclnnes v. Onslow-Fane.

Of those of SHYSTER's chosen cases which were decided before
Ackroyd v. Whitehouse, only Mclnnes v. Onslow-Fane was cited
in argument — none was cited in the Court's judgment. However,
the legal expert approves of SHYSTER'S choice of cases in the Ex-
pectation area. SHYSTER issues no warnings.

Hodgens v. Gunn 1989

In Hodgens v. Gunn. ex parte Hodgens, 206 Hodgens had been con-
victed of ill-treating dogs. The Magistrates Court had made an
order under s. 19(2) of the Animals Protection Act 1925 (Qld)
permanently prohibiting him from having a dog. He continued to
breed dogs, which were seized by the police. The Minister made
an order under s. 11(4) of the Act forfeiting the dogs to the Crown.
Hodgens appealed to the Full Court of the Supreme Court of
Queensland, claiming that the principles of natural justice required
that he should have been given a hearing before the Minister made
his order. The Court agreed: since the Minister's order involved
the expropriation of property without compensation, and the Min-
ister had absolute discretion in the exercise of this power, natural

§4.5 Case studies 225

justice was implied. (However, Hodgens was denied the remedy he
sought because he had waited six months before seeking a Court
order.)

SHYSTERS fact vectors are (NNNNNN). (YNNN) and (YYNYYNY).

In the Expectation area, SHYSTER concludes that Hodgens did
not have a legitimate expectation which was affected by the de-
cision. This is appropriate; legitimate expectation was not an
issue in Hodgens v. Gunn. The nearest neighbour is Minister v.
Peko -Walls end: the nearest other is Cole v. Cunningham.

In the Affected area. SHYSTER agrees with the Court that the
decision affected Hodgens's property, right, interest, status, or
legitimate expectation. The nearest neighbour is Bread Manufac-
turers v. Evans] the nearest other is Minister v. Peko-Wallsend.

However, in the Natural area, SHYSTER concludes (unlike the
Supreme Court) that a duty to observe natural justice is not im-
plied. The nearest neighbour is CCSU v. Minister] the nearest
others are FAI v. Winneke and Haoucher v. Minister.

Interestingly, all of SHYSTER's unweighted measures suggest
that FAI v. Winneke and Haoucher v. Minister should be the
nearest neighbours. Furthermore the ideal points, centroids, and
specified and ideal point directions all suggest that the result
should be Implied, but only the specified directions give rise to
a warning. 207

Of SHYSTER's chosen cases, only FAI v. Winneke was judicially
cited. 208 The legal expert agrees with SHYSTER's choice of cases
in the Affected area, and with its choice of nearest neighbour in the
Expectation area. However, she disagrees with its choice of Cole v.
Cunningham as nearest other in the Expectation area because that
case did not concern property rights.

WA v. Bropho 1991

In Western Australia v. Bropho, 209 land on the Swan River was
converted into a Crown reserve and a proposal put forward to ren-
ovate an old brewery on the site. However, the Aboriginal Heritage
Act 1972 (WA) required that a Committee be asked to recommend
to the Minister as to whether there was any Aboriginal site on the
land. The Committee invited public submissions, and recommen-
ded that the site was a significant Aboriginal site and that the
development should not be allowed. Nevertheless, the Minister

226 A pragmatic legal expert system §4-5

gave her consent to the development. Bropho was a custodian of
the site, in accordance with the customs of his people. He claimed
that the Minister was obliged to give him a hearing before making
her decision.

The Full Court of the Supreme Court of Western Australia held
that the fact that Bropho had not made a submission to the Com-
mittee, before it made its recommendation to the Minister, meant
that he had no legitimate expectation of a hearing before the Min-
ister. The Minister was obliged to follow natural justice principles
in coming to her decision but, Bropho having had an opportunity
to be heard before the Committee, this obligation had been met.

Malcolm CJ stated that Bropho had a "special interest in the
subject matter of the action" and was "more particularly affected
by the proposed development than ordinary members of the
public". 210 But Anderson J. with whom Franklyn J agreed, was
not so sure. He thought that Bropho's interest in the land was "in
the nature of a spiritual interest" which might not be sufficient. 211

SHYSTER'S fact vectors are (NNNMN). (NNNN) and (NYYNYYN).

In the Expectation area. SHYSTER agrees with the Supreme Court
that Bropho did not have a legitimate expectation which was
affected by the decision. The nearest neighbour is Minister v.
Peko -Walls end: the nearest other is Cole v. Cunningham.

In the Affected area, SHYSTER concludes that the decision did
not affect Bropho's property, right, interest, status, or legitim-
ate expectation. This conclusion is in line with the reservations
expressed by Anderson J. The nearest neighbour is Minister v.
Peko -Walls end: the nearest other is Bread Manufacturers v. Evans.

In the Natural area. SHYSTER concludes, as did the Court, that a
duty to observe natural justice is implied. The nearest neighbour
is Marine Hull v. Hurford; the nearest other is Minister v. Peko-
W alls end.

(Of course, if a decision does not affect the interests of an ap-
plicant then natural justice should never be implied. This fact
is captured, in a limited sense, by the specified direction towards
Not-Implied of a value of NO for A x — and SHYSTER warns that the
specified directions suggest Not-Implied in WA v. Bropho. 212 )

None of SH YSTER's chosen cases was cited in the published judg-
ment.

§4.5 Case studies 221

Ainsworth v. CJC 1992

In Ainsworth v. Criminal Justice Commission, 213 the Commis-
sion's advice had been sought on the introduction of poker ma-
chines into Queensland. Its report dealt with matters of general
concern and with particular poker machine suppliers and manu-
facturers, including Ainsworths companies. It was very critical
of the conduct of his companies, and recommended that they not
be permitted to participate in the gaming machine industry in
Queensland.

The Commission had formed its opinion on the basis of reports
of other bodies. Ainsworth had not been informed of the Commis-
sion's interest in his companies, and was not given any opportunity
to make representations on the matter before the Commission
made its report. Ainsworth claimed that the business reputation
of his companies was an interest which was adversely affected by
the Commission's report.

The Full Court of the High Court unanimously agreed. It fol-
lowed Annetts v. McCann and held that the Commission ought to
have applied natural justice principles. 214

The report files for Ainsworth v. CJC (one for each area) are given
as examples in § A. 5. SHYSTER's fact vectors are (NHNNHN), (NNYN)
and (YYYYNNM). 215

In the Expectation area, SHYSTER concludes that the applicant
did not have a legitimate expectation which was affected by the
decision. This is appropriate; legitimate expectation was not an
issue in Ainsworth v. CJC. The nearest neighbour is Minister v.
Peko -Walls end: the nearest other is Cole v. Cunningham.

In the Affected area, SHYSTER concludes, as did the High Court.
that the decision affected the property, right, interest, status, or
legitimate expectation of the applicant. The nearest neighbour
is Annetts v. McCann: the case followed by the High Court; the
nearest other is Minister v. Peko-Wallsend.

In the Natural area, SHYSTER agrees with the Court that a duty
to observe natural justice is implied. The nearest neighbour is also
Annetts v. McCann: the nearest other is Mclnnes v. Onslow-Fane.

However. SHYSTER warns that these two cases are equidistant
from Ainsworth v. CJC: it chooses Annetts v. McCann because it

228 A pragmatic legal expert system §4-5

is a decision of five justices of the High Court whereas Mclnnes v.
Onslow-Fane is merely a decision of the Chancery Division of the
English High Court. It also warns that both the weighted as-
sociation and weighted correlation coefficients prefer Mclnnes v.
Onslow-Fane as nearest neighbour.

4.5.4 Generated tests

The legal expert provided two examples for generating tests using
the Natural area of the Natural specification: generated tests 7
and 8.

Generated test 7

The legal expert advises that, if the decision did not affect the
property, right, interest, status, or legitimate expectation of the
applicant, then natural justice will not be implied. The fact vector
is (NUUUUUU). 210 SHYSTER generates 64 instantiations, half the
search space, and in 56 of them the choice of result is good (i.e. Not-
Implied).

In one of the eight instantiations in which the choice of result is
bad. SHYSTER warns that an ideal point suggesting the good result
is at least as near to the instantiation as is the nearest neighbour.
In all eight, the decision is apt to have a discrete impact on the
interests of the applicant (A 2 = yes). However, this does not
justify SHYSTER's choice of result in those instantiations.

Generated test 8

The legal expert also advises that, if the decision is not apt to
have a discrete impact on the interests of the applicant, natural
justice will not be implied. The required fact vector is (UNUUUUU).
But this cannot be given to SHYSTER directly because the value
of A\ cannot be unknown: it is an external attribute defined by
reference to the Affected area. So. this test is divided into two.
In the first test ("8a" in figure 4.13), the value of A x is yes; in
the second ("8b") it is no: i.e. the fact vectors are (YNUUUUU)
and (NNUUUUU). 217 Each of these two tests generates half of the
required search space.

§4.5 Case studies 229

In 55 of the total of 64 instantiations generated for test 8, the
choice of result is good (i.e. Not-Implied). In one of the nine in-
stantiations in which the choice of result is bad, SHYSTER issues
an ideal point warning. In each of these nine instantiations, the
decision affected the property, right, interest, status, or legitimate
expectation of the applicant (A x = yes). However, this does not
justify SHYSTER's choice of result in those instantiations.

4.5.5 Conclusion: the Natural specification

The test cases described above indicate that, using the Natural
specification, SHYSTER is exceptionally good at coming to the
right conclusion (95.40% good results), though not very good at
choosing cases to use in argument (40% good cases).

As shown in figure 4.13, SHYSTER'S choice of result is good in
86.72% of the instantiations in both generated tests. If a bad
choice of result is considered good where an ideal point warning
is issued, SHYSTER's success rate rises only slightly to 89.06%
and 85.94% for each test: 87.50% overall. (In none of the in-
stantiations in these tests does SHYSTER issue a warning about a
nearer ideal point when the choice of result is good: i.e. none of
SHYSTER's ideal point warnings is bad.)

As discussed above. 218 three pairs of cases in the Natural area have
identical facts yet different results. Because all of the leading cases
in the Natural area were correctly decided, this inconsistency indic-
ates a need for more attributes to distinguish between cases in that
area. Extra attributes were not added, however; the seven attrib-
utes in the Natural area are based on the seven factors identified
by McMillan as being important. 219

The Natural area is a good example of an area with attributes
which require further definition by reference to other areas. The
Natural specification represents — and represents effectively — an
area of law that is quite fluid and expanding rapidly: as discussed
above. 220 both Twist v. Randwick and Salemi v. MacKellar would
be decided differently today — -nearly twenty years after they were
heard by the High Court.

SHYSTER (when using the Natural specification) produces
fairly good advice in the test cases, and chooses good results in
the vast majority' of the generated tests.

230 A pragmatic legal expert system §4.6

4.6 Conclusion

The results of the testing described in this chapter — test cases and
generated tests — and of reflexive testing are discussed below. Gen-
eral conclusions are drawn from these results in the next chapter.

4.6.1 Test cases

The results of the testing using test cases and ideal points are
summarized in figure 4.12. Several special symbols are used in
that figure. A tick is used to indicate a "good" result or choice
of cases; a cross indicates a "bad" result or choice of cases. 221 A
dash indicates that a chosen case was decided after the test case
and that it is not possible to evaluate that choice. A blank space
indicates that no symbol is applicable.

The "Warnings" column summarizes the safeguard warnings that
were issued during each test. A warning that weighted similarity
measures suggested a different result is indicated by the appropri-
ate symbol: S' for the weighted association coefficients; r' for the
weighted correlation coefficients. A => indicates that the specified
directions suggested a different result. A = indicates that the test
produced equidistant results.

A fraction in the warnings column indicates that one or more
unknown values was instantiated. The fraction is the number of in-
stantiations that produced a different result to that of the nearest
neighbour, over the total number of instantiations (SHYSTER only
issues a warning if an instantiation has a different result). At the
right of the warnings column is a tick or a cross which indicates
whether the warning (or absence thereof) was "good" or "bad."

As figure 4.12 illustrates, SHYSTER has proven itself exceptionally
able to choose good results: a 95.24% success rate (including ideal
point tests). However, its ability to choose good cases with which
to argue varies dramatically between the four different specifica-
tions.

For the Finder specification, SHYSTER'S choice of cases is ex-
cellent, but there is only one test case for that specification. For
the Authorization specification, just over 83% of its chosen cases
are good. However, only 41.67% and 40% of the cases chosen from
the Employee and Natural specifications are good.

§4.6

Case studies 231

Case

Result

Nearest
n'bours

Nearest
others

Warning

Parker v. BA

—

CBS v. Ames

WE A v. Hanimex

CBS v. Amstrad

B
^

A PR A v. Jain

x
-

Hypothetical case 1

Hypothetical case 2

r'^ =

-
<

*Auth

•^Not-Auth

/
/

-^Liable

r'l

Narich v. CPT

Stevens v. Brodribb

Re Porter

1
4

BWIUv. Odco

—

Hypothetical case 3

r' -

' 2

Hypothetical case 4

■'Employee

'Contractor

r'l

Twist v. Randwick

///

- /

Salemi v. MacKellar

///

r'°-

Heatley v. TRGC

x -

S' t' = §

Cole v. Cunningham

/ -

X X

S' r' =

Ackroyd v. Whitehouse

///

/ X X

/ - /

Hodgens v. Gunn

// X

/ X

WA v. Bropho

///

XXX

Ainsworth v. CJC

///

X X

S' t' =

•'Implied

S' r'

'Not-Implied

Figure 4.12: A summary of the testing using test cases dis-
cussed in this chapter. The meaning of the symbols used here
is explained in §4.6.1.

Symbols for tests in the Natural specification refer to the
Expectation, Affected and Natural areas, respectively. Safeguard
warnings for the Natural specification are from the Natural
area — except those marked with an asterisk, which are from
the Expectation area.

232 A pragmatic legal expert system §4.6

One reason for this discrepancy may be the greater number
of leading cases in the Employee and Natural specifications
(14 and 17. 222 respectively) than in the Finder and Authoriz-
ation specifications (8 and 9). Judges and lawyers will often refer
only to a handful of cases; only rarely does a judge discuss ex-
haustively all of the important cases in a field. 223 Given an area
of law in which there were comparatively few leading cases, SHY-
STER would be more likely to chose a good case (all other things
being equal).

The appropriateness of SHYSTER'S warnings also varies between
specifications. In the single test case for the Finder specifica-
tion, the (absence of a) warning is good. But warnings for the
Authorization, Employee and Natural specifications are gener-
ally bad: only 22.22%, 50% and 30%. respectively, good warnings.

For the Authorization specification, all but two of the seven
bad warnings are solely due to specified directions. Across all four
specifications, the rate of good warnings would rise from 35.71%
to 60.71% if the specified direction warnings were ignored. Yet
specified directions are derived from the expert's opinion as to
which values for certain attributes suggest certain results; warn-
ings based upon the specified directions should be sound.

It is possible that the preponderance of bad specified direction
warnings in tests of the Authorization specification may be due
to the fact that the Authorization area has more than two results.
However, that is not the case in the other specifications.

In each of the nine tests in which a warning was issued due to an
extra similarity measure (i.e. S' or r') the warning was bad. This
indicates that the choice of similarity measure made in chapter 2 224
was a good one: using known and unknown distance measures
(variations on the weighted distance measure d r - k ), SHYSTER pro-
duces better results than it would using either of the other two
weighted measures.

In only one of the tests did an instantiation of the instant case
have a different result to that of the uninstantiated instant case:
Re Porter, 225 a case used to test the Employee specification. As
explained above, 226 the legal expert agrees with SHYSTER that
that instantiation would indeed lead to a different result.

§4.6

Case studies 233

No.

/ Results

x Results
IP Wrn

/ Results
IP Wrn

/ Results
and warnings

14/16

87.50

14/16

87.50

12/16

75.00

16/16

100.00

51/64

79.69

64/64

100.00

77/96

80.21

94/96

97.92

363/512

70.90

145

508/512

99.22

4198/6144

68.33

861

492

144

111

4579/6144

74.53

875/1024

85.45

112

987/1024

96.39

5436/7680

70.78

902

749

144

111

6074/7680

79.09

56/64

87.50

57/64

89.06

23/32

71.88

23/32

71.88

32/32

100.00

32/32

100.00

55/64

85.94

55/64

85.94

111/128

86.72

112/128

87.50

Figure 4.13: A summary of the generated testing performed
in this chapter. These tests are described in §4.3.4, §4.4.4
and §4.5.4. This table is explained in §4.6.2.

4.6.2 Generated tests

The results of the generated testing are summarized in figure 4.13.
The "/ Results" column gives the number of instantiations in which
SHYSTER's choice of result was good, as a fraction of the total
number of instantiations. That fraction is also expressed as a
percentage. For tests in which SHYSTER's choice of result was bad
(i.e. "x Results") the number of tests in which ideal point warnings
were issued is given in the "Wrn" column. The "IP" column gives
the number of tests in which the nearest ideal point was that of a
different result, but in which no warning was issued because that
ideal point was not at least as near to the instantiation as was the
nearest neighbour. Corresponding figures for those tests in which
SHYSTER's choice of result was good are given in the next column.

234 A pragmatic legal expert system

In the "/ Results and warnings" column, a bad choice is considered
good if an ideal point warning was issued. The number of good
choices is also expressed as a fraction of the total number of gen-
erated tests, and as a percentage. Except in one generated test,
SHYSTER never issued an ideal point warning when a good res-
ult was chosen. The bad warnings issued in that test — generated
test 5 — -are accounted for in this column because a good choice of
result is considered bad if an ideal point warning was issued.

A total for each column is given after each group of tests. Ig-
noring warnings, SHYSTER has a success rate of 80.21%, 70.78%
and 86.72%. respectively, for each of the three specifications. If
warnings are taken into account the success rate rises to 97.92%,
79.09% and 87.50%.

4.6.3 Reflexive tests

As explained in chapter 2, reflexive tests do not test SHYSTER's
approach to case law. However, if the choice of result is good
in a large proportion of a specification's reflexive tests, that is
indicative of that specification's suitability to handle new cases. 227
A reflexive test was carried out for every leading case in each of
the four specifications. 228 For both the Finder and Authoriza-
tion specifications, 75% of the choices of result were good. For the
Employee specification, 71.43% were good. But only 53.33% of
the reflexive tests of the Natural specification yielded good res-
ults. These tests indicate that the Finder. Authorization and
Employee specifications are capable of handling new fact situ-
ations, but that the Natural specification is less capable.

Notes

1 Wiener 1962 at 1023-4.

2 Gilbert 1977 at 212.

3 193 US 197 at 400.

4 Complete details of all four specifications are given in appendix A to
my doctoral thesis (Popple 1993a).

5 §2.13.

6 §2.13.4.

7 §2.13.5.

8 §1.5.2.

9 (1722) 1 Str 505; 93 ER 604.

10 Bridges v. Hawkesworth (1851) 21 LJQB 75 at 77 per Patteson J.

11 (1851) 21 LJQB 75.

Case studies 235

12 (1886) 33 Ch D 502.

13 [1896] 2 QB 44.

14 [1945] KB 509.

15 Hibbert v. McKiernan [1948] 2 KB 142 at 149 (footnotes omitted). Lord
Goddarcl CJ, Humphreys and Pritchard J J decided that the law of trover did
not apply in that case.

16 Tyree 1989 at 119.

17 [1982] QB 1004.

18 ibid, at 1017.

19 ibid, at 1017-18. Tyree 1989 says that Donaldson LJ's statement of
the principles to be applied is "a good candidate for a direct translation into a
rule-based system'' (at 119). He develops some rules to reflect this statement
and concludes that "building even small rule-based systems is not a task
which may be completed easily" (at 126). The rule base is considerably more
complicated than both FINDER and SHYSTER/s specification of FINDER.

20 Hibbert v. McKiernan [1948] 2 KB 142 at 149.

21 §4.4.2.

22 Tyree 1989 at 131 n. 12.

23 In City of London Corporation v. Appleyard (l) [1963] 1 WLR 982, "the
finders . . . were not the servants of Yorkwin but the servants of Wates Ltd, who
were engaged by Yorkwin as independent contractors" (at 988 per McNair J).
In South Staffordshire v. Sharman [1896] 2 QB 44, the finder was "employed"
by the company. It is not clear whether the finder was an employee or an
independent contractor, although some of what Lord Russell of Killowen CJ
says (at 46) suggests the latter.

24 (1722) 1 Str 505; 93 ER 664: C 3 in figure 4.3.

25 (1851) 21 LJQB 75: C 2 .

26 (1886) 33 Ch D 562: C s .

27 [1896] 2 QB 44: C 7 .

28 [1945] KB 509: C x .

29 [1963] 1 WLR 982: C 5 and C 6 .

30 [1969] 2 QB 152: C 4 .

31 In fact, the premises were leased by Venture Property and Development
Co. Ltd which held the premises on trust and to the order of Yorkwin. Venture
and Yorkwin were "associated companies." This added complication is not
relevant for FINDER's (or SHYSTER's) purposes, and is ignored.

32 In FINDER these "two cases" are referred to as Yorkwin v. Appleyard
and Corporation of London v. Yorkwin, respectively. In the Finder area of
SHYSTER's FINDER specification they are C 5 and Cq.

33 §2.14.1.

34 A% and A§ in Armory v. Delamirie.

35 A\o in Armory v. Delamirie, and A$ in London v. Appleyard (2): both
are YESs and both should be NOs.

36 In the Finder area, vj± = w$ > wq = wj = u'g = u;io > w\ = w% =
w'5 > u*2 (where Wi is the weight of Ai).

37 [1969] 2 QB 152.

38 [1982] QB 1004.

236 A pragmatic legal expert system

39 Donaldson LJ put it somewhat melodramatically: "On November 15,
1978, the plaintiff, Alan George Parker, had a date with fate — and perhaps
with legal immortality" (ibid, at 1007).

40 [1982] QB 1004 at 1011-13 per Donaldson LJ.

41 §4.2.1. His Lordship was surprised to find the law so complex: "It is
astonishing that there should be any doubt as to who is right. But there
is. Indeed, it seems that the academics have been debating this problem for
years" (ibid, at 1008).

42 [1982] QB 1004 at 1011-14.

43 ibid, at 1021.

44 [1963] 1 WLR 982 at 986 per McNair J.

45 Tyree 1986 at 18, 1989 at 160; Tyree, Greenleaf and Mowbray 1988
at 246-7, 1989 at 50-1.

46 e.g. University of New South Wales v. Moorhouse (1975) 133 CLR 1;
RCA Corporation v. John Fairfax and Sons Ltd [1981] 1 NSWLR 251: WEA In-
ternational Inc. v. Hanimex Corporation Ltd (1987) 17 FCR 274; Australasian
Performing Right Association Ltd v. Jain (1990) 26 FCR 53.

47 [1924] 1 KB 1 at 12.

48 [1926] 2 KB 474 at 491.

49 Adelaide Corporation v. Australasian Performing Right Association
Ltd (1928) 40 CLR 481 at 489.

50 Ricketson 1984 at 229.

51 (1975) 133 CLR 1 at 12-13.

52 ibid, at 12, citing Adelaide v. A PR A (1928) 40 CLR 481 at 497-8, 503.
Although Adelaide v. APRA is cited as authority for several propositions in
this area, it is not one of the leading cases in the AUTHORIZATION specification
because it was not actually a case about authorization. The question in that
case was whether the City of Adelaide had "permitted'" a song to be sung
contrary to s. 2(3) of the Copyright Act 1911 (UK), in force in Australia by
virtue of the Copyright Act 1912 (Cth).

53 (1975) 133 CLR 1 at 12-13.

54 ibid, at 13.

55 ibid, at 12.

56 Ricketson 1984 at 229-34.

57 Peter Drahos, Lecturer, Faculty of Law, The Australian National Uni-
versity.

58 Ricketson 1984 states the law as it was on 1 May 1983.

59 An employer is vicariously liable for an employee's infringing act if the
employee is acting within the course of her/his employment.

60 See §2.14.1.

61 The EMPLOYEE specification would have to be modified before it could
be linked to the AUTHORIZATION specification in this fashion. At present, the
Employee area assumes that there is a worker who is either an employee or an
independent contractor (§4.4.2). In the Authorization area, the infringer may
not work for the accused at all.

62 §4.5.2.

63 UNSW v. Moorhouse (1975) 133 CLR 1 at 13 per Gibbs J.

64 See §5.3 for a discussion of conditional attributes.

65 §5.2.4.

Case studies 231

00 [1924] 1 KB 1: C 7 in figure 4.6.

07 [1926] 2 KB 474: G 5 .

08 [1979] FSR 1: C 8 .

09 [1940] AC 491: C 4 .

70 [1946] VLR 338: C* 3 .

71 [1962] NSWR405: C 9 .

72 [1964-05] NSWR 138: C 2 .

73 (1975) 133 CLR 1: C x .

74 [1981] 1 NSWLR251: C 6 .

75 [1917-23] MacG Cop Cas 309.

76 See §4.5.2 for a discussion of identical cases.

77 In the Authorization area, it's > U 'Q > w 'i > ^-'2 > ^'4 > ^'3 = w '7 (where
itJj is the weight of Ai).

78 [1982] Ch 91.

79 id.

80 [1945] AC 108, a decision of the Judicial Committee of the Privy Coun-
cil on appeal from the Supreme Court of Canada.

81 [1982] Ch 91 at 95-6.

82 ibicl. at 95.

83 (1987) 17 FCR 274.

84 ibid, at 286-7.

85 ibid, at 287.
80 ibid, at 288.

87 [1981] 1 NSWLR 251 at 257 per Kearney J.

88 [1988] AC 1013.

89 (1990) 20 FCR 53.

90 ibid, at 61.

91 id.

92 The facts of A PR A v. Jain are summarized in the third paragraph of
SHYSTER's report file in §A.3: "In the instant case . . ."

93 A value of YES for A 7 in the Authorization area of the AUTHORIZATION
specification is directed towards Auth or Liable.

94 Creighton, Ford and Mitchell 1983 at 25.

95 Brooks 1990 at 1.
90 id.

97 Creighton et al. 1983 at 25.

98 This simplified statement of the law is incorporated in the help strings
for A\ and A2 in the Authorization area of the AUTHORIZATION specification.

99 (1978) 21 ALR 388 at 399-402 per Smithers, Evatt and Keely J J.

100 Creighton et al. 1983 at 25-44.

101 Brooks 1990; Smith 1992.

102 Phillipa Weeks. Senior Lecturer, Faculty of Law, The Australian Na-
tional University.

103 Creighton et al. 1983 state the law as it was on 31 March 1982.

104 "Pay as you earn/

105 (1940) 14 ALJ 162: C 2 in figures 3.9 and 4.7.
100 (1978) 21 ALR 388: Ciq.

107 (1980) 29 ALR 322: C 4 .

108 (1944) 09 CLR 227: C 3 .

238 A pragmatic legal expert system

109 (1945) 70 CLR 539: C 9 .

110 [1968] 2 QB 497: C 14 .

111 (1949) 79 CLR 389: C 8 .

112 (1955) 93 CLR 561: Ci-

113 [1924] 1 KB 762: C 7 .

114 [1952] 1 TLR 101: C and Ci 3 .

115 [1976] 1 WLR 1213: C 5 .

116 (1978) 18 ALR385: C u .

117 [1978] 1 WLR 676: G 12 .

118 C13 and C§, respectively.

119 In the Employee area, w\s = w\4 = w\q > w\j > ti'15 > u'2 = it's =
u'10 > it'g > u'lg > u'12 > u'e > tfi > u'g > u?4 = u's = W7 = tt'n (where Wi is
the weight of Ai).

120 (1983) 50 ALR417.

121 (1986) 160 CLR 16.

122 ibid, at 24, 26 per Mason J (with whom Brennan and Deane JJ gen-
erally agreed), at 38 per Wilson and Dawson JJ.

123 (1989) 34 IR 179.

124 (1954) 94 CLR 409.

125 [1985] VR577.

126 (1989) 34 IR 179 at 181.

127 There were five different owner-drivers in Re Porter, and the circum-
stances for all five were very similar. These attribute values are for the first
driver mentioned in the judgment— Newman: (1989) 34 IR 179 at 188-92.
The facts of Re Porter are summarized in the third paragraph of SHYSTER's
report file in §A.4: "In the instant case . . ."

128 (1991) 29 FCR 104.

129 Odco was trading as "Trouble Shooters Available."

130 (1991) 29 FCR 104 at 107.

131 ibid, at 126-7.

132 The facts of BWIU v. Odco are summarized in the third paragraph of
SHYSTER'S report file in §B.7: "In the instant case . . ."

133 Individual Employment Law, 1991, Faculty of Law, The Australian
National University, question 4.

134 [1976] 1 WLR 1213 at 1217-19.

135 Generated test 5 contributes 80% of the instantiations in these three
tests.

136 §5.2.5.

137 Clark and Wedderburn 1983 at 153.

138 McMillan 1991 at 63.

139 (1985) 159 CLR 550 at 584.

140 McMillan 1991 at 68.

141 Cole v. Cunningham (1983) 49 ALR 123 at 131 per Bowen CJ, Shep-
pard and Morling JJ.

142 McMillan 1991 at 70, citing Haoucher v. Minister (1990) 169 CLR
648 at 653.

143 (1987) 15 FCR 274 at 306.

144 [1967] 2 AC 337.

145 (1977) 137 CLR 396.

Case studies 239

146 (1982) 151 CLR 342.

147 McMillan 1991 at 71.

148 ibid, at 71-4.

149 The test case in question is Twist v. Randwick Municipal Council
(1976) 136 CLR 106 (see §4.5.3).

150 ibid, at 117. The section references are to the Local Government Act
1919 (NSW).

151 (1985) 10 FCR 234 at 248.

152 (1986) 10 FCR 476 at 480.

153 (1990) 21 ALD 730 at 735.

154 (1991) 5 WAR 75.

155 See §4.5.3, where WA v. Bropho is used as a test case.

156 Robin Creyke, Senior Lecturer, Faculty of Law, The Australian Na-
tional University.

157 McMillan 1991.

158 See also figure 2.1, and the discussion in §2.5.1 on the relationship
between areas in a specification.

159 The significance of this attribute is discussed in §4.5.1. For reasons
given there, the specified directions direct a value of YES for Aq towards Not-
Implied, and a value of NO towards Implied. Note that there are only two
leading cases in the Natural area with a value of YES for Aq — Commissioner
of Police v. Tanos (C$ in figure 4.11) and Marine Hull v. Hurford {Cq) — -and
natural justice was implied in both, but for other reasons.

160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182

1990) 169 CLR 648.

1983] 2 AC 629.

1958) 98 CLR 383: C 5 in figure 4.11.

(1981) 180 CLR 404: C u .

(1982) 151 CLR 342: Cj.

(1985) 159 CLR 550: C 4 .
(1987) 163 CLR 378: Ci .
(1990) 169 CLR 648: C 2 .
(1990) 170 CLR 596: C 3 .

(1986) 10 FCR 476: C 6 .

(1987) 15 FCR 274: C 12 .
(1981) 36 ALR 215: <7i 3 .
(1987) 9 NSWLR268: C 7 .
[1967] 2 AC 337: C* 9 .
[1983] 2 AC 629: C 8 .
[1985] AC 374: C 14 .
[1978] 1 WLR 1520: C ]5 .
(1977) 137 CLR 487.

(1983) 49 ALR 123.
§4.5.3.

(1976) 136 CLR 106.

(1977) 137 CLR 396.

Annetts v. McCann (C3) and Mclnnes v. Onslow -Fane (C15), Mac-
rae v. AG (Cj) and SA v. O'Shea (C\q). and Durayappah v. Fernando (Cg)
and Nashua v. Channon (C13).

240 A pragmatic legal expert system

183 In the Natural area, w\ = wj > wi = it'e > w$ > W4 > W3 (where wi
is the weight of Ai).

184 This is the basis of generated test 7 (§4.5.4).

185 In the Affected area, W2 = W3 > w\ = w±.

180 In the Expectation area, W4 > w% = wq > w\ = it's > it'2-

187 (1992) 175 CLR 564.

188 (1976) 136 CLR 106.

189 [1964] AC 40.

190 (1890) 24 QBD 703.

191 (1976) 136 CLR 106 at 111-12 per Barwick CJ, at 117 per Mason J.

192 ibid, at 118-19.

193 §4.5.1.

194 (1977) 137 CLR 396.

195 [1969] 2 Ch 149.

196 (1977) 137 CLR 396 per Barwick CJ, Gibbs, Stephen, Jacobs and
Aickin JJ, Murphy J dissenting.

197 ibid, per Barwick CJ, Gibbs and Aickin JJ.

198 (1977) 137 CLR 487.

199 ibid, at 494 per Mason J, at 495-6 per Murphy J, at 510 per Aickin J
(with whom Stephen J agreed), Barwick CJ dissenting.

200 ibid, at 509.

201 (1983) 49 ALR 123.

202 ibid, at 125.

203 ibid, at 133.

204 (1985) 2 NSWLR 239.

205 ibid, at 246.

206 [1990] lQdRl.

207 §2.12.2-§2.12.4.

208 [1990] 1 Qd R 1 at 3 per Thomas J (with whom Shepherdson and
Williams JJ generally agreed).

209 (1991) 5 WAR 75.

210 ibid, at 79.

211 ibid, at 87.

212 See §5.3 for a discussion of conditional attributes.

213 (1992) 175 CLR 564.

214 ibid, at 578 per Mason CJ, Dawson, Toohey and Gaudron JJ, at 585
per Brennan J.

215 The facts of Ainsworth v. CJC are summarized in SHYSTER's report
files in §A.5 (in the second paragraph of the reports for the Natural and
Expectation areas; in the first paragraph of the report for the Affected area):
"In the instant case . . ."

216 A value of NO for A\ (an external attribute) in the Natural area is
achieved as a result of a fact vector of (NNNN) in the Affected area. Similarly,
a value of NO for A4 (an external attribute) in the Affected area is achieved
as a result of a fact vector of (NNNNNN) in the Expectation area.

217 The way in which A\ in the Natural area is given a value of NO is
explained in the previous note. Similarly, fact vectors of (YYYY) and (YYYYYY)
for the Affected and Expectation areas result in a value of YES for A\ in the
Natural area.

Case studies 241

218 §4.5.2.

219 McMillan 1991 at 71-4 (see §4.5.1).

220 §4.5.3.

221 These terms are defined in §2.13.2.

222 The NATURAL specification has 15 leading cases in the Natural area,
9 in the Affected area, and 8 in the Expectation area. There are 17 different
leading cases in the specification.

223 For example, Donaldson LJ's discussion of finder cases in Parker v.
BA [1982] QB 1004 at 1008-16 (see §4.2.1 and §4.2.3). and Gummow J's
discussion of authorization cases in WEA v. Hanimex (1987) 17 FCR 274
at 281-8 (see §4.3.3).

224 §2.9.3.

225 (1989) 34 IR 179.

226 §4.4.3.

227 §2.13.5.

228 Complete results of reflexive testing are given in appendix D to my
doctoral thesis (Popple 1993a).

Conclusion

Attending a Cabinet when there was a tendency to Mutiny in the Fleet,
Sir Thomas Troubridge, . . . who was a most excellent Officer, was asked his
opinion what was best to be done. He said let me hang a hundred Lawyers,
and we shall hear no more of the business. I asked what he could mean — what
were these People that he called Lawyers. He replied, Fellows that can read
and write. They are the Fellows, that I call Lawyers, and make the whole of
the Mischief.

Lord Eldon (1827)
Lord Eldon's Anecdote Book

Donne, I suppose, was such another
Who found no substitute for sense,
To seize and clutch and penetrate:
Expert beyond experience . . .

T.S. Eliot (1920)
Whispers of Immortality

This is the tale of the 1401,

The law clerk that was nobody's son.

It spends its days in a furious hunt

For authorities, dictum, and argumunt.

But after it found them, it burned with shame;

The Supreme Court reversed it just the same.

T.H. Lassagne (19G3) 3

243

244 A pragmatic legal expert system §5-1

5.1 Introduction

I developed SHYSTER to test the main argument of this book:
that a useful, working legal expert system can be based upon a
pragmatic approach to the law. In this chapter, SHYSTER and its
approach to case law are evaluated (§5.2), and avenues of future
research are identified (§5.3). The contribution made by the SHY-
STER project to the field of legal expert systems development is
discussed in §5.4.

5.2 Evaluating SHYSTER

This evaluation of SHYSTER is carried out under five headings: its
usefulness, its generality, the quality of its advice, its limitations,
and possible enhancements that could be made to it. Although
there is some overlap between these five topics — and in the iden-
tification of areas of future research below — it is convenient to
evaluate SHYSTER in this fashion.

5.2.1 A useful, working system

Susskind claims that in 198G there was "an embarrassing lack of
demonstrable expert systems in law''. 4 Although several systems
have been developed since then, the embarrassment has continued.
In 1990. Kowalski and Sergot announced that they had decided not
to finish their system because their experience suggested that the
final stage of the development process "can involve a considerable
amount of work and extra programming effort."' That decision
made feeble their claim that "there are no outstanding technical
obstacles which need to be overcome 11 , and left them particularly
vulnerable to Moles' s suggestion that the real reason for the aban-
donment of their project was a realization of the inadequacy of
their approach.'

The development of (at least) a working prototype of a sys-
tem is an important part of any expert system project. In order
to demonstrate the efficacy of a pragmatic approach to case law.
SHYSTER's case-based system has been fully implemented, as ex-
plained in chapter 3. SHYSTER has been tested, as explained in
chapter 4. That testing was not as comprehensive as is desirable
but was as comprehensive as possible within the restrictions of
the project, and compares favourably with the testing of other
comparable systems. 8

§5.2 Conclusion 245

Ashley lists the following tasks as being general to all case-based
reasoning:

... (1) ordering relevant cases and potentially relevant cases in
terms of how analogous they are to the problem situation, (2) se-
lecting the most analogous cases, (3) identifying configurations of
counterexamples, (4) hypothetically modifying the problem situ-
ation to explore contingencies, and (5) comparing case-based ana-
lyses of different problem situations to explain differences. 9

SHYSTER performs all five of these tasks, within the restrictions
placed upon it by its simple knowledge representation.

SHYSTER's usefulness for the user — a lawyer — is due to the fact
that it is based upon a pragmatic model of legal reasoning; it
operates at the same pragmatic level of abstraction as do lawyers.
SHYSTER's Reporter module goes to some considerable trouble
to ensure that its reports are in a form that might be produced by
a lawyer. Details of SHYSTER's distance calculations are not part
of its reports.

During the development and testing of SHYSTER, an unexpected
aspect of the system's usefulness became apparent. I developed
three of the four specifications described in chapter 4 in consulta-
tion with lawyers with expertise in the relevant fields. Two of
those legal experts found that the process of specifying an area
of law for SHYSTER, and examining and evaluating its reports,
changed the way that they looked at that area. They felt that the
process had contributed to their knowledge and understanding of
the field. For example, the legal expert whom I consulted about
the Employee specification found that the process showed that
that area of law was more systematic than she had first thought.

Ashley says that a case-baseel legal expert system could be useful
as part of a legal tutoring system. 10 The experiences of the legal
experts during the development of its specifications suggest that
SHYSTER has the potential to assist users to gain similar insights.

5.2.2 A general approach

The general applicability of SHYSTER's approach to case law is an
open question. SHYSTER has been designed so that it can provide
advice in different areas of case law, specified by legal experts.
However, as explained in chapter 2, 11 the model of legal reasoning
adopted for SHYSTER reflects the way in which lawyers reason with
statutes and cases in areas of private law. SHYSTER has not been

246 A pragmatic legal expert system §5.2

designed to deal with areas of public law in which the doctrine of
precedent is given less weight. 12

Nevertheless, the four specifications described and tested in
chapter 4 suggest a wide application for SHYSTER's approach to
case law. The specifications are quite different from each other:
they are of differing sizes (in terms of the number of attributes
and the number of cases), and they represent disparate areas of
the law.

The Finder specification represents a completely case-based
area of law. The Authorization specification represents an area
of case law which has developed from specific statutory provisions.
The Employee specification represents an area of case law which
has developed alongside several different statutes. In the cases
that make up the Natural specification, judges were overtly con-
cerned with matters of policy. That specification is more like a
public law area than any of the other three specifications.

Menclelson applied dimensional analysis (as performed by HYPO)
to the law governing government appeals in US criminal cases. 13
His goal was to determine how "domain-sensitive" was HYPO'a
approach to case law. His dimensional analysis was unsuccessful,
because, he concludes, the area of law "is unsettled, fact-sensitive
and conflicting." 14 As explained in chapter 4, 15 the law embodied
in the Natural specification shares all three of these charac-
teristics. This makes SHYSTER's success with that specification
(although not as consistent as its success with other specifications)
particularly interesting.

Although results vary between the test specifications, they are
generally good in all four. These results do not demonstrate the
generality of SHYSTER's approach to private case law, but they do
suggest its broad applicability.

5.2.3 Good advice

As discussed in chapter 2, 10 the quality of SHYSTER'S advice is
evaluated on the basis of its prediction as to the likely result, and
its choice of cases for use in argument. Within the limitations
imposed by its internal representation of case law, SHYSTER is
capable of producing quite sophisticated reports — but these re-
ports can only be as good as the cases it chooses. 1.

As is shown in chapter 4, SHYSTER has proved remarkably good
at choosing a result — in test cases and in generated tests. It is.

§5.2 Conclusion 247

however, quite inconsistent (across the four specifications) in the
quality of its choice of cases.

It should be remembered, however, that the test cases used in
chapter 4 are difficult tests. As explained in chapter 2, 1S the same
filters in the legal system which restrict the number of cases that
are reported also ensure that those cases that are reported make
difficult tests. 19

Furthermore, it is a little harsh to rule SHYSTER's choice of a
case "bad" on the basis that that case was not cited in judgment
or argument by a judge or by counsel. Some judges will not look
beyond the cases cited in argument by counsel. For all sorts of
non-legal reasons (lack of time, incompetence, etc.) counsel may
not cite all the good cases in argument. And, of course, opinion
as to what is a good case to use in argument differs. But. harsh as
it may be. some method of evaluating the quality of SHYSTER'S
choice of cases had to be adopted. 20

5.2. 4 Limitations

Although SHYSTER has proved quite successful, it has several lim-
itations.

As well as being restricted (for the most part 21 ) to areas of
private law, there are some areas of private law to which SHYSTER
is not suited. These restrictions upon domains — restrictions which
apply to all legal expert systems — are examined in chapter 2. 22

"Reasonableness." for example, is a difficult concept. It plays
an important part in the Authorization specification, yet it is so
nebulous a concept that it would probably be impossible to specify
for SHYSTER. Furthermore, the meaning of ^reasonableness" may
well be very different within different areas of the law; even if
it could be specified for one area, that specification might not
be appropriate for another area within which the concept is also
important.

SHYSTER's solution to this problem in the Authorization spe-
cification is to leave the user to answer the question of whether the
accused took reasonable steps. 23 This is not a major limitation: it
is quite possible that the user (a lawyer) knows the answer to that
question, but not to the larger question of whether the accused
authorized the infringement. In any event, the user has the option
of answering unknown to difficult questions, forcing SHYSTER to
examine instantiations.

248 A pragmatic legal expert system §5-2

As discussed in the previous chapter. 24 if many leading cases are
specified then SHYSTER's ability to choose good cases is reduced.
This problem can be avoided by increasing the threshold within
which SHYSTER will consider two cases to be equidistant. 25 If such
a change is made. SHYSTER will tend to choose more cases to use
in argument. This increases the likelihood of SHYSTER choosing
good cases, but also increases the number of bad cases chosen.

The fact that SHYSTER will construct arguments for all possible
results means that sometimes it will produce extraneous argu-
ments. For example, given an authorization case, SHYSTER will
argue about the Liable result even if the question of direct or vicari-
ous liability does not arise. 20 In an area where there are several
results, this problem w r ould be exacerbated. One solution to this
problem would be to restrict SHYSTER to arguing only about the
nearest n results (where n is 2 or more): i.e. the result of the
nearest neighbour, and the results of the n — 1 nearest others. This
solution reduces the likelihood of SHYSTER building an irrelevant
argument, but increases the likelihood of it missing a relevant one.

The simplicity of SHYSTER's knowledge representation means that
it is unable to distinguish between differences and distinctions. In
the example report discussed in chapter 3, 2 ' SHYSTER lists the
differences between Building Workers' Industrial Union of Aus-
tralia v. Odco Pty Ltd 28 (the instant case) and Humberstone v.
Northern Timber Mills 29 (the nearest neighbour) in which Hum-
berstone was held to be an independent contractor. One of these
differences is that, in Humberstone v. NTM, Humberstone and his
employer did not express any intention that the relationship would
be one of principal and independent contractor. But the fact that
the workers and the employer did express such an intention in
BWIU v. Odco makes the argument even stronger that the build-
ing workers were independent contractors. This difference is really
an additional reason to conclude that the workers in BWIU v. Odco
were independent contractors, not a point of distinction between
the two cases.

When arguing with hypothetical in BWIU v. Odco, 30 SHYSTER
states that it would be more strongly of the opinion that the build-
ing workers were independent contractors if the facts of the case
were changed so that (amongst other things) the employer and the
workers did not express any intention that the relationship would
be one of principal and independent contractor. This change would

§5.2 Conclusion 249

make BWIU v. Odco more like Humber stone v. NTM. but it would
not make BWIU v. Odco a stronger case for the result in Humber-
stone v. NTM: i.e. Contractor. 31

It might be possible to modify SHYSTER so that it avoids such
embarrassing statements, by reference to attribute direction. But
this would require a significant change to the way that SHYSTER
treats attribute direction which, at present, is used only as a safe-
guard. And. in any event, attribute direction would be of little
assistance to SHYSTER in this example: although an expression of
intention as to the nature of the relationship is directed towards
the appropriate result (Employee or Contractor) the absence of such
an expression is not — and ought not to be — directed towards either
result. 32

5.2.5 Possible enhancements

The model of legal reasoning adopted for SHYSTER allows for
open-textured concepts to be defined by reference to case law.
However, there may be circumstances in which a case-based open-
textured concept is defined by reference to statute law.

SHYSTER's model of legal reasoning does not allow for such
circumstances, though it could be augmented to allow resolution of
open texture using legislation. Implementing such an augmented
model should not pose any major difficulties. SHYSTER'S case law
specification language would have to be extended to allow external
attributes to be linked to rules in the rule base in a similar way to
that in which external attributes are presently linked to areas in
the case base.

SHYSTER allows the specification of no more than one ideal point
for each result in each area. Yet, there might be several, equally
ideal combinations of attribute values for a given result. There is
no reason why SHYSTER could not be modified to allow multiple
ideal points for the same result.

As demonstrated in chapter 4, 33 SHYSTER can produce clumsily
worded sentences. The quality of SHYSTER's reports could be im-
proved by changing the way in which it uses attribute strings. The
number of attribute strings that are written could be restricted on
the basis of the relative importance of the attributes (determined
by reference to their weights), though it must not be forgotten
that the user may sometimes be interested in a less important

250 A pragmatic legal expert system §5-3

attribute. Attribute strings could be combined in more sophistic-
ated ways: rather than writing "x is not true; and y is not true."
SHYSTER could write "neither x nor y is true." This would require
changes to SHYSTER'S case law specification language, as well as
to the Reporter module.

Similarly, the number of questions that SHYSTER asks of the
user could be restricted. Attribute questions could be asked in
order of importance until the answers provided made a particular
result inevitable, at which stage the attribvites still unanswered
could be set to unknown. However, the advantage of asking fewer
questions of the user would have to be balanced against the effect
of the extra, unknown attribute values upon SHYSTER'S reports.

5.3 Future research

The most obvious avenue of future research using SHYSTER is
the development of a rule-based system, and the linking together
of that rule-based system with the existing case-based system to
form a hybrid system. As argued in chapter l, 34 a hybrid ap-
proach is desirable where the system seeks to represent statute
law and case law. SHYSTER has been designed to facilitate such
an articulation. 30

There are also several areas of further research which could be
carried out using the existing SHYSTER system. New specifications
could be written, and tested in the same manner as are the four
specifications described in chapter 4. As well as writing different
specifications, different legal experts could be asked to specify the
same area of law. It would be interesting to examine the differences
(if any) between the specifications and the ways in which those
specifications make SHYSTER behave.

Research could also be done into the process of updating spe-
cifications to take account of changes in the law. Most changes
could be captured by the addition of new cases to, and the removal
of out-of-date cases from, the specification — both operations are
easily performed using SHYSTER's case law specification language.
More significant changes in the law might require the addition of
new attributes, or the complete rewriting of the specification.

SHYSTER's method of assigning weights to attributes deserves fur-
ther examination. As explained in chapter 2, 30 SHYSTER adopts

§5.3 Conclusion 251

FIXDER's method of attribute weighting. This method contrib-
utes significantly to SHYSTER'S distance calculations, and so has
an important effect on SHYSTER's opinions. SHYSTER's opinions
are generally good which suggests that its method of attribute
weighting is also a good one. Yet, as demonstrated in chapter 4, 37
SHYSTER's weighting of attributes is not always appropriate.

Some experiments that I have performed indicate that weighting
attributes according to the inverse of the variance of each attribute
across the leading cases (as SHYSTER does) produces better results
than weighting according to the variance (the standard approach
adopted in statistical classification problems 38 ) or weighting each
attribute equally.

Further testing could be performed. SHYSTER's Adjuster
provides a practical tool for experimenting with different weighting
schemes.

The choice of SHYSTER's similarity measure, made in chapter 2, 39
is vindicated by the number of bad warnings generated using the
other similarity measures. 40 Nevertheless, further research could
also focus on the refinement of SHYSTER's similarity measure: its
known and unknown distance. SHYSTER makes only limited use
of the hierarchy of courts that can be part of any specification. 41
It would be possible to take account of the rank of a leading case
when quantifying the distance between it and the instant case.
Another possibility would be to determine the court in which the
instant case is likely to be heard, and to apply rigorously the
doctrine of precedent which holds that a court is bound only by
the decisions of courts higher in its hierarchy. 42

As demonstrated in several of the examples in chapter 4, there
are occasions when specifying an area of law seems to call for a
more disjunctive structure than SHYSTER permits: i.e. occasion-
ally there is a need for a conditional attribute.

For example. A Ti in the Authorization area of the Authorization
specification concerns whether reasonable steps were taken to avoid
an infringement of copyright. If the accused does take reasonable
steps to avoid an infringement then she/he will not be said to
have authorized the infringement, 43 although she/he may still be
directly or vicariously liable. Other examples can be found in the
generated tests discussed in chapter 4: each of these relies on some
attribute value or combination of attribute values implying a given
result or results.

252 A pragmatic legal expert system

The idea of conditional attributes can be captured, in a limited
way. by attribute direction. 44 But the idea of allowing a given
result to be conditional upon certain attribute values, in a strict
sense, could be investigated. This would represent a major change
to SHYSTER's operation, and raises questions of attribute depend-
ence.

5.4 The contribution of the SHYSTER project

I have argued that a legal expert system need not be based upon
a complex or deep model of legal reasoning in order to be success-
ful. I have recommended, instead, a pragmatic approach to legal
expert system design. SHYSTER is a working example — just one
possible example — of such an expert system. Despite its simple
knowledge representation structure, it has shown itself capable of
producing good advice. And its simple structure has facilitated
the specification of several different areas of law.

For SHYSTER, the approach taken by the developers of FINDER
has been adopted, and expanded so as to incorporate some of the
capabilities of more sophisticated systems (e.g. HYPO) without
greatly increasing the complexity of the underlying model.

SHYSTER incorporates a simple model of the way in which law-
yers argue with cases. However, no attempt has been made to
model the way in which lawyers decide which cases to use in
those arguments; I do not claim that lawyers choose the cases
that they use in argument by reference to distance calculations
in n-dimensional space. SHYSTER's is not a deep model of legal
reasoning, but as Ashley points out:

Even an only partially successful computational theory of jurispru-
dence may still prove useful. For example LEXIS and WESTLAW
[legal retrieval systems 45 ] represent a theory about legal reason-
ing: that relevance can be assessed by the appearance of keywords
in past cases. No one would accept that theorj' as adequate, and
yet it yields a practical tool. 46

SHYSTER, too, is a practical tool. It demonstrates that a prag-
matic approach to legal expert system development can be a
successful one.

Conclusion 253

Notes

1 Quoted by Muir 1970 at 81. Muir does not date this quotation but,
according to Twiss 1844 at 16, Lord Eldon finished his anecdote book in
1827.

2 Eliot 1963 at 55.

3 Quoted by Lawlor 1963 at 344.

4 Susskind 1989b at 24.

5 Kowalski and Sergot 1990 at 208.

6 id.

7 Moles 1991 at 162-3.

8 §2.13.3.

9 Ashley 1990 at 127, as quoted in §1.5.

10 Ashley 1985 at 116 (see §1.5).

11 §2.3.1 and §2.3.3.

12 The distinction between private and public law is discussed in §2.3.1.

13 Mendelson 1989.

14 ibid, at 137.

15 §4.5.1.
1C §2.13.2.

17 What is meant by a "good" choice of cases or a "good" result is defined
in §2.13.2.

18 §2.13.3.

19 The hypothetical test cases are also difficult tests, although the ideal
points (which are also used as test cases) are not.

20 The harshness of this method is mitigated somewhat by taking the
legal expert's opinion into account: see, for example, the discussions on Cole v.
Cunningham (1983) 49 ALR 123 and Ackroyd v. Whitehouse (1985) 2 NSWLR
239 in §4.5.3.

21 Note the comments on the NATURAL specification in §5.2.2.

22 §2.13.1.

23 See the discussion on A$ in the Authorization area of the AUTHORIZA-
TION specification (§4.3.2).

24 §4.6.1.

25 §3.12.1.
20 §4.3.3.

27 §3.13.1.

28 (1991) 29 FCR 104. See the discussion on BWIU v. Odco in §4.4.3.

29 (1949) 79 CLR 389.

30 §3.13.3.

31 However, SHYSTER's argument could be interpreted (somewhat char-
itably) along these lines: "Imagine that there had been no statement of
intention in BWIU v. Odco. Without such a statement, BWIU v. Odco is

254 A pragmatic legal expert system

even more like Humberstone v. NTM. The court in Humberstone v. NTM de-
cided that Humberstone was a contractor even though there was no statement
of intention. All the more reason to come to the same conclusion in BWIU v.
Odco. where there was." This interpretation does, however, read rather a lot
into SHYSTER'S report.

32 See A\-? and A\% in the dump file for the EMPLOYEE specification
in §B.3.

33 §4.4.5.

34 §1.8.

35 The mechanism by which this could be achieved is explained in §3.3.

36 §2.7.

37 See the discussion on weights in §4.2.2, §4.3.2, §4.4.2 and §4.5.2.

38 §2.7.

39 §2.9.3.

40 §4.0.1.

41 The rank of courts is used as the first step in the resolution of equi-
distance (§2.10.3), and when referring to the relative importance of different
leading cases (§3.13.1).

42 §1.1.3.

43 University of New South Wales v. Moorhouse (1975) 133 CLR 1 at 13
per Gibbs J. This is the basis of generated test 3 (§4.3.4).

44 See, for example, the discussion on Western Australia v. Bropho (1991)
5 WAR 75 in §4.5.3.

45 LEXIS: Stanley 1980; WESTLAW: Herman 1980.

46 Ashley 1990 at 254, capitalization added.

Example reports

The law is not a series of calculating machines where definitions and answers
come tumbling out when the right levers are pushed.

William 0. Douglas (1948) 1

An argument which does not convince yourself, may convince the Judge to
whom you urge it; and if it does convince him, why, then. Sir, you are wrong,
and he is right.

Samuel Johnson (1768)

255

256 A pragmatic legal expert system §A.l

A.l Introduction

This appendix presents six example reports from four test cases,
demonstrating the use of each of the four specifications described
in chapter 4. Each report appears exactly as produced by SHY-
STER, after processing by the I^TgX document processor. 3

Parker v. British Airways Board 4 is used as a test case for the
Finder specification. SHYSTER's opinion on that case is evaluated
in §4.2.3. and its report file is in §A.2.

Australasian Performing Right Association Ltd v. Jain 5 is used
in §4.3.3 as a test case for the Authorization specification, and
SHYSTERS advice is evaluated there. SHYSTER'S report file for
APRA v. Jain is in §A.3.

Re Porter; Re Transport Workers Union of Australia is used
as a test case for the Employee specification. SHYSTER's opinion
on Re Porter is evaluated in §4.4.3, and its report file is in §A.4.

Ainsworth v. Criminal Justice Commission 7 is used in §4.5.3
as a test case for the Natural specification, and SHYSTER's ad-
vice is evaluated there. SHYSTER prepares a report file for each
of the three areas in that specification: the Natural, Affected and
Expectation areas. All three reports, combined, are in §A.5. 8

For each example report, SHYSTER was requested to hypothes-
ize with a limit of two changed attribute values, and to report on
up to one hypothetical per result.

(The report file for Building Workers 7 Industrial Union of Aus-
tralia v. Odco Pty Ltd 9 is given in §B.7 as part of the complete
example, using the Employee specification, in appendix B.)

Notes

1 Douglas 1948 at 105.

2 Quoted by Boswell 1960 at 342.

3 Lamport 1986, 1994.

4 [1982] QB 1004.

5 (1990) 26 FCR 53.

6 (1989) 34 IR 179.

7 (1992) 175 CLR 564.

8 The REPORTER module will not summarize the same case twice in the
same report file. However, there is some repetition in §A.5 because there
are three separate report files for Ainsworth v. CJC (one for each area in
the NATURAL specification): Minister for Arts, Heritage and Environment v.
Peko-Wallsend Ltd (1987) 15 FCR 274 and Annetts v. McCann (1990) 170
CLR 596 are both summarized twice.

9 (1991) 29 FCR 104.

§A.2 Example reports 257

A. 2 Report file for Parker v. BA

Finder area

Instant case

In the instant case, the finder was not the occupier of the premises
where the chattel was found; the chattel was not attached; the other
claimant was not the owner of the premises where the chattel was
found; the other claimant was not the true owner of the chattel and
was not claiming through the rights of the true owner; the finder handed
over the chattel to the other claimant after the finding; neither party
relied on the terms of an agreement regarding the right to the chattel;
the finder was not a servant of the other claimant; the chattel was not
hidden and was not in a position so as to be difficult to find; an attempt
was made to find the true owner of the chattel or, alternatively, the
chattel was clearly abandoned; and neither party knew of the existence
of the chattel prior to the finding.

In my opinion— following Bridges v. Hawkesworth — -the finder wins.

In Bridges v. Hawkesworth. 1 an 1851 decision of the Queen's Bench
Division of the English High Court, the plaintiff found a bundle of
banknotes on the floor of the public area of a shop. He handed the notes
to the shopkeeper in order that the true owner of the notes might be
found. Although the owner was never found, the shopkeeper refused to
return the notes to the finder. The Court found for the finder, holding
that there is a "general right of [a] finder to any article which has been
lost as against all the world except the true owner". 2 It was further
noted that the notes had never been in the custod}' of the shopkeeper
nor within the protection of his house as might be the case had they
intentionally been deposited there.

There are several significant similarities between the instant case and
Bridges v. Hawkesworth: the finder was not the occupier of the premises
where the chattel was found; the chattel was not attached; the other
claimant was not the true owner of the chattel and was not claiming
through the rights of the true owner; the finder handed over the chattel
to the other claimant after the finding; neither party relied on the terms
of an agreement regarding the right to the chattel; the finder was not a
servant of the other claimant; the chattel was not hidden and was not
in a position so as to be difficult to find; an attempt was made to find

] (1851) 21 LJQB 75.
"ibid, at 77 per Patteson J.

258 A pragmatic legal expert system §A.2

the true owner of the chattel or, alternatively, the chattel was clearly
abandoned; and neither party knew of the existence of the chattel prior
to the finding.

However, the instant case is not on all fours with Bridges v. Hawkes-
worth. In that case the other claimant was the owner of the premises
where the chattel was found.

Nevertheless, I believe that Bridges v. Hawkesworth should be fol-
lowed.

If City of London Corporation v. Appleyard (1) is followed then the
finder loses.

In City of London Corporation v. Appleyard (1)? a 1963 decision of
the Queen's Bench Division of the English High Court, workmen em-
ployed by Wates Ltd were engaged in cutting a key- way into a cellar
wall for the purposes of securing a foundation when the)'' found an old
wall-safe built into a recess of the old wall. Inside was a wooden box
which contained a large number of Bank of England notes. The notes
were handed over to the City of London police who sought interpleader
proceedings to determine who was entitled to the possession of the
notes.

The Court followed the decision in South Staffordshire Water Co. v.
Sharman 4 in holding that the occupier is, in the absence of a better
title elsewhere, entitled to the possession of objects which are attached
to or under the land. Consequently, since the notes were in a wooden
box within a safe built into the wall of the old building, the safe formed
part of the demised premises. Yorkwin, being in lawful possession of the
premises, was in de facto possession of the safe, even though ignorant
of its existence.

Although Yorkwin was entitled to possession as against the finders,
it in turn was displaced by the City of London which relied successfully
on a term in the lease which granted it the right to certain objects found
on the premises.

There are several similarities between the instant case and London v.
Appleyard (1): the finder was not the occupier of the premises where
the chattel was found; the other claimant was not the owner of the
premises where the chattel was found; the other claimant was not the
true owner of the chattel and was not claiming through the rights of the

3 [1963] 1 WLR982.
4 [1896] 2 QB 44.

§A.2 Example reports 259

true owner; neither party relied on the terms of an agreement regarding
the right to the chattel; an attempt was made to find the true owner
of the chattel or, alternatively, the chattel was clearly abandoned; and
neither party knew of the existence of the chattel prior to the finding.

However, there are several significant differences between the instant
case and London v. Appleyard (1). In that case the chattel was at-
tached; the finder did not hand over the chattel to the other claimant
after the finding; the finder was a servant of the other claimant; and
the chattel was hidden or was in a position so as to be difficult to find.

Despite the fact that London v. Appleyard (1) and Bridges v. Hawkes-
worth are both decisions of the Queen 's Bench Division of the English
High Court, there is nothing in London v. Appleyard (1) to warrant
any change in ixry conclusion.

Hypothetical 1

Consider the instant case changed so that the following is true: the
other claimant was the owner of the premises where the chattel was
found.

If that were so then I would be more strongly of the opinion that —
following Bridges v. Hawkesworth — the finder wins.

Details of Bridges v. Hawkesworth are summarized above. The hypo-
thetical case is on all fours with Bridges v. Hawkesworth.

If City of London Corporation v. Appleyard (2) or South Staffordshire
Water Co. v. Sharman are followed then the finder loses.

In City of London Corporation v. Appleyard (2),° a 1963 decision of
the Queen's Bench Division of the English High Court, workmen em-
ployed by Wates Ltd were engaged in cutting a key- way into a cellar
wall for the purposes of securing a foundation when the}'' found an old
wall-safe built into a recess of the old wall. Inside was a wooden box
which contained a large number of Bank of England notes. The notes
were handed over to the City of London police who sought interpleader
proceedings to determine who was entitled to the possession of the
notes.

Wates Ltd was an independent contractor engaged by Yorkwin In-
vestments Ltd for a construction project. Yorkwin was lessee in pos-
session of the property which was owned in fee simple by the City of
London. The Court found that the safe formed part of the demised
premises and that, consequently, Yorkwin was entitled to the notes as
against the workmen.

} [1963] 1 WLR982.

260 A pragmatic legal expert system §A.2

The lease contained a clause which purported to grant the rights
to "every relic or article of antiquity rarity or value" to the City of
London. The sole issue was to determine if the notes fell into that
description. The Court could find no reason for limiting the generality
of the words and so found for the City of London.

There are several similarities between the hypothetical case and Lon-
don v. Appleyard (2): the other claimant was the owner of the premises
where the chattel was found; the other claimant was not the true owner
of the chattel and was not claiming through the rights of the true owner:
the finder handed over the chattel to the other claimant after the find-
ing; the finder was not a servant of the other claimant: an attempt was
made to find the true owner of the chattel or, alternatively, the chattel
was clearly abandoned; and neither party knew of the existence of the
chattel prior to the finding.

However, there are several significant differences between the hypo-
thetical case and London v. Appleyard (2). In that case the finder was
the occupier of the premises where the chattel was found: the chattel
was attached; one of the parties relied on the terms of an agreement
made with the other which purported to give her/him the right to the
chattel; and the chattel was hidden or was in a position so as to be
difficult to find.

Despite the fact that London v. Appleyard (2) and Bridges v. Hawkes-
worth are both decisions of the Queen's Bench Division of the English
High Court, there is nothing in London v. Appleyard (2) to warrant
an}' change in my conclusion.

In 1896, the Queen's Bench Division of the English High Court also de-
cided South Staffordshire Water Co. v. SharmanP (Note, however, that
London v. Appleyard (2) is 67 years more recent than South Stafford-
shire v. Sharman.)

In South Staffordshire v. Sharman, the defendant was a workman
employed by the plaintiff to clean out a pool located on land owned
by the plaintiff. During the operation the defendant found two gold
rings embedded in the mud at the bottom of the pool. Although the
plaintiff demanded the rings, the defendant refused to give them up.
He placed them in the hands of police authorities who unsuccessfully
endeavoured to find the owners of the rings. The police returned the
rings to the defendant who was then sued in detinue for the recovery
of the rings.

It was proved at the trial that there was no special contract between
the parties which called upon the defendant to give up any articles
which might be found.

s [1896] 2 QB 44.

§A . 2 Example reports 261

Although the county court held in favour of the defendant on the
basis of Bridges v. Hawkesworth, 7 the appeal found for the plaintiff on
the basis that it had, as owner of the land and pool, the right to exercise
control over the same. Bridges v. Hawkesworth was distinguished on
the grounds that the notes in that case were in a public part of the
shop and the shopkeeper did not in any sense control them.

The Court stated a general principle: where a person has possession
of a house or land with a manifest intention to exercise control over it
and the things which may be upon or in it, then there is a presumption
that things found there are in the possession of the owner.

There are several similarities between the hypothetical case and
South Staffordshire v. Sharman: the finder was not the occupier of
the premises where the chattel was found; the other claimant was the
owner of the premises where the chattel was found; the other claimant
was not the true owner of the chattel and was not claiming through the
rights of the true owner; neither party relied on the terms of an agree-
ment regarding the right to the chattel; an attempt was made to find
the true owner of the chattel or, alternatively, the chattel was clearly
abandoned; and neither party knew of the existence of the chattel prior
to the finding.

However, there are several significant differences between the hy-
pothetical case and South Staffordshire v. Sharman. In that case the
chattel was attached; the finder did not hand over the chattel to the
other claimant after the finding; the finder was a servant of the other
claimant; and the chattel was hidden or was in a position so as to be
difficult to find.

Despite the fact that South Staffordshire v. Sharman and Bridges v.
Hawkesworth are both decisions of the Queen's Bench Division of the
English High Court, there is nothing in South Staffordshire v. Sharman
to warrant any change in my conclusion.

Hypothetical 2

Consider the instant case changed so that the following is true: the
finder was a servant of the other claimant; and the chattel was hidden
or was in a position so as to be difficult to find.

If that were so then my opinion would be that' — following City of Lon-
don Corporation v. Appleyard (1) — the finder loses.

Details of London v. Appleyard (1) are summarized above. There are
several significant similarities between the Irypothetical case and Lon-
don v. Appleyard (1): the finder was not the occupier of the premises

7 (1851) 21 LJQB 75.

262 A pragmatic legal expert system §A.2

where the chattel was found; the other claimant was not the owner of
the premises where the chattel was found; the other claimant was not
the true owner of the chattel and was not claiming through the rights
of the true owner; neither party relied on the terms of an agreement
regarding the right to the chattel; the finder was a servant of the other
claimant; the chattel was hidden or was in a position so as to be diffi-
cult to find; an attempt was made to find the true owner of the chattel
or, alternatively, the chattel was clearly abandoned; and neither party
knew of the existence of the chattel prior to the finding.

However, the hypothetical case is not on all fours with London v.
Appleyard (1). In that case the chattel was attached; and the finder
did not hand over the chattel to the other claimant after the finding.

Nevertheless, I believe that London v. Appleyard (1) should be fol-
lowed.

If Hannah v. Peel is followed then the finder wins.

In Hannah v. Peel* a 1945 decision of the King's Bench Division of
the English High Court, a brooch was found by the plaintiff who was a
lance-corporal stationed in a house owned b} r the defendant. The house
had been requisitioned b} r the army during the war and had never been
occupied by the defendant.

The plaintiff was adjusting the black-out curtains when he touched
something on the top of the window-frame. He thought the object to
be a piece of dirt or plaster and he dropped it on the outside window
ledge. On the following morning, he saw that it was a brooch and, on
the advice of his commanding officer, turned it over to the police for the
purpose of finding the owner. In the following year, the police returned
the brooch to the defendant who sold it to a jeweller. The plaintiff at
all times maintained his rights to the brooch against all persons other
than the true owner.

The Court found for the plaintiff on the basis of Bridges v. Hawkes-
worth 9 after a thorough review of the authorities. The Court further
noted that the defendant was never in possession of the premises, that
the brooch was never his, and that he had no knowledge of it until it
whs brought to his notice by the finder.

There are several similarities between the hypothetical case and Han-
nah v. Peel: the finder was not the occupier of the premises where the
chattel was found; the chattel was not attached; the other claimant
was not the true owner of the chattel and was not claiming through
the rights of the true owner; neither party relied on the terms of an

8 [1945] KB 509.
9 (1851) 21 LJQB 75.

§A.3 Example reports 263

agreement regarding the right to the chattel; the chattel was hidden
or was in a position so as to be difficult to find; an attempt was made
to find the true owner of the chattel or, alternatively, the chattel was
clearly abandoned; and neither party knew of the existence of the chat-
tel prior to the finding.

However, there are several significant differences between the hypo-
thetical case and Hannah v. Peel. In that case the other claimant was
the owner of the premises where the chattel was found; the finder did
not hand over the chattel to the other claimant after the finding; and
the finder was not a servant of the other claimant.

Despite the fact that Hannah v. Peel is a decision of the King's Bench
Division of the English High Court (and as good authority as a case
decided by the Queen's Bench Division of the English High Court-
like London v. Appleyard (1)) : there is nothing in Hannah v. Peel to
warrant any change in my conclusion.

A. 3 Report file for APRA v. Jain

Authorization area

Instant case

The notion of authorization extends beyond the authority given to an
agent. The word "authorize 11 should be "understood in its ordinary
dictionary sense of 'sanction, approve, and countenance. ,Ml

"[A] person who has under his control the means by which an in-
fringement of copyright may be committed . . . and who makes it avail-
able to other persons, knowing, or having reason to suspect, that it is
likely to be used for the purpose of committing an infringement, and
omitting to take reasonable steps to limit its use to legitimate purposes,
would authorize any infringement that resulted from its use.""

In the instant case, the infringer was not an employee of the accused;
the infringer was an independent contractor to the accused; the accused
did not sell or hire the infringer the means of infringing; the accused
had the power to prevent the infringement; the accused did not take
reasonable steps to avoid the infringement: the accused knew, or had
reason to anticipate or suspect, that the infringing act was to be, or
was likely to be, done; and the specific infringement was not causally
related to an incitement to infringe on the part of the accused.

1 Falcon v. Famous Players Film Co. [1920] 2 KB 474 at 491 per Bankes LJ.

2 University of New South Wales v. Moorhouse (1975) 133 CLR 1 at 13 per
Gibbs J.

264 A pragmatic legal expert system §A.3

In my opinion — following Mellor v. Australian Broadcasting Commis-
sion — the accused authorized the infringement.

In Mellor v. Australian Broadcasting Commission? a 1940 decision of
the Judicial Committee of the Privy Council. Mellor and others held
the sole right to perform in public in Australia some musical works
arranged for performance by brass and military bands. They published
and distributed advertising pamphlets which included a statement that
all of their sheet music was " 'Free for Public Performance' anywhere
. . . We have paid for the performing rights of every piece we issue." 4
The ABC engaged bands to play some of this music, and broadcast the
bands' performances on radio.

The Privy Council held that the ABC had authorized the bands to
perform the musical works within the meaning of s. 1(2) of the Cop} r -
right Act 1911 (UK) which was in force in Australia by virtue of the
Copyright Act 1912 (Cth). However, the ABC had not infringed the
plaintiffs' sole right to authorize public performance because the state-
ments made in the pamphlets amounted to consent.

There are several significant similarities between the instant case and
Mellor v. ABC: the infringer was not an employee of the accused; the
infringer was an independent contractor to the accused; the accused
did not sell or hire the infringer the means of infringing; the accused
had the power to prevent the infringement: the accused did not take
reasonable steps to avoid the infringement: and the accused knew, or
had reason to anticipate or suspect, that the infringing act was to be,
or was likely to be, done.

However, the instant case is not on all fours with Mellor v. ABC. In
that case the specific infringement was causally related to an incitement
to infringe on the part of the accused.

Nevertheless, I believe that Mellor v. ABC should be followed.

If RCA Corporation v. John Fairfax and Sons Ltd is followed then the
accused did not authorize the infringement.

In RCA Corporation v. John Fairfax and Sons Ltd? a 1981 decision
of the Supreme Court of New South Wales, the Fairfax newspaper the
Sun-Herald carried an article which pointed out that, using cassette
tapes and good quality taping equipment, the same album can be taped
by many people. It also discussed how the advent of FM radio had made
it easy for people to tape new album and single releases without buying

3 [1940] AC 491.
4 ibid. at 498-9.
5 fl98ll 1 NSWLR 2131.

§A.3 Example reports 265

the discs: "Why spend nearly $10 on the new David Bowie album when
you can tape it from 2JJJ?"

Kearnej' J held that "authorization involves some element of causa-
tion — and hence the necessity for some relationship creating a link or
connection however tenuous between the authorizer and the infringer." '
There was no such link, so Fairfax had not authorized any infringement
within the meaning of s. 13(2) of the Copyright Act 1968 (Cth).

There are several similarities between the instant case and RCA v.
Fairfax: the infringer was not an employee of the accused; the accused
did not sell or hire the infringer the means of infringing; the accused
did not take reasonable steps to avoid the infringement; the accused
knew, or had reason to anticipate or suspect, that the infringing act
was to be, or was likely to be, done; and the specific infringement was
not causally related to an incitement to infringe on the part of the
accused.

However, there are two very significant differences between the in-
stant case and RCA v. Fairfax. In that case the infringer was not an
independent contractor to the accused; and the accused did not have
the power to prevent the infringement.

Despite the fact that RCA v. Fairfax is a decision of the Supreme
Court of New South Wales (and better authority than a case decided
by the Judicial Committee of the Privy Council— like Mellor v. ABC).
there is nothing in RCA v. Fairfax to warrant any change in my con-
clusion.

If Australasian Performing Right Association Ltd v. Miles is followed
then the accused is liable (directly or vicariously) for the infringement.

In Australasian Performing Right Association Ltd v. Miles, 8 a 1961
decision of the Supreme Court of New South Wales, the Dee Why RSL
Club engaged a band to play at a dance held at the club. During
the dance the band pla3'ed Vve Got a Lovely Bunch of Coconuts, the
copyright in which was owned by the Australasian Performing Right
Association.

Jacobs J held that the members of the band were servants of the club,
because "the club through its officers was exercising a control over the
work performed in such a way as to show that there was an authority
to command the orchestra in its performance."* So the members of
the club, through the band, performed the musical work and infringed
the copyright under s. 2(1) of the Copyright Act 1911 (UK) which was
in force in Australia by virtue of the Copyright Act 1912 (Cth).

6 ibid. at 252.
7 ibid. at 259.
8 [1962] NSWR405.
9 ibid. at 407.

266 A pragmatic legal expert system §A.3

There are several similarities between the instant case and APRA v.
Miles: the accused did not sell or hire the infringer the means of in-
fringing; the accused had the power to prevent the infringement; the
accused did not take reasonable steps to avoid the infringement; and
the accused knew, or had reason to anticipate or suspect, that the
infringing act was to be, or was likely to be, done.

However, there are several significant differences between the instant
case and APRA v. Miles. In that case the infringer was an employee
of the accused; the infringer was not an independent contractor to
the accused; and the specific infringement was causally related to an
incitement to infringe on the part of the accused.

Despite the fact that APRA v. Miles is a decision of the Supreme
Court of New South Wales (and better authority than a case decided
by the Judicial Committee of the Privy Council— like Mellor v. ABC).
there is nothing in APRA v. Miles to warrant any change in my con-
clusii m.

Hypothetical 1

Consider the instant case changed so that the following is true: the
specific infringement was causally related to an incitement to infringe
on the part of the accused.

If that were so then I would be more strongly of the opinion that —
following Mellor v. Australian Broadcasting Commission—the accused
authorized the infringement.

Details of Mellor v. ABC are summarized above. The hypothetical case
is on all fours with Mellor v. ABC.

If RCA Corporation v. John Fairfax and Sons Ltd is followed then the
accused did not authorize the infringement.

Details of RCA v. Fairfax are summarized above. There are several
similarities between the hypothetical case and RCA v. Fairfax: the
infringer was not an employee of the accused; the accused did not sell
or hire the infringer the means of infringing; the accused did not take
reasonable steps to avoid the infringement; and the accused knew, or
had reason to anticipate or suspect, that the infringing act was to be,
or was likely to be, done.

However, there are several significant differences between the hypo-
thetical case and RCA v. Fairfax. In that case the infringer was not
an independent contractor to the accused; the accused did not have
the power to prevent the infringement; and the specific infringement
was not causally related to an incitement to infringe on the part of the
accused.

§A.3 Example reports 261

Despite the fact that RCA v. Fairfax is a decision of the Supreme
Court of New South Wales (and better authority than a case decided
by the Judicial Committee of the Privy Council — like Mellor v. ABC).
there is nothing in RCA v. Fairfax to warrant any change in my con-
clusion.

If Australasian Performing Right Association Ltd v. Miles is followed
then the accused is liable (directly or vicariously) for the infringement.

Details of APRA v. Miles are summarized above. There are several
similarities between the hypothetical case and APRA v. Miles: the
accused did not sell or hire the infringer the means of infringing; the
accused had the power to prevent the infringement; the accused did
not take reasonable steps to avoid the infringement; the accused knew,
or had reason to anticipate or suspect, that the infringing act was to
be, or was likely to be, done; and the specific infringement was causally
related to an incitement to infringe on the part of the accused.

However, there are two very significant differences between the hy-
pothetical case and APRA v. Miles. In that case the infringer was an
employee of the accused; and the infringer was not an independent
contractor to the accused.

Despite the fact that APRA v. Miles is a decision of the Supreme
Court of New South Wales (and better authority than a case decided
by the Judicial Committee of the Privy Council — like Mellor v. ABC).
there is nothing in APRA v. Miles to warrant any change in my con-
clusion.

Hypothetical 2

Consider the instant case changed so that the following is true: the
accused did not know, and had no reason to anticipate or suspect, that
the infringing act was to be, or was likely to be, done.

If that were so then my opinion would be that — following Performing
Right Society Ltd v. Ciryl Theatrical Syndicate Ltd — the accused did
not authorize the infringement.

In Performing Right Society Ltd v. Ciryl Theatrical Syndicate Ltd, 10 a
1923 decision of the English Court of Appeal, the syndicate was the
lessee of a theatre. The managing-director of the syndicate produced
a play at that theatre, and engaged a band to perform at the theatre
under the direction of a bandmaster. In the absence of the managing-
director, and without his knowledge, the band performed works the
copyright in which was owned by the Performing Right Society.

[1924] 1 KB 1.

268 A pragmatic legal expert system §A.3

Bankes, Scrutton and Atkin LJJ held that the managing-director
had not authorized the infringing performances, within the meaning
of s. 1(2) of the Copyright Act 1911 (UK), because the infringement
occurred without his knowledge and he had no reason to anticipate
or suspect that the band was likely to give performances which would
breach copyright.

The hypothetical case is on all fours with PRS v. Ciryl.

If Mellor v. Australian Broadcasting Commission is followed then the
accused authorized the infringement.

Details of Mellor v. ABC are summarized above. There are several
similarities between the hypothetical case and Mellor v. ABC: the
infringer was not an employee of the accused; the infringer was an
independent contractor to the accused; the accused did not sell or hire
the infringer the means of infringing; the accused had the power to
prevent the infringement; and the accused did not take reasonable steps
to avoid the infringement.

However, there are two very significant differences between the hy-
pothetical case and Mellor v. ABC. In that case the accused knew, or
had reason to anticipate or suspect, that the infringing act was to be,
or was likely to be, done; and the specific infringement was causally
related to an incitement to infringe on the part of the accused.

Despite the fact that Mellor v. ABC is a decision of the Judicial
Committee of the Privy Council (and better authority than a case
decided by the English Court of Appeal like PRS r. Ciryl) t there is
nothing in Mellor v. ABC to warrant any change in my conclusion.

If Australasian Performing Right Association Ltd v. Miles is followed
then the accused is liable (directly or vicariously) for the infringement.

However, there are several significant differences between the hypo-
thetical case and APRA v. Miles. In that case the infringer was an
employee of the accused; the infringer was not an independent con-
tractor to the accused; the accused knew, or had reason to anticipate
or suspect, that the infringing act was to be, or was likely to be, done;
and the specific infringement was causally related to an incitement to
infringe on the part of the accused.

Despite the fact that APRA v. Miles is a decision of the Supreme
Court of New South Wales (and better authority than a case decided
by the English Court of Appeal — like PRS v. Ciryl), there is nothing
in APRA v. Miles to warrant any change in my conclusion.

§A.3 Example reports 269

Hypothetical 3

Consider the instant case changed so that the following is true: the
infringer was an employee of the accused; and the infringer was not an
independent contractor to the accused.

If that were so then my opinion would be that — following Australasian
Performing Right Association Ltd v. Miles — the accused is liable (dir-
ectly or vicariously) for the infringement.

Details of APRA v. Miles are summarized above. There are several sig-
nificant similarities between the hypothetical case and APRA v. Miles:
the infringer was an employee of the accused; the infringer was not an
independent contractor to the accused; the accused did not sell or hire
the infringer the means of infringing; the accused had the power to
prevent the infringement; the accused did not take reasonable steps to
avoid the infringement; and the accused knew, or had reason to anti-
cipate or suspect, that the infringing act was to be, or was likely to be,
done.

However, the hypothetical case is not on all fours with APRA v.
Miles. In that case the specific infringement was causally related to an
incitement to infringe on the part of the accused.

Nevertheless, I believe that APRA v. Miles should be followed.

If University of New South Wales v. Moorhouse is followed then the
accused authorized the infringement.

In University of New South Wales v. Moorhouse, 11 a 1975 decision
of three justices of the High Court of Australia, a graduate of the
Universit}' used a photocopy machine in the University library to make
two copies of a story from a library copy of a book of short stories.

McTiernan ACJ, Gibbs and Jacobs J J held that the University had
authorized the infringement within the meaning of s. 36(1) of the Copy-
right Act 1968 (Cth); it had the power to prevent infringements, but
had not taken reasonable steps to prevent them. 12 Gibbs J's state-
ment about what constitutes authorization of an infringement is quoted
ab< ive.

There are several similarities between the hypothetical case and
UNSW v. Moorhouse: the infringer was not an independent contractor
to the accused; the accused had the power to prevent the infringement;
the accused did not take reasonable steps to avoid the infringement;
the accused knew, or had reason to anticipate or suspect, that the

"(1975) 133 CLR 1.

The addition, in 1980, of s. 39A to the Copyright Act ameliorated the
effect of UNSW v. Moorhouse as far as photocopying in libraries is concerned.

270 A pragmatic legal expert system §A.4

infringing act was to be, or was likely to be, done; and the specific
infringement was not causally related to an incitement to infringe on
the part of the accused.

However, there are two very significant differences between the hy-
pothetical case and UNSW v. Moorhouse. In that case the infringer
was not an employee of the accused; and the accused sold or hired the
infringer the means of infringing.

Despite the fact that UNSW v. Moorhouse is a decision of three
justices of the High Court of Australia (and better authority than a
case decided by the Supreme Court of New South Wales^like APRA v.
Miles), there is nothing in UNSW v. Moorhouse to warrant any change
in my conclusion.

If RCA Corporation v. John Fairfax and Sons Ltd is followed then the
accused did not authorize the infringement.

Details of RCA v. Fairfax are summarized above. There are several
similarities between the hypothetical case and RCA v. Fairfax: the
infringer was not an independent contractor to the accused; the accused
did not sell or hire the infringer the means of infringing; the accused
did not take reasonable steps to avoid the infringement; the accused
knew, or had reason to anticipate or suspect, that the infringing act
was to be, or was likely to be, done; and the specific infringement was
not causally related to an incitement to infringe on the part of the
accused.

However, there are two very significant differences between the hy-
pothetical case and RCA v. Fairfax. In that case the infringer was not
an employee of the accused; and the accused did not have the power
to prevent the infringement.

Despite the fact that RCA v. Fairfax and APRA v. Miles are both
decisions of the Supreme Court of New South Wales, there is nothing
in RCA v. Fairfax to warrant any change in my conclusion.

A. 4 Report file for Re Porter

Employee area

Instant case

The law distinguishes between a contract of service (between employer
and employee) and a contract for services (between principal and in-
dependent contractor). This distinction affects the terms that will be
implied in the absence of an express agreement, the liability of the
employer to third parties, the applicability of industrial awards, the
applicability of statutes which may affect workers' compensation, oc-
cupational health and safetj', long-service leave, fringe benefits tax, etc.

§A.4 Example reports 271

The terms "employer" and "worker" are used here to mean "em-
ployer" and "employee" (in the case of a contract of service) or "prin-
cipal" and "independent contractor" (in the case of a contract for
services).

In the instant case, the employer directed the manner in which the
work was to be done; the worker was not allowed to use her/his own
discretion in doing an aspect of the work that was not specified before-
hand; the worker was an integral part of the employer's business; the
worker owned the tools or provided the transport with which she/he
performed the work; it is not known whether the employer would make
a profit/loss if the work performed bj' the worker cost less/more than
expected; the work was not performed on the employer's premises; it
is not known whether the employer supervised or inspected the work;
the worker was not in business on her/his own account; the worker was
not allowed to employ others to assist with her/his work; the worker
was obliged to work only for the employer; the worker was not required
to work at specified times; the employer paid the worker by time; the
money that the employer paid to the worker was not stated to be a
"fee"; the money that the employer paid to the worker was not stated
to be "wages" or "salary"; the employer deducted PAYE tax instal-
ments from the worker's pay; the employer paid the worker neither
sick pay nor holidaj' pay; the employer and the worker did not express
any intention that the relationship would be one of employer and em-
ployee; and the employer and the worker expressed an intention that
the relationship would be one of principal and independent contractor.

In my opinion— following Ferguson v. John Dawson and Partners (Con-
tractors) Ltd — the worker is an employee.

In Ferguson v. John Dawson and Partners (Contractors) Ltd, 1 a 1976
decision of the English Court of Appeal, Ferguson fell off a roof while re-
moving some scaffolding boards. He claimed damages against Dawson
(the building contractors) for breach of statutory duty relying on the
Construction (Working Places) Regulations 1966 (UK). This duty
would only be owed if Ferguson was an employee of Dawson.

Megaw and Browne LJJ held that, despite the fact that both parties
labelled Ferguson a "self-employed labour only subcontractor", the real-
ity of the relationship between them was that of employer and employ-

ee."

There are several significant similarities between the instant case and
Ferguson v. Dawson: the employer directed the manner in which the
work was to be done; the worker was not allowed to use her/his own

] [1976] 1 WLR 1213.

2 ibicl. at 1219 per Megaw LJ, at 1228-9 per Browne LJ.

272 A pragmatic legal expert system §A.4

discretion in doing an aspect of the work that was not specified be-
forehand; the worker was an integral part of the employer's business;
the work was not performed on the employer's premises; the worker
was not in business on her/his own account; the worker was not al-
lowed to employ others to assist with her/his work; the employer paid
the worker bj r time; the money that the employer paid to the worker
was not stated to be a "fee"; the money that the employer paid to
the worker was not stated to be "wages" or "salary"; the employer
paid the worker neither sick pay nor holidaj' pay; the employer and
the worker did not express any intention that the relationship would
be one of employer and employee; and the employer and the worker
expressed an intention that the relationship would be one of principal
and independent contractor.

However, the instant case is not on all fours with Ferguson v. Dawson.
In that case the worker neither owned the tools nor provided the trans-
port with which she/he performed the work; the employer would make
a profit/loss if the work performed by the worker cost less/more than
expected; the employer supervised or inspected the work; the worker
was not obliged to work only for the employer; the worker was required
to work at specified times; and the employer did not deduct PAYE tax
instalments from the worker's pa}'.

Nevertheless, I believe that Ferguson v. Dawson should be followed.

If Ready Mixed Concrete (South East) Ltd v. Minister of Pensions
and National Insurance is followed then the worker is an independent
contractor.

In Ready Mixed Concrete (South East) Ltd v. Minister of Pensions and
National Insurance. 3 a 1967 decision of the Queen's Bench Division of
the English High Court, Latimer worked for Read}' Mixed as an "owner-
driver." He was paid at mileage rates, and was obliged to buy the truck
through a financial organization associated with Ready Mixed. The
truck was painted in the company's colours, and he had to wear a Ready
Mixed uniform. Latimer was obliged to meet the costs of maintenance,
repair and insurance of the truck (and the attached mixing unit, which
belonged to Ready Mixed). The Minister determined that Latimer was
employed under a contract of service, and was therefore an "employed
person" under s. 1(2) of the National Insurance Act 1965 (UK), making
Ready Mixed liable to make weekly contributions.

MacKenna J examined the contract and held that the rights it con-
ferred, and the duties it imposed, between Latimer and Ready Mixed
were not such as to make it a contract of service.

'[1968] 2 QB 497.

§A.4 Example reports 273

There are several similarities between the instant case and Ready
Mixed v. Minister: the worker was an integral part of the employer's
business; the worker owned the tools or provided the transport with
which she/he performed the work; the work was not performed on the
emploj'er's premises; the worker was not in business on her/his own
account; the worker was obliged to work only for the employer; the
worker was not required to work at specified times; the money that the
employer paid to the worker was not stated to be a "fee"; the money
that the employer paid to the worker was not stated to be "wages"
or "salary"; the employer paid the worker neither sick pay nor holiday
pa} 1- ; the employer and the worker did not express any intention that the
relationship would be one of employer and employee; and the employer
and the worker expressed an intention that the relationship would be
one of principal and independent contractor.

However, there are several significant differences between the instant
case and Ready Mixed v. Minister. In that case the employer did not
direct the manner in which the work was to be done; the worker was
allowed to use her/his own discretion in doing an aspect of the work
that was not specified beforehand; the employer would not make a
profit/loss if the work performed by the worker cost less/more than
expected; the employer neither supervised nor inspected the work; the
worker was allowed to employ others to assist with her/his work; the
employer did not pay the worker by time; and the employer did not
deduct PAYE tax instalments from the worker's pa}'. Note also that
Ready Mixed v. Minister is only a decision of the Queen's Bench Di-
vision of the English High Court and not as good authority as a case
decided by the English Court of Appeal — like Ferguson v. Dawson.

Consequently, there is nothing in Ready Mixed v. Minister to warrant
an}' change in my conclusion.

Instantiation 4

It may be that the following is true of the instant case: the employer
would not make a profit/loss if the work performed by the worker
cost less/more than expected; and the employer neither supervised nor
inspected the work.

If that is so then in my opinion— following Ready Mixed Concrete
(South East) Ltd v. Minister of Pensions and National Insurance —
the worker is an independent contractor.

Details of Ready Mixed v. Minister are summarized above. There are
several significant similarities between the instantiated case and Ready
Mixed v. Minister: the worker was an integral part of the employer's
business; the worker owned the tools or provided the transport with
which she/he performed the work; the employer would not make a

274 A pragmatic legal expert system §A.4

profit/loss if the work performed by the worker cost less/more than
expected; the work was not performed on the employer's premises;
the employer neither supervised nor inspected the work; the worker
was not in business on her/his own account; the worker was obliged
to work only for the employer; the worker was not required to work
at specified times; the money that the employer paid to the worker
was not stated to be a "fee"; the money that the employer paid to
the worker was not stated to be "wages" or "salary"; the employer
paid the worker neither sick pay nor holidaj' pay; the employer and
the worker did not express any intention that the relationship would
be one of employer and employee: and the employer and the worker
expressed an intention that the relationship would be one of principal
and independent contractor.

However, the instantiated case is not on all fours with Ready Mixed v.
Minister. In that case the employer did not direct the manner in which
the work was to be done; the worker was allowed to use her/his own
discretion in doing an aspect of the work that was not specified before-
hand; the worker was allowed to employ others to assist with her/his
work; the employer did not pa}' the worker by time; and the employer
did not deduct PAYE tax instalments from the worker's pay.

Nevertheless, I believe that Ready Mixed v. Minister should be fol-
lowed.

If Ferguson v. John Dawson and Partners (Contractors) Ltd is followed
then the worker is an employee.

Details of Ferguson v. Dawson are summarized above. There are several
similarities between the instantiated case and Ferguson v. Dawson:
the employer directed the manner in which the work was to be done;
the worker was not allowed to use her/his own discretion in doing an
aspect of the work that was not specified beforehand; the worker was
an integral part of the employer's business; the work was not performed
on the employer's premises; the worker was not in business on her/his
own account; the worker was not allowed to employ others to assist
with her/his work; the employer paid the worker bj f time; the money
that the employer paid to the worker was not stated to be a "fee";
the money that the employer paid to the worker was not stated to be
"wages" or "salary"; the employer paid the worker neither sick pay nor
holiday pay; the employer and the worker did not express any intention
that the relationship would be one of employer and employee: and the
employer and the worker expressed an intention that the relationship
would be one of principal and independent contractor.

However, there are several significant differences between the instan-
tiated case and Ferguson v. Dawson. In that case the worker neither
owned the tools nor provided the transport with which she/he per-
formed the work; the employer would make a profit/loss if the work

§A.4 Example reports 275

performed by the worker cost less/more than expected; the employer
supervised or inspected the work; the worker was not obliged to work
only for the employer; the worker was required to work at specified
times; and the employer did not deduct PAYE tax instalments from
the worker's pay.

Despite the fact that Ferguson v. Dawson is a decision of the Eng-
lish Court of Appeal (and better authority than a case decided by
the Queen's Bench Division of the English High Court — like Ready
Mixed v. Minister), there is nothing in Ferguson v. Dawson to warrant
an} 1- change in my conclusion.

Hypothetical 1

Consider the instant case changed so that the following is true: the
worker was not obliged to work only for the employer; and the employer
did not deduct PAYE tax instalments from the worker's pay.

If that were so then I would be more strongly of the opinion that —
following Ferguson v. John Dawson and Partners (Contractors) Ltd—
the worker is an emploj'ee.

Details of Ferguson v. Dawson are summarized above. There are several
significant similarities between the hypothetical case and Ferguson v.
Dawson: the employer directed the manner in which the work was to be
done; the worker was not allowed to use her/his own discretion in doing
an aspect of the work that was not specified beforehand; the worker was
an integral part of the employer's business; the work was not performed
on the employer's premises; the worker was not in business on her/his
own account; the worker was not allowed to emplo}' others to assist
with her/his work; the worker was not obliged to work only for the
employer; the employer paid the worker by time; the money that the
employer paid to the worker was not stated to be a "fee"; the money
that the employer paid to the worker was not stated to be "wages" or
"salary"; the employer did not deduct PAYE tax instalments from the
worker's pay; the employer paid the worker neither sick pay nor holiday
pay; the employer and the worker did not express any intention that the
relationship would be one of employer and employee; and the employer
and the worker expressed an intention that the relationship would be
one of principal and independent contractor.

However, the hypothetical case is not on all fours with Ferguson v.
Dawson. In that case the worker neither owned the tools nor provided
the transport with which she/he performed the work; the employer
would make a profit/loss if the work performed by the worker cost
less/more than expected; the employer supervised or inspected the
work; and the worker was required to work at specified times.

Nevertheless, I believe that Ferguson v. Dawson should be followed.

276 A pragmatic legal expert system §A.4

If Ready Mixed Concrete (South East) Ltd v. Minister of Pensions
and National Insurance is followed then the worker is an independent
contractor.

Details of Ready Mixed v. Minister are summarized above. There are
several similarities between the hypothetical case and Ready Mixed v.
Minister: the worker was an integral part of the employer's business;
the worker owned the tools or provided the transport with which she/he
performed the work; the work was not performed on the emploj'er's
premises; the worker was not in business on her/his own account; the
worker was not required to work at specified times; the money that the
employer paid to the worker was not stated to be a "fee"; the money
that the employer paid to the worker was not stated to be "wages" or
"salary"; the employer did not deduct PAYE tax instalments from the
worker's pay; the employer paid the worker neither sick pay nor holiday
pay; the employer and the worker did not express any intention that the
relationship would be one of employer and employee; and the employer
and the worker expressed an intention that the relationship would be
one of principal and independent contractor.

However, there are several significant differences between the hypo-
thetical case and Ready Mixed v. Minister. In that case the employer
did not direct the manner in which the work was to be done; the worker
was allowed to use her/his own discretion in doing an aspect of the
work that was not specified beforehand; the employer would not make
a profit/loss if the work performed by the worker cost less/more than
expected; the employer neither supervised nor inspected the work; the
worker was allowed to employ others to assist with her/his work; the
worker was obliged to work only for the employer; and the employer
did not pa}' the worker by time. Note also that Ready Mixed v. Min-
ister is only a decision of the Queen's Bench Division of the English
High Court and not as good authority as a case decided by the English
Court of Appeal — like Ferguson v. Dawson.

Consequently, there is nothing in Ready Mixed v. Minister to warrant
anj'' change in my conclusion.

Hypothetical 2

Consider the instant case changed so that the following is true: the
worker was allowed to use her/his own discretion in doing an aspect of
the work that was not specified beforehand; and the employer did not
deduct PAYE tax instalments from the worker's pay.

If that were so then my opinion would be that — following Ready Mixed
Concrete (South East) Ltd v. Minister of Pensions and National In-
surance — the worker is an independent contractor.

§A.4 Example reports 21

Details of Ready Mixed v. Minister are summarized above. There are
several significant similarities between the hypothetical case and Ready
Mixed v. Minister: the worker was allowed to use her/his own discretion
in doing an aspect of the work that was not specified beforehand; the
worker was an integral part of the employer's business; the worker
owned the tools or provided the transport with which she/he performed
the work; the work was not performed on the employer's premises; the
worker was not in business on her/his own account; the worker was
obliged to work only for the employer; the worker was not required to
work at specified times; the money that the employer paid to the worker
was not stated to be a "fee"; the money that the employer paid to the
worker was not stated to be "wages" or "salary"; the employer did
not deduct PAYE tax instalments from the worker's pay; the employer
paid the worker neither sick pay nor holidaj' pay; the employer and
the worker did not express any intention that the relationship would
be one of employer and employee; and the employer and the worker
expressed an intention that the relationship would be one of principal
and independent contractor.

However, the hypothetical case is not on all fours with Ready Mixed v.
Minister. In that case the employer did not direct the manner in which
the work was to be done; the employer would not make a profit/loss if
the work performed by the worker cost less/more than expected; the
employer neither supervised nor inspected the work; the worker was
allowed to employ others to assist with her/his work; and the employer
did not pay the worker by time.

Nevertheless, I believe that Ready Mixed v. Minister should be fol-
lowed.

If Ferguson v. John Dawson and Partners (Contractors) Ltd is followed
then the worker is an employee.

Details of Ferguson v. Dawson are summarized above. There are sev-
eral similarities between the hypothetical case and Ferguson v. Dawson:
the employer directed the manner in which the work was to be done;
the worker was an integral part of the employer's business; the work
was not performed on the employer's premises; the worker was not in
business on her/his own account; the worker was not allowed to em-
ploy others to assist with her/his work; the employer paid the worker
by time; the mone}' that the employer paid to the worker was not stated
to be a "fee"; the money that the employer paid to the worker was not
stated to be "wages" or "salary"; the employer did not deduct PAYE
tax instalments from the worker's pay; the emploj'er paid the worker
neither sick pay nor holiday pay; the employer and the worker did not
express any intention that the relationship would be one of employer
and employee; and the employer and the worker expressed an inten-

278 A pragmatic legal expert system §A.5

tion that the relationship would be one of principal and independent
contractor.

However, there are several significant differences between the hypo-
thetical case and Ferguson v. Dawson. In that case the worker was not
allowed to use her/his own discretion in doing an aspect of the work
that was not specified beforehand; the worker neither owned the tools
nor provided the transport with which she/he performed the work; the
employer would make a profit/loss if the work performed by the worker
cost less/more than expected; the employer supervised or inspected the
work; the worker was not obliged to work only for the employer; and
the worker was required to work at specified times.

A. 5 Report files for Ainsworth v. CJC

Natural area
Instant case

In recent years courts have tended to imply a duty to observe the
principles of natural justice. It has been said that "[t]he law has now
developed to a point where it may be accepted that there is a common
law duty to act fairly, in the sense of according procedural fairness,
in the making of administrative decisions which affect rights, interests
and legitimate expectations, subject only to the clear manifestation of a
contrary statutory intention." 1 However, there are some circumstances
in which a duty to observe natural justice will not be implied: "the law
has not yet reached the stage of applying the obligation of natural
justice to ever}' - decision which disadvantages individuals.""

In the instant case, the decision affected the property, right, interest,
status, or legitimate expectation of the applicant; the decision is apt to
have a discrete impact on the interests of the applicant; the power is of
a nature that would suggest that procedural fairness would be applied;
the statutory or factual criteria focused on matters which were discrete
to the interests of the applicant; the decision-maker was not a high-

1 Kioa v. West (1985) 159 CLR 550 at 584 per Mason J.
Minister for Arts, Heritage and Environment v. Peko-Wallsend Ltd (1987)
15 FCR 274 at 306 per Wilcox J.

§A.5 Example reports 279

level policy-maker; there is no statutory right to appeal against the
decision; and there were no circumstances which would have made an
obligation to observe natural justice inappropriate.

In my opinion — following Annetts v. McCann—a, duty to observe nat-
ural justice is implied.

In Annetts v. McCann, 3 a 1990 decision of five justices of the High
Court of Australia, a coroner had been conducting an inquest into the
death of a 16-year old boy. The boj^'s parents (Mr and Mrs Annetts)
sought to make a submission before the coroner made a finding. The
coroner decided that the Coroner's Act 1920 (WA) gave him the dis-
cretion (which he chose to exercise) to disallow their submission. The
Annettses appealed.

The High Court (Mason CJ, Brennan, Deane, Toohey and McHugh
JJ) held that their son's reputation gave the Annettses an interest in
the Coroner's inquiry. "A finding in an inquest into a death is naturally
likely to deal with the conduct of the deceased leading to death. An
unfavourable reflection on the deceased is usually a matter of concern
to her or his parents, spouse or children and, if they choose to appear
at the inquest in order to safeguard the reputation of the deceased, the
familial relationship suffices, in my view, to establish the deceased's
reputation as a relevant interest which should not be adversely affected
without according natural justice to those who are seeking to safeguard
that reputation." 4

The Court held that the fact that the coroner's decision was merely
recommendatory (whether or not to prosecute) was not sufficient to
avoid the implication of natural justice; the coroner was bound to hear
the Annettses before making airy finding adverse to them or their son. 5

The instant case is on all fours with Annetts v. McCann.

If Mclnnes v. Onslow-Fane is followed then a duty to observe natural
justice is not implied.

In Mclnnes v. Onslow-Fane, 6 a 1978 decision of the Chancery Division
of the English High Court, Mclnnes had held, at various times, licences
to promote, train and act as master of ceremonies in professional box-
ing. All his licences were revoked bj f the British boxing board of control.

3 (1990) 170 CLR 596.
ibid, at 012 per Brennan J.

d ibid. at 603 per Mason CJ, Deane and McHugh JJ. at 612 per Brennan J,
at 621 per Toohey J. Note, however, that Brennan and Toohey JJ would
have dismissed the appeal because they believed that the decision of the Full
Court of the Supreme Court of Western Australia (from which the Annettses
appealed) was right on the material before it.

6 [1978] 1 WLR 1520.

280 A pragmatic legal expert system §A.5

He made five unsuccessful applications for a manager's licence. With
his sixth application he requested an oral hearing and prior notification
of anything that might prevent the area council (to which he applied)
making a favourable recommendation to the board. The board refused
his applications without giving him an oral hearing or informing him
of the case against him.

Megarry V-C held that the board was under no duty to provide reas-
ons to Mclnnes or to allow him a hearing: "This is not a case in which
there has been any suggestion of the board considering any alleged dis-
honesty or morally culpable conduct of the plaintiff. A man free from
any moral blemish may nevertheless be wholly unsuitable for a partic-
ular type of work ... In such circumstances, in the absence of anything
to suggest that the board have been affected by dishonesty or bias or
caprice, or that there is any other impropriety, I think that the board
are full}' - entitled to give no reasons for their decision, and to decide the
application without any preliminary indication to the plaintiff of those
reasons. The board are the best judges of the desirability of granting
the licence, and in the absence of any impropriety the court ought not
to interfere." 7

The instant case is on all fours with Mclnnes v. Onslow-Fane. Note,
however, that Mclnnes v. Onslow-Fane is only a decision of the Chan-
cery Division of the English High Court and not as good authority
as a case decided bj' five justices of the High Court of Australia — like
Annetts v. McCann.

Consequently, there is nothing in Mclnnes v. Onslow-Fane to war-
rant any change in my conclusion.

Hypothetical 1

Consider the instant case changed so that the following is true: the
statutory or factual criteria focused on matters of policy or public in-
terest; and the decision-maker was a high-level policy-maker.

If that were so then my opinion would be that — following South Aus-
tralia v. O'Shea — a duty to observe natural justice is not implied.

In South Australia v. O'Shea, 8 a 1987 decision of five justices of the
High Court of Australia, O'Shea had been convicted of two offences
of indecent assault of 3'oung children. He was released on licence and
remained at liberty after the licence expired. Over a year later, after
allegations had been made against him, O'Shea was apprehended and
detained. The parole board recommended his release on licence on

7 ibid. at 1535.
8 (1987) 163 CLR 378.

§A.5 Example reports 281

various conditions, but the Governor in Council resolved to take no
action. O'Shea had been given a hearing by the parole board, but he
claimed he was entitled to a further hearing before the Governor in
Council could exercise his discretionary powers under s. 77a(7a) of the
Criminal Law Consolidation Act, 1935 (SA).

Mason CJ, Wilson, Brennan and Toohey JJ (Deane J dissenting)
held that O'Shea was not entitled to a further hearing. "Given the
nature of this decision, it cannot be said that Mr O'Shea could have
more than a hope that the Governor would be prepared to act on
the recommendation of the Board. Hope, of itself, is not sufficient to
ground an expectation that will attract legal consequences. So far as
the concept of legitimate expectation is concerned, Mr O'Shea must be
taken to know that the Act committed to the Governor, with the advice
and consent of the Executive Council, the responsibility for determining
where the public interest lay . . . The nature of the decision that they
were required to make was such that participation by Mr O'Shea was
inappropriate." 9

The hypothetical case is on all fours with SA v. O'Shea.

If Macrae v. Attorney- General for New South Wales is followed then a
duty to observe natural justice is implied.

In Macrae v. Attorney- General for New South Wales, 10 a 1987 decision
of the New South Wales Court of Appeal, five magistrates who had
been appointed under the Justices Act 1902 (NSW) were not appointed
under the Local Courts Act 1982 (NSW). The new Act had reorganized
the magistracy in NSW, and magistrates appointed under the old Act
were entitled to apply for appointment as magistrates under the new
Act. The five had applied and were interviewed. Allegations were made
privately to the Attorney-General claiming that they were unfit to be
appointed, but these allegations were not brought to their notice at the
time of the interviews.

The Court of Appeal held that the Attorney-General's decision not
to recommend the appointment of the magistrates was void because
the}' - were denied their legitimate expectation of procedural fairness.
"They have not been treated fairly." 11

The hypothetical case is on all fours with Macrae v. AG. Note,
however, that Macrae v. AG is only a decision of the New South Wales
Court of Appeal and not as good authority as a case decided by five
justices of the High Court of Australia — like SA v. O'Shea.

Consequently, there is nothing in Macrae v. AG to warrant any
change in my conclusion.

ibid, at 402 per Wilson and Toohey JJ.
10 (1987) 9 NSWLR268.
u ibid. at 283 per Kirby P.

282 A pragmatic legal expert system §A.5

Affected area

Instant case

In the instant case, the decision did not affect a financial, property
or occupational interest of the applicant; the decision did not affect
the applicant's personal liberty; the decision affected the applicant's
reputation; and the applicant did not have a legitimate expectation
which was affected h}' the decision.

In my opinion- — following Annetts v. McCann — the decision affected
the property, right, interest, status, or legitimate expectation of the
applicant.

In Annetts v. McCann, 12 a 1990 decision of five justices of the High
Court of Australia, a coroner had been conducting an inquest into the
death of a 16-year old boy. The boj^'s parents (Mr and Mrs Annetts)
sought to make a submission before the coroner made a finding. The
coroner decided that the Coroner's Act 1920 (WA) gave him the dis-
cretion (which he chose to exercise) to disallow their submission. The
Annettses appealed.

The instant case is on all fours with Annetts v. McCann.

(1990) 170 CLR 590.

ibid, at 612 per Brennan J.

ibid, at 603 per Mason CJ, Deane and McHugh JJ. at 612 per Brennan J,
at 621 per Toohey J. Note, however, that Brennan and Toohey JJ would
have dismissed the appeal because they believed that the decision of the Full
Court of the Supreme Court of Western Australia (from which the Annettses
appealed) was right on the material before it.

§A.5 Example reports 283

If Minister for Arts, Heritage and Environment v. Peko-Wallsend Ltd
is followed then the decision did not affect the property, right, interest,
status, or legitimate expectation of the applicant.

In Minister for Arts, Heritage and Environment v. Peko-Wallsend
Ltd, 15 a 1987 decision of the Full Court of the Federal Court of Aus-
tralia, Peko-Wallsend held various mining interests in Stage 2 of Kakadu
National Park. Federal Cabinet decided to nominate Stage 2 for inclu-
sion in the World Heritage List, so it became "identified property"
within the meaning of s. 3(2) of the World Heritage Properties Conser-
vation Act 1983 (Cth). This meant that the Governor-General could,
by proclamation, make mining operations unlawful in the area. The
decision did not affect Peko-Wallsend 's mining rights which were pre-
served under s. 8b of the National Parks and Wildlife Conservation Act
1975 (Cth).

Before Cabinet's decision, Peko-Wallsend had lobbied Ministers and
other officials extensively, seeking to preserve its mining interests. After
the decision the companj' commenced proceedings to prevent the Gov-
ernment from taking any further steps to have Stage 2 nominated on
the World Heritage List, claiming that Cabinet was bound by the rules
of natural justice and had failed to give Peko-Wallsend an opportunity
to be heard. Beaumont J (a Federal Court judge) agreed, and held the
Cabinet decision void. 10

The Full Court of the Federal Court disagreed. Bowen CJ decided
that "it would ... be inappropriate for this Court to intervene to
set aside a Cabinet decision involving such complex policy consider-
ations". 17 Both Sheppard and Wilcox J J held that Peko-Wallsend had
had adequate opportunity to put its case to relevant Ministers and offi-
cials before the Cabinet decision, and was not denied natural justice. 18
Further, Wilcox J (with whose reasons the other two judges generally
agreed) held that the Cabinet's decision in this case did not attract the
obligations of natural justice. 19

There are several similarities between the instant case and Minis-
ter v. Peko-Wallsend: the decision did not affect a financial, property
or occupational interest of the applicant; the decision did not affect the
applicant's personal liberty; and the applicant did not have a legitimate
expectation which was affected by the decision.

15 (1987) 15 FCR 274.

Peko-Wallsend Ltd v. Minister for Arts, Heritage and Environment (1986)
13 FCR 19.

17 (1987) 15 FCR 274 at 279.

18 ibid. at 282 per Sheppard J, at 308 per Wilcox J.

19:

ibid, at 308.

284 A pragmatic legal expert system §A.5

However, there is one extremely significant difference between the
instant case and Minister v. Peko-Wallsend. In that case the decision
did not affect the applicant's reputation. Note also that Minister v.
Peko-Wallsend is only a decision of the Full Court of the Federal Court
of Australia and not as good authority as a case decided by five justices
of the High Court of Australia- — like Annetts v. McCann.

Consequently, there is nothing in Minister v. Peko-Wallsend to war-
rant any change in my conclusion.

Hypothetical 1

Consider the instant case changed so that the following is true: the
decision did not affect the applicant's reputation.

If that were so then my opinion would be that — -following Minister for
Arts, Heritage and Environment v. Peko-Wallsend Ltd — the decision
did not affect the property, right, interest, status, or legitimate expect-
ation of the applicant.

Details of Minister v. Peko-Wallsend are summarized above. The hy-
pothetical case is on all fours with Minister v. Peko-Wallsend.

If Bread Manufacturers of New South Wales v. Evans is followed then
the decision affected the property, right, interest, status, or legitimate
expectation of the applicant.

In Bread Manufacturers of New South Wales v. Evans, 20 a 1981 decision
of five justices of the High Court of Australia, the bread manufacturers
claimed that an order made by the Prices Commission was void. The
order affected the classification of bread products and had an incid-
ental effect on the price of hamburger buns. The bread manufacturers
complained that they should have been given the right to put their case
to the Commission.

The Prices Regulation Act 1948 (NSW) provided that a public in-
quiry had to be held before an order could be made setting prices,
except where the Minister consented to dispensing with the inquiry.
The Minister had dispensed with an inquiry before this order was made.
Hence, "[t]he argument that the Commission was bound to disclose to
the [bread manufacturers] the fact that it proposed to make an order
which would have the incidental effect of reducing the price of ham-
burger buns can only succeed if the Commission, although not bound
to hold an inquiry, was bound to observe the rules of natural justice". 21

20 (1981) 180 CLR 404.
21 ibid. at 414 per Gibbs CJ.

§A.5 Example reports 285

The High Court held that there was no denial of natural justice in
relation to the order, because "the reduction of the maximum price in
respect of one item was simply a minor incident in a major revision of
the price framework covering the whole range of bread products. The
effect of that major revision was generally to increase prices. There
was, in our opinion, no obligation on the Commission to give advance
notice of this development or of the possibility of its occurrence."

There are several similarities between the hypothetical case and
Bread Manufacturers v. Evans: the decision did not affect the ap-
plicant's personal liberty; the decision did not affect the applicant's
reputation; and the applicant did not have a legitimate expectation
which was affected by the decision.

However, there is one extremely significant difference between the
hypothetical case and Bread Manufacturers v. Evans. In that case the
decision affected a financial, property or occupational interest of the
applicant.

Despite the fact that Bread Manufacturers v. Evans is a decision of
five justices of the High Court of Australia (and better authority than
a case decided by the Full Court of the Federal Court of Australia — like
Minister v. Peko -Walls end) , there is nothing in Bread Manufacturers v.
Evans to warrant any change in my conclusion.

Expectation area

Instant case

If the applicant had a legitimate expectation which was affected by the
decision, natural justice may be implied. " '[Ljegitimate expectations*
. . . are capable of including expectations which go beyond enforceable
legal rights, provided they have some reasonable basis". 23

In the instant case, the decision-maker did not break a promise or un-
dertaking; the decision-maker did not go against an established course
of practice; the decision did not involve a refusal to renew an existing
interest; neither the decision-maker nor a statutory provision suggested
that an initial interest would be granted; the decision did not affect an
established liberty or interest; and there was no standard administrat-
ive procedure which the decision-maker should have followed.

""ibid, at 435 per Mason and Wilson JJ, with whom Murphy and Aickin JJ
agreed on this point.

23 Cole v. Cunningham (1983) 49 ALR 123 at 131 per Bowen CJ, Sheppard
and Morlins JJ.

286 A pragmatic legal expert system §A.5

In my opinion- — -following Minister for Arts, Heritage and Environ-
ment v. Peko -Walls end Ltd — the applicant did not have a legitimate
expectation which was affected by the decision.

In Minister for Arts, Heritage and Environment v. Peko-Wallsend
Ltd? 4 a 1987 decision of the Full Court of the Federal Court of Aus-
tralia, Peko-Wallsend held various mining interests in Stage 2 of Kakadu
National Park. Federal Cabinet decided to nominate Stage 2 for inclu-
sion in the World Heritage List, so it became "identified property"
within the meaning of s. 3(2) of the World Heritage Properties Conser-
vation Act 1983 (Cth). This meant that the Governor-General could,
by proclamation, make mining operations unlawful in the area. The
decision did not affect Peko-Wallsend's mining rights which were pre-
served under s. 8b of the National Parks and Wildlife Conservation Act
1975 (Cth).

Before Cabinet's decision, Peko-Wallsend had lobbied Ministers and
other officials extensively, seeking to preserve its mining interests. After
the decision the company commenced proceedings to prevent the Gov-
ernment from taking any further steps to have Stage 2 nominated on
the World Heritage List, claiming that Cabinet was bound by the rules
of natural justice and had failed to give Peko-Wallsend an opportunity
to be heard. Beaumont J (a Federal Court judge) agreed, and held the
Cabinet decision void. 25

The Full Court of the Federal Court disagreed. Bowen CJ decided
that "it would ... be inappropriate for this Court to intervene to
set aside a Cabinet decision involving such complex policy consider-
ations". 20 Both Sheppard and Wilcox JJ held that Peko-Wallsend had
had adequate opportunity to put its case to relevant Ministers and offi-
cials before the Cabinet decision, and was not denied natural justice. 27
Further, Wilcox J (with whose reasons the other two judges generally
agreed) held that the Cabinet's decision in this case did not attract the
obligations of natural justice. 28

The instant case is on all fours with Minister v. Peko-Wallsend.

24 (1987) 15 FCR 274.

'^Peko-Wallsend Ltd v. Minister for Arts, Heritage and Environment (1986)
13 FCR 19.

26 (1987) 15 FCR 274 at 279.

27;

28;

ibid, at 282 per Sheppard J, at 308 per Wilcox J.
ibid, at 308.

§A.5 Example reports 287

If Cole v. Cunningham is followed then the applicant had a legitimate
expectation which was affected by the decision.

In Cole v. Cunningham 29 a 1983 decision of the Full Court of the Fed-
eral Court of Australia, Cunningham had been encouraged to resign
from the public service because his superiors believed he had been
guilty of misconduct in the performance of his duties. He had formed
an attachment and begun to live with a Fijian woman whose applica-
tion for an extension of her temporary entry permit he had processed.
He was threatened with criminal prosecution for harbouring an illegal
immigrant, and told that "[i]f you resign now it will be a normal resig-
nation and you'll leave with a clean record." 30 He resigned.

About eighteen months later, Cunningham sought reappointment to
the public service and was told that he would be offered a position
subject to police and security clearances. The next day he was told
that he had been given an unsatisfactory report based on the earlier
events.

Bowen CJ, Sheppard and Morling JJ held that, in general, applic-
ants for appointment or reappointment to the public service are not
entitled to have natural justice principles applied because those applic-
ants have no legitimate expectation which can be affected by a refusal
to appoint. However, Cunningham did have a legitimate expectation
that any decision to reappoint him would not be made on the basis of
his past record.

There are several similarities between the instant case and Cole v.
Cunningham: the decision-maker did not go against an established
course of practice; the decision did not involve a refusal to renew an
existing interest; neither the decision-maker nor a statutory provision
suggested that an initial interest would bo grunto'l: the decision did not
affect an established liberty or interest; and there was no standard ad-
ministrative procedure which the decision-maker should have followed.

However, there is one extremely significant difference between the
instant case and Cole v. Cunningham. In that case the decision-maker
broke a promise or undertaking.

Despite the fact that Cole v. Cunningham and Minister v. Peko-
Wallsend are both decisions of the Full Court of the Federal Court
of Australia, there is nothing in Cole v. Cunningham to warrant any
change in my conclusion.

■>!';

(1983) 49 ALR 123.

ibid, at 125.

288 A pragmatic legal expert system

Hypothetical 1

Consider the instant case changed so that the following is true: the
decision-maker broke a promise or undertaking.

If that were so then my opinion would be that— following Cole v.
Cunningham — the applicant had a legitimate expectation which was
affected by the decision.

Details of Cole v. Cunningham are summarized above. The hypothet-
ical case is on all fours with Cole v. Cunningham.

If Minister for Arts, Heritage and Environment v. Peko-Wallsend Ltd
is followed then the applicant did not have a legitimate expectation
which was affected b}' the decision.

Details of Minister v. Peko-Wallsend are summarized above. There
are several similarities between the hypothetical case and Minister v.
Peko-Wallsend: the decision-maker did not go against an established
course of practice; the decision did not involve a refusal to renew an
existing interest; neither the decision-maker nor a statutorj' provision
suggested that an initial interest would b< i granted; the decision did not
affect an established liberty or interest; and there was no standard ad-
ministrative procedure which the decision-maker should have followed.

However, there is one extremely significant difference between the
hypothetical case and Minister v. Peko-Wallsend. In that case the
decision-maker did not break a promise or undertaking.

Despite the fact that Minister v. Peko-Wallsend and Cole v. Cun-
ningham are both decisions of the Full Court of the Federal Court of
Australia, there is nothing in Minister v. Peko-Wallsend to warrant
any change in my conclusion.

A complete example

... in order to enter the magic circle of employees, the worker has to solve not
only the riddle of "contract" but also that of "employment/' And no sooner
has the hapless worker tried to answer the riddles, but the judicial wizards
have changed the question.

Bob Hepple (1986)
Restructuring Employment Rights

Fury said to a mouse, That he met in the house,
"Let us both go to law: / will prosecute you.
Come, I'll take no denial: We must have the trial;
For really this morning I've nothing to do."
Said the mouse to the cur, "Such a trial, dear sir,
With no jury or judge, would be wasting our breath."
"I'll be judge, I'll be jury," said cunning old Fury:
"I'll try the whole cause, and condemn you to death."

Lewis Carroll (1865)
Alice's Adventures in Wonderland'

289

290 A pragmatic legal expert system §B.l

B.l Introduction

This appendix gives a complete example of SHYSTER'S output for
one of the cases used to test the Employee specification in §4.4.3:
Building Workers' Industrial Union of Australia v. Odco Pty Ltd. 3

The log file (§B.2) summarizes SHYSTER'S operation for this test
case. The only input to SHYSTER is the Employee specification
Employee. els and the test case's fact vector which is entered by
the user, one attribute value at a time, in response to questions
asked by SHYSTER. The fact vector for BWIU v. Odco is given in
the log file.

SHYSTER was requested to hypothesize with a limit of two
changed attribute values, and to report on up to one hypothet-
ical per result.

Those of its output files that SHYSTER writes in IAT E X format
are shown in this appendix as they appear after processing by the
I^TgX document processor: 4 viz. the dump file Dump.tex (§B.3),
the probabilities file Probabilities.tex (§B.4), the weights file
Weights.tex (§B.5). the distances file Distances-Employee.tex
(§B.6) and the report file Report-Employee.tex (§B.7).

Notes

1 Hepple 1986 at 71.

2 Carroll 1982 at 19.

3 (1991) 29 FCR 104.

4 Lamport 1986, 1994. The L^TgX output is complete, and unaltered—
except that (so that all of the output would fit within the margins of this
book) the attribute value matrix near the beginning of §B.3 has been rotated
anti-clockwise by 90 degrees, the probabilities matrix in §B .4 has been reduced
in size, and all seven distance tables in §B.6 have been rotated and reduced.

?.2 A complete example 291

B.2 Log file for BWIU v. Odco

SHYSTER version 1.0

Reading case law specification from "Employee .els" ...

9 courts in the hierarchy.

Employee area:

2 results

18 attributes

14 cases

2 ideal points

Case law specification is valid.

Writing dump to "Dump.tex".

Writing probabilities to "Probabilities.tex".

WARNING (Checker) : evidence of stochastic dependence between A3
and A17 in Employee area.

WARNING (Checker) : functional dependence (inverse) between A4
and A5 in Employee area.

WARNING (Checker) : evidence of stochastic dependence between A4
and All in Employee area.

WARNING (Checker) : evidence of stochastic dependence between A5
and All in Employee area.

WARNING (Checker) : evidence of stochastic dependence between A7
and A9 in Employee area.

WARNING (Checker) : evidence of stochastic dependence between A9
and All in Employee area.

WARNING (Checker): evidence of stochastic dependence between All
and A12 in Employee area.

Writing weights to "Weights .tex" .

Case-based system called with area identifier "Employee" .

Area is Employee.

Writing distances to "Distances-Employee.tex".

Writing report to "Report-Employee.tex".

Fact vector is (NYNYNNNUNNYUNNNNNY) .

Nearest neighbours :

Contractor :

C8 Humberstone v. NTM

Nearest others:

Employee :

C5 Ferguson v. Dawson

Nearest result for the instant case is Contractor.

292 A pragmatic legal expert system §B-2

Instantiation 1 is (NYNYNNNYNNYYNNNNNY) .

A A
Nearest neighbours :

Contractor :

C8 Humberstone v. NTM

Nearest others:

Employee :

C5 Ferguson v. Dawson

Safeguards :

Distance measures :

* C6 Stevenson v. Macdonald (2) (Employee)

+ Cll AMP v. Chaplin

+ C13 Stevenson v. Macdonald (1)

Association coefficients :
+ Cll AMP v. Chaplin

Correlation coefficients :

- C8 Humberstone v. NTM
+ Cll AMP v. Chaplin

Weighted correlation coefficients:

- C8 Humberstone v. NTM
+ Cll AMP v. Chaplin

Nearest result for instantiation 1 is Contractor.

Instantiation 2 is (NYNYNNNYNNYNNNNNNY) .

A A
Nearest neighbours :

Contractor :

C8 Humberstone v. NTM

Nearest others:

Employee :

C2 Cam v. Sargent

Safeguards :

Distance measures :

+ Cll AMP v. Chaplin

Association coefficients :
+ Cll AMP v. Chaplin

Correlation coefficients :

- C8 Humberstone v. NTM
+ Cll AMP v. Chaplin

Weighted correlation coefficients:

- C8 Humberstone v. NTM
+ Cll AMP v. Chaplin

Nearest result for instantiation 2 is Contractor.

?.2 A complete example 293

Instantiation 3 is (NYNYNNNNNNYYNNNNNY) .

A A
Nearest neighbours :

Contractor :

C8 Humberstone v. NTM

Nearest others:

Employee :

C5 Ferguson v. Dawson

Safeguards :

Distance measures :

* C6 Stevenson v. Macdonald (2) (Employee)
+ C13 Stevenson v. Macdonald (l)

Nearest result for instantiation 3 is Contractor.

Instantiation 4 is (NYNYNNNNNNYNNNNNNY) .

A A
Nearest neighbours :

Contractor :

C8 Humberstone v. NTM

Nearest others:

Employee :

C2 Cam v. Sargent

Nearest result for instantiation 4 is Contractor.

Hypothetical 1 is (NYNYNNNUYNYUNNNNNN) .

A A
Nearest neighbours :

Contractor :

C8 Humberstone v. NTM

Nearest others:

Employee :

C2 Cam v. Sargent

Nearest result for hypothetical 1 is Contractor.

Hypothetical 2 is (NNYYNNNUNNYUNNNNNY) .
AA
Nearest neighbours :

Employee :

C5 Ferguson v. Dawson

Nearest others:

Contractor :

Cll AMP v. Chaplin

C14 Ready Mixed v. Minister

Safeguards :

Distance measures :

+ C6 Stevenson v. Macdonald (2)

* Cll AMP v. Chaplin (Contractor)

* C14 Ready Mixed v. Minister (Contractor)

294 A pragmatic legal expert system

Association coefficients :

* Cll AMP v. Chaplin (Contractor)

* C14 Ready Mixed v. Minister (Contractor)
Correlation coefficients :

* Cll AMP v. Chaplin (Contractor)

* C14 Ready Mixed v. Minister (Contractor)

Ideal points :
Contractor

Centroids :

Contractor

Specified directions :
Contractor

Ideal point directions:

Contractor
Centroid directions :

Contractor

WARNING (Reporter) : the specified directions suggest a
different result or results .

Nearest result for hypothetical 2 is Employee.

All 4 instantiations have the same nearest result as does the
instant case.

Reported on 2 hypotheticals of 136 (limit of 2 changes) .

Case-based system returned result identifier "Contractor" .

Finished.

B.3 Dump file for the EMPLOYEE specification

Hierarchy

Court

five justices of the High Court of Australia

four justices of the High Court of Australia

three justices of the High Court of Australia

a single justice of the High Court of Australia

three judges of the Federal Court of Australia

the Judicial Committee of the Privy Council

the English Court of Appeal

the King's Bench Division of the English High Court

the Queen's Bench Division of the English High Court

Employee area

A complete example 295

•*-J

■»-»

>
o

4-»

c
LU

i—l

i— i

•

t-
i— i

•

i— 1

•

i— 1

•

i— 1

•

i— i

•

i— i

•

-O

-t* o

•

^ a

•

i— I

•

■1.

■*-•

<i)

l—l

•o

i- _2

f-H

-1

T m

O |

296 A pragmatic legal expert system §B-3

Opening

The terms "employer" and "worker" are used here to mean "employer**
and "employee" (in the case of a contract of service) or "principal" and
"independent contractor" (in the case of a contract for services).

Results

Employee: the worker is an employee.

Contractor: the worker is an independent contractor.

Attributes

Ai m . Did the employer direct not only what work was to be done, but
also the manner in which it was to be done?

YES: the employer directed the manner in which the work was
to be done.
^* Employee

NO: the employer did not direct the manner in which the work
was to be done.
^* Contractor

UNKNOWN: it is not known whether the employer directed the
manner in which the work was to be done.

If the employer had a right of control over how the worker did
the work then the employer had the power to direct not only
what work was to be done, but also the manner in which it was
to be done.

Ao'. Was the worker allowed to use her/his own discretion in doing an
aspect of the work that was not specified beforehand?

YES: the worker was allowed to use her/his own discretion in
doing an aspect of the work that was not specified before-
hand.
=> Contractor

1.3 A complete example 291

NO: the worker was not allowed to use her/his own discretion in
doing an aspect of the work that was not specified before-
hand.

^* Employee

UNKNOWN: it is not known whether the worker was allowed to
use her/his own discretion in doing an aspect of the work
that was not specified beforehand.

A 3 : Was the worker an integral part of the employer's business?

YES: the worker was an integral part of the employer's business.
^ Employee

NO: the worker was not an integral part of the employer's busi-
ness, but was accessory to it.
^ Contractor

UNKNOWN: it is not known whether the worker was an integral
part of the employer's business or was merely accessory to
it.

If the worker was "part and parcel" of the employer's business
then she/he was an integral part of the business, not merely
accessory to it.

A+: Did the worker own the tools or provide the transport with which
she/he performed the work?

YES: the worker owned the tools or provided the transport with
which she/he performed the work.
^ Contractor

NO: the worker neither owned the tools nor provided the trans-
port with which she/he performed the work.
^ Employee

UNKNOWN: it is not known whether the worker owned the tools
or provided the transport with which she/he performed the
work.

A 5 : Would the employer make a profit/loss if the work performed by
the worker cost less/more than expected?

YES: the emplo3 j 'er would make a profit/loss if the work per-
formed by the worker cost less/more than expected.

NO: the employer would not make a profit/loss if the work per-
formed by the worker cost less/more than expected.

UNKNOWN: it is not known whether the employer would make a
profit/loss if the work performed by the worker cost less/
more than expected.

298 A pragmatic legal expert system §B-$

Aq: Was the work performed on the employer's premises?

YES: the work was performed on the emploj^er's premises.
NO: the work was not performed on the employer's premises.
^* Contractor

UNKNOWN: it is not known whether the work was performed on
the employer's premises.

A-j\ Did the employer supervise or inspect the work?

YES: the employer supervised or inspected the work.

^ Employee
NO: the employer neither supervised nor inspected the work.

^ Contractor

UNKNOWN: it is not known whether the employer supervised or
inspected the work.

.A3: Was the worker in business on her/his own account?

YES: the worker was in business on her/his own account.

=> Contractor
NO: the worker was not in business on her/his own account.

^ Employee

UNKNOWN: it is not known whether the worker was in business
on her/his own account.

Aq: Was the worker allowed to employ others to assist with her/his
work?

YES: the worker was allowed to employ others to assist with her/
his work.
^ Contractor

NO: the worker was not allowed to employ others to assist with
her/his work.
^ Employee

UNKNOWN: it is not known whether the worker was allowed to
employ others to assist with her/his work.

A 10: Was the worker obliged to work only for the employer?

YES: the worker was obliged to work only for the employer.

^ Employee
NO: the worker was not obliged to work only for the employer.

^* Contractor

UNKNOWN: it is not known whether the worker was obliged to
work only for the employer.

§B.3 A complete example 299

An'. Was the worker required to work at specified times?

YES: the worker was required to work at specified times.

^ Employee
NO: the worker was not required to work at specified times.

^ Contractor

UNKNOWN: it is not known whether the worker was required to
work at specified times.

.A 10: Did the employer pay the worker by time?

YES: the employer paid the worker by time.

^ Employee
NO: the employer did not pay the worker by time.

=> Contractor

UNKNOWN: it is not known whether the employer paid the worker
by time.

The employer could pay the worker by time (e.g. by the hour, or
by the week) or by results.

A13'. Was the money that the employer paid to the worker stated to
be a "fee"?

YES: the money that the employer paid to the worker was stated
to be a "fee".

^ Contractor

NO: the monej' that the employer paid to the worker was not
stated to be a "fee".
^ Employee

UNKNOWN: it is not known whether the money that the employer
paid to the worker was stated to be a "fee".

A i4 : Was the money that the employer paid to the worker stated to
be "wages" or "salary"?

YES: the money that the employer paid to the worker was stated
to be "wages" or "salary".
^ Employee

NO: the monej' that the employer paid to the worker was not
stated to be "wages" or "salary".
^* Contractor

UNKNOWN: it is not known whether the money that the employer
paid to the worker was stated to be "wages" or "salary".

300 A pragmatic legal expert system §B-3

Ais'. Did the employer deduct PAYE tax instalments from the worker's
pay?

YES: the employer deducted PAYE tax instalments from the
worker's p&y.
^ Employee

NO: the employer did not deduct PAYE tax instalments from the
worker's pay.

^* Contractor

UNKNOWN: it is not known whether the employer deducted PAYE
tax instalments from the worker's pay.

.A ie: Did the employer pay the worker sick pay or holiday pay?

YES: the employer paid the worker sick pay or holiday pay.
^ Employee

NO: the employer paid the worker neither sick pay nor holiday
pay.
^* Contractor

UNKNOWN: it is not known whether the employer paid the worker
sick pay or holiday pa}'.

A 17 : Did the employer and the worker express an intention that the
relationship would be one of employer and emploj'ee?

YES: the employer and the worker expressed an intention that
the relationship would be one of employer and employee.
^ Employee

NO: the employer and the worker did not express any intention
that the relationship would be one of employer and em-
ployee.

UNKNOWN: it is not known whether the employer and the worker
expressed an intention that the relationship would be one
of employer and employee.

For example, if the employer and the worker characterized their
agreement as being a "contract of service," that would be an
expression of an intention that the relationship would be one of
employer and employee.

A ls : Did the employer and the worker express an intention that the re-
lationship would be one of principal and independent contractor?

YES: the employer and the worker expressed an intention that
the relationship would be one of principal and independent
contractor.
=> Contractor

1.3 A complete example 301

NO: the employer and the worker did not express any intention
that the relationship would be one of principal and inde-
pendent contractor.

UNKNOWN: it is not known whether the employer and the worker
expressed an intention that the relationship would be one
of principal and independent contractor.

For example, if the emplo} r er and the worker characterized their
agreement as being a "contract for services," that would be an
expression of an intention that the relationship would be one of
principal and independent contractor.

Cases in which the worker is an employee

C x : Zuijs v. Wirth Brothers Pty Ltd (1955) 93 CLR 561 ("Zuijs v.
Wirfh")

A\ m . the employer did not direct the manner in which the work
was to be done.

Ao'. the worker was allowed to use her/his own discretion in doing
an aspect of the work that was not specified beforehand.

A 3 : the worker was an integral part of the employer's business.

A^\ the worker neither owned the tools nor provided the trans-
port with which she/he performed the work.

A^: the employer would make a profit/loss if the work performed
by the worker cost less/more than expected.

Aq: the work was performed on the employer's premises.

Aj m . the employer supervised or inspected the work.

As' the worker was not in business on her/his own account.

A$: the worker was not allowed to employ others to assist with
her/his work.

Aiq\ the worker was not obliged to work only for the employer.

An: the worker was required to work at specified times.

A 12: the employer paid the worker by time.

A 13: the money that the employer paid to the worker was not
stated to be a "fee".

A 14: the money that the employer paid to the worker was not
stated to be "wages" or "salary".

A 15: the employer deducted PAYE tax instalments from the
worker's pay.

A i<3 : the employer paid the worker neither sick pa} r nor holiday
pay.

302 A pragmatic legal expert system §B-3

An', the employer and the worker did not express any inten-
tion that the relationship would be one of employer and
employee.

Ais' the employer and the worker did not express any intention
that the relationship would be one of principal and inde-
pendent contractor.

In Zuijs v. Wirth Brothers Pty Ltd, 1 a 1955 decision of five
justices of the High Court of Australia, Zuijs was an acrobat who
fell during a trapeze act at one of Wirth Brothers' circuses. He
sought compensation under the Worker's Compensation Act 1926
(NSW), claiming to be an employee of Wirth Brothers. Wirth
Brothers claimed that, because of the high degree of skill and
personal judgment that he had to exercise in his work, Zuijs was
an independent contractor and therefore not entitled to compens-
ation.

The High Court unanimously agreed with Zuijs. "Even if [one of
the circus managers] could not interfere in the actual technique
of the acrobats and in the character of the act, no reason appears
why [Zuijs] should not be subject to his directions in all other
respects . . . There are countless examples of highly specialized
functions in modern life that must as a matter of practical ne-
cessity and sometimes even as a matter of law be performed on
the responsibility of persons who possess particular knowledge
and skill and who are accordingly qualified. But those engaged
to perform the functions may nevertheless work under a contract
of service." 2

C 2 : Cam and Sons Pty Ltd v. Sargent (1940) 14 ALJ 162 ("Cam v.
Sargent")

Ax", the employer directed the manner in which the work was to
be done.

A 2 '. the worker was allowed to use her/his own discretion in doing
an aspect of the work that was not specified beforehand.

^4 3 : the worker was an integral part of the employer's business.

A 4 : the worker neither owned the tools nor provided the trans-
port with which she/he performed the work.

A$: the employer would make a profit/loss if the work performed
by the worker cost less/more than expected.

A 6 : the work was not performed on the employer's premises.

] (1955) 93 CLR 561.
ibid, at 571-2 per Dixon CJ, Williams, Webb and Taylor JJ.

1.3 A complete example 303

A 7
A 8

A 9

the employer neither supervised nor inspected the work.

the worker was not in business on her/his own account.

the worker was allowed to employ others to assist with her/
his work.

a 11

A\2

A 13

the worker was not obliged to work only for the employer.

the worker was not required to work at specified times.

the employer did not pay the worker by time.

the money that the employer paid to the worker was not

stated to be a "fee".
A i^: the money that the employer paid to the worker was not

stated to be "wages" or "salary".
A in: the employer did not deduct PAYE tax instalments from

the worker's pay.
.Aig: the employer paid the worker neither sick pay nor holiday

pay.
An: the employer and the worker did not express any inten-
tion that the relationship would be one of employer and

employee.

A ls : the employer and the worker did not express any intention
that the relationship would be one of principal and inde-
pendent contractor.

In Cam and Sons Pty Ltd v. Sargent? a 1940 decision of four
justices of the High Court of Australia, Sargent was the master
of a ship. He entered into an agreement with Cam and Sons that
claimed that the ship was hired by Cam and Sons to Sargent
and his fellow contractors (called "the partnership*'). However,
it was doubtful whether that agreement actually deprived Cam
and Sons of any control over the ship. The partnership was to use
the ship only to carr}' coal from Swansea to Sydney. Cam and
Sons were sole agents of the partnership for securing cargoes for
the ship, and for collecting monej' due to the partnership. The
partnership paid nothing for the "hire" of the ship, but received
a specified sum for each return trip of a certain tonnage plus (in
certain circumstances) 5% of the earnings, the balance of which
was retained by Cam and Sons. Cam and Sons had to approve
people employed by the partnership.

Sargent claimed that he (and others in the partnership) were
employed by Cam and Sons, and therefore came within the terms
of an industrial award. Cam and Sons claimed that members of
the partnership were independent contractors.

J (1940) 14 ALJ 162.

304 A pragmatic legal expert system

The High Court unanimously agreed with Sargent. Rich J came
to the conclusion that the agreement was an attempt to evade
the terms of the industrial award. 4

C3: Federal Commissioner of Taxation v. J. Walter Thompson (Aus-
tralia) Pty Ltd (1944) 69 CLR 227 ("FCT v. Thompson")

A\ m . the employer directed the manner in which the work was to
be done.

A 2 '. the worker was not allowed to use her/his own discretion in
doing an aspect of the work that was not specified before-
hand.

A3: the worker was an integral part of the employer's business.
A 4 : the worker neither owned the tools nor provided the trans-
port with which she/he performed the work.

A$: the employer would make a profit/loss if the work performed
by the worker cost less/more than expected.

A 6 : the work was performed on the employer's premises.

Aj: the employer supervised or inspected the work.

A s : it is not known whether the worker was in business on her/
his own account.

Aq: the worker was not allowed to employ others to assist with

her/his work.
.A 10 : the worker was not obliged to work only for the employer.
An: the worker was required to work at specified times.
A X2 '- the employer did not pay the worker by time.

Ai 3 : the money that the employer paid to the worker was stated

to be a "fee".
-A 14 : the money that the employer paid to the worker was not

stated to be "wages" or "salary".

A15: the employer did not deduct PAYE tax instalments from

the worker's pay.
A 16 : the employer paid the worker neither sick pa} r nor holiday

pay.

A 17: the employer and the worker did not express any inten-
tion that the relationship would be one of employer and
employee.

A is', the employer and the worker did not express any intention
that the relationship would be one of principal and inde-
pendent contractor.

4 ibid. at 103.

1.3 A complete example 305

In Federal Commissioner of Taxation v. J. Walter Thompson
(Australia) Pty Ltd, a 1944 decision of a single justice of the
High Court of Australia, the FCT claimed that payments made
to radio artists by Thompson were "wages" within the meaning of
the Pay-roll Tax Assessment Act 1941 (Cth) and therefore tax-
able. The artists were selected by a producer and paid to appear
in radio plaj's. The}'' were paid a "fee" for each performance,
but were paid nothing for attending (compulsor}' - ) rehearsals.
Thompson claimed that the artists were presumed to know their
work and "to render services in the same manner as a profes-
sional man, such as a surgeon or an architect, not being subject
... to detailed control as to the manner in which those services
are to be performed." Hence, Thompson claimed, they were
independent contractors.

Latham CJ held that the radio actors were employed "to co-
operate with others in a team under the control of the producer
to bring about a result, the details of which must in great meas-
ure be determined by the producer." Hence the artists were
employed by Thompson; the fee they were paid was subject to
payroll tax.

C4: Australian Timber Workers Union v. Monaro Sawmills Pty Ltd
(1980) 29 ALR322 ("ATWU v. Monaro")

Ax", the employer directed the manner in which the work was to
be done.

A2'. the worker was not allowed to use her/his own discretion in
doing an aspect of the work that was not specified before-
hand.

.A3: the worker was an integral part of the employer's business.

A4: the worker owned the tools or provided the transport with
which she/he performed the work.

A$: the employer would not make a profit/loss if the work per-
formed by the worker cost less/more than expected.

A 6 : the work was performed on the employer's premises.

A T : the employer supervised or inspected the work.

A s : the worker was not in business on her/his own account.

A Q : it is not known whether the worker was allowed to employ
others to assist with her/his work.

(1944) 69 CLR 227
5 ibid. at 231.
ibid, at 232.

306 A pragmatic legal expert system §B-$

Aiq: the worker was not obliged to work only for the employer.

An: the worker was not required to work at specified times.

A 12'. the employer did not pay the worker by time.

^4 13 : the money that the employer paid to the worker was not
stated to be a "fee".

Ai^: the money that the employer paid to the worker was not
stated to be "wages" or "salary".

A 15: the employer did not deduct PAYE tax instalments from
the worker's pay.

A 10: the employer paid the worker neither sick pa}' nor holiday
pay.

.A 17: the employer and the worker did not express any inten-
tion that the relationship would be one of employer and
employee.

.A is: the employer and the worker did not express any intention
that the relationship would be one of principal and inde-
pendent contractor.

In Australian Timber Workers Union v. Monaro Sawmills Pty
Ltd* a 1980 decision of three judges of the Federal Court of
Australia, Wales was a tree feller who cut timber exclusively for
Monaro Sawmills. He performed his work in an area allotted to
him by Monaro Sawmills. He, and other fellers, were paid by
the amount of millable wood the)'' cut. Wales provided his own
tools and transport, but was (with the other fellers) covered by
Monaro Sawmill's workers' compensation policy.

The union sought an order that a penalty be imposed on Mon-
aro Sawmills for breaching the Timber Industries Consolidated
Award 1974 by failing to pay Wales money in lieu of annual
leave. Monaro Sawmills claimed that Wales was an independent
contractor, and so was not subject to the award.

Sweeney and Evatt JJ examined the circumstances of Wales's
employment and held that those circumstances clearly pointed
to the existence of a relationship of employer and employee. They
could not see "any sense in which it could be said that Wales was
conducting some sort of business of his own." 9

*(1980) 29 ALR 322.
ibid, at 329.

1.3

A complete example 301

C5: Ferguson v. John Dawson and Partners (Contractors) Ltd [1976]
1 WLR 1213 ("Ferguson v. Dawson")

A\: the employer directed the manner in which the work was to
be done.

the worker was not allowed to use her/his own discretion in
doing an aspect of the work that was not specified before-
hand.

the worker was an integral part of the employer's business.
the worker neither owned the tools nor provided the trans-
port with which she/he performed the work.

A$: the employer would make a profit/loss if the work performed
by the worker cost less/more than expected.

} : the work was not performed on the employer's premises.

fi the employer supervised or inspected the work.

i m . the worker was not in business on her/his own account.

>: the worker was not allowed to employ others to assist with
her/his work.

>: the worker was not obliged to work only for the employer.
t : the worker was required to work at specified times.
>: the employer paid the worker by time.

3 : the money that the employer paid to the worker was not
stated to be a "fee".

the money that the employer paid to the worker was not
stated to be "wages" or "salary".

the employer did not deduct PAYE tax instalments from
the worker's pay.

the employer paid the worker neither sick pay nor holiday
pay.

the employer and the worker did not express any inten-
tion that the relationship would be one of employer and
employee.

the employer and the worker expressed an intention that
the relationship would be one of principal and independent
contractor.

A 3 :
A 4 :

A 6
A 7 :
As
A 9 :

Aio:
A u :
Ai 2 :
A l3 .

11-

1." -

Ai 6 :
A l7 :

La-

in Ferguson v. John Dawson and Partners (Contractors) Ltd, 10
a 1976 decision of the English Court of Appeal, Ferguson fell
off a roof while removing some scaffolding boards. He claimed
damages against Dawson (the building contractors) for breach

[1976] 1 WLR 1213.

308 A pragmatic legal expert system §B-3

of statutory duty relying on the Construction (Working Places)
Regulations 1966 (UK). This duty would only be owed if Fer-
guson was an employee of Dawson.

Megaw and Browne LJJ held that, despite the fact that both
parties labelled Ferguson a "self-emploj'ed labour only subcon-
tractor", the reality of the relationship between them was that of
employer and employee. 11

Cq: Stevenson Jordan and Harrison Ltd v. Macdonald and Evans (2)
[1952] 1 TLR 101 ("Stevenson v. Macdonald (2)")

A\: the employer did not direct the manner in which the work
was to be done.

Ao'. the worker was allowed to use her/his own discretion in doing
an aspect of the work that was not specified beforehand.

.A3: the worker was an integral part of the employer's business.

A 4 : the worker neither owned the tools nor provided the trans-
port with which she/he performed the work.

A$: the employer would make a profit/loss if the work performed
by the worker cost less/more than expected.

Aq: the work was not performed on the employer's premises.

Aj: it is not known whether the employer supervised or inspected
the work.

As', the worker was not in business on her/his own account.

A 9 : the worker was not allowed to employ others to assist with
her/his work.

the worker was not obliged to work only for the employer.

the worker was required to work at specified times.

A 10
A12

it is not known whether the employer paid the worker by
time.

A 13 : it is not known whether the money that the employer paid
to the worker was stated to be a "fee".

A l4 : it is not known whether the money that the employer paid
to the worker was stated to be "wages" or "salary"

A in: it is not known whether the employer deducted PAYE tax
instalments from the worker's pay.

11:

ibid, at 1219 per Megaw LJ, at 1228-9 per Browne LJ.

1.3 A complete example 309

Aiq: it is not known whether the employer paid the worker sick
pay or holiday pay.

An: it is not known whether the employer and the worker ex-
pressed an intention that the relationship would be one of

employer and employee.

A is', the employer and the worker did not express any intention
that the relationship would be one of principal and inde-
pendent contractor.

In Stevenson Jordan and Harrison Ltd v. Macdonald and Evans
(2), 12 a 1951 decision of the English Court of Appeal, Evans-
Hemming was an accountant who had been employed (first as a
servant, then as an executive officer) by Macdonald and Evans.
Shortly after he left them, he wrote a textbook on business man-
agement and submitted the manuscript to Stevenson Jordan and
Harrison (a firm of publishers). He died before the book was pub-
lished. Macdonald and Evans claimed that the book was written
while Evans-Hemming was their employee, and so they owned
the copyright in the work under s. 5(l)(b) of the Copyright Act
1911 (UK).

The book was divided into five sections. The second section was
written in its final form while Evans-Hemming was employed by
Macdonald and Evans. The Court of Appeal held that he wrote
the second section as an employee, and hence the copyright in
the second section was in Macdonald and Evans.

C 7 : Performing Right Society Ltd v. Mitchell and Booker (Palais de
Danse) Ltd [1924] 1 KB 762 ("PRS v. Palais de Danse")

A\: the employer directed the manner in which the work was to
be done.

Ao'. the worker was allowed to use her/his own discretion in doing
an aspect of the work that was not specified beforehand.

-A 3 : the worker was an integral part of the employer's business.

^4: the worker owned the tools or provided the transport with
which she/he performed the work.

A^\ the employer would not make a profit/loss if the work per-
formed by the worker cost less/more than expected.

Aq\ the work was performed on the employer's premises.

A 7 : the employer supervised or inspected the work.

A 8 : the worker was not in business on her/his own account.

[1952] 1 TLR 101.

310 A pragmatic legal expert system §B-3

Aq: the worker was not allowed to employ others to assist with
her/his work.

the worker was obliged to work only for the employer.

the worker was required to work at specified times.

the employer paid the worker by time.

■4 10
An

A 12

A 13

the money that the employer paid to the worker was not
stated to be a "fee".

A i^: the money that the employer paid to the worker was stated
to be "wages" or "salary".

^4 15 : the employer did not deduct PAYE tax instalments from
the worker's pay.

Axe', the employer paid the worker neither sick pa}' nor holiday
pay.

A 17: the employer and the worker did not express any inten-
tion that the relationship would be one of employer and
employee.

A ls : the employer and the worker did not express any intention
that the relationship would be one of principal and inde-
pendent contractor.

In Performing Right Society Ltd v. Mitchell and Booker (Palais
de Danse) Ltd, 13 a 1924 decision of the King's Bench Division
of the English High Court, the defendant was the occupier of a
dance hall. It engaged a band to provide music in the hall. The
agreement provided that the band should not infringe copyright,
and that the band would be liable for damages and costs caused
by any such infringement. There was also a notice displayed in
the hall stating that "[o]nly such music as may be played without
fee or licence is allowed to be plaj'ed in this Hall." 14

The band performed several pieces of music, the copyright in
which was held by the Performing Right Society, without its
permission. The defendant did not know, and had no reasonable
grounds for suspecting, that the infringement was to take place.

The PRS abandoned its earlier claim that the defendant had
"permitted" the infringement under s. 2(3) of the Copyright Act
1911 (UK). However, it claimed that the band members were
the defendant's emploj^ees, and so the defendant was vicariously
liable for the infringement.

13
14-

[1924] 1 KB 762.
ibid, at 764.

1.3 A complete example 311

McCardie J examined the agreement and found that it gave to
the defendant "the right of continuous, dominant, and detailed
control on every point, including the nature of the music to be
played". 15 Hence the band members were employees of the de-
fendant, which was liable for the infringement.

-^Employee (the ideal case in which the worker is an employee):

A x : the employer directed the manner in which the work was to
be done.

A2'. the worker was not allowed to use her/his own discretion in
doing an aspect of the work that was not specified before-
hand.

A 3 : the worker was an integral part of the employer's business.

A4: the worker neither owned the tools nor provided the trans-
port with which she/he performed the work.

A 5 : the employer would make a profit/loss if the work performed
by the worker cost less/more than expected.

Aq: the work was performed on the employer's premises.

Aj: the employer supervised or inspected the work.

As', the worker was not in business on her/his own account.

Ag: the worker was not allowed to employ others to assist with
her/his work.

the worker was obliged to work only for the employer.

the worker was required to work at specified times.

the employer paid the worker by time.

A10
A u

A\2

A 13

the money that the employer paid to the worker was not
stated to be a "fee".

A l4 : the money that the employer paid to the worker was stated
to be "wages" or "salary".

.A 15: the employer deducted PAYE tax instalments from the
worker's pay.

Ai 6 : the employer paid the worker sick pay or holiday pay.

A 17: the employer and the worker expressed an intention that
the relationship would be one of employer and employee.

Ais'. it is not known whether the employer and the worker ex-
pressed an intention that the relationship would be one of

principal and independent contractor.

15 ibid. at 771.

312 A pragmatic legal expert system §B-3

Cases in which the worker is an independent contractor

C 8 : Humberstone v. Northern Timber Mills (1949) 79 CLR 389 ("Hum-
berstone v. NTM")

Ax", the employer did not direct the manner in which the work
was to be done.

Ao'. the worker was allowed to use her/his own discretion in doing
an aspect of the work that was not specified beforehand.

.A3: the worker was not an integral part of the emploj'er's busi-
ness, but was accessory to it.

A 4 : the worker owned the tools or provided the transport with
which she/he performed the work.

A$: the employer would not make a profit/loss if the work per-
formed by the worker cost less/more than expected.

A 6
A 7
A 8
A 9

the work was not performed on the employer's premises.
the employer neither supervised nor inspected the work.
the worker was not in business on her/his own account,
the worker was allowed to employ others to assist with her/
his work.

A10
An

A\2

A 13

the worker was not obliged to work only for the employer.

the worker was not required to work at specified times.

the employer did not pay the worker by time.

the money that the employer paid to the worker was not

stated to be a "fee".
A i 4 : the money that the employer paid to the worker was not

stated to be "wages" or "salary".
-A 10 : the employer did not deduct PAYE tax instalments from

the worker's pay.
A 10: the employer paid the worker neither sick pay nor holiday

pay.
A 17: the employer and the worker did not express any inten-
tion that the relationship would be one of employer and

employee.

A ls : the employer and the worker did not express any intention
that the relationship would be one of principal and inde-
pendent contractor.

In Humberstone v. Northern Timber Mills, 16 a 1949 decision of
three justices of the High Court of Australia, Humberstone car-
ried goods for NTM. He had originally held himself out as a

(1949) 79 CLR 389.

1.3 A complete example 313

carrier, prepared to carry for arryone, but for over twenty years
he had carried goods solely for NTM (although he would, infre-
quently, carry back-loads for NTM's customers). Humberstone
owned the truck, and paid for petrol and repairs. He was paid
weekly on a weight-mileage basis. He was a licenced carrier, and
had his name printed on the side of his truck with the description
"carrier."

On the way back from a job, he had a puncture. He went home
to change the wheel, but exerted himself so strenuously in trying
to remove the tyre from the wheel that he became ill and later
lapsed into a coma, from which he did not recover. Section 3 of
the Worker's Compensation Act 1928 (Vic) had been amended
about a year before Humberstone's death so as to include inde-
pendent contractors in its definition of a "worker" covered bj' the
Act. However, the High Court held that the amendment applied
only to contracts entered into after it came into operation. Fur-
ther, the Court decided that Humberstone was not an employee
of NTM. Hence, he was not a "worker" under the Act, and his
widow was not entitled to compensation under the Act.

Cg: Queensland Stations Pty Ltd v. Federal Commissioner of Taxation
(1945) 70 CLR 539 ("Queensland Stations v. FCT")

A\: the employer did not direct the manner in which the work
was to be done.

Ao'. the worker was allowed to use her/his own discretion in doing
an aspect of the work that was not specified beforehand.

A 3 : the worker was not an integral part of the employer's busi-
ness, but was accessory to it.

A^\ the worker owned the tools or provided the transport with
which she/he performed the work.

^5: the employer would not make a profit/loss if the work per-
formed by the worker cost less/more than expected.

A 6
A 7
As
A g

Am
A u
A 12
A 13

the work was not performed on the employer's premises.

the employer supervised or inspected the work.

the worker was in business on her/his own account.

the worker was allowed to employ others to assist with her/
his work.

the worker was not obliged to work only for the employer.

the worker was not required to work at specified times.

the employer did not pay the worker by time.

the money that the employer paid to the worker was not
stated to be a "fee".

314 A pragmatic legal expert system §B-3

A\^: the money that the employer paid to the worker was not
stated to be "wages" or "salary".

A 15: the employer did not deduct PAYE tax instalments from
the worker's pay.

A is : the employer paid the worker neither sick pay nor holiday
pay.

An: the employer and the worker expressed an intention that
the relationship would be one of employer and employee.

A ls : the employer and the worker did not express any intention
that the relationship would be one of principal and inde-
pendent contractor.

In Queensland Stations Pty Ltd v. Federal Commissioner of
Taxation, 11 a 1945 decision of three justices of the High Court
of Australia, agreements were entered into between Queensland
Stations and some drovers. The agreements stated that the
drovers would "serve** Queensland Stations and take charge of a
specified number of cattle, and deliver them to a specified place.
The drovers were paid a specified rate per head of cattle success-
fully delivered. Each drover was responsible for hiring help, and
paying for feed for the cattle. The drovers were to "obej' and
carry out all lawful instructions and to use the whole of [their]
time, energy and ability in the careful droving of the stock." 18
The FCT claimed that payments made to drovers were "wages'*
within the meaning of the Pay-roll Tax Assessment Act 1941
(Cth), and that Queensland Stations was liable to payroll tax.

The High Court held that the drovers were independent con-
tractors, so the payments were not "wages." Rich J pointed out
that drovers were traditionally free from the control of owners of
cattle. "The obligation imposed on the drover to obej' and carry
out all lawful instructions is not a reservation of detailed control
and possession having regard to the terms of the agreement as a
whole." 19

Ciq: Price v. Grant Industries Pty Ltd (1978) 21 ALR 388 {"Price v.
Grant")

A\: the employer did not direct the manner in which the work
was to be done.

Ao'. the worker was allowed to use her/his own discretion in doing
an aspect of the work that was not specified beforehand.

17 (1945) 70 CLR 539.
18 ibid. at 540.
19 ibid. at 549.

1.3

A complete example 315

^3: the worker was an integral part of the employer's business.

A 4 : the worker owned the tools or provided the transport with
which she/he performed the work.

Ao'. the employer would not make a profit/loss if the work per-
formed by the worker cost less/more than expected.

the work was not performed on the employer's premises.

the employer neither supervised nor inspected the work.

the worker was not in business on her/his own account.

the worker was allowed to employ others to assist with her/
his work.

: the worker was not obliged to work only for the employer.

: the worker was not required to work at specified times.

: the employer did not pay the worker by time.

: the money that the employer paid to the worker was not
stated to be a "fee".

the money that the employer paid to the worker was not
stated to be "wages" or "salary".

the employer deducted PAYE tax instalments from the
worker's pay.

the employer paid the worker neither sick pay nor holiday
pay.

the employer and the worker did not express any inten-
tion that the relationship would be one of employer and
employee.

the employer and the worker did not express any intention
that the relationship would be one of principal and inde-
pendent contractor.

In Price v. Grant Industries Pty Ltd, 20 a 1978 decision of three
judges of the Federal Court of Australia, Grant Industries manu-
factured and sold wardrobes, which Price (and others) delivered
and installed. Price and each of the other "contractors" (as Grant
Industries called them) had to provide and maintain a suitable
truck to deliver the wardrobes, and provide the tools required to
install them. Price sought an order that a penalty be imposed
on Grant Industries for breaching the Furnishing Trades (Con-
solidated) Award 1975 b}' not paying him the appropriate rate of
wages, and not giving him annual leave. The award only applied
to "employees" of specified employers.

A 6
A 7
A 8
A 9

Aid:
An:
A 12 :
A 13 :

A 14 :
A 15 :

IR-

IT-

IS-

(1978) 21 ALR 388.

316 A pragmatic legal expert system §B-3

The Federal Court examined the facts, and the provisions of the
agreement, and held that Price was an independent contractor
and, therefore, not subject to the award.

C\\: Australian Mutual Provident Society v. Chaplin (1978) 18 ALR
385 ( U AMP v. Chaplin")

A\\ the employer did not direct the manner in which the work
was to be done.

A 2 '. the worker was allowed to use her/his own discretion in doing
an aspect of the work that was not specified beforehand.

A 3 : the worker was an integral part of the employer's business.

A 4 : the worker owned the tools or provided the transport with
which she/he performed the work.

A%\ the employer would not make a profit/loss if the work per-
formed by the worker cost less/more than expected.

A 6 : the work was not performed on the employer's premises.

A 7 : the employer neither supervised nor inspected the work.

A 8 : the worker was in business on her/his own account.

Aq: the worker was allowed to employ others to assist with her/
his work.

A l0 ; the worker was obliged to work only for the employer.

An: the worker was not required to work at specified times.

A i2'. the employer did not pay the worker by time.

.A 13: the money that the employer paid to the worker was not

stated to be a "fee".
A 14: the money that the employer paid to the worker was not

stated to be "wages" or "salary".
A in: the employer did not deduct PAYE tax instalments from

the worker's pay.
^4ig: the employer paid the worker neither sick pa}' nor holiday

pay.
An: the employer and the worker did not express any inten-
tion that the relationship would be one of employer and

employee.
A ls : the employer and the worker expressed an intention that

the relationship would be one of principal and independent

contractor.

In Australian Mutual Provident Society v. Chaplin, 21 a 1978 de-
cision of the Judicial Committee of the Privy Council, Chap-
lin was a representative of AMP. A clause of the agreement

21 (1978) 18 ALR 385.

1.3 A complete example 311

between them stated that the relationship was one of "prin-
cipal and agent" and not one of "master and servant." Chaplin
claimed that he was employed under a contract of service, and
was therefore a "worker" under the Long Service Leave Act, 1967
(SA) and entitled to certain benefits.

The Privy Council found that there was no reason to think that
the clause was not a genuine statement of the parties* inten-
tions. Examining the agreement, their Lordships concluded that
it provided for a contract of agency. The fact that Chaplin was
given the power of unlimited delegation of the whole performance
of his work was "almost conclusive against the contract being a
contract of service." 22

C l2 : Massey v. Crown Life Insurance Co. [1978] 1 WLR 676 ("Mas-
sey v. Crown Li/e")

A\\ the employer directed the manner in which the work was to
be done.

A2'. the worker was allowed to use her/his own discretion in doing
an aspect of the work that was not specified beforehand.

.A3: the worker was an integral part of the employer's business.
A 4 : the worker neither owned the tools nor provided the trans-
port with which she/he performed the work.

A$: the employer would make a profit/loss if the work performed
by the worker cost less/more than expected.

Aq: the work was performed on the employer's premises.
A?: the employer neither supervised nor inspected the work.
As', the worker was in business on her/his own account.
A Q : the worker was allowed to employ others to assist with her/
his work.

A 10: the worker was not obliged to work only for the employer.
An: the worker was required to work at specified times.
A12'. the employer paid the worker by time.

A13: the money that the employer paid to the worker was not

stated to be a "fee".
-A 14 : the money that the employer paid to the worker was not

stated to be "wages" or "salary".

A in: the employer did not deduct PAYE tax instalments from
the worker's pay.

Ai$: the employer paid the worker sick pay or holida}' pay.

318 A pragmatic legal expert system §B-3

An', the employer and the worker did not express any inten-
tion that the relationship would be one of employer and
employee.

A is', the employer and the worker expressed an intention that
the relationship would be one of principal and independent
contractor.

In Massey v. Crown Life Insurance Co., 23 a 1977 decision of the
English Court of Appeal, Massey was the manager of a branch
of Crown Life. He had been an employee for two years, then he
and Crown Life entered into a new agreement whereby Massey
continued to perform the same duties as before, but was self-
employed. This arrangement had tax advantages for Massey.
After a further two years, Crown Life terminated the agreement
and Massey sought compensation for unfair dismissal under the
Trade Union and Labour Relations Act 1974 (UK). Compensa-
tion was only payable if Massey was employed under a contract
of service.

Lord Denning MR stated that "if the true relationship of the
parties is that of master and servant under a contract of service,
the parties cannot alter the truth of that relationship by putting
a different label upon it." 24 However, Lord Denning (and the rest
of the Court of Appeal) held that the agreement was genuinely
intended to establish Massey as being self-employed; he was an
independent contractor.

C13: Stevenson Jordan and Harrison Ltd v. Macdonald and Evans (1)
[1952] 1 TLR 101 {"Stevenson v. Macdonald (1)")

A\\ the employer did not direct the manner in which the work
was to be done.

Ao'. the worker was allowed to use her/his own discretion in doing
an aspect of the work that was not specified beforehand.

^4 3 : the worker was not an integral part of the employer's busi-
ness, but was accessory to it.

A4: the worker neither owned the tools nor provided the trans-
port with which she/he performed the work.

A$\ the employer would make a profit/loss if the work performed
by the worker cost less/more than expected.

^4 : the work was not performed on the employer's premises.

Aj: it is not known whether the employer supervised or inspected
the work.

23 [1978] 1 WLR676.
24 ibicl. at 679.

1.3 A complete example 319

As', the worker was not in business on her/his own account.
Aq: the worker was not allowed to employ others to assist with

her/his work.

the worker was not obliged to work only for the employer.

the worker was required to work at specified times.

it is not known whether the employer paid the worker by

time.

■^10
A n
A 12

A 13 : it is not known whether the money that the employer paid

to the worker was stated to be a "fee".
A l4 : it is not known whether the money that the employer paid

to the worker was stated to be "wages" or "salary".

A 15: it is not known whether the employer deducted PAYE tax
instalments from the worker's pay.

Aiq: it is not known whether the employer paid the worker sick
pay or holiday pay.

An: the employer and the worker expressed an intention that
the relationship would be one of employer and employee.

A 18 : the employer and the worker did not express any intention
that the relationship would be one of principal and inde-
pendent contractor.

In Stevenson Jordan and Harrison Ltd v. Macdonald and Evans
(1), a 1951 decision of the English Court of Appeal, Evans-
Hemming was an accountant who had been employed (first as a
servant, then as an executive officer) by Macdonald and Evans.
Shortly after he left them, he wrote a textbook on business man-
agement and submitted the manuscript to Stevenson Jordan and
Harrison (a firm of publishers). He died before the book was pub-
lished. Macdonald and Evans claimed that the book was written
while Evans-Hemming was their employee, and so they owned
the copyright in the work under s. 5(l)(b) of the Copyright Act
1911 (UK).

The book was divided into five sections. The first section con-
sisted of the text of three public lectures that Evans-Hemming
had given while employed by Macdonald and Evans. The Court
of Appeal held that he had given these lectures as an independ-
ent contractor. As Denning LJ said, "under a contract of service,
a man is employed as part of the business, and his work is done
as an integral part of the business; whereas, under a contract for
services, his work, although done for the business, is not integ-
rated into it but is onty accessory to it . . . The lectures were, in

[1952] 1 TLR 101.

320 A pragmatic legal expert system §B-3

a sense, part of the services rendered by Mr Evans-Hemming for
the benefit of the company. But they were in no sense part of
his service. It follows that the copyright in the lectures was in
Mr Evans-Hemming." 26

C14: Ready Mixed Concrete (South East) Ltd v. Minister of Pensions
and National Insurance [1968] 2 QB 497 ("Ready Mixed v. Min-
ister")

A\: the employer did not direct the manner in which the work
was to be done.

Ao'. the worker was allowed to use her/his own discretion in doing
an aspect of the work that was not specified beforehand.

.A3: the worker was an integral part of the employer's business.

A4: the worker owned the tools or provided the transport with
which she/he performed the work.

At>: the employer would not make a profit/loss if the work per-
formed by the worker cost less/more than expected.

Aq: the work was not performed on the employer's premises.

Aj: the employer neither supervised nor inspected the work.

A s : the worker was not in business on her/his own account.

Aq: the worker was allowed to employ others to assist with her/
his work.

.A 10: the worker was obliged to work only for the employer.

An: the worker was not required to work at specified times.

A V2 : the employer did not pay the worker by time.

A 13: the money that the employer paid to the worker was not

stated to be a "fee".

A i 4 : the money that the employer paid to the worker was not

stated to be "wages" or "salary".
A i5 : the employer did not deduct PAYE tax instalments from

the worker's pay.

Ai$: the employer paid the worker neither sick pa}' nor holiday
pay.

A 17: the employer and the worker did not express any inten-
tion that the relationship would be one of employer and
employee.

A iS : the employer and the worker expressed an intention that
the relationship would be one of principal and independent
contractor.

26;

ibid, at 111.

1.3 A complete example 321

In Ready Mixed Concrete (South East) Ltd v. Minister of Pen-
sions and National Insurance, a 1967 decision of the Queen 's
Bench Division of the English High Court, Latimer worked for
Read\ f Mixed as an "owner-driver." He was paid at mileage rates,
and was obliged to buy the truck through a financial organiza-
tion associated with Ready Mixed. The truck was painted in the
company's colours, and he had to wear a Read}' Mixed uniform.
Latimer was obliged to meet the costs of maintenance, repair and
insurance of the truck (and the attached mixing unit, which be-
longed to Ready Mixed). The Minister determined that Latimer
was employed under a contract of service, and was therefore an
"employed person" under s. 1(2) of the National Insurance Act
1965 (UK), making Ready Mixed liable to make weekly contri-
butions.

MacKenna J examined the contract and held that the rights it
conferred, and the duties it imposed, between Latimer and Ready
Mixed were not such as to make it a contract of service.

-^Contractor (the ideal case in which the worker is an independent con-
tractor):

Ax', the employer did not direct the manner in which the work
was to be done.

A 2 '. the worker was allowed to use her/his own discretion in doing
an aspect of the work that was not specified beforehand.

A 3 : the worker was not an integral part of the emploj'er's busi-
ness, but was accessory to it.

A 4 : the worker owned the tools or provided the transport with
which she/he performed the work.

A$: the employer would not make a profit/loss if the work per-
formed by the worker cost less/more than expected.

A 6 : it is not known whether the work was performed on the
employer's premises.

A 9

the employer neither supervised nor inspected the work.

the worker was in business on her/his own account.

the worker was allowed to employ others to assist with her/
his work.

.Aio: the worker was not obliged to work only for the employer.

An: the worker was not required to work at specified times.

A lo: the employer did not pay the worker by time.

[1968] 2 QB 497.

322 A pragmatic legal expert system

A\^: the money that the employer paid to the worker was stated
to be a "fee".

A\±: the money that the employer paid to the worker was not
stated to be "wages" or "salary".

A 15: the employer did not deduct PAYE tax instalments from
the worker's pay.

A 10: the employer paid the worker neither sick pa}' nor holiday
pay.

.A 17: it is not known whether the employer and the worker ex-
pressed an intention that the relationship would be one of
employer and employee.

A ls : the employer and the worker expressed an intention that
the relationship would be one of principal and independent
contractor.

B.4 Probabilities file for the EMPLOYEE specification

Employee area

A 2

A 3

■1 .

■-U

A e

A 7

A a

A 9

A lit

A12

A l3

A,.,

•-*16

■ lir

Aj8

0.05

1.00

1.00
0.15

0.30
0.95

0.95
0.30

1.00
0.06

0.96
0.28

0.69

0.80

0.41
0.91

: i52
0.85

0.95
0.30

0.97
0.27

1.00
0.50

0.23
1.00

1.00
0.50

0.27
1.00

0.83
0.59

A s

0.45
1. 00

0.90
0.50

0.50
0.90

0.27
0.97

0.09
1.00

1.00
0.58

1.00
0.19

1.00
0.45

0.50
0.90

0.75
0.76

0.25
1.00

1.00
0.75

1.00
0.55

1.00
0.75

1.00
0.58

0.67
0.82

A 2

0.50
0.90

0.90
0.50

1.00
0.23

0.77
0.77

0.58
0.89

0.56
0.88

1.00
0.45

0.90
0.50

1. 00
0.42

1.00
0.83

1.0(
0.83

1.00
0.68

1.00
0.83

0.04*
1.00

1.00
0.33

A 3

0.00*
1.00

0.50
0.87

0.50
0.88

0.88
0.56

1.00
0.08

1.00
0.10

0.01«
1.00

0.15
0.99

0.42
1.00

1.00
0.58

0.68
0.85

0.42
1.00

0.73

0.81

0.72
0.72

A 4

0.87
0.50

0.88
0.50

0.56
0.88

0.08
1.00

0.10
1.00

1.00
0.0 1«

0.99
0.15

1.00
0.42

0.58
1.00

0.85
0.68

1.00
0.42

0.81
0.73

0.72
0.72

0.99
0.12

. 8 :
0.71

0.22
0.98

0.73
0.77

0.99
0.13

0.99
0.15

1.00
0.42

0.85
0.68

1.00
0.42

0.36

1.00

0.55
0.87

A B

0.58
0.88

0.02*
1.00

0.50
0.91

0.99
0.12

0.97
0.27

1.00
0.50

0.77
0.77

0.50
1.00

1.00
0.50

0.27
0.97

A 7

1.00
0.16

0.89

0.58

0.56
0.88

0.72
0.79

1.00
1.00

0.73
1.00

0.51
1.00

1.00
0.27

0.95
0.45

0.99
0.20

A s

0.88
0.56

0.00«
1.00

0.09
1.00

0.36
1.00

0.62
0.89

1.00
0.64

0.68
0.85

0.95
0.34

A 9

0.50
0.90

0.76
0.75

0.75
1.00

1.00
0.25

0.55
1.00

0.75
1.00

0.58
1.00

0.99
0.18

A lQ

1.00
0.01«

1.00
0.42

0.85
0.68

1.00
0.42

0.81
0.73

0.72
0.72

0.67
1.00

1.00
0.33

0.91
0.58

1.00
0.33

0.67
1.00

0.93
0.41

A 12

0.92
1.00

0.83
1.00

0.92
1.00

0.67
1.00

Aiz

0.83
1.00

0.92
1.00

0.67
1.00

0.83
1.00

0.83

1.00

0.42
1.00

0.92
1.00

1.00
0.33

A i0

0.46
1.00

A 17

§B.6

A complete example 323

B.5 Weights file for the EMPLOYEE specification

Employee area

Attr.

mployee

tr w

Contract

or
W

0.71

0.20

4.90

0.14

0.12

8.17

0.43

0.24

4.08

A 2

0.57

0.24

4.08

1.00

0.00

0.79

0.17

5.94

A 3

1.00

0.00

0.57

0.24

4.08

0.79

0.17

5.94

0.29

0.20

4.90

0.71

0.20

4.90

0.50

0.25

4.00

A 5

0.71

0.20

4.90

0.29

0.20

4.90

0.50

0.25

4.00

A 6

0.57

0.24

4.08

0.14

0.12

8.17

0.36

0.23

4.36

A 7

0.83

0.14

7.20

0.17

0.14

7.20

0.50

0.25

4.00

A a

0.00

0.43

0.24

4.08

0.23

0.18

5.63

A 9

0.17

0.14

7.20

0.86

0.12

8.17

0.54

0.25

4.02

-^10

0.14

0.12

8.17

0.29

0.20

4.90

0.21

0.17

5.94

A u

0.71

0.20

4.90

0.29

0.20

4.90

0.50

0.25

4.00

A 12

0.50

0.25

4.00

0.17

0.14

7.20

0.33

0.22

4.50

A 13

0.17

0.14

7.20

0.00

0.08

13.09

A l4

0.17

0.14

7.20

0.00

0.08

13.09

A 15

0.17

0.14

7.20

0.17

0.14

7.20

0.17

0.14

7.20

-^16

0.00

0.17

0.14

7.20

0.08

13.09

A 17

0.00

0.29

0.20

4.90

0.15

0.13

7.68

■^18

0.14

0.12

8.17

0.43

0.24

4.08

0.29

0.20

4.90

B.6 Distances file for BWIU v. Odco

Employee area

324 A pragmatic legal expert system

§B.6

Instant case

'-•'.

—

-"■

—
-11

>-i

---II

-11

1ft

O) (O

—

to fO

<o CI

■;

7. --

—

O CI

—

o o

■<r

C» CI

ci «o

;

o o

O CZi

;

CD CD

« CI

»o

1^ CM

»o

CO CO

—

CI CO

O CI

—

■-< CM

■v

CM CI

■^

>o *o

;

o o

1 1

O O

;

o o

O) so

:•'.

O O

- 1

CO »o

CO <c

CI 1*

T? ~P

—

CI CI

—

■H

CI — t

o o

;

CD O

-r o

O T

—

■H rH

■V

■;

■o

i-( >o

-T

-r «o

-■■".

TJ* -*

—

?) CO

-i

CI CI

;

o o

O O

;

o ■-:-

;

o ■:■

■■-

r- oi

•V N

>o >o

-V

CO •*

co co

--".■

OJ CO

.-:

CO CO

-"-".■

:'.

■H

N CO

p
•a

—

rH rH

I— •

CI —t

—

■-I i-H

—

i-i

■;

■H

■-H i-H

o o

;

so o

O CZ>

■;

ci CD

—

■H i-H

—

(D —t

—

i-H i-H

—

■H

CI i-t

»Q —t

30 CO

- 1

—( fO

i-t eo

O CO

9 o

•v

■o

— »

■4

O O

» ci

- i

-V o

■;

rH W

CO »o

— T

.-".

—

CM CI

—

CI —t

CM T

I- CO

CO >o

■

•

X X

•

X X

■

■-■

X X

•

• X

-r

■Tj

X X

•

■

• X

•

■■

X X

•

■

■■

X X

•

X X

«i

•

X X

•

■

■■

X •

X X

■t;

■

X X

•

■

-■

X X

■t;

■

X •

■

-■

X X

• X

■t;

•

X X

•

-■

X X

•

X X

•

X •

•

• •

•

■■

X X

•

X X

* „

t> °

1 ^

X X

X •

•

■

X X

•

■'.

•

X X

■

• X

X X

■

•

• •

•

• •

X X

■

• X

•

• X

•t;

•

X •

•

■-■

• X

X X

■-'

•

X •

•

X •

•

■-.

X X

•

• *

•

• X

•

■■■

X X

■

•

X X

•

X X

•

X •

■

•

• •

•

• •

•

• •

•

• •

•

■

X •

•

X •

•

• X

•i;

• •

•

X •

•

■-■

X X

•

X X

•—

° c7

2 E

§B.6

A complete example 325

Instantiation 1

—

:-.

-11

■H

aft

a
u

-11

-T O

O CM

O N

■o

fC T

o co

—

—t o

—

'-•'■

m . °

•r

CM -T

o o
1 I

r
1

o o

z
1

o o

» »

t- o

-"■".

■"■

rt O)

■o

N CO

;

— ei

—

o ■--

■V

'-■'■

CM O

—

-T -r

o o
1 1

r
1

o o

z
1

O O

;

O O

■o

■;, o

fQ CM

lO !-•

—

■f

»o «o

CO o

"* "*

.'-:

" 1

C> CO

CM CM

o o

;

O O

—

O O)

O ~V

—

CO O)

•O CO

-T

to ID

«o

■* ■*

*-_

- 1

CO CO

CM CO

■:■■ ■■:■■

o o

O O

O) o

c»

« oo

—

■c

(O |^

■o

-T fO

■o

| to

o ,

1 1

°. '

—

-r

■V

—

w —

•o

i- -T

—

■-< o

•v

■-< O

o w

■a* en

— CM

-r

-- o

r-i -V

'-'■

■a* N

■V

-■--

CJ (D

•v

;

•6 en

-T «0

CM -tf

- 1

CM CO

CM CM

CM -V

I- 00

CO «

■

•

X X

•

X X

■

-■

X X

•

• X

-r

■Tj

X X

•

■

• X

•

X X

•

■

X X

•

X X

•

X X

•

■

■■

X •

X X

■-'

X X

•

■

-■

X X

■t;

■-'

X •

-■

X X

• X

•

X X

•

X X

•

X X

•

X •

•

• •

•

■■■

X X

•

X X

* „

k> °

* *^

X X

X •

•

■

X X

•

X X

■

• X

X X

■

•

• •

•

• •

•

X X

■

• X

•

• X

•t;

•

X •

•

■-■

• X

X X

■-'

•

X •

•

X •

•

■

X X

•

• *

•

• X

•

■■■

X X

■

•

X X

•

X X

•

X •

■",

■

•

• •

•

• •

•

• •

•

• •

•

■-.

X •

•

X •

•

• X

■t;

• •

•

X •

•

■-■

X X

•

X X

' :

T( %

,_!

PI 1-

•r

II)

\B t-

■r

; °

2 E

a
O

° c7

J i

326 A pragmatic legal expert system

§B.6

Instantiation 2

coo

.2 a a ci

•5 (I -ft all

o o o o o o o

n n h (. <o
— — o o o

o o o o o o o

a pi o «i to ro >o

m co -v co co ■* -*

00*0*0* o o o*

O ^ W N "1" "O O

O ^ O ^" ■* T »0

o o o o o o o

CJ) W O) X 00 O OS

o n ^r

X X X X m X X

X X X X X

X X X X

X X X X X

X X • X X

X X X X X X •
X • X XXX

XX X X X X

X X X • XX

X X
X X

'■■ ■

6 CJ tJ d

c +

U CI

ft -ft

•3 1 *0 OS OS -3 1

o o o o o o o

OS O 1- CI -^ CI O

CO CO i-H « CM iH CO
o o o o o o o

—

CI —

;

o o

CM
CM

CO CI

i-l CO T CI

— - -

-T -V CI
o o o o o o o

■* O fO •# CO o o

D i-l O O OS CM T

"O (D N OS OS CM IO

00 ^ r-i oi CO f6

H d "1 H T? d

CO CO O O N 1^ 00

X X X X

X X
X X

X X
X X
X X

X X X X

tJ tJ o o

§B.6

A complete example 321

Instantiation 3

z
-'.
:•-.

+
X

-H

z
—

,1!

•J

=
z.
>o

CO
to

»o
4

-11

to
to
to
to

ail

o o

t~ CM

rH ^

i«

fO CI

- •

O CO

—

CI — •

—

rH rH

—

— -T

d d

z
I

;

d d

o o

;

■;

d d

— «

rH rH

OJ N

•v

■o

!-• !-•

—

— CI

—

Ct CO

—

'-■'■

O —t

CO -T

o o

1 1

;

O O

o o

;

O O

rH o

■fl*

to (^

■o

o ':

•v

J-t

o o

CO «o

.""-

to co

—

CO CI

CO CM

d d

o o

;

d d

-r o*

'-■'

■; r.

—

- 1

OJ CO

-"'"

—

— w

-r «o

to to

*- m

--.

- 1

to CO

■^

c«

CO CM

d d

o o

;

d d

X o

■V N

—

I- CO

■o

»o »o

■o

-r

p
■a

| to

o ,

1 1

°. '

—

•V

■o

—

« r-t

TV 1-H

- 1

-t o

-T

--".

>o to

-r re

30 CO

—

**: *?

—

■o

•v

CI CO

•v

>6 -V

no ci

—

■'-

-r d

»o

-Z

•v

i-i co

to o

rH ^

—

to CO

■V

CO CM

CM -*

I- CO

to >o

■

•

X X

•

X X

■

-■

X X

•

• X

■Tj

X X

•

■

• X

•

'■

X X

•

■

■■

X X

•

X X

•

X X

•

■

X •

X X

•*

■-'

X X

•

■

-■

X X

■t;

■-'

X •

-■

X X

• X

•

X X

•

X X

•

X X

•

X •

•

• •

•

■■■

X X

•

X X

* „

k> °

* *^

X X

X •

•

■

X X

•

X X

■

• X

X X

■

•

• •

•

• •

«i^

■"

X X

■

• X

•

• X

•t;

•

X •

•

■-■

• X

X X

■-'

•

X •

•

X •

•

■-.

X X

•

• *

•

• X

•

■

X X

■

•

X X

•*

•

X X

•

X •

■",

■

•

• •

•

• •

•

• •

•

• •

•

■-.

X •

•

X •

•

• X

■t;

• •

•

X •

•

■-■

X X

•

X X

' :

T( %

,_!

PI 1-

•r

II)

\B t-

■r

; a

2 E

a
O

tj tj

•—

;

° c7

J i

328 A pragmatic legal expert system

§B.6

Instantiation 4

& -

"* f>l

i +

+ l :

E R

UJ u

8 art

-II

to to CI

O) I-

—t —t C* — O NO

o o o o o o o

o o o o o

-"-:■

CO CO i-t

CI CO -<J«

o o o

co co ■*
o o o o o o o

» N O) f- N -V CO

CO O O CO CI CO a

CO CO O CO CO —t -T

X X X X • X X

X X X X X

X X X X

X X X X X

X X • X X

X X X X X X •
X • X XXX

XX X X X X

X X X • XX

X X
X X

'■■ ■

6 CJ tJ d

e + + =

= X X "*

■3 5 ci *ft
n -n

CO O »0 O O — T.

n ci ci tt i-t m »t
o o o o o o o

o o o o o o o

co ^ N

V CO H

o o o

o o o o o o o

ro o o o tn ^* ^*

-1

CO
CO

-1

n ^ a t co co o

OJ CI o »o

ci d d o

—t CO CI CI

(DOS)

oi O O)

^ N i-l

CO CO O CO N 1^ CO

X X X X

X X
X X

X X
X X
X X

X X X X

tJ tJ o o

§B.6

A complete example 32 £

Hypothetical 1

t~ m

z
—

CD
*3

z
■o

*ft

™

—

-11

a
U

-1

--JI

-11

—

-T 'O

i- ai

no o

—

•v

—

•o w

co ci

Cfl

CI o

■* -v

■^

—

CI fD

d d
1

z
1

d d

;

d d

CO CI

1- CI

-T -T

,_,

ci »o

CI CI

—

-H CI

o »o

—

>o t-

d d
1

d d

;

d d

e«

•-» N

» o

—

O CO

-T Cft

CI T

-"-".■

CO CO

i-H i-H

-"■".

CI o

;

O O

;

d d

■;

d d

•v

— t CD

O -V

—

ot a>

-v

■;

■o

-H CO

to »o

—

■* ■*

—

■-« i-H

CI -H

d d

c d

—

d d

■'-

■O O)

T- f-

CO CO

■V

•v

CO CI

CO CO

OJ CO

:-"

--.

V) CO

.'.

■H

I- CO

i— 1

i— 1 rH

—

CI -H

;

—

i— t i-H

—

■;

-H —

O O

—

Z-

so d

o o

□

;

CI o

"-•

rH rH

—

[O -H

"-•

—

■-H i-H

"-•

r-t

CI -H

" i

CI CO

(D CO

■V

CO -V

ai -r

O -T

Oi -t

."■".

—

(D -H

-"-".

7. 7.

ci ci

N i-i

>Q N

—i C)

CI -v

i-H ~r

i-H I-H

■v

•-t

CI ■*

j- m

P) >o

■t;

•

X X

•

X X

■-■

X X

•

• X

■t;

X X

•

■■■

• X

•

•i;

X X

•

■

-■

X X

•

X X

•

X X

•

■■

X •

X X

•

■■

X X

'■.

X X

—

■t;

■-'

X •

■

■■

X X

• X

•t;

•

X X

•

-■

X X

•

X X

u Z

•

X •

•

• •

•

-■

X X

•

X X

3 "^

■Q

b °

5 ^

■-■

X X

■:

X •

•

■

-■

X X

•

X X

■^

•

• X

X X

■

■-■

•

• •

•

• •

X X

-■

• X

•

• X

■t;

•

X •

•

■■

• X

X X

■-■

•

X •

•

X •

•

-■

X X

•

■■

■t;

•

• •

•

• X

•

■■

X X

•

X X

•

X X

•

X •

■

•

• •

•

• •

•

• •

•

• •

•

X •

•

X •

•*

•

• X

■t;

• •

•

X •

•

X X

•

X X

1 a

m m

„,

to t-

en a

■ '

2 |

■-_

J J

—•

330 A pragmatic legal expert system

§B.6

Hypothetical 2

z
—

—

•v

■H

CO
CO

-It

aft

-11

aft

a* -r

CO co

—

1^ CI

■ -

•v

CO -T

—

o o

:-t

CM o

i-i

.-".

—t CI

—

O CI

o o

o o
1

00 N

«D

t- CM

00 CO

OJ —t

o o

■**!

—t CI

;

O CI

•r

■H CO

;

d d
1

d d

O CO

,_,

o o

•V

o »o

m oo

Cfl

CI CO

—

** "^

- i

CO CI

—

-"'".

—

CO <-<

o o

;

o o

■V

•V "?

O -V

—

•v —

•v

■o

CO »o

-T -T

.-"

■* "*

—

'-■'■

■* **?

■o

CO CI

o o

;

o o

■'-

t- t~

•v

~? N

N >0

—

■^

»o ■*

■a

CO CO

OJ CO

CO CO

—

N CO

1— 1 1— 1

r-i

CI —t

—

i-H i-H

■-H

—

1— —

;

d d

—

co d

d d

;

ci d

—

rH rH

— '

—

CO i-t

—

iH rH

—

■;

■H

CI —*

>o CO

—

00 CO

-tf

00 to

■;

OJ CO

:-.

Ol -T

:-:

■A

Tj* O

—

■v

O 00

00 CI

—

CD <3

CI 00

CO CO

r " 1

1-1 Tj<

.-:

" I

CO CI

—

1-1

CO <-<

CI ■*

I- 00

CO «

•r;

•

X X

•

X X

■

■■

X X

•

• X

X X

•

■■

• X

•

•i;

X X

•

X X

•

X X

■Tj

•

X X

•

■

-■

X •

X X

■Tj

X X

•

■

-■

X X

X •

■■

X X

• X

•

X X

•

-■

X X

•

X X

u Z
1 ^

•

X •

•

• •

•

-■

X X

•

X X

l *^

X X

X •

•

■

■■

X X

•

X X

•t;

■"

• X

X X

■

•

• •

•

• •

X X

■

-■

• X

•

• X

■t;

•

X •

•

• X

X X

•

X •

•

X •

•

■-■

X X

•

*t*

•

• ♦

•

• X

•

X X

•

X X

<q*

•

X X

•

X •

■

•

• *

•

• •

■t;

•

• •

•

• •

•

X •

•

X •

•

• X

"»•

• *

•

X •

•

■■■

X X

•

X X

to t-

1 3

1 I

fc 6*

J J

*i a

'-.

§B.7 A complete example 331

B.7 Report file for BWIU v. Odco

Employee area

Instant case

The law distinguishes between a contract of service (between employer
and employee) and a contract for services (between principal and in-
dependent contractor). This distinction affects the terms that will be
implied in the absence of an express agreement, the liability of the
employer to third parties, the applicability of industrial awards, the
applicability of statutes which may affect workers 1 compensation, oc-
cupational health and safet}', long-service leave, fringe benefits tax, etc.
The terms "employer" and "worker" are used here to mean "em-
ployer" and "employee" (in the case of a contract of service) or "prin-
cipal" and "independent contractor" (in the case of a contract for
services).

In the instant case, the emploj'er did not direct the manner in which the
work was to be done; the worker was allowed to use her/his own discre-
tion in doing an aspect of the work that was not specified beforehand;
the worker was not an integral part of the employer's business, but was
accessory to it; the worker owned the tools or provided the transport
with which she/he performed the work; the employer would not make
a profit/loss if the work performed by the worker cost less/more than
expected; the work was not performed on the employer's premises; the
emploj'er neither supervised nor inspected the work; it is not known
whether the worker was in business on her/his own account; the worker
was not allowed to employ others to assist with her/his work; the worker
was not obliged to work only for the employer; the worker was required
to work at specified times; it is not known whether the employer paid
the worker by time; the money that the employer paid to the worker
was not stated to be a "fee"; the money that the employer paid to
the worker was not stated to be "wages" or "salary"; the employer did
not deduct PAYE tax instalments from the worker's pay; the employer
paid the worker neither sick pay nor holidaj' pay; the employer and
the worker did not express any intention that the relationship would
be one of employer and employee; and the employer and the worker
expressed an intention that the relationship would be one of principal
and independent contractor.

332 A pragmatic legal expert system §B-7

In my opinion- — -following Humberstone v. Northern Timber Mills — the
worker is an independent contractor.

In Humberstone v. Northern Timber Mills, 1 a 1949 decision of three
justices of the High Court of Australia, Humberstone carried goods
for NTM. He had originally held himself out as a carrier, prepared
to carry for anyone, but for over twenty j'ears he had carried goods
solely for NTM (although he would, infrequently, carr} r back-loads for
NTM's customers). Humberstone owned the truck, and paid for petrol
and repairs. He was paid weekly on a weight-mileage basis. He was a
licenced carrier, and had his name printed on the side of his truck with
the description "< arrier."

On the way back from a job, he had a puncture. He went home
to change the wheel, but exerted himself so strenuously in trying to
remove the tyre from the wheel that he became ill and later lapsed
into a coma, from which he did not recover. Section 3 of the Worker's
Compensation Act 1928 (Vic) had been amended about a year before
Humberstone's death so as to include independent contractors in its
definition of a "worker" covered by the Act. However, the High Court
held that the amendment applied only to contracts entered into after
it came into operation. Further, the Court decided that Humberstone
was not an employee of NTM. Hence, he was not a "worker" under the
Act, and his widow was not entitled to compensation under the Act.

There are several significant similarities between the instant case
and Humberstone v. NTM: the employer did not direct the manner in
which the work was to be done; the worker was allowed to use her/his
own discretion in doing an aspect of the work that was not specified
beforehand; the worker was not an integral part of the employer's busi-
ness, but was accessor}' - to it; the worker owned the tools or provided
the transport with which she/he performed the work; the employer
would not make a profit/loss if the work performed by the worker cost
less/more than expected; the work was not performed on the employer's
premises; the employer neither supervised nor inspected the work; the
worker was not obliged to work only for the employer; the money that
the employer paid to the worker was not stated to be a "fee"; the money
that the employer paid to the worker was not stated to be "wages" or
"salary": the employer did not deduct PAYE tax instalments from the
worker's pay; the employer paid the worker neither sick pay nor holiday
pay; and the employer and the worker did not express any intention
that the relationship would be one of employer and employee.

However, the instant case is not on all fours with Humberstone v.
NTM. In that case the worker was not in business on her/his own ac-
count; the worker was allowed to employ others to assist with her/his

x (1949) 79 CLR 389.

?.7 A complete example 333

work; the worker was not required to work at specified times; the em-
ployer did not pay the worker by time; and the employer and the worker
did not express any intention that the relationship would be one of
principal and independent contractor.

Nevertheless, I believe that Humberstone v. NTM should be followed.

If Ferguson v. John Dawson and Partners (Contractors) Ltd is followed
then the worker is an employee.

In Ferguson v. John Dawson and Partners (Contractors) Ltd,~ a 1976
decision of the English Court of Appeal, Ferguson fell off a roof while re-
moving some scaffolding boards. He claimed damages against Dawson
(the building contractors) for breach of statutory duty relying on the
Construction (Working Places) Regulations 1966 (UK). This duty
would only be owed if Ferguson was an employee of Dawson.

There are several similarities between the instant case and Fer-
guson v. Dawson: the work was not performed on the emp^^er's
premises; the worker was not allowed to employ others to assist with
her/his work; the worker was not obliged to work only for the employer;
the worker was required to work at specified times; the money that the
employer paid to the worker was not stated to be a "fee"; the money
that the employer paid to the worker was not stated to be "wages" or
"salary"; the employer did not deduct PAYE tax instalments from the
worker's pay; the employer paid the worker neither sick pay nor hol-
iday pay; the employer and the worker did not express an}' intention
that the relationship would be one of employer and employee; and the
emploj'er and the worker expressed an intention that the relationship
would be one of principal and independent contractor.

However, there are several significant differences between the instant
case and Ferguson v. Dawson. In that case the emploj^er directed the
manner in which the work was to be done; the worker was not allowed to
use her/his own discretion in doing an aspect of the work that was not
specified beforehand; the worker was an integral part of the employer's
business; the worker neither owned the tools nor provided the transport
with which she/he performed the work; the employer would make a
profit/loss if the work performed by the worker cost less/more than
expected; the employer supervised or inspected the work; the worker
was not in business on her/his own account; and the employer paid

2 [1976] 1 WLR 1213.

3 ibid. at 1219 per Megaw LJ, at 1228-9 per Browne LJ.

334 A pragmatic legal expert system §B-7

the worker by time. Note also that Ferguson v. Dawson is only a
decision of the English Court of Appeal and not as good authority as
a case decided b} r three justices of the High Court of Australia- — like
Humberstone v. NTM.

Consequently, there is nothing in Ferguson v. Dawson to warrant
an}' change in my conclusion.

Hypothetical 1

Consider the instant case changed so that the following is true: the
worker was allowed to employ others to assist with her/his work; and
the employer and the worker did not express any intention that the
relationship would be one of principal and independent contractor.

If that were so then I would be more strongly of the opinion that —
following Humberstone v. Northern Timber Mills — the worker is an
independent contractor.

Details of Humberstone v. NTM are summarized above. There are
several significant similarities between the Irypothetical case and Hum-
berstone v. NTM: the employer did not direct the manner in which the
work was to be done; the worker was allowed to use her/his own discre-
tion in doing an aspect of the work that was not specified beforehand;
the worker was not an integral part of the employer's business, but was
accessory to it; the worker owned the tools or provided the transport
with which she/he performed the work; the employer would not make
a profit/loss if the work performed bj' the worker cost less/more than
expected; the work was not performed on the employer's premises; the
emploj'er neither supervised nor inspected the work; the worker was
allowed to employ others to assist with her/his work; the worker was
not obliged to work only for the employer; the money that the employer
paid to the worker was not stated to be a "fee"; the money that the
employer paid to the worker was not stated to be "wages" or "salary";
the employer did not deduct PAYE tax instalments from the worker's
pay; the employer paid the worker neither sick pay nor holidaj' pay; the
employer and the worker did not express any intention that the rela-
tionship would be one of employer and employee; and the employer and
the worker did not express any intention that the relationship would
be one of principal and independent contractor.

However, the Irypothetical case is not on all fours with Humber-
stone v. NTM. In that case the worker was not in business on her/his
own account; the worker was not required to work at specified times;
and the employer did not pay the worker by time.

Nevertheless, I believe that Humberstone v. NTM should be followed.

?.7 A complete example 335

If Cam and Sons Pty Ltd v. Sargent is followed then the worker is an
employee.

In Cam and Sons Pty Ltd v. Sargent* a 1940 decision of four justices
of the High Court of Australia, Sargent was the master of a ship. He
entered into an agreement with Cam and Sons that claimed that the
ship was hired by Cam and Sons to Sargent and his fellow contractors
(called "the partnership"). However, it was doubtful whether that
agreement actually deprived Cam and Sons of any control over the
ship. The partnership was to use the ship only to carry coal from
Swansea to Sydney. Cam and Sons were sole agents of the partnership
for securing cargoes for the ship, and for collecting money due to the
partnership. The partnership paid nothing for the "hire" of the ship,
but received a specified sum for each return trip of a certain tonnage
plus (in certain circumstances) 5% of the earnings, the balance of which
was retained b} r Cam and Sons. Cam and Sons had to approve people
emploj'ed by the partnership.

Sargent claimed that he (and others in the partnership) were em-
ployed bj' Cam and Sons, and therefore came within the terms of an
industrial award. Cam and Sons claimed that members of the partner-
ship were independent contractors.

The High Court unanimously agreed with Sargent. Rich J came to
the conclusion that the agreement was an attempt to evade the terms
of the industrial award.

There are several similarities between the hypothetical case and
Cam v. Sargent: the worker was allowed to use her/his own discre-
tion in doing an aspect of the work that was not specified beforehand;
the work was not performed on the employer's premises; the employer
neither supervised nor inspected the work; the worker was allowed to
employ others to assist with her/his work; the worker was not obliged
to work only for the employer; the money that the employer paid to
the worker was not stated to be a "fee"; the money that the employer
paid to the worker was not stated to be "wages" or "salar} 1- "; the em-
ployer did not deduct PAYE tax instalments from the worker's pay;
the employer paid the worker neither sick pay nor holiday pay; the
employer and the worker did not express any intention that the rela-
tionship would be one of employer and employee: and the employer and
the worker did not express any intention that the relationship would
be one of principal and independent contractor.

However, there are several significant differences between the hypo-
thetical case and Cam v. Sargent. In that case the employer directed

4 (1940) 14 ALJ 162.
5 ibid. at 103.

336 A pragmatic legal expert system §B-7

the manner in which the work was to be done; the worker was an integ-
ral part of the employer's business; the worker neither owned the tools
nor provided the transport with which she/he performed the work; the
employer would make a profit/loss if the work performed by the worker
cost less/more than expected; the worker was not in business on her/his
own account; the worker was not required to work at specified times;
and the employer did not pay the worker by time.

Despite the fact that Cam v. Sargent is a decision of four justices of
the High Court of Australia (and better authority than a case decided
by three justices of the High Court of Australia— like Humberstone v.
NTM), there is nothing in Cam v. Sargent to warrant any change in
my conclusion.

Hypothetical 2

Consider the instant case changed so that the following is true: the
worker was not allowed to use her/his own discretion in doing an aspect
of the work that was not specified beforehand; and the worker was an
integral part of the employer's business.

If that were so then my opinion would be that — -following Ferguson v.
John Dawson and Partners (Contractors) Ltd — the worker is an em-
ployee.

Details of Ferguson v. Dawson are summarized above. There are several
significant similarities between the hypothetical case and Ferguson v.
Dawson: the worker was not allowed to use her/his own discretion
in doing an aspect of the work that was not specified beforehand; the
worker was an integral part of the employer's business; the work was not
performed on the employer's premises; the worker was not allowed to
employ others to assist with her/his work; the worker was not obliged
to work only for the employer; the worker was required to work at
specified times; the money that the employer paid to the worker was
not stated to be a "fee"; the money that the employer paid to the
worker was not stated to be "wages" or "salary"; the employer did
not deduct PAYE tax instalments from the worker's pay: the employer
paid the worker neither sick pay nor holidaj' pay; the employer and
the worker did not express any intention that the relationship would
be one of employer and employee: and the employer and the worker
expressed an intention that the relationship would be one of principal
and independent contractor.

However, the hypothetical case is not on all fours with Ferguson v.
Dawson. In that case the employer directed the manner in which the
work was to be done; the worker neither owned the tools nor provided
the transport with which she/he performed the work; the employer

?.7 A complete example 331

would make a profit/loss if the work performed by the worker cost
less/more than expected; the employer supervised or inspected the
work; the worker was not in business on her/his own account; and
the employer paid the worker by time.

Nevertheless, I believe that Ferguson v. Dawson should be followed.

If Australian Mutual Provident Society v. Chaplin or Ready Mixed Con-
crete (South East) Ltd v. Minister of Pensions and National Insurance
are followed then the worker is an independent contractor.

In Australian Mutual Provident Society v. Chaplin® a 1978 decision of
the Judicial Committee of the Privy Council, Chaplin was a represent-
ative of AMP. A clause of the agreement between them stated that the
relationship was one of "principal and agent" and not one of "master
and servant." Chaplin claimed that he was employed under a contract
of service, and was therefore a "worker" under the Long Service Leave
Act, 1967 (SA) and entitled to certain benefits.

The Privy Council found that there was no reason to think that
the clause was not a genuine statement of the parties' intentions. Ex-
amining the agreement, their Lordships concluded that it provided for
a contract of agency. The fact that Chaplin was given the power of
unlimited delegation of the whole performance of his work was "almost
conclusive against the contract being a contract of service."

There are several similarities between the hypothetical case and
AMP v. Chaplin: the employer did not direct the manner in which
the work was to be done; the worker was an integral part of the em-
ployer's business; the worker owned the tools or provided the transport
with which she/he performed the work; the employer would not make
a profit/loss if the work performed bj' the worker cost less/more than
expected; the work was not performed on the employer's premises; the
employer neither supervised nor inspected the work; the money that
the employer paid to the worker was not stated to be a "fee"; the money
that the employer paid to the worker was not stated to be "wages" or
"salary"; the employer did not deduct PAYE tax instalments from the
worker's pay; the employer paid the worker neither sick pay nor hol-
iday pay; the employer and the worker did not express anj r intention
that the relationship would be one of employer and employee: and the
employer and the worker expressed an intention that the relationship
would be one of principal and independent contractor.

However, there are several significant differences between the hypo-
thetical case and AMP v. Chaplin. In that case the worker was allowed
to use her/his own discretion in doing an aspect of the work that was
not specified beforehand; the worker was in business on her/his own

3 (1978) 18 ALR 385.

ibid, at 391.

338 A pragmatic legal expert system §B-7

account; the worker was allowed to emploj' others to assist with her/his
work; the worker was obliged to work only for the employer; the worker
was not required to work at specified times; and the employer did not
pay the worker by time.

Despite the fact that AMP v. Chaplin is a decision of the Judicial
Committee of the Privy Council (and better authority than a case de-
cided by the English Court of Appeal — like Ferguson v. Dawson), there
is nothing in AMP v. Chaplin to warrant any change in my conclusion.

In 1967, Ready Mixed Concrete (South East) Ltd v. Minister of Pen-
sions and National Insurance 8 was decided by the Queen's Bench Divi-
sion of the English High Court. (A case decided by the Queen's Bench
Division of the English High Court is not as good authority as a case
decided by the Judicial Committee of the Privy Council- — like AMP v.
Chaplin: furthermore Ready Mixed v. Minister is 11 years older than
AMP v. Chaplin.)

In Ready Mixed v. Minister, Latimer worked for Ready Mixed as
an "owner-driver." He was paid at mileage rates, and was obliged to
buy the truck through a financial organization associated with Ready
Mixed. The truck was painted in the company's colours, and he had to
wear a Ready Mixed uniform. Latimer was obliged to meet the costs of
maintenance, repair and insurance of the truck (and the attached mix-
ing unit, which belonged to Ready Mixed). The Minister determined
that Latimer was employed under a contract of service, and was there-
fore an "employed person" under s. 1(2) of the National Insurance Act
1965 (UK), making Ready Mixed liable to make weekly contributions.

MacKenna J examined the contract and held that the rights it con-
ferred, and the duties it imposed, between Latimer and Ready Mixed
were not such as to make it a contract of service.

There are several similarities between the hypothetical case and
Ready Mixed v. Minister: the employer did not direct the manner in
which the work was to be done; the worker was an integral part of the
employer's business; the worker owned the tools or provided the trans-
port with which she/he performed the work; the employer would not
make a profit/loss if the work performed by the worker cost less/more
than expected; the work was not performed on the employer's premises;
the employer neither supervised nor inspected the work; the money that
the employer paid to the worker was not stated to be a "fee"; the money
that the employer paid to the worker was not stated to be "wages" or
"salary"; the employer did not deduct PAYE tax instalments from the
worker's pay; the employer paid the worker neither sick pay nor hol-
iday pay; the employer and the worker did not express an}' intention

'[1968] 2 QB 497.

?.7 A complete example 339

that the relationship would be one of employer and employee; and the
employer and the worker expressed an intention that the relationship
would be one of principal and independent contractor.

However, there are several significant differences between the hypo-
thetical case and Beady Mixed v. Minister. In that case the worker
was allowed to use her/his own discretion in doing an aspect of the
work that was not specified beforehand; the worker was not in business
on her/his own account; the worker was allowed to employ others to
assist with her/his work; the worker was obliged to work onty for the
employer; the worker was not required to work at specified times; and
the employer did not pay the worker by time. Note also that Ready
Mixed v. Minister is only a decision of the Queen's Bench Division of
the English High Court and not as good authority as a case decided by
the English Court of Appeal — like Ferguson v. Dawson.

Consequently, there is nothing in Ready Mixed v. Minister to warrant
any change in my conclusion.

Bibliography

ALDENDERFER, Mark S. and BLASHFIELD, Roger K. 1984, Cluster
Analysis, Sage University Paper Series on Quantitative Applic-
ations in the Social Sciences, no. 07-044, Sage, Beverty Hills.
ISBN 8039 2376 7.

ALLEN, Layman E. 1963, "Beyond document retrieval toward in-
formation retrieval", Minnesota Law Review, vol. 47, pp. 713-67.
ISSN 0026-5535.

ALLEN, Layman E. 1965, "Usefulness of modern logic to the readers
and writers of legal documents", in Communication Sciences and
Law: Reflections from the Jurimetrics Conference, eds Layman E.
Allen and Mary E. Caldwell, Bobbs-Merrill, Indianapolis, ch. 13,
pp. 83-112.

From the Conference on the Implications of Developments in the Com-
munication Sciences for Legal Education in the Next Decade of the
Jurimetrics Committee of the Association of American Law Schools. Yale
Law School, New Haven, Connecticut, 5-7 September 1963.

ALLEN, Layman E. 1968, "A language-normalization approach to
information retrieval in law", Jurimetrics Journal, vol. 9, no. 1,
September, pp. 41-56. ISSN 0022-6793.

ALLEN, Layman E. 1981, "Analysis of law by symbolic logic", in Com-
puters and the Law: An Introductory Handbook (third edition), ed.
Robert P. Bigelow, American Bar Association, pp. 84-90.

ALLEN, Layman Edward 1982, "Towards a normalized language
to clarify the structure of legal discourse", in Deontic Logic.
Computational Linguistics and Legal Information Systems, ed. Ant-
onio A. Martino, vol. II, North-Holland, Amsterdam, pp. 349-407.
ISBN 444 86415 6, 444 86413 X (set).

From the International Conference on "Logic, Informatics, Law", Florence,
Italy, April 1981.

341

342 A pragmatic legal expert system

Allen. Layman E., BROOKS, Robin B. S. and James, Patricia A. 1962,
Automatic Retrieval of Legal Literature: Why and How, Walter E.
Meyer Research Institute of Law, New Haven, Connecticut.

Allen, Layman E. and Caldwell, Mary Ellen 1963, "Modern lo-
gic and judicial decision making: A sketch of one view", Law
and Contemporary Problems, vol. 28, no. 1, Winter, pp. 213-70.
ISSN 0023-9186.

Republished in Baade 1903.

ALLEN, Layman E. and SAXON, Charles S. 1985, "Computer aided
normalizing and unpacking: Some interesting machine-processable
transformations of legal rules", in Computing Power and Legal Reas-
oning, ed. Charles Walter, West Publishing Company, St Paul,
ch. 20, pp. 495-572. ISBN 314 95570 4.

ALLEN, Layman and SAXON, Charles 1988, "Exploring computer-
aided generation of questions for normalizing legal rules", in
Computer Power and Legal Language: The Use of Computational
Linguistics, Artificial Intelligence, and Expert Systems in the Law,
ed. Charles Walter, Quorum, New York, ch. 19, pp. 243-316.
ISBN 89930 306 4.

From the Second Annual Conference on Law and Technology, University
of Houston, 24-28 June 1985.

AsHLEY, Kevin D. 1985, "Reasoning by analogy: A survey of se-
lected AI research with implications for legal expert systems", in
Computing Power and Legal Reasoning, ed. Charles Walter, West
Publishing Company, St Paul, ch.6, pp. 105-27.
ISBN 314 95570 4.

AsHLEY, Kevin D. 1989a, "Toward a computational theor}' of ar-
guing with precedents: Accomodating [sic] multiple interpretations
of cases", in Proceedings of the Second International Conference on
Artificial Intelligence and Law (ICAIL-89), University of British
Columbia, Vancouver, 13-16 June, pp. 93-102. ISBN 89791 322 1.

ASHLEY, Kevin D. 1989b, "Defining salience in case-based arguments",
in Proceedings of the Eleventh International Joint Conference on
Artificial Intelligence (IJCAI-89), Detroit, Michigan, 20-25 August,
vol. 1, pp. 537-42. ISBN 1 55860 094 9. ISSN 1045-0823.

ASHLEY, Kevin D. 1990, Modeling Legal Argument: Reasoning with
Cases and Hypothetical, Artificial Intelligence and Legal Reas-
oning Series, MIT Press (Bradford), Cambridge, Massachusetts.
ISBN 262 01114 X.

Bibliography 343

Ashley, Kevin D. and Aleven, Vincent 1991, "Toward an intelli-
gent tutoring system for teaching law students to argue with cases",
in Proceedings of the Third International Conference on Artificial
Intelligence and Law (ICAIL-91), St Catherine's College, Oxford,
25-28 June, pp. 42-52. ISBN 89791 399 X.

Ashley, Kevin D. and RlSSLAND, Edwina L. 1986, "Toward model-
ling legal argument", in Automated Analysis of Legal Texts: Logic,
Informatics, Law, eds Antonio A. Mart in o and Fiorenza Socci
Natali, Elsevier Science (North-Holland), Amsterdam, pp. 19-30.
ISBN 444 701117.

From the Second International Conference on 'Logic, Informatics, Law*.
Florence, Italy, September 1985.

Ashley, Kevin D. and RlSSLAND, Edwina L. 1987a, "But, see, ac-
cord: Generating ; blue book* citations in HYPO", in Proceedings
of the First International Conference on Artificial Intelligence and
Law (ICAIL-87), Boston, Massachusetts, 27-29 May, pp. 67-74.
ISBN 89791230 6.

Ashley, Kevin D. and RlSSLAND, Edwina L. 1987b, "Compare and
contrast, a test of expertise", in Proceedings of the Sixth National
Conference on Artificial Intelligence (AAAI-87), Seattle, Washing-
ton, 13-17 July, vol. 1, pp. 273-8. ISBN 934613 42 7.

ASHLEY, Kevin D. and RlSSLAND, Edwina L. 1988, "Waiting on
weighting: A symbolic least commitment approach", in Proceed-
ings of the Seventh National Conference on Artificial Intelligence
(AAAI-88), St Paul, Minnesota, 21-26 August, vol. 1, pp. 239-44.
ISBN 929280 00 8.

AliBERT, Vilhelm 1963, "Conscientious objectors before Norwegian
military courts", in Judicial Decision- Making, ed. Glendon Schu-
bert, International Yearbook of Political Behavior Research, vol. 4,
Free Press of Glencoe, New York, ch. vii, pp. 201-19.

AUSTIN, John 1885, Lectures on Jurisprudence or The Philosophy of
Positive Law (fifth edition), John Murray, London.

Two volumes, revised and edited by Robert Campbell.

BAADE, Hans W. (ed.) 1963, Jurimetrics, Basic Books, New York.

Originally published as Law and Contemporary Problems, vol. 28, no. 1,
Winter 1963.

BAIN, William M. 1986, "A case-based reasoning system for subjective
assessment", in Proceedings of the Fifth National Conference on
Artificial Intelligence (AAAI-86), Philadelphia, Pennsylvania, 11-
15 August, vol. 1, pp. 523-7. ISBN 934613 13 3.

344 A pragmatic legal expert system

BANKOWSKI, Zenon and MACCORMICK, D. Neil 1991, "Statutory-
interpretation in the United Kingdom", in Interpreting Statutes:
A Comparative Study, eds D. Neil MacCormick and Robert S.
Summers, Applied Legal Philosophy Series, Dartmouth, Alder shot,
ch. 10, pp. 359-406. ISBN 1 85521 183 1.

BENCH-CAPON, T.J.M. 1987, "Support for policy makers: Formulat-
ing legislation with the aid of logical models", in Proceedings of
the First International Conference on Artificial Intelligence and
Law (ICAIL-87), Boston, Massachusetts, 27-29 May, pp. 181-9.
ISBN 89791230 6.

BENCH-CAPON, T. J. M. 1989, "Deep models, normative reasoning and
legal expert systems", in Proceedings of the Second International
Conference on Artificial Intelligence and Law (ICAIL-89), Uni-
versity of British Columbia, Vancouver, 13-16 June, pp. 37-45.
ISBN 89791322 1.

Bench-Capon, T.J.M. (ed.) 1991, Know ledge- Based Systems and
Legal Applications, APIC Series, no. 36, Academic Press, London.
ISBN 12 086441 X.

Bench-Capon, T.J.M., Robinson, CO., Routen, T.W. and Ser-
GOT, M.J. 1987, "Logic programming for large scale applications
in law: A formalisation of supplementary benefit legislation", in
Proceedings of the First International Conference on Artificial In-
telligence and Law (ICAIL-87), Boston, Massachusetts, 27-29 May,
pp. 190-8. ISBN 89791 230 6.

BENCH-CAPON, Trevor and SERGOT, Marek 1988, "Toward a rule-
based representation of open texture in law", in Computer Power
and Legal Language: The Use of Computational Linguistics, Ar-
tificial Intelligence, and Expert Systems in the Law, ed. Charles
Walter, Quorum, New York, ch. 6, pp. 39-60. ISBN 89930 306 4.

From the Second Annual Conference on Law and Technology, University
of Houston, 24-28 June 1985.

BERMAN, Donald H. 1991, "Developer's choice in the legal domain: The
Sisyphean journey with CBR or down hill with rules", in Proceed-
ings of the Third International Conference on Artificial Intelligence
and Law (ICAIL-91), St Catherine's College, Oxford, 25-28 June,
pp. 307-9. ISBN 89791 399 X.
A working paper for the Case-Rules Panel.

BERMAN, Donald H. and HAFNER, Carole D. 1991, "Incorporating
procedural context into a model of case-based legal reasoning",
in Proceedings of the Third International Conference on Artificial
Intelligence and Law (ICAIL-91), St Catherine's College, Oxford,
25-28 June, pp. 12-20. ISBN 89791 399 X.

Bibliography 345

BlNG, Jon 1980, "Legal norms, discretionary rules and computer
programs", in Computer Science and Law, ed. Bryan Niblett, Cam-
bridge University Press, Cambridge, ch. 7, pp. 119-36.
ISBN 521234514.

BlNG, Jon (ed.) 1984a, Handbook of Legal Information Retrieval, El-
sevier Science (North-Holland), Amsterdam. ISBN 444 87576 X.

BlNG, Jon 1984b, "Legal information services: Some trends and
characteristics", in Data Processing and the Law, ed. Colin Camp-
bell, Sweet and Maxwell, London, pp. 29-45. ISBN 421 32000 1.

BlNG, Jon 1987, "Designing text retrieval systems for 'conceptual
searching' ", in Proceedings of the First International Conference on
Artificial Intelligence and Law (ICAIL-87), Boston, Massachusetts,
27-29 May, pp. 43-51. ISBN 89791 230 6.

BlNG, Jon 1989, "The law of the books and the law of the files: Pos-
sibilities and problems of legal information systems", in Advanced
Topics of Law and Information Technology, ed. G.P. V. Vanden-
berghe, Computer/Law Series, no. 3, Kluwer Law and Taxation
Publishers, Deventer, pp. 151-82. ISBN 90 6544 391 6.

BlNG, Jon 1992, [Review of Ashley 1990], Artificial Intelligence and
Law, vol. 1, no. 1, pp. 103-7. ISSN 0924-8463.

BlNG, Jon and SELMER, Knut S. (eds) 1980, 4 Decade of Computers
and Law, Universitetsforlaget, Oslo. ISBN 82 00 05376 8.

BJARUP, Jes 1988, "Kripke's case: Some remarks on rules, their in-
terpretation and application", Rechtstheorie, vol. 19, pp. 39-49.
ISSN 0034-1398.

BLACKSTONE, William 1773, Commentaries on the Laws of England
(fifth edition), Oxford (Clarendon).
Four volumes.

BODEN, Margaret A. 1985, "Panel: Artificial intelligence and legal
responsibility", in Proceedings of the Ninth International Joint
Conference on Artificial Intelligence (IJCAI-85), Los Angeles, Cali-
fornia, 18-23 August, vol. 2, pp. 1267-8.
ISBN 934613 02 8. ISSN 1045-0823.

BORCHGREVINK, Mette and HANSEN, Johannes 1980, "SARA: A sys-
tem for the analysis of legal decisions", in A Decade of Computers
and Law, eds Jon Bing and Knut S. Selmer, Universitetsforlaget,
Oslo, ch. 18, pp. 342-75. ISBN 82 00 05376 8.

BOSWELL, James I960?, The Life of Samuel Johnson, vol. I, Heron
Books, London.
First published in 1791.

346 A pragmatic legal expert system

BOTTOMLEY, Stephen, GUNNINGHAM, Neil and PARKER, Stephen
1991, Law in Context, Federation Press, Leichhardt.
ISBN 186287 0616.

B RANTING, L. Karl 1989, "Representing and reusing explanations
of legal precedents", in Proceedings of the Second International
Conference on Artificial Intelligence and Law (ICAIL-89), Uni-
versity of British Columbia, Vancouver, 13-16 June, pp. 103-10.
ISBN 89791322 1.

BRANTING, L. Karl 1991, "Reasoning with portions of precedents",
in Proceedings of the Third International Conference on Artificial
Intelligence and Law (ICAIL-91), St Catherine's College, Oxford,
25-28 June, pp. 145-54. ISBN 89791 399 X.

BROOKS, Adrian 1990, "Employment", in Australian Commentary on
Halsbury's Laws of England (fourth edition), ed. Garfield Barwick,
Butterworths, Sydney, ch. 56. ISBN 409 30930 3.

BUCHANAN, Bruce G. and HEADRICK, Thomas E. 1970, "Some spec-
ulation about artificial intelligence and legal reasoning", Stanford
Law Review, vol. 23, no. 1, November, pp. 40-62. ISSN 0038-9765.

BUCHANAN, Bruce G. and SHORTLIFFE, Edward H. (eds) 1984,
Rule-Based Expert Systems: The MYCIN Experiments of the Stan-
ford Heuristic Programming Project, Addison- Wesley Series in
Artificial Intelligence, Addison- Wesley, Reading, Massachusetts.
ISBN 201 10172 6.

BURTON, Steven J. 1985, An Introduction to Law and Legal Reasoning,
Little, Brown and Company, Boston. ISBN 316 11786 2.

CAMPBELL, J. A. 1984, "The expert computer and professional neg-
ligence: who is liable?", in Artificial Intelligence: Human Effects,
eds Masoud Yazdani and Ajit Narayanan, Ellis Horwood Series
in Artificial Intelligence, Ellis Horwood, Chichester, pp. 37-51.
ISBN 85312 577 5, 470 20092 8.

CAPPER, Phillip and SUSSKIND, Richard E. 1988, Latent Damage
Law — The Expert System, Butterworths, London.

ISBN 406 02362 X.

CARROLL, Lewis 1982, The Complete Illustrated Works of Lewis Car-
roll, Avenel Books, New York. ISBN 517 38566 X.

ClESIELSKI, Victor B. 1990, Some Experiments in the Use of Expert
Systems for Legislation and Regulations, Technical Report 21, Re}'
Centre for Knowledge Based Systems, Departments of Computer
Science, Royal Melbourne Institute of Technology and The Uni-
versity of Melbourne, 27 April.

Bibliography 347

CLARK, Andrew 1988, [Review of Susskind 1987a], Journal of Law and
Society, vol. 15, no. 4, Winter, pp. 426-30. ISSN 0263-323X.

CLARK, Jon and WEDDERBURN, Kenneth 1983, "Modern labour law:
Problems, functions, and policies", in Labour Law and Industrial Re-
lations: Building on Kahn-Freund, eds Kenneth Wedderburn, Roy
Lewis and Jon Clark, Oxford University Press (Clarendon), Oxford,
ch. 6, pp. 127-242. ISBN 19 825393 1.

CLARKE, Roger 1988, "Legal aspects of knowledge-based technology",
Journal of Information Technology, vol. 3, no. 1, March, pp. 9-16.
ISSN 0268-3962.

CLAUDY, John G. 1972, "A comparison of five variable weighting
procedures", Educational and Psychological Measurement, vol. 32,
no. 2, Summer, pp. 311-22. ISSN 0013-1644.

CONOVER, W.J. 1971, Practical Nonparametric Statistics, John Wiley,
New York. ISBN 0471 16851 3.

Cook, Sandra, HAFNER, Carole D., McCarty, L. Thorne, MELDMAN,
Jeffrey A., PETERSON, Mark, SPROWL, James A., SRIDHARAN,
N.S. and WATERMAN, D.A. 1981, "The applications of artificial
intelligence to law: A survej' of six current projects", AFIPS Con-
ference Proceedings, vol. 50, pp. 689-96. ISSN 0449-1173.

From the National Computer Conference, Chicago, Illinois, 4-7 May 1981.

Cover, T.M. and Hart, P. E. 1967, ;; Nearest neighbor pattern
classification", IEEE Transactions on Information Theory, vol. IT-
13, no. 1, January, pp. 21-7. ISSN 0018-9448.

Creighton, W.B., Ford, W.J. and Mitchell, R.J. 1983, Labour
Law: Materials and Commentary, Law Book Company, Sydney.
ISBN 455 20413 6.

CROSS, Rupert 1961, Precedent in English Law, Clarendon Law Series,
Oxford University Press (Clarendon), London.

CROYDON, Stanley O., Jr 1980, "JURIS: A tool for legal research",
in Legal and Legislative Information Processing, ed. Beth Krevitt
Eres, Greenwood Press, Westport, Connecticut, ch. 11, pp. 163-72.
ISBN 313 21343 7.

D'AMATO, Anthony 1977, "Can/should computers replace judges?",
Georgia Law Review, vol. 11, no. 5, September, pp. 1277-1301.
ISSN 0016-8300.

DEBESSONET, Cary G. and CROSS, George R. 1985, "Representation
of some aspects of legal causality", in Computing Power and Legal
Reasoning, ed. Charles Walter, West Publishing Company, St Paul,
ch. 11, pp. 205-14. ISBN 314 95570 4.

348 A pragmatic legal expert system

DEBESSONET. Cary G. and CROSS, George R. 1986, "Conceptual
retrieval and legal decision-making", in Automated Analysis of
Legal Texts: Logic, Informatics, Law, eds Antonio A. Martino and
Fiorenza Socci Natali, Elsevier Science (North-Holland), Amster-
dam, pp. 219-27. ISBN 444 70111 7.

From the Second International Conference on 'Logic, Informatics, Law*.
Florence, Italy, September 1985.

DEBESSONET, Cary and CROSS, George 1988, "Distinguishing legal
language-types for conceptual retrieval", in Computer Power and
Legal Language: The Use of Computational Linguistics, Artificial
Intelligence, and Expert Systems in the Law, ed. Charles Walter,
Quorum, New York, ch. 14, pp. 155-66. ISBN 89930 306 4.

From the Second Annual Conference on Law and Technology, University
of Houston, 24-28 June 1985.

DlCK, Judith P. 1991, "Representation of legal text for conceptual
retrieval", in Proceedings of the Third International Conference on
Artificial Intelligence and Law (ICAIL-91), St Catherine's College,
Oxford, 25-28 June, pp. 244-53. ISBN 89791 399 X.

DOUGLAS, William O. 1948, "The dissent: A safeguard of democracy",
Journal of the American Judicature Society, vol. 32, December,
pp. 104-7. ISSN 0022-5800.

DRAHOS, Peter and PARKER, Stephen 1992, "Rule following, rule scep-
ticism and indeterminacy in law: A conventional account", Ratio
Juris, vol. 5, no. 1, March, pp. 109-19. ISSN 0952-1917.

DWORKIN, Ronald 1977, Taking Rights Seriously, Duckworth, London.
ISBN 7156 0715 4.

DWORKIN, Ronald 1978, "No right answer?", New York University Law
Review, vol. 53, no. 1, April, pp. 1-32. ISSN 0028-7881.

ELIOT, T. S. 1963, Collected Poems 1909-1962, Faber and Faber, Lon-
don. ISBN 57105549 4.

EVERITT, Brian 1974, Cluster Analysis, Social Science Research Coun-
cil Reviews of Current Research, no. 11, Heinemann Educational
Books, London. ISBN 435 82297 7.

FEIGENBAUM, E. A. 1981, "Expert systems in the 1980s", in Machine
Intelligence, ed. A.H. Bond, Infotech State of the Art Report
Series 9, no. 3, Pergamon Infotech, Maidenhead, Berkshire, pp. 2 19—
29. ISBN 008 028556 2. ISSN 0734-8487.

FEIGENBAUM, Edward A. and McCoRDUCK, Pamela 1983, The
Fifth Generation: Artificial Intelligence and Japan's Computer
Challenge to the World, Addison-Wesley, Reading, Massachusetts.
ISBN 201 11519 0.

Bibliography 349

FlSHER. Ronald A. 1970, Statistical Methods for Research Workers
(fourteenth edition), Hafner, New York.

Fix, E. and HODGES, J.L., Jr 1951, Discriminatory Analysis, Non-
parametric Discrimination, Project 21-49-004, Report 4, United
States Air Force School of Aviation Medicine, Randolph Field,
Texas, February.

FRANK, Jerome 1930, Law and the Modern Mind, Brentano's, New
York.

FRANK, Jerome 1949, Courts on Trial: Myth and Reality in American
Justice, Princeton University Press, Princeton, New Jersey.

GALSWORTHY, John 1941, Ten Famous Plays, Duckworth, London.

GARDNER, Anne von der Lieth 1983, "The design of a legal analysis
program", in Proceedings of the National Conference on Artificial
Intelligence (AAAI-83), Washington, DC, 22-26 August, pp. 114-
18. ISBN 86576 065 9.

GARDNER, Anne von der Lieth 1984, An Artificial Intelligence
Approach to Legal Reasoning, PhD Thesis STAN-CS-85-1045, De-
partment of Computer Science, Stanford University, Stanford,
California, June.

GARDNER, Anne v.d.L. 1985, "Overview of an artificial intelligence
approach to legal reasoning", in Computing Power and Legal Reas-
oning, ed. Charles Walter, West Publishing Company, St Paul,
ch. 13, pp. 247-74. ISBN 314 95570 4.

Gardner, Anne von der Lieth 1987, An Artificial Intelligence
Approach to Legal Reasoning, Artificial Intelligence and Legal Reas-
oning Series, MIT Press (Bradford), Cambridge, Massachusetts.
ISBN 262 07104 5.

Gay, John 1967, Fables, William Andrews Clark Memorial Library,
University of California, Los Angeles.
Reprint of the 1727 (vol. I) and 1738 (vol. II) editions.

GILBERT, W.S. 1977, The Savoy Operas: Being the Complete Text of
the Gilbert and Sullivan Operas as originally produced in the years
1875-1896, Macmillan, London. ISBN 333 06901 3.

GOLD, David I. and SUSSKIND, Richard E. 1986, "Expert systems
in law: A jurisprudential and formal specification approach", in
Automated Analysis of Legal Texts: Logic, Informatics, Law, eds
Antonio A. Martino and Fiorenza Socci Natali, Elsevier Science
(North-Holland), Amsterdam, pp. 625-42. ISBN 444 70111 7.

From the Second International Conference on 'Logic, Informatics, Law*.
Florence, Italy, September 1985.

350 A pragmatic legal expert system

Goldman, Seth R., Dyer, Michael G. and Flowers, M argot 1987,
"Precedent-based legal reasoning and knowledge acquisition in
contract law: A process model", in Proceedings of the First Inter-
national Conference on Artificial Intelligence and Law (ICAIL-87),
Boston, Massachusetts, 27-29 May, pp. 210-21.
ISBN 89791230 6.

GOLDMAN, Sheldon and Sarat, Austin (eds) 1978, American Court
Systems: Readings in Judicial Process and Behavior, W.H. Free-
man, San Francisco. ISBN 7167 0061 1.

GREENLEAF, G.W., Mowbray, A.S. and Lewis, D.P. 1988, Aus-
tralasian Computerised Legal Information Handbook, Butterworths,
Sydney. ISBN 409 49445 3.

GREENLEAF, Graham, Mowbray, Andrew and Tyree, Alan L. 1987,
"Expert systems in law: The DataLex project", in Proceedings of
the First International Conference on Artificial Intelligence and
Law (ICAIL-87), Boston, Massachusetts, 27-29 May, pp. 9-17.
ISBN 89791230 6.

GREENLEAF, Graham, MOWBRAY, Andrew and Tyree, Alan 1991,
"The DataLex legal workstation— integrating tools for lawyers",
in Proceedings of the Third International Conference on Artificial
Intelligence and Law (ICAIL-91), St Catherine's College, Oxford,
25-28 June, pp. 215-24. ISBN 89791 399 X.

GRUNBAUM, W.F. 1988, [Review of Gardner 1987], Computing Re-
views, vol. 29, no. 12, December, p. 639. ISSN 0010-4884.

Haar, Charles M., Sawyer, John P., Jr and CUMMINGS, Stephen J.
1977, "Computer power and legal reasoning: A case stud}' of ju-
dicial decision prediction in zoning amendment cases", American
Bar Foundation Research Journal, no. 3, Summer, pp. 651-768.
ISSN 0361-9486.

HAFNER, Carole D. 1981, "Representation of knowledge in a legal in-
formation retrieval system", in Information Retrieval Research, eds
R.N. Oddy, S.E. Robertson, C.J. van Rijsbergen and P.W. Willi-
ams, Butterworths, London, ch. 9, pp. 139-53. ISBN 408 10775 8.

From a symposium "Research and Development in Information Retrieval",
St John's College. Cambridge, June 1980.

HAFNER, Carole D. 1987, "Conceptual organization of case law know-
ledge bases", in Proceedings of the First International Conference
on Artificial Intelligence and Law (ICAIL-87), Boston, Massachu-
setts, 27-29 May, pp. 35-42. ISBN 89791 230 6.

HARRIS, J. W. 1980, Legal Philosophies, Butterworths, London.
ISBN 406 593612.

Bibliography 35 1

HART, H.L. A. 1961, The Concept of Law, Clarendon Law Series, Ox-
ford University Press (Clarendon), London.

HART, H.L. A. 1983, Essays in Jurisprudence and Philosophy, Oxford
University Press (Clarendon), London. ISBN 19 825387 7.

Hayes-Roth, Frederick, WATERMAN, Donald A. and Lenat, Doug-
las B. (eds) 1983, Building Expert Systems, Teknowledge Series in
Knowledge Engineering, Addison- Wesley, Reading, Massachusetts.
ISBN 201 10686 8.

HEPPLE, Bob 1986, "Restructuring employment rights", The Industrial
Law Journal, vol. 15, pp. 69-83. ISBN 421 37340 7.

HERMAN, Theodor 1980, "WESTLAW: Computerized legal research
program of West Publishing Company", in Legal and Legislative
Information Processing, ed. Beth Krevitt Eres, Greenwood Press,
Westport, Connecticut, ch. 10, pp. 157-61. ISBN 313 21343 7.

HOLMES, O.W. 1897, "The path of the law", Harvard Law Review,
vol. 10, no. 8, 25 March, pp. 457-78. ISSN 0017-811X.

JOHNSON, Peter and MEAD, David 1991, "Legislative knowledge
base systems for public administration — Some practical issues",
in Proceedings of the Third International Conference on Artificial
Intelligence and Law (ICAIL-91), St Catherine's College, Oxford,
25-28 June, pp. 108-17. ISBN 89791 399 X.

JONES, R. 1988, [Review of Gardner 1987], Computers and Law, no. 56,
June, p. 32. ISSN 0140-3249.

KAYTON, Irving 1964, "Can jurimetrics be of value to jurisprudence?",
The George Washington Law Review, vol. 33, no. 1, October,
pp. 287-317. ISSN 0016-8076.

KELMAN, Mark 1987, A Guide to Critical Legal Studies, Harvard Uni-
versity Press, Cambridge, Massachusetts. ISBN 674 36755 3.

KELSEN, Hans 1945, General Theory of Law and State, 20th Cen-
tury Legal Philosophy Series, vol. I, Russell and Russell, New York.
ISBN 8462 0215 8.

KELSEN, Hans 1967, Pure Theory of Law, University of California
Press, Berkeley.

Translation of Reine Rechtslehre (I960, second revised edition).

KERNIGHAN, Brian W. and RITCHIE, Dennis M. 1988, The C Program-
ming Language (second edition), Prentice Hall, Englewood Cliffs,
New Jersey. ISBN 13 110362 8.'

352 A pragmatic legal expert system

KNUTH, Donald E. 1984, The TfiXbook, Addison- Wesley, Reading, Mas-
sachusetts. ISBN 201 13448 9.

KORT, Fred 1963a, "Content analysis of judicial opinions and rules of
law", in Judicial Decision-Making, ed. Glendon Schubert, Interna-
tional Yearbook of Political Behavior Research, vol. 4, Free Press
of Glencoe, New York, ch. vi, pp. 133-97.

KORT, Fred 1963b, "Simultaneous equations and boolean algebra in
the analysis of judicial decisions", Law and Contemporary Problems.
vol. 28, no. 1, Winter, pp. 143-63. ISSN 0023-9186.

Republished in Baade 1963.

KOWALSKI, Andrzej 1991, "Case-based reasoning and the deep struc-
ture approach to knowledge representation", in Proceedings of the
Third International Conference on Artificial Intelligence and Law
(ICAlL-91), St Catherine's College, Oxford, 25-28 June, pp. 21-30.

ISBN 89791 399 X.

KOWALSKI, Robert 1989, "The treatment of negation in logic programs
for representing legislation", in Proceedings of the Second Interna-
tional Conference on Artificial Intelligence and Law (ICAIL-89),
University of British Columbia, Vancouver, 13-16 June, pp. 11-15.
ISBN 89791322 1.

KOWALSKI, Robert and SERGOT, Marek 1985, "Computer represent-
ation of the law", in Proceedings of the Ninth International Joint
Conference on Artificial Intelligence (IJCAI-85), Los Angeles, Cali-
fornia, 18-23 August, vol. 2, pp. 1269-70.
ISBN 934613 02 8. ISSN 1045-0823.

KOWALSKI, Robert and SERGOT, Marek 1990, "The use of logical
models in legal problem solving", Ratio Juris, vol. 3, no. 2, July,
pp. 201-18. ISSN 0952-1917.

KRIPKE, Saul A. 1982, Wittgenstein on Rules and Private Language:
An Elementary Exposition, Basil Blackwell, Oxford.
ISBN 631 13077 2.

KRYGIER, Martin 1986, "Julius Stone: Leeways of choice, legal tradi-
tion and the declaratory theory of law", University of New South
Wales Law Journal, vol. 9, no. 2, pp. 26-38. ISSN 0313-0096.

LAMBERT, Kenneth A. and GRUNEWALD, Mark H. 1989, "LESTER:
Using paradigm cases in a quasi-precedential legal domain", in
Proceedings of the Second International Conference on Artificial
Intelligence and Law (ICAIL-89), University of British Columbia,
Vancouver, 13-16 June, pp. 87-92. ISBN 89791 322 1.

Bibliography 353

LAMBERT, Kenneth A. and GRUNEWALD, Mark H. 1991, "Legal the-
ory and case-based reasoners: The importance of context and the
process of focusing", in Proceedings of the Third International
Conference on Artificial Intelligence and Law (ICAIL-91). St Cath-
erine's College, Oxford, 25-28 June, pp. 191-5. ISBN 89791 399 X.

LAMPORT, Leslie 1986, IATgX: A Document Preparation System, Addi-
son- Wesley, Reading, Massachusetts. ISBN 201 15790 X.

Describes I£T E X 2.09.

LAMPORT, Leslie 1994, UT^X: A Document Preparation System (second
edition), Addison- Wesley, Reading, Massachusetts.
ISBN 20152983 1.

Describes I^TgX 2g.

Lasswell, Harold D. 1955, "Current studies of the decision process:
Automation versus creativity", The Western Political Quarterly.
vol. 8, no. 3, September, pp. 381-99. ISSN 0043-4078.

LAWLOR, Reed C. 1963, "What computers can do: Analysis and pre-
diction of judicial decisions", American Bar Association Journal.
vol. 49, no. 4, April, pp. 337-44. ISSN 0002-7596.

LAWLOR, Reed C. 1968, "Personal stare decisis", Southern California
Law Review, vol. 41, pp. 73-118. ISSN 0038-3910.

LAWLOR, Reed C. 1972, "Fact content of cases and precedent — A mod-
ern theory of precedent", Jurimetrics Journal, vol. 12, no. 4, June,
pp. 245-70. ISSN 0022-6793.

Excerpts from a paper presented at the September 1971 Annual Meeting
of the American Political Science Association.

LAWLOR, Reed C. 1981, "Analysis and prediction of judicial decisions",
in Computers and the Law: An Introductory Handbook (third edi-
tion), ed. Robert P. Bigelow, American Bar Association, pp. 81-4.

LEITH, Philip 1985a, "An IKBS implementation", Software — Practice
and Experience, vol. 15, no. 1, January, pp. 65-86. ISSN 0038-0644.

LEITH, Philip 1985b, Law and Computer Program: The Limits to Logic.
Paper, Department of Jurisprudence, Queen's University, Belfast,
March.

LEITH, Philip 1986a, "Clear rules and legal expert systems", in
Automated Analysis of Legal Texts: Logic, Informatics, Law, eds
Antonio A. Martino and Fiorenza Socci Natali, Elsevier Science
(North-Holland), Amsterdam, pp. 661-79. ISBN 444 70111 7.

From the Second International Conference on 'Logic, Informatics, Law*.
Florence, Italy, September 1985.

354 A pragmatic legal expert system

LEITH, Philip 1986b, [Review of Michie and Johnston 1984, and
Yazdani and Narayanan 1984], Northern Ireland Legal Quarterly,
vol. 37, no. 1, Spring, pp. 97-100. ISSN 0029-3105.

LEITH, Philip 1987, "The Emperor's new expert system", The Modern
Law Review, vol. 50, no. 1, January, pp. 128-32. ISSN 0026-7961.

LEVI, Edward H. 1949, An Introduction to Legal Reasoning, University
of Chicago Press (Phoenix), Chicago.

LLEWELLYN, Karl N. 1931, "Some realism about realism— Responding
to Dean Pound", Harvard Law Review, vol. 44, no. 8, June, pp. 1222-
64. ISSN 0017-811X.

LLEWELLYN, K.N. 1951, The Bramble Bush: On Our Law and its
Study, Oceana Publications, New York.

Llewellyn, Karl N. 1962, Jurisprudence: Realism in Theory and

Practice, University of Chicago Press, Chicago.

LOEVINGER, Lee 1949, "Jurimetrics: The next step forward", Min-
nesota Law Review, vol. 33, no. 5, April, pp. 455-93.
ISSN 0026-5535.

LOEVINGER, Lee 1961, "Jurimetrics: Science and prediction in the field
of law", Minnesota Law Review, vol.46, no. 2, December, pp. 255-
75. ISSN 0026-5535.

LOEVINGER, Lee 1963, "Jurimetrics: The methodology of legal in-
quiry", Law and Contemporary Problems, vol. 28, no. 1, Winter,
pp. 5-35. ISSN 0023-9186.
Republished in Baade 19G3.

LOEVINGER, Lee 1964, "Jurimetrics", in Judicial Behavior: A Reader in
Theory and Research, ed. Glendon Schubert, Rand McNally Polit-
ical Science Series, Rand McNally and Company, Chicago, ch. I.C.7,
pp. 72-6.
Extracts from Loevinger 1949.

LUCAS, E.V. (ed.) 1935, The Letters of Charles Lamb to which are
added those of his sister Mary Lamb, vol. Ill, Dent and Methuen,
London.

McCARTY, L. Thorne 1977, "Reflections on TAXMAN: An experiment
in artificial intelligence and legal reasoning", Harvard Law Review,
vol. 90, no. 5, March, pp. 837-93. ISSN 0017-811X.

McCARTY, L. Thorne 1980a, Some Requirements for a Computer-
Based Legal Consultant, Technical Report LRP-TR-8, Laboratory
for Computer Science Research, Rutgers University, New Brun-
swick, New Jersey, 1 July.
Published, with minor abbreviations, as McCarty 1980b.

B i bliography 355

McCARTY, L. Thome 1980b, "Some requirements for a computer-
based legal consultant", in Proceedings of the First Annual National
Conference on Artificial Intelligence (AAAI-80), Stanford Uni-
versity, 18-21 August, pp. 298-300.

McCARTY, L. Thorne 1980c, "The TAXMAN project: Towards a cog-
nitive theory of legal argument", in Computer Science and Law.
ed. Bryan Niblett, Cambridge University Press, Cambridge, ch. 3,
pp. 23-43. ISBN 521 23451 4.

McCARTY, L. Thorne 1982, "A computational theory of Eisner v.
Macomber", in Artificial Intelligence and Legal Information Sys-
tems., ed. Constantino Ciampi, vol. I, North-Holland, Amsterdam,
pp. 329-55. ISBN 444 86414 8, 444 86413 X (set).

From the International Conference on "Logic, Informatics, Law", Florence,
Italy, April 1981.

McCARTY, L. Thome 1983, "Intelligent legal information systems:
Problems and prospects", Rutgers Computer and Technology Law
Journal, vol. 9, no. 2, pp. 265-94. ISSN 0735-8938.
Republished as McCarty 1984.

McCARTY, L. Thome 1984, "Intelligent legal information systems:
Problems and prospects", in Data Processing and the Law, ed. Colin
Campbell, Sweet and Maxwell, London, pp. 125-51.
ISBN 42132000 1.

McCARTY, L. Thorne 1986, "Permissions and obligations: An in-
formal introduction", in Automated Analysis of Legal Texts: Logic,
Informatics, Law, eds Antonio A. Mart in o and Fiorenza Socci
Natali, Elsevier Science (North-Holland), Amsterdam, pp. 307-37.
ISBN 444 701117.

From the Second International Conference on 'Logic, Informatics, Law",
Florence, Italy, September 1985.

McCARTY, L. Thorne 1989, "A language for legal discourse: I. Basic
features", in Proceedings of the Second International Conference on
Artificial Intelligence and Law (ICAIL-89), University of British
Columbia, Vancouver, 13-16 June, pp. 180-9. ISBN 89791322 1.

McCARTY, L. Thorne 1990, "Artificial intelligence and law: How to
get there from here", Ratio Juris, vol. 3, no. 2, July, pp. 189-200.
ISSN 0952-1917.

McCARTY, L. Thorne 1991, "On the role of prototypes in ap-
pellate legal argument (abstract)", in Proceedings of the Third
International Conference on Artificial Intelligence and Law (ICAIL-
91), St Catherine's College, Oxford, 25-28 June, pp. 185-90.
ISBN 89791 399 X.

356 A pragmatic legal expert system

McCARTY, L. Thome and SRIDHARAN, N.S. 1981, "The repres-
entation of an evolving system of legal concepts: II. Prototypes
and deformations", in Proceedings of the Seventh International
Joint Conference on Artificial Intelligence (IJCAI-81), University
of British Columbia, Vancouver, British Columbia, Canada, 24-28
August, vol. I, pp. 246-53. ISBN 86576 059 4. ISSN 1045-0823.

MACCORMICK, Neil 1978, Legal Reasoning and Legal Theory, Clar-
endon Law Series, Oxford University Press (Clarendon), Oxford.
ISBN 19 876080 9.

MACKAAY, Ejan and ROBILLARD, Pierre 1974, "Predicting judicial de-
cisions: The nearest neighbour rule and visual representation of case
patterns", Datenverarbeitung im Recht, vol. 3, no. 3/4, November,
pp. 302-31. ISSN 0301-2980.

McMillan, John 1991, "Developments under the ADJR Act: The
grounds of review", Federal Law Review, vol. 20, no. 1, pp. 50-82.
ISSN 0067-205X.

Revised version of a paper presented at the Conference "Ten Years of
the Federal Administrative Decisions (Judicial Review) Act", Canberra,
September 1990.

MAGGS, Peter B. and DEBESSONET, Cary G. 1972, "Automated logical
analysis of systems of legal rules", Jurimetrics Journal, vol. 12, no. 3,
March, pp. 158-69. ISSN 0022-6793.

MEHL, Lucien 1959, "Automation in the legal world: From the machine
processing of legal information to the 'law machine' ", in Mechanisa-
tion of Thought Processes, vol. II, Her Majesty's Stationery Office,
London, pp. 755-87.

From a symposium held at the National Physical Laboratory, 24-27
November 1958.

MELDMAN, Jeffrey A. 1977, "A structural model for computer-aided
legal analysis", Rutgers Journal of Computers and the Law, vol. 6,
no. 1, pp. 27-71. ISSN 0735-8938.

MENDELSON, Simon 1989, "An attempted dimensional analysis of the
law governing government appeals in criminal cases", in Proceedings
of the Second International Conference on Artificial Intelligence
and Law (ICAIL-89), Universit}' of British Columbia, Vancouver,
13-16 June, pp. 128-37. ISBN 89791 322 1.

MlCHAELSEN, Robert and MlCHlE, Donald 1983, "Expert systems
in business", Datamation, vol. 29, no. 11, November, pp. 240-6.
ISSN 0011-6963.

MlCHIE, Donald and JOHNSTON, Rory 1984, The Creative Computer:
Machine Intelligence and Human Knowledge, Penguin (Viking),
Harmondsworth. ISBN 670 80060 0.

Bibliography 357

Moles, Robert N. 1987, Definition and Rule in Legal Theory: A
Reassessment of H. L. A. Hart and the Positivist Tradition, Basil
Blackwell, Oxford. ISBN 631 15342 X.

MOLES, Robert N. 1991, "Logic programming — An assessment of its
potential for artificial intelligence applications in law", Journal of
Law and Information Science, vol. 2, no. 2, pp. 137-64.
ISSN 0729-1485.

MOLES, Robert N. and DAYAL, Surendra 1992, "There is more to life
than logic", Journal of Law and Information Science, vol. 3, no. 2,
pp. 188-218. ISSN 0729-1485.

MONTESQUIEU 1973, Be VEsprit des lots, vol. I, Garnier Freres, Paris.
First published in 1748.

MORRIS, Gwen, Cook, Catriona, CREYKE, Robin and Geddes,
Robert 1992, Laying Down the Law: The Foundations of Legal
Reasoning, Research and Writing in Australia and New Zealand
(third edition), Butterworths, Sydney. ISBN 409 30541 3.

MuiR, Frank 1976, The Frank Muir Book: An Irreverent Companion
to Social History, Heinemann, London. ISBN 434 48150 5.

MURPHY, Lionel 1980, "The responsibility of judges", in Law. Politics
and the Labor Movement, ed. Gareth Evans, Legal Service Bul-
letin, Law Faculty, Monash University, Clayton, ch. 1, pp. 2-10.
ISBN 9594727 3.

NAGEL, Stuart 1960, "Using simple calculations to predict judicial
decisions**, American Behavioral Scientist, vol. 4, no. 4, December,
pp. 24-8. ISSN 0002-7642.

NAGEL, Stuart S. 1963, "OfT-the-bench judicial attitudes", in Judicial
Decision- Making, ed. Glendon Schubert, International Yearbook of
Political Behavior Research, vol. 4, Free Press of Glencoe, New
York, ch. ii, pp. 29-53.

NAGEL, Stuart S. 1986, "Microcomputers and judicial prediction", in
Automated Analysis of Legal Texts: Logic, Informatics, Law, eds
Antonio A. Martino and Fiorenza Socci Natali, Elsevier Science
(North-Holland), Amsterdam, pp. 681-702. ISBN 444 70111 7.

From the Second International Conference on 'Logic, Informatics, Law",
Florence, Italy, September 1985.

NAGEL, Stuart S. 1989, Decision- Aiding Software and Legal Decision-
Making: A Guide to Skills and Applications Throughout the Law,
Quorum, New York. ISBN 89930 382 X.

358 A pragmatic legal expert system

Neter, John, WASSERMAN, William and WHITMORE, G. A. 1982,
Applied Statistics (second edition), Allyn and Bacon, Boston.
ISBN 205 07613 0.

NlBLETT, Bryan 1981, "Expert systems for lawyers", Computers and
Law, no. 29, August, pp. 2-4. ISSN 0140-3249.

NlBLETT, Bryan 1988, [Review of Susskind 1987a], Computers and
Law, no. 56, June, pp. 32-3. ISSN 0140-3249.

OSKAMP, A. 1989, "Knowledge representation and legal expert sys-
tems", in Advanced Topics of Law and Information Technology, ed.
G.P. V. Vandenberghe, Computer/Law Series, no. 3, Kluwer Law
and Taxation Publishers, Deventer, pp. 195-211.
ISBN 90 6544 3916.

Oskamp, A., Walker, R. F., Schrickx, J. A. and van den Berg,
P.H. 1989, "PROLEXS, divide and rule: a legal application", in
Proceedings of the Second International Conference on Artificial
Intelligence and Law (ICAIL-89), University of British Columbia,
Vancouver, 13-16 June, pp. 54-62. ISBN 89791 322 1.

PATTISON, Michael and ClESIELSKI, Victor B. 1990, Using Expert Sys-
tems and Conceptual Graphs to Review Legal Contracts, Technical
Report 22, Key Centre for Knowledge Based Systems, Departments
of Computer Science, Royal Melbourne Institute of Technology and
The University of Melbourne, 10 May.

PEARCE, Dennis, CAMPBELL, Enid and HARDING, Don 1987, Austral-
ian Law Schools: A Discipline Assessment for the Commonwealth
Tertiary Education Commission, vol. 1, Australian Government
Publishing Service, Canberra.
ISBN 644 06182 0, 644 06186 3 (set).

PEARCE, D.C. and GEDDES, R. S. 1988, Statutory Interpretation in
Australia (third edition), Butterworths, Sydney.
ISBN 409 49525 5.

POPP, Walter G. and SCHLINK, Bernhard 1975, "JUDITH, a computer
program to advise lawyers in reasoning a case", Jurimetrics Journal,
vol. 15, no. 4, Summer, pp. 303-14. ISSN 0022-6793.

POPPLE, James 1993a, SHYSTER: A Pragmatic Legal Expert System,
PhD thesis, The Australian National University, Canberra, April.
ISBN 7315 1827 6.

POPPLE, James 1993b, SHYSTER: The Program, Technical Report
TR-CS-93-13, Department of Computer Science, Faculty of En-
gineering and Information Technology, The Australian National
University, Canberra, December.

Bibliography 35 £

POUND, Roscoe 1908, "Mechanical jurisprudence", Columbia Law Re-
view, vol. 8, no. 8, December, pp. 605-23. ISSN 0010-1958.

QuiNLAN, J.R. 1986, "Induction of decision trees", Machine Learning.
vol. 1, no. 1, pp. 81-106. ISSN 0885-6125.

QuiNLAN, J.R. 1988, "Induction, knowledge and expert systems", in
Artificial Intelligence Developments and Applications, eds John S.
Gero and Robin Stanton, Elsevier Science (North-Holland), Ams-
terdam, ch. 17, pp. 253-71. ISBN 444 70465 5.

From the Australian Joint Artificial Intelligence Conference, Sydney. Aus-
tralia, 2-4 November 1987.

Quinlan, J.R., Compton, P. J., Horn, K. A. and Lazarus, L. 1987,
"Inductive knowledge acquisition: a case study", in Applications of
Expert Systems, ed. J. Ross Quinlan, Addison- Wesley, Sydnej', ch. 9,
pp. 157-73. ISBN 201 17449 9.

Based on the proceedings of the Second Australian Conference on Applic-
ations of Expert Systems, May 1980.

Rawson, Hugh 1991, A Dictionary of Invective: A Treasury of Curs-
es, Insults, Put-Downs, and Other Formerly Unprintable Terms
from Anglo-Saxon Times to the Present, Robert Hale, London.
ISBN 7090 4399 6.

RlCKETSON, Staniforth 1984, The Law of Intellectual Property, Law
Book Company, Sydney. ISBN 455 20553 1.

RlSSLAND, Edwina L. 1983, "Examples in legal reasoning: Legal
hypotheticals", in Proceedings of the Eighth International Joint
Conference on Artificial Intelligence (IJCAI-83), Karlsruhe, West
Germany, 8-12 August, vol. 1, pp. 90-3.
ISBN 86576 064 0. ISSN 1045-0823.

RlSSLAND, Edwina L. 1985, "Argument moves and hypotheticals", in
Computing Power and Legal Reasoning, ed. Charles Walter, West
Publishing Company, St Paul, ch. 7, pp. 129-43.
ISBN 314 95570 4.

RlSSLAND, Edwina L. 1989, "Dimension-based analysis of hypothet-
icals from Supreme Court oral argument*', in Proceedings of the
Second International Conference on Artificial Intelligence and Law
(ICAIL-89), Universitj' of British Columbia, Vancouver, 13—16
June, pp. 111-20. ISBN 89791 322 1.

RlSSLAND, Edwina L. 1990, "Artificial intelligence and law: Stepping
stones to a model of legal reasoning'*, The Yale Law Journal, vol. 99,
no. 8, June, pp. 1957-81. ISSN 0044-0094.

360 A pragmatic legal expert system

RlSSLAND, Edwina L. and ASHLEY, Kevin D. 1987, "A case-based sys-
tem for trade secrets law", in Proceedings of the First International
Conference on Artificial Intelligence and Law (ICAIL-87), Boston,
Massachusetts, 27-29 May, pp. 60-6. ISBN 89791 230 6.

RlSSLAND, Edwina L. and ASHLEY, Kevin D. 1989, "HYPO: A
precedent-based legal reasoner", in Advanced Topics of Law and In-
formation Technology, ed. G.P. V. Vandenberghe, Computer/Law
Series, no. 3, Kluwer Law and Taxation Publishers, Deventer,
pp. 213-34. ISBN 90 6544 391 6.

RlSSLAND, Edwina L. and SKALAK, David B. 1989a, "Interpreting
statutory predicates", in Proceedings of the Second International
Conference on Artificial Intelligence and Law (ICAIL-89), Uni-
versity of British Columbia, Vancouver, 13—16 June, pp. 46-53.
ISBN 89791322 1.

RlSSLAND, Edwina L. and SKALAK, David B. 1989b, "Combining
case-based and rule-based reasoning: A heuristic approach", in
Proceedings of the Eleventh International Joint Conference on Ar-
tificial Intelligence (IJCAI-89), Detroit, Michigan, 20-25 August,
vol. 1, pp. 524-30. ISBN 1 55860 094 9. ISSN 1045-0823.

RlSSLAND, Edwina L., VALCARCE, Eduardo M. and Ashley, Kevin D.
1984, "Explaining and arguing with examples", in Proceedings of
the National Conference on Artificial Intelligence (AAAI-84)- Uni-
versity of Texas at Austin, 6-10 August, pp. 288-94.
ISBN 86576 080 2.

SANDERS, Kathryn E. 1991, "Representing and reasoning about open-
textured predicates", in Proceedings of the Third International
Conference on Artificial Intelligence and Law (ICAIL-91). St Cath-
erine's College, Oxford, 25-28 June, pp. 137-44.
ISBN 89791 399 X.

SCHMIDT, Frank L. 1971, "The relative efnciencj' of regression and
simple unit predictor weights in applied differential psychology",
Educational and Psychological Measurement, vol. 31, no. 3, Au-
tumn, pp. 699-714. ISSN 0013-1644.

SCHUBERT, Glendon 1963a, "Civilian control and stare decisis in the
Warren Court", in Judicial Decision-Making, ed. Glendon Schubert,
International Yearbook of Political Behavior Research, vol.4, Free
Press of Glencoe, New York, ch. iii, pp. 55-77.

SCHUBERT, Glendon 1963b, "Judicial attitudes and voting beha-
vior: The 1961 term of the United States Supreme Court", Law
and Contemporary Problems, vol. 28, no. 1, Winter, pp. 100-42.
ISSN 0023-9186.
Republished in Baade 1963.

Bibliography 36 1

SCHUBERT, Glendon 1968, "The importance of computer technology to
political science research injudicial behavior", Jurimetrics Journal.
vol. 8, no. 3, March, pp. 56-63. ISSN 0022-6793.

SCHUBERT, Glendon 1975, Human Jurisprudence: Public Law as Polit-
ical Science, University Press of Hawaii, Honolulu.
ISBN 8248 0294 2.

SERGOT, Marek, Cory, Therese, HAMMOND, Peter, KOWALSKI,
Robert, KRIWACZEK, Frank and SADRI, Fariba 1986a, "Formalisa-
tion of the British Nationality Act", in Yearbook of Law Computers
and Technology, ed. Christopher Arnold, vol. 2, Butterworths, Lon-
don, pp. 40-52. ISBN 406 18701 0.

Sergot, M.J., Sadri, F., Kowalski, R. A., Kriwaczek, F., Ham-
mond, P. and CORY, H.T. 1986b, "The British Nationality Act as
a logic program", Communications of the ACM, vol. 29, no. 5, May,
pp. 370-86. ISSN 0001-0782.

SHANNON, David T. and Golshani, Forouzan 1988, "On the automa-
tion of legal reasoning", Jurimetrics Journal, vol. 28, no. 3, Spring,
pp. 305-15. ISSN 0897-1277.

SHERMAN, David M. 1987, "A PROLOG model of the Income Tax Act
of Canada", in Proceedings of the First International Conference on
Artificial Intelligence and Law (ICAIL-87), Boston, Massachusetts,
27-29 May, pp. 127-36. ISBN 89791 230 6.

SHERMAN, David M. 1989, "Expert systems and ICAI in tax law:
Killing two birds with one AI stone", in Proceedings of the Se-
cond International Conference on Artificial Intelligence and Law
(ICAIL-89), University of British Columbia, Vancouver, 13—16
June, pp. 74-80. ISBN 89791 322 1.

SIMPSON, Brian 1986, "The common law and legal theory", in Legal
Theory and Common Law, ed. William Twining, Basil Blackwell,
Oxford, ch. 2, pp. 8-25. ISBN 631 14477 3.

SKALAK, David B. 1989, "Taking advantage of models for legal
classification", in Proceedings of the Second International Confer-
ence on Artificial Intelligence and Law (ICAIL-89), University of
British Columbia, Vancouver, 13-16 June, pp. 234-41.
ISBN 89791322 1.

SKALAK, David B. and RlSSLAND, Edwina L. 1991, "Argument moves
in a rule-guided domain", in Proceedings of the Third International
Conference on Artificial Intelligence and Law (ICAIL-91), St Cath-
erine's College, Oxford, 25-28 June, pp. 1-11. ISBN 89791 399 X.

362 A pragmatic legal expert system

SKALAK, David B. and RisSLAND, Edwina L. 1992, "Arguments and
cases: An inevitable intertwining", Artificial Intelligence and Law.
vol. 1, no. 1, pp. 3-44. ISSN 0924-8463.

SMITH, Ian 1992, "Emplojanent", in Halsbury's Laws of England
(fourth edition reissue), ed. Lord Hailsham of St Marylebone,
vol. 16, Butterworths, London.
ISBN 6 406 03465 5, 406 03400 1 (set).

SNEATH, Peter H.A. and SOKAL, Robert R. 1973, Numerical Tax-
onomy: The Principles and Practice of Numerical Classification,
W. H. Freeman, San Francisco. ISBN 7167 0697 0.

SPAETH, Harold J. 1963, "Warren Court attitudes toward business:
The 4 B : scale", in Judicial Decision-Making, ed. Glendon Schubert,
International Yearbook of Political Behavior Research, vol.4, Free
Press of Glencoe, New York, ch. iv, pp. 79-108.

SPENGLER, Joseph J. 1963, "Machine-made justice: Some implica-
tions'', Law and Contemporary Problems, vol. 28, no. 1, Winter,
pp. 36-52. ISSN 0023-9186.

Republished in Baade 1963.

SPROWL, James A. 1979, "Automating the legal reasoning process:
A computer that uses regulations and statutes to draft legal
documents", American Bar Foundation Research Journal, no. 1,
Winter, pp. 1-81. ISSN 0361-9486.

STAMPER, Ronald 1980, "LEGOL: Modelling legal rules by computer",
in Computer Science and Law, ed. Bryan Niblett, Cambridge Uni-
versity Press, Cambridge, ch. 4, pp. 45-71. ISBN 521 23451 4.

STAMPER, Ronald 1986, "A non-classical logic for based on the
structures of behaviour [sic]", in Automated Analysis of Legal
Texts: Logic, Informatics, Law, eds Antonio A. Mart in o and
Fiorenza Socci Natali, Elsevier Science (North-Holland), Amster-
dam, pp. 115-39. ISBN 444 70111 7.

From the Second International Conference on 'Logic, Informatics, Law',
Florence, Italy, September 1985.

STAMPER, Ronald, Tagg, Clare, Mason, Peter, Cook, Sandra and
MARKS, Jo 1982, "Developing the LEGOL semantic grammar",
in Artificial Intelligence and Legal Information Systems, ed. Con-
stantino Ciampi, vol. I, North- Holland, Amsterdam, pp. 357-79.
ISBN 444 86414 8, 444 86413 X (set).

From the International Conference on "Logic, Informatics, Law", Florence,
Italy, April 1981.

Bibliography 363

STANLEY, Karlyn D. 1980, "LEXIS: Legal research and litigation
support", in Legal and Legislative Information Processing, ed.
Beth Krevitt Eres, Greenwood Press, Westport, Connecticut, ch. 9,
pp. 149-56. ISBN 313 21343 7.

STONE, Julius 1964a, Legal System and Lawyers' Reasonings, M ait-
land, Sydnej^.

STONE, Julius 1964b, "Man and machine in the search for justice",
Stanford Law Review, vol. 16, May, pp. 515-60. ISSN 0038-9765.

STONE, Julius 1966, Social Dimensions of Law and Justice, Maitland,
Sydney.

STONE, Julius 1985, Precedent and Law: Dynamics of Common Law
Growth, Butterworths, Sydney. ISBN 409 49304 X.

SUSSKIND, Richard E. 1986, "Expert systems in law: A jurisprudential
approach to artificial intelligence and legal reasoning", The Modern
Law Review, vol. 49, no. 2, March, pp. 168-94. ISSN 0026-7961.

SUSSKIND, Richard E. 1987a, Expert Systems in Law: A Jurispru-
dential Inquiry, Oxford University Press (Clarendon), Oxford.
ISBN 19 825582 9.

SUSSKIND, Richard E. 1987b, "Expert systems in law: Out of the
research laboratory and into the marketplace", in Proceedings
of the First International Conference on Artificial Intelligence
and Law (ICAIL-87), Boston, Massachusetts, 27-29 May, pp. 1-8.
ISBN 89791230 6.

SUSSKIND, Richard 1989a, "Pragmatism and purism in artificial in-
telligence and legal reasoning", AI and Society, vol. 3, no. 1,
January-March, pp. 28-38. ISSN 0951-5666.

SUSSKIND, Richard E. 1989b, "The Latent Damage System: A jur-
isprudential analysis", in Proceedings of the Second International
Conference on Artificial Intelligence and Law (ICAIL-89), Uni-
versity of British Columbia, Vancouver, 13-16 June, pp. 23-32.
ISBN 89791322 1.

SUSSKIND, Richard E. 1990, [Review of Gardner 1987], in Yearbook of
Law Computers and Technology, ed. Kenneth Russell, vol. 4, But-
terworths, London, pp. 221-8. ISBN 406 18703 7.

TANENHAUS, Joseph, SCHICK, Marvin, MURASKIN, Matthew and
ROSEN, Daniel 1963, "The Supreme Court's certiorari jurisdiction:
Cue theory", in Judicial Decision-Making, ed. Glendon Schubert,
International Yearbook of Political Behavior Research, vol. 4, Free
Press of Glencoe, New York, ch. v, pp. 111-32.

364 A pragmatic legal expert system

TAPPER, Colin 1963, "Lawyers and machines", The Modern Law Re-
view, vol. 26, no. 2, March, pp. 121-37. ISSN 0026-7961.

TAPPER, Colin 1973, Computers and the Law, Law in Context, Weiden-
feld and Nicolson, London. ISBN 297 76571 X.

TAPPER, Colin 1984, "An experiment with citation vectors", in Data
Processing and the Law, ed. Colin Campbell, Sweet and Maxwell,
London, pp. 90-109. ISBN 421 32000 1.

THOMAS, Bart T. 1979, "Unauthorized practice and computer aided
legal analysis systems", Jurimetrics Journal, vol. 20, no. 1, Fall,
pp. 41-51. ISSN 0897-1277.

TORSUN, I.S. 1987, "PAYE: A tax expert system", in Research and
Development in Expert Systems III, ed. M.A. Bramer, British
Computer Society Workshop Series, Cambridge University Press,
Cambridge, pp. 69-80. ISBN 521 34145 X.

From Expert Systems "86: the Sixth Annual Technical Conference of the
British Computer Society Specialist Group on Expert Systems, Brighton,
15-18 December 1980.

TURING, A.M. 1950, "Computing machinery and intelligence", Mind,
vol. 59, no. 236, October, pp. 433-60. ISSN 0026-4423.

TwiSS, Horace 1844, The Public and Private Life of Lord Chancellor
Eldon, with Selections from his Correspondence (second edition),
vol. Ill, John Murray, London.

TYREE, Alan L. 1977, "The geometiy of case law", Victoria University
of Wellington Law Review, vol. 8, pp. 403-20. ISSN 0042-5117.

TYREE, Alan L. 1980, "Can a 'deterministic* computer judge overrule
himself?", Rutgers Journal of Computers, Technology and the Law,
vol. 7, no. 2, pp. 381-4. ISSN 0735-8938.

TYREE, Alan L. 1981, "Fact content analysis of case law: Methods
and limitations", Jurimetrics Journal, vol. 22, no. 1, Fall, pp. 1-33.
ISSN 0897-1277.'

TYREE, A. 1985, "FINDER: An expert system", in Proceedings of the
Fortieth Annual Conference of the Australasian Universities Law
Schools Association, University of Adelaide, 26-29 August.

TYREE, Alan 1986, "Expert S3 f stems and the law: Will justice fall to
bits?", Current Affairs Bulletin, vol. 62, no. 10, March, pp. 13-18.
ISSN 0011-3182.

Bibliography 36 5

TYREE, Alan 1989, Expert Systems in Law, Prentice Hall. New York.
ISBN 13 295650 0.

TYREE, Alan L. 1992, "The logic programming debate", Journal of Law
and Information Science, vol. 3, no. 1, pp. 111-15. ISSN 0729-1485.

TYREE, Alan L., GREENLEAF, Graham and MOWBRAY, Andrew 1988,
"Legal reasoning: the problem of precedent", in Artificial Intelli-
gence Developments and Applications, eds John S. Gero and Robin
Stanton, Elsevier Science (North-Holland), Amsterdam, ch. 16,
pp. 231-47. ISBN 444 70465 5.

From the Australian Joint Artificial Intelligence Conference, Sydney. Aus-
tralia, 2-4 November 1987.

TYREE, Alan L., GREENLEAF, Graham and Mo[w]BRAY, Andrew
1989, "Generating legal arguments", Know ledge- Based Systems,
vol. 2, no. 1, March, pp. 46-51. ISSN 0950-7051.

ULMER, S. Sidney 1963, "Quantitative analysis of judicial processes:
Some practical and theoretical applications", Law and Contempor-
ary Problems, vol. 28, no. 1, Winter, pp. 164-84. ISSN 0023-9186.

Republished in Baade 1963.

ULMER, S. Sidney 1964, "Homeostasis in the Supreme Court", in Ju-
dicial Behavior: A Reader in Theory and Research, ed. Glendon
Schubert, Rand McNaliy Political Science Series, Rand McNally
and Company, Chicago, ch. II. B. 6, pp. 162-80.

van Opdorp. G. J., Walker, R.F., Schrickx, J. A., Groendijk, C.
and VAN DEN BERG, P. H. 1991, "Networks at work: A connection-
ist approach to non-deductive legal reasoning", in Proceedings of the
Third International Conference on Artificial Intelligence and Law
(ICAIL-91), St Catherine's College, Oxford, 25-28 June, pp. 278-
87. ISBN 89791 399 X.

VOSSOS, George, ZELEZNIKOW, John, DILLON, Tharam and VOSSOS.
Vivian 1991, "An example of integrating legal case based reasoning
with object-oriented rule-based systems: IKBALS II", in Proceed-
ings of the Third International Conference on Artificial Intelligence
and Law (ICAIL-91), St Catherine's College, Oxford, 25-28 June,
pp. 31-41. ISBN 89791 399 X.

Walker, R.F., Zeinstra, P.G. M. and van den Berg, P.H. 1989,
"A model to model knowledge about knowledge or implementing
meta-knowledge in PROLEXS", in Advanced Topics of Law and In-
formation Technology, ed. G.P. V. Vandenberghe, Computer/Law
Series, no. 3, Kluwer Law and Taxation Publishers, Deventer,
pp. 235-60. ISBN 90 6544 391 6.

366 A pragmatic legal expert system

Waterman. Donald A., Paul, Jody and Peterson, Mark 1987,
"Expert systems for legal decision making", in Applications of Ex-
pert Systems, ed. J. Ross Quinlan, Addison- Wesley, Sydney, ch. 2,
pp. 23-47. ISBN 201 17449 9.

Based on the proceedings of the Second Australian Conference on Applic-
ations of Expert Systems, May 1986.

Waterman, D. A. and Peterson, Mark 1980, "Rule-based models of
legal expertise", in Proceedings of the First Annual National Con-
ference on Artificial Intelligence (AAAI-80), Stanford University,
18-21 August, pp. 272-5.

WEIZENBAUM, Joseph 1976, Computer Power and Human Reason:
From Judgment to Calculation. W.H. Freeman, San Francisco.
ISBN 7167 0464 1.

WlENER, Frederick Bernays 1962, "Decision prediction by computers:
Nonsense cubed — and worse", American Bar Association Journal.
vol.48, no. 11, November, pp. 1023-8. ISSN 0002-7596.

WILLIAMS, W.T. 1971, "Principles of clustering", Annual Review of
Ecology and Systematica, vol. 2, pp. 303-26. ISSN 0066-4162.

WlLLICK, Marshal S. 1985, "Professional malpractice and the unau-
thorized practice of professions: Some legal and ethical aspects of
the use of computers as decision-aids", in Computing Power and
Legal Reasoning, ed. Charles Walter, West Publishing Company,
St Paul, ch. 28, pp. 817-63. ISBN 314 95570 4.

WlRTH, Niklaus 1977, "What can we do about the unnecessary di-
versity of notation for syntactic definitions?", Communications of
the ACM, vol. 20, no. 11, November, pp. 822-3. ISSN 0001-0782.

WITTGENSTEIN, Ludwig 1974, Philosophical Investigations (third edi-
tion), Basil Blackwell, Oxford. ISBN 631 14670 9.

Translation of Philosophische Untersuchungen (1953).

WOLSTENHOLME, David E. 1989, "Amalgamating regulation- and case-
based advice systems through suggested answers", in Proceedings of
the Second International Conference on Artificial Intelligence and
Law (ICAIL-89), University of British Columbia, Vancouver, 13-16
June, pp. 63-7. ISBN 89791 322 1.

YABLON, Charles M. 1987, "Law and metaphysics", The Yale Law
Journal, vol. 96, no. 3, January, pp. 613-36. ISSN 0044-0094.

Review of Kripke 1982.

YAZDANI, Masoud and NARAYANAN, Ajit (eds) 1984, Artificial In-
telligence: Human Effects, Ellis Horwood Series in Artificial
Intelligence, Ellis Horwood, Chichester.
ISBN 85312 577 5, 470 20092 8.

Cases

Bold page numbers refer to summaries of case details. Italicized
page numbers refer to SHYSTER output.

ALU Records Inc. v. Audio Magnetics Inc. (UK) Ltd [1979] FSR 1:

180, 182-6
Ackroyd v. Whitehouse (Director of National Parks and Wildlife

Service) (1985) 2 NSWLR 239: 223-4, 231, 253
Adelaide Corporation v. Australasian Performing Right Association

Ltd (1928) 40 CLR481: 236
AG v. Ng: see Attorney-General of Hong Kong v. Ng Yuen Shiu
Ainsworth v. Criminal Justice Commission (1992) 175 CLR 564: 217,

227, 231, 240, 256, 278-88
Amoco Production Co. v. Lindley (1980) 609 P 2d 733: 123
AMP v. Chaplin: see Australian Mutual Provident Societ}' v. Chaplin
Annetts v. McCann (1990) 170 CLR 596: 213, 222-3, 227, 239, 256,

279-80, 282, 284
APRA v. Canterbury-Bankstown: see Australasian Performing Right

Association Ltd v. Canterbury-Bankstown League Club Ltd
APRA v. Jain: see Australasian Performing Right Association Ltd v.

Jain
APRA v. Miles: see Australasian Performing Right Association Ltd v.

Miles
Armory v. Delamirie (1722) 1 Str 505; 93 ER 664: 165, 168, 235
Attorney-General of Hong Kong v. Ng Yuen Shiu [1983] 2 AC 629:

213, 220
ATWU v. Monaro: see Australian Timber Workers Union v. Monaro

Sawmills Pt}' Ltd
Australasian Performing Right Association Ltd v. Canterbury-
Bankstown League Club Ltd [1964-65] NSWR 138: 180
Australasian Performing Right Association Ltd v. Jain (1990) 26 FCR

53: 186-7, 231, 236-7, 256, 263-70
Australasian Performing Right Association Ltd v. Miles [1962] NSWR

405: 180, 183-8, 265-70

367

368 A pragmatic legal expert system

Australian Mutual Provident Society v. Chaplin (1978) 18 ALR 385:

153, 157, 195, 197, 199, 292~4, 316-17, 337-8
Australian Timber Workers Union v. Monaro Sawmills Pty Ltd

(1980) 29 ALR 322: 194, 305-6

B v. B [1978] Fam 26: 13

Barro Group Pty Ltd v. Fraser [1985] VR 577: 199

Betts v. Brady (1942) 316 US 455: 20

Bread Manufacturers of New South Wales v. Evans (1981) 180 CLR

404: 119, 213, 218, 225-6, 284~5
Bridges v. Hawkesworth (1851) 21 LJQB 75: 123, 165, 168, 171-4,

234, 257-62
Bropho v. Western Australia (1990) 21 ALD 730: 209
Building Workers* Industrial Union of Australia v. Odco Pt} r Ltd

(1991) 29 FCR 104: 148, 150-1, 153-7, 200-2, 206, 231, 238,

248-9, 253, 256, 290, 291-4, 323-39
BWIU v. Odco: see Building Workers' Industrial Union of

Australia v. Odco Pty Ltd

Cam and Sons Pty Ltd v. Sargent (1940) 14 ALJ 162: 154, 194, 199,

292-3, 302, 303-4, 335-6
Cantliff v. Jenkins [1978] Fam 47: 13, 53
CBS Inc. v. Ames Records and Tapes Ltd [1982] Ch 91: 181-2, 184-6,

231
CBS Songs Ltd v. Amstrad Consumer Electronics Pic [1988] AC 1013:

186, 231
CCSU v. Minister: see Council of Civil Service Unions v. Minister for

the Civil Service
City of London Corporation v. Appleyard [1963] 1 WLR 982: 169,

173-4, 235, 258-60, 261-3
Cole v. Cunningham (1983) 49 ALR 123: 214, 222-3, 225-7, 231,

238, 253, 285, 287-8
Commissioner of Police v. Tanos (1958) 98 CLR 383: 213, 218-20,

224, 239
Council of Civil Service Unions v. Minister for the Civil Service [1985]

AC 374: 213, 223-5
Crown Industries Inc. v. Kawneer Co. (1971) 335 F Supp 749: 123

Daniels v. R. White and Sons Ltd [1938] 4 All ER 258: 53

Davis v. Johnson [1979] AC 264: 53

Durayappah v. Fernando [1967] 2 AC 337: 208, 213, 239

Eisner v. Macomber (1920) 252 US 189: 55, 355
Elwes v. Brigg Gas Co. (1886) 33 Ch D 562: 165, 169

Cases 369

FAI Insurances Ltd v. Winneke (1982) 151 CLR 342: 208, 213, 222,

224-5
Falcon v. Famous Players Film Co. [1926] 2 KB 474: 176, 180, 182-6,

188, 263
FCT v. Thompson: see Federal Commissioner of Taxation v.

J. Walter Thompson (Australia) Ply Ltd
Federal Commissioner of Taxation v. J. Walter Thompson (Australia)

Pty Ltd (1944) 69 CLR 227: 194, 304~5
Ferguson v. John Dawson and Partners (Contractors) Ltd [1976] 1

WLR 1213: 153-4, 156-7, 195, 197-8, 200-2, 271-8, 291-3,

307-8, 333-4, 336-9

Gideon v. Wainwright (1963) 372 US 335: 20-1

Hannah v. Peel [1945] KB 509: 165, 169, 262-3

Haoucher v. Minister for Immigration and Ethnic Affairs (1990) 169

CLR 648: 208, 213, 218, 221, 225, 238
Heatley v. Tasmanian Racing and Gaming Commission (1977) 137

CLR 487: 214, 218, 221-3, 231
Hibbert v. McKiernan [1948] 2 KB 142: 235

Hodgens v. Gunn, ex parte Hodgens [1990] 1 Qd R 1: 224-5, 231
Humberstone v. Northern Timber Mills (1949) 79 CLR 389: 151-2,

154-5, 157, 195, 199, 201, 248-9, 254, 291-3, 312-13, 332-4, 336

Kioa v. West (1985) 159 CLR 550: 207, 213, 221, 278

London v. Apple3 f ard: see City of London Corporation v. Appleyard

Mabo v. Queensland (1988) 166 CLR 186: 51

Mabo v. Queensland [No. 2] (1992) 175 CLR 1: 51

Mclnnes v. Onslow-Fane [1978] 1 WLR 1520: 213, 218-19, 222-4,

227-8, 239, 279-80
Macrae v. Attorney-General for New South Wales (1987) 9 NSWLR

268: 213, 239, 281
Marine Hull and Liability Insurance Co. Ltd v. Hurford (1985) 10

FCR234: 209
Marine Hull and Liability Insurance Co. Ltd v. Hurford (1986) 10

FCR476: 209, 213, 226, 239
Massey v. Crown Life Insurance Co. [1978] 1 WLR 676: 154, 195,

197-8, 201-2, 317-18
Mellor v. Australian Broadcasting Commission [1940] AC 491: 180,

187, 264-8
Minister for Arts, Heritage and Environment v. Peko-Wallsend Ltd

(1987) 15 FCR274: 208, 213, 218, 221-7, 256, 278, 283-5, 286-8
Moffatt v. Kazana [1969] 2 QB 152: 160, 169-70

370 A pragmatic legal expert system

Narich Pty Ltd v. Commissioner of Pay-roll Tax (1983) 50 ALR 417:

121, 162, 196-7, 198, 231
Nashua Australia Pty Ltd v. Channon (1981) 36 ALR 215: 213, 239
Northern Securities Co. v. United States (1904) 193 US 197: 163

Parker v. British Airways Board [1982] QB 1004: 165, 171-2, 173-^,

231, 241, 256, 257-63
Peko-Wallsend Ltd v. Minister for Arts, Heritage and Environment

(1986) 13 FCR 19: 283, 286
Performing Right Society Ltd v. Bradford Corporation [1917-23]

MacG Cop Cas 309: 180
Performing Right Society Ltd v. Ciryl Theatrical Syndicate Ltd

[1924] 1 KB 1: 176, 180, 267-8
Performing Right Society Ltd v. Mitchell and Booker (Palais de

Danse) Ltd [1924] 1 KB 762: 195, 201, 309, 310-11
Porter, Re; Re Transport Workers Union of Australia (1989) 34 IR

179: 162, 199-200, 231-2, 238, 256, 270-8
Price v. Grant Industries Pty Ltd (1978) 21 ALR 388: 192, 194, 314,

31516
PRS v. Bradford: see Performing Right Societj' Ltd v. Bradford

Corporation
PRS v. Ciryl: see Performing Right Society Ltd v. Ciryl Theatrical

Syndicate Ltd
PRS v. Palais de Danse: see Performing Right Society Ltd v. Mitchell

and Booker (Palais de Danse) Ltd

Queensland v. Commonwealth (1977) 139 CLR 585: 51
Queensland Stations Pty Ltd v. Federal Commissioner of Taxation
(1945) 70 CLR 539: 137, 141, 194, 313-14

R. v. Hampden (1637) 3 How St Tr 825: 61

RCA Corporation v. John Fairfax and Sons Ltd [1981] 1 NSWLR 251:

180, 185, 187-8, 236, 264~5. 266-7, 270
Ready Mixed Concrete (South East) Ltd v. Minister of Pensions and

National Insurance [1968] 2 QB 497: 153, 155, 160, 194, 199-200,

272-8, 293-4, 320-1, 337, 338-9
Ridge v. Baldwin [1964] AC 40: 217

SA v. O'Shea: see South Australia v. O'Shea

Salemi v. MacKellar [No. 2] (1977) 137 CLR 396: 107, 123, 208, 214,

220-1, 229, 231
Schmidt v. Secretary of State for Home Affairs [1969] 2 Ch 149: 220
South Australia v. O'Shea (1987) 163 CLR 378: 213, 218, 220-1, 224,

239, 280-1
South Staffordshire Water Co. v. Sharman [1896] 2 QB 44: 165, 169,

171, 173, 235, 258-9, 2601

Cases 371

Stevens v. Brodribb Sawmilling Co. Pty Ltd (1986) 160 CLR 16:

199-201, 206, 231
Stevenson Jordan and Harrison Ltd v. Macdonald and Evans [1952] 1

TLR 101: 154, 157, 195, 201, 292-3, 308-9, 318, 319-20
Structural Dynamics Research Corporation v. Engineering Mechanics

Research Corporation (1975) 401 F Supp 1102: 123

Twist v. Randwick Municipal Council (1976) 136 CLR 106: 123,
209-10, 214, 217, 219-20, 223, 229, 231, 239

University of New South Wales v. Moorhouse (1975) 133 CLR 1: 176,

180, 182, 185, 187, 236, 254, 263, 269-70
UNSW v. Moorhouse: see University of New South Wales v.

Moorhouse
USM Corporation v. Marson Fastener Corporation (1979) 379 Mass

90; 393 NE 2d 895: 123

Vestry of St James and St John, Clerkenwell v. Feary (1890) 24 QBD

703: 217
Vigneux v. Canadian Performing Right Society Ltd [1945] AC 108:

182

WA v. Bropho: see Western Australia v. Bropho

WEA International Inc. v. Hanimex Corporation Ltd (1987) 17 FCR

274: 183, 185-6, 231, 236, 241
Western Australia v. Bropho (1991) 5 WAR 75: 210, 225-6, 231, 239,

254
Western Australia v. Commonwealth (1975) 134 CLR 201: 51
Western Australia v. Commonwealth (1995) 183 CLR 373: 51
Winstone v. Wurlitzer Automatic Phonograph Co. of Australia Pty

Ltd [1946] VLR 338: 180, 182
Wright v. Attorney-General for Tasmania (1954) 94 CLR 409: 199

Zuijs v. Wirth Brothers Pty Ltd (1955) 93 CLR 561: 119, 195, 301-2

Statutes

Italicized page numbers refer to SHYSTER output.

Aboriginal Heritage Act 1972 (WA): 225

Administrative Decisions (Judicial Review) Act 1977 (Cth): 356

Animals Protection Act 1925 (Qld)

s. 11(4): 224

s. 19(2): 224

British Nationality Act 1981 (UK): 30-1, 38, 361

s. 1(2): 175-6, 264, 268

s. 2(1): 265

s. 2(3): 236, 310

s.5(l)(b): 309, 319
Copyright Act 1912 (Cth): 175, 180, 236, 264~5
Copyright Act 1956 (UK)

s. 1(2): 182
Copyright Act 1968 (Cth): 173, 175-7, 180, 183, 190

s. 13(2): 175, 190, 265

s. 36(1): 175, 190, 269

s. 39a: 269

s. 101(1): 175, 190
Coroner's Act 1920 (WA): 279, 282
Criminal Law Consolidation Act 1935 (SA)

s. 77a(7a): 281

Domestic Violence and Matrimonial Proceedings Act 1976 (UK)
s. 13(2): 52

Income Tax Act (Canada): 361

Justices Act 1902 (NSW): 281

373

374 A pragmatic legal expert system

Local Courts Act 1982 (NSW): 281
Local Government Act 1919 (NSW): 239

s. 317b: 209, 217
Long Service Leave Act 1967 (SA): 317, 337

Migration Act 1958 (Cth): 220
s. 18: 220

National Insurance Act 1965 (UK)

s. 1(2): 272, 321, 338
National Parks and Wildlife Act 1974 (Cth)

s. 134(1): 223

s. 135: 223
National Parks and Wildlife Conservation Act 1975 (Cth)

s. 8B: 283, 286

Pay-roll Tax Act 1971 (NSW): 197

Pay-roll Tax Assessment Act 1941 (Cth): 305, 314

Prices Regulation Act 1948 (NSW): 284

Racing and Gaming Act 1952 (Tas)
s. 39(3): 221

Supplementary Benefits Act 1976 (UK): 30

Trade Practices Act 1974 (Cth)

s.45D: 200
Trade Union and Labour Relations Act 1974 (UK): 318

Worker's Compensation Act 1926 (NSW): 302
Worker's Compensation Act 1928 (Vic)

s. 3: 313, 332
World Heritage Properties Conservation Act 1983 (Cth)

s. 3(2): 283, 286

Index

Bold page numbers are more important references. Italicized page
numbers refer to SHYSTER output.

abstraction, level of

appropriate for legal expert

systems: 7, 44, 47, 49-50,
63, 245
at which open-textured

concepts fully denned:
68-9, 71-3
accuracy confused with precision:

50, 115
AI: see artificial intelligence
analogical reasoning: 43, 68-9.

99, 113
analysis systems: see legal

analysis systems
ANF: 35-6

areas: 70-2, 74-5, 126, 133, 135
Affected: 210-11, 212, 214,

215-16, 228, 240-1, 282-5
Authorization: 124, 177-81,

182, 184, 190, 206, 236-7,
251, 263-70
Employee: 119, 122-4, 139,
145-tf, 148, 150-1, 157,
178-9, 192-6, 198, 206,
236, 238, 270-8, 291-339
Expectation: 136, 140, 210,
212-13, 214, 215-16,
217, 240-1, 285-8

areas (continued)

external: 72, 136, 140, see also

attributes, external;

linkage between areas in

case base
forward and recursive

references: 160
Finder: 167-71, 174, 235,

257-63
Natural: 119, 210-11, 212-14,

215-16, 219, 228-9,

239-41, 278-81
argument: 41, 67-8, 73, 78, 92,

106
and prediction: 41, 49, 65, 78
choosing cases for use in:

93-5, 106, 230-2, 246-8
for other results: 153-4, 248
for the nearest result: 151-3,

248
using cases in: 37, 68, 95-9,

107
with hypotheticals: 39, 58, 64,

92, 99-100, 111-12, 115,

122, 130, 151, 155-6,

161-2, 248, 259-63,

266-70, 275-8, 280-1,

284-5, 288, 293-4, 334~9,