The Role of Semantics in Legal Expert Systems and Legal Reasoning*


Ratio Juris. Vol. 4 No. 2 July 1991 (219-44) 
copyright 0 Ronald Stamper 1991 

NOTES DISCUSSION BOOK REVIEWS 

The Role of Semantics in Legal Expert Systems and 
Legal Reasoning* 

RONALD K. STAMPER 

1. Semantics is Central to Legal Reasoning 

The consensus among legal philosophers is probably that rule-based legal expert 
systems leave much to be desired as aids in legal decision-making. Why? What can 
we do about it? 

A bureaucrat administering some set of complex rules will ascertain the facts and 
apply the rules to them in order to discover their consequences for the case in hand. 
This process of deductive reasoning is characteristically bureaucratic.' 

If the client or subject of the decision, after he has checked the deductions, does not 
like the apparent consequences of the rules, he will question their interpretation. This 
is not a deductive process. He will examine the meanings of the words in the rules and 
those used to characterise his case, looking for adjustments that lead to a more 
favourable decision. 

* This paper is based upon an invited contribution in May 1989 to the Conference on Legal 
Expert Systems organised by Prof. Enrico Pattaro as a part of the Ninth Centennial Celebrations 
of the University of Bologna. The contributions to this research programme by past and present 
members of the team are gratefully acknowledged, especially those by Peter Mason, Susan 
Jones, Clare Tagg, Sandra Cook, Martin Kolkmann and Liu Kechen. 

Weber's characterisation of bureaucracy is found in his Essays in Sociology where, in the 
EngIish translation (1946), he says: "The fully developed bureaucratic mechanism compares 
with other organisations exactly as does the machine with the non-mechanical modes of 
production." This allows the administration to be performed "precisely, unambiguously, 
continuously and with as much speed as possible." He stresses the "objective [. . .] discharge 
of business according to calculable rules" which are "of paramount importance for modem 
bureaucracy. " 



220 Ronald K .  S t a m p e r  

If he turns to an expert for advice, he will call in a lawyer. Of course it may be worth 
rechecking the deductions but that is a relatively trivial part of his skilL2 The lawyer 
applies his real expertise when he looks for variant reading of the rules, from those 
readings already made by the bureaucracy, to readings that might be made by a judge 
or tribunal. The dispute will only be resolved when the inclusion or exclusion of 
different rules has been settled and when the precise meanings of the included rules 
are agreed. Of course, the lawyer will also help his client by managing the procedural 
options in his favour. The procedures are also governed by rules and we have but 
another decision problem that divides into issues of bureaucratic choice and issues of 
interpretation. In his shoes, I should not be pleased with a lawyer who can reason 
about the deductive part of these problems but not the interpretive issues. 

So-called “legal expert systems” that fail to handle the problems of interpretation 
do not deserve the epithet “ e ~ p e r t . ” ~  At best they can be called “bureaucratic 
expert systems,” which is not to deny their potential value, only to recognise 
honestly their limitations. 

We shall examine the extent to which expert systems can handle meanings, the root 
of all problems of interpretation. We need to uncover the semantics of the system, 
those principles, tacit or explicitly stated, that link the elements of a knowledge-base 
or the text of a body of rules to the features of the world they signify. 

2 .  Misleading Metaphors 
In the paper, “Expert Systems: Lawyers Beware!”, Stamper (1988) draws attention 
to several metaphors regularly and un uestioningly employed in the discussion of 

1 .  The conduit metaphor of language treats words, expressions and sentences as carriers 
of meanings; detached from people, words go from place to place, or are ”stored” in 
books and computers, “carrying” with them this abstract “content“ we call 
“meaning.“ Whenever we talk about language this metaphor tends to be used (see 
Reddy 1979; Stamper 1985). 
2 .  The chemical engineering metaphor of data-processing reveals itself in such commonplace 
descriptive phrases as ”the extraction of information from data” and “the distillation 
of meaning from information.” They lull one into thinking that data, information and 
meaning, like chemical materials, exist independently of their users.‘ 
3 .  The set metuphor of reality, basic to the treatment of meaning in classical logic, 
regards the world as composed of individuals which can be assembled into sets, 
for example, the set of all red  individual^.^ This metaphor is appropriate in the 
mathematical world of timeless abstractions but is not justified in the world of 
practical affairs where individuality and class membership can be open to dispute. 

expert-systems and the related fields o 4 problem analysis. 

In their book How To Do Things With Rules, Twining and Myers (1976), whilst acknowledging 
the importance of deductive reasoning in legal argumentation, nevertheless devote only six 
pages to that topic in their 270 pages volume. 

See Dreyfus and Dreyfus (1986) for a thorough criticism of the misuse of the word ”expert” in 
this context. 

In the expert system field this metaphor is used. Michie and Johnston (1984) write of ”a novel 
type of industrial plant, the ’knowledge refinery’, which would take in specialist knowledge in 
its existing form . . . and turn out knowledge that is precise, tested and certified correct.” 

For examples of mathematical logic spilling over into non-mathematical domains see 
Whitehead and Russell (1910127). Carnap (1942) is wholehearted in confirming this extension 
and Tarski (1965) in formulating it more precisely. One of the important points later in this paper 
is that to use mathematical logic in the domain of social and legal affairs requires justification of 
a non-mathematical kind. 



Semantics in Legal Expert Systems 221 

4 .  The correspondence metaphor of meaning6 is also needed if one defines the meaning 
of “red” as that set of a l l  red individuals - an idea that leads to making another 
metaphysical assumption about the existence of ”possible worlds” without which 
some expressions (”your next contract”) would have nothing to correspond with. 
This illustrates 
5. The platonic metaphor which most of us accept readily enough as a result of the 
indoctrination we receive in our mathematicslessons. We do not balk at assuming the 
existence of worlds populated by abstract objects and mappings between them.’ 
Finally, bringing together all these metaphors, we have number 6 .  
6. The information-processing metaphor of mind, which equates minds and information 
processing devices, this helps us to accept the notion of knowledge as a saleable 
commodity packaged in an expert system.’ 

So commonplace are these metaphors that they seem l i e  common sense. Even 
when someone’s attention is drawn to them and their shortcomings have been 
acknowledged, they can be difficult to abandon. Always the question is asked: What 
can we put in their place? We cannot answer by offering to eliminate metaphors but 
by supplying others, as we shall explain below. In order to deal more successfully 
with social systems, a new kind of common sense, new metaphors, a new paradigm, 
can be introduced. 

3. Propositional Logic as a Basis for Legal Expert Systems 
To illustrate the semantic problems that arise in constructing legal expert systems, we 
shall examine rule-based systems built upon classical logics as their theoretical 
foundation. Equal attention should, perhaps, be given to systems based on some 
form of semantic net but their diversity and relative informality makes such systems 
difficult to target critically. In every case, the basis of the analysis should be to ask 
what is the underlying theory of meaning employed to relate the character-strings in 
the computer to the reality in which the systems serve their users. Logic-based 
systems are familiar enough to serve our purposes. 

The simplest logical tool is Propositional Logic. The smallest meaningful character- 
strings are elementary propositions, P,Q,R . . . etc. Their meanings are not analysed 
any further than to determine whether they are true or false (t or f ) .  So we can 
characterise the underlying semantics by a function, u, which provides a truth value 
for each proposition. 

a(l‘)=t, u(Q)=f, a(R)=t, . . . . . 
The logic allows one to construct compound propositions which have meanings, in 
the rudimentary sense of the function u, computable from the truth-functional 
meanings of their constituents. Thus Z = (PorQ)&R is a compound proposition for 
which 

.(Z) = t . 

This is fully worked out in “Montague Semantics” to which Dowty et a l .  (1981) provide an 
excellent introduction. 
’ For a discussion of the value of platonism as an aid to the mathematical imagination, see 
Davis and Hersh (1983). 

Psychology needs metaphors in its search for a scientific understanding of the mind. 
Computer science provides the cognitive psychologist with a useful modern mechanical 
analogy, but the computer scientist cannot then use the presumption that minds are like 
machines to justlfy his quest for artificial intelligence. 



222 Ronald K. Stamper 

Using Propositional Logic knowledge of the law can be stored in the form of rules 
such as 

(PorQ)&R- S 

so that, when the complex condition (PorQ)&R is met, the conclusion, S, can be 
asserted to be true also. This simple idea is just a re-writing of the familiar 
decision-table. 

This kind of legal expert system is described by Susskind (1987). Here is a slightly 
simplified fragment from his analysis of the Scottish law of divorce. 

legal production number 1 
IF AND ONLY IF 
the marriage has broken down irretrievably 
A.s.l(l) of The Divorce (Scotland) Act 1976 
AND NOT 
decree is to be withheld in respect of 
action under s.l(Z)(e) 
A.s.1(5) of The Divorce (Scotland) Act 1976 
THEN 
the court may decide that [permission] 
d e a e e  of divorce is to be granted 
legal production number 2 
IF 
the defender has committed adultery 
time limitation: since the date of the marriage 
A.s.l(Z)(a) of The Divorce (Scotland) Act 1976 
AND NOT 
the adultery has been connived at in such a way 
as to raise the defence of lenocinium 
A.s.1(3) of The Divorce (Scotland) Act 1976 
AND NOT 
the adultery had been condoned 
A.s.1(3) of The Divorce (Scotland) Act 1976 
THEN 
it shall be established that [obligation] 
the marriage has broken down irretrievably 

A&-B- [PIC 

A 

B 

C 
(D&D')%(-E&-F)- [o]G 

D 
D' 

E 

F 

G 

The basic structure here comes from Propositional Logic with some important 
additional features, in particular the deontic operators (permission and obligation) 
and the references to the relevant authorities, which we shall examine later. What 
does such a system "know" about the meanings of the words used in the law? 

"Not much" must be the answer. Whoever assigns the truth-values to the 
elementary propositions makes the interpretations from which the consequences are 
mechanically determined. This kind of system has virtually no semantics, nor does it 
pretend to have any. 

Its strength lies in the analysis of the constituent propositions and their 
organisation into rule structures that draw upon a mixture of statute law, case law 
and legal principles, supported by a clear statement of the authority for each element. 
In addition to the logical structure, such a system could provide an indexing scheme 
to a diversity of relevant material for use by the person who is deciding on the truth- 
values of the propositions. 

Thus the semantic knowledge that is the most important part of the lawyers' 
expertise has to be left out of an expert system based on propositional logic because 



Semantics in Legal Expert Systems 223 

such a logic has no structural machinery to embody semantic information except the 
truth values of whole propositions. Propositional logic does not recognise smaller 
units than whole propositions, certainly not individual words. 

4. Predicate Logic for Expert Systems 
Predicate Logic goes much further in the structures it recognises so it can exploit a 
more sophisticated semantics. Expert systems with this foundation appear to 
embody a substantial amount of knowledge of meanings. We shall examine the 
extent of this knowledge. 

Predicate logic deals with the inner structure of each proposition. Each proposition 
has the simple grammatical structure 

< subject > is c predicate > 
the subject can be the name of a single individual and the predicate the name of a 
property, or the subject can be a pair, a triple, . . . of names of individuals and the 
predicate names a relationship between them: 

John is a lawyer 
John and Mary are married. 

The crucial idea is that this logical system can distinguish not only propositions but 
also names of individuals and names of predicates. But what do these names mean? 

In order that predicate logic should not be an abstract mathematical tool, we have 
to provide it with a semantics that can relate its symbolic expressions to entities in the 
world of practical affairs. This requires two basic assumptions. First an ontological 
assumption that the world is composed of individuals each of which can be iden- 
tified; and also an epistemological assumption that we can know to which individual 
each name applies. 

The semantics of predicate logic uses the idea of a truth function, as does 
propositional logic. The meaning of a proposition is given by its mapping onto a 
truth-value: 

u(John is a lawyer) = true 

for example. But this truth value can be determined by referring to a more 
fundamental information. "John" is the name of an actual person (let us signlfy him 
by john) whilst "lawyer" is the name of the class of all individuals of that kind (let us 
signify this class of real individuals by LAWYER). The above proposition is true if and 
only if, referring to actual set membership, we find 

john E LAWYER 

We do something similar to give a precise meaning to "married" by making it mean 
the set of all ordered pairs of individuals of the appropriate kind, called MARRIED. 
s o  

< john,mary > E MARRIED 
has to be tested to find the truth of the proposition "John and Mary are married." 

Therefore, to know the meanings of the words in the formulae of predicate logic, 
we have to assume a knowledge of which individuals have the properties or fall'into 
the classes named by the predicates, and which pairs, triples, . . ., of individuals are 
related in ways named by the higher order predicates. Each allocation of individuals 
to the relevant classes is called a "model." 



224 Ronald K. Stamper 

The legal rules will be regarded as an axiomatic theory which can have any number 
of interpretations supplied by different models, that is, sets of individuals and their 
groupings into sets of individuals, sets of ordered pairs, sets of ordered triples . . . 
etc. This elegant mathematical theory can draw upon all the sophisticated machinery 
of set theory. In its richest form, initiated by Richard Montague (1974; see Dowty et al. 
1981), it can handle the semantics of a logical formalism that approximates to natural 
language. How then does it serve our purposes in building legal expert systems? 

5. Theoretical Semantic Problems 
We have no need to look at the application of this kind of logic to law to see that it 
will raise some important semantic problems. Let us make a note of them before 
examining the additional problems arising in applications. 

Propositional logic, which is a component of predicate logic, employs the concept 
of truth-functional semantics. It assumes that, at the level of whole propositions, you 
need only know the truth-values of each of them in any situation to know their 
meanings and the meanings of the logical expressions derived from them. This 
reliance on truth as a simple, basic concept for a theory of meaning seems to work in 
domains which are free from dispute (routine engineering, or routine natural science) 
where we can perform a reliable operation of checking the truth of a fact-statement. 
In the case of a legal dispute, truth is what we arrive at, not what we start from. In 
such circumstances the operational basis of the semantics must be different so that we 
can treat truth as the point we reach following negotiation (among the members of a 
jury or between the parties to a dispute). 

Predicate logic uses the concept of an individual. Outside the mathematical 
domain, this concept is rather a sophisticated one which gives rise to the kinds of 
paradoxes that suggest we should be cautious about making it the basis of a 
semantics. What is and is not one and the same individual has been the subject of 
paradoxes since ancient times (the river of Heraclitus). The semantics of predicate 
logic does not elucidate these problems, is merely expects that the user has solved 
them to his own satisfaction. There are many important entities, such as water, gold 
and other substances that may be difficult to treat within the model of the world as a 
collection of individuals. Disputed identity may cause legal conflict, as in the case, for 
example, of the written-off car which appears on the road again - do we have one car 
that has been repaired or two cars, the second having been constructed from raw 
materials taken from the first when that one went out of existence. 

The meaning of each predicate is defined as a set of individuals, or pairs of 
individuals, or triples . . . etc. If the membership of the set changes, then the mean- 
ing of the predicate changes, unless one is prepared to abandon the set-theoretic 
definition of identical sets in terms of the one-to-one identity of their members (by 
extension) in favour of an appeal to defining properties (by intension). But doing this 
requires a different ontological position in which the set is independent of the 
membership. The problem has to be solved: What happens as individuals are born 
and die? We have to live with an every-changing meaning of “person”! To escape 
from this dilemma, we can take as our defining set, the set of all persons past and 
future and in any imaginable world. (This illustrates the comfort to be derived from 
a faith in the Platonic Reality.) In its turn, this entails our treating all instants of time 
as identifiable individuals. 

There is an escape. We can forget about a semantic theory that maps linguistic 
expressions onto sets of real objects. Instead we can put our trust in a purely syntactic 
theory. The logic contains rules of inference. These enable us, given one set of 
propositions (premises) to deduce, by mechanical operations on those premises, any 



Semantics in Legal Expert Systems 225 

number of conclusions (theorems). To bother with what lies outside the system is 
regarded as irrelevant. This route simply dismisses semantics as not a real problem 
(Kowalski 1979,9, for example) - a justifiable point of view if you are only interested 
in the closed world of deductive processes and have no wish to devote time to 
justifying their relationship with the untidy world outside. 

To escape into the study of purely syntactic problems does not make the semantic 
problems disappear. On deciding to apply abstract logical formalisms to practical 
affairs, assumptions about their semantic justification should be made explicit. As we 
shall see, there is often too much reliance upon the users supplying the semantics 
intuitively. 

6 .  A Fragment of an Expert System 
To illustrate the semantics of expert systems based on predicate logic, a fragment of 
a rule-system developed by Sergot et al. (1986) will be examined. It deals with the 
British Nationality Act 1981, and it typifies the kind of system that interests us. 

The first sub-section of the British Nationality Act 1981, 

l.-(l) A person born in the United Kingdom after commencement shall be a British citizen if at 
the time of birth his father or mother is 

(a) a British citizen; or 
(b) settled in the United Kingdom. 

They express as follows, captured by the knowledge engineer in a number of Horn Clauses in 
Prolog 
x acquires British citizenship by section 1.1 on date y 

if x is born in the U.K. 
and x was born on date y 
and y is after or on commencement 
and x has a parent who qualifies under 1.1 on date y 

if z is a parent of x 
and z is a British citizen on date y 

if z is a parent of x 
and z is settled in the U.K. on date y 

if z is the mother of x 

if z is the father of x 

x has a parent who qualifies under 1.1 on date y 

x has a parent who qualifies under 1.1 on date y 

z is a parent of x 

z is a parent of x 

A system based on such rules can either give us their logical consequences if we feed 
in the facts e.g.: 

Matthew was born in the United Kingdom 
Matthew was born on 10 January 1987 
Ronald is the father of Matthew 
10 Jan 87 is after commencement 
Ronald was a British citizen on 10 Jan 1987 

or, given a goal such as finding out whether Matthew has British citizenship or not, 
the system can generate the questions needed to elicit the facts before chaining 
forward to compute the answer. 



226 Ronald K. Stamper 

Most of us would describe this kind of application of the law as a function of an ideal 
Weberian bureaucracy. Straightforward facts generate unequivocal consequences. 
The process does cot em loy much of a lawyer’s powers of reasoning although the 
bureaucracy may employ P awyers to perform this kind of deduction, especially to deal 
with very complex and seldom used chains of rules. Normal legal practice could 
not be performed only using the skills of deductive reasoning; neither could real 
bureaucrats do their work using only such banal reasoning processes. What is 
missing? 

7. Semantic Problems Arising in Expert Systems Based on Predicate Logic 
The bureaucrat has to put the facts into words. Even when his job is limited to 
mechanical, deductive decision-making, he soon encounters some obvious semantic 
problems. For example, if I gave him the facts relevant to my son‘s claim for citizen- 
ship as I did above, then the system could not reason from those postulates. They 
would have to be revised as follows: 

not: 
Matthew was born in the United Kingdom 
but, to match “ x i s  born in the U.K.”: 
Matthew is born in the U.K. 
not: 
Matthew was born on 10 January 1987 
but, to match ”x was born on date y ” :  
Matthew was born on date 10 January 1987 
not: 
10 ]an 87 is after commencement 
but, to match ” y  is after or on commencement”: 
10 January 1987 is after or on commencement 
not: 
Ronald is Matthew’s father 
but, to match “z is the father of x”: 
Ronald is the father of Matthew 
not: 
Ronald was a British citizen on 10 Jan 1987 
but, to match “z is a British citizen on date y”: 
Ronald is a British citizen on date 10 January 1987 

Of course the system would not have been designed to accept the input in any form 
but rather to ask for the facts to be ”filled in” using the formats prescribed by the 
analyst. This solution means that this system has its own private way of collecting 
relevant data - just the problem that caused so much trouble in the early days of data- 
processing when dozens of stand-alone applications could not exchange data. In 
administration this is a serious fault. 

8. Semantics and Large Systems 
The problems caused by inventing a private language for each application not only 
affects the exchange of data but it makes it impossible to integrate legal expert 
systems by the obvious, simple method of merging rules. This would not be a serious 
problemif different areas of the law could be placed in water-tight compartments, but 
as the law is a single fabric, this is a serious issue when considering strategy of 
developing integrated systems. 



Semantics in Legal Expert Systems 227 

At the root of this problem is the fact that predicate logic is weak methodologically. 
Analysis of a legal text and its expression for manipulation in predicate logic does 
not elucidate meanings, but rather obscures them. Nothing in the structures 
d i s h  ished b this logic will lead two different knowledge engineers to the same 
paraprase of t i e  original text. Thus, the same original concept appearing in two 
different pieces of legislation analysed by two different engineers will only result in 
the identical predicate expression by a happy accident. We should not have to rely 
upon such chance agreements to preserve semantic integrity across a growing corpus 
of legal knowledge-bases. 

Returning to the example above, we can observe another semantic problem. The 
predicate names (in bold type) must have exactly the same form whoever uses them 
because they function as single symbols. A more honest representation of the above 
ten predicates created by the analyst would use single symbols, or at least strings that 
do not masquerade as natural language: 

for x acquires British citizenship by section 1.1 on date y 
for x is born in the U.K. 
for x was born on date y 
for y is after or on commencement 
for x has a parent who qualifies under 1.1 on date y 
for z is a parent of x 
for z is a British citizen on day y 
for z is settled in the U.K. on date y 
for z is the mother of x 
for z is the father of x 

The analyst who ran out of symbols could use longer strings, following the general 
practice in programming. This simple change emphasizes that we should not forget 
the fact that the analyst invents a new, artificial language when he creates an expert 
system using a language such as Prolog. 

9. The Humpty Dumpty Syndrome 
These "fat predicates," the elongated symbols that look like natural language, are 
constructed by trial and error by the anal ~ t . ~  He invests them with meaning. The 

and other symbols precisely the meanings he wants them to have. The Humpty 
Dumpty syndrome is not only a disease transmitted by Prolog, it infects virtually all 
computer applications. It has some serious consequences. 

First, by actually allowing the analyst to invent an artificial language, over and 
above the already complex language of the law, we reduce the chance of anyone 
understanding correctly the contents of the "knowledge-base." 

Second, "fat predicates" serve to deceive the naive customer looking for an expert 
system. He may imagine - as an unscrupulous expert system vendor may intend - 
that the computer understands the meanings of the natural language words. It does 
not. We should not hesitate to criticise a mechanical engineer who supplies cardboard 
boxes painted grey leaving the customer to assume that they are steel girders, so let 
us demand similar standards from the knowledge engineer. 

Third, the analyst assumes that the original text of the statute contains a meaning 
which he has "captured" (note the metaphor) equally well in his formal version. 

Sergot et al. (1986) explain and appear to recommend this procedure in the section entitled 

analyst behaves like Humpty Dumpty i n A  r ice Through the Looking Glass, giving words 

"Formalisation by Trial and Error." 



228 Ronald K. Stamper 

Lawyers, however, take great care to preserve forms of words that have withstood 
the test of scrutiny by courts over many years. They will recognise the danger in this 
casual attitude towards language. The same problem will arise if the original text is 
analysed into its constituents by even the most sophisticated semantic analysis 
method, but, in this case, there is a reasonable chance that the analytical process will 
tend to improve the drafting of legal texts from a semantic point of view. 

We have noted the ease with which semantic confusions can be introduced to the 
knowledge-base in this kind of system. What can we do to remove the potential 
confusions introduced by a dozen Humpty Dumpties working on different parts of 
a huge corpus of formalised law? 

The only notion of meaning accessible to the computer depends upon the purely 
syntactic equivalence of one sign to another. Of course, the analyst working on a 
lengthy problem finds it difficult to remember the exact form of each elaborate 
predicate expression he has defined. Where we fear that several different formal 
predicates express the same legal concept, we might attempt to uncover these 
accidental semantic confusions by searching predicate strings for common terms. 
This cannot guarantee success. One can easily find paraphrases that contain no 
common significant terms! For example, the original 

x acquires British citizenship by section 1.1 on date y 

x acquires British citizenship on date y by sect. 1.1 
British citizenship of x acquired on y under sub-section l.-(l) 
x is a citizen from y by virtue of British statute 1981c61,1.-(1) 
x is a natural born Briton commencing on y (ref UK 1981~61) 

may appear elsewhere in the following guises 

Unfortunately, as one can see from the last version, paraphrases need not include any 
common term. 

Clearly, this problem of matching different versions of what are supposed to be the 
same legal concept will give rise to errors and confusions when systems have to be 
extended and amended. It will prove an obstacle to the difficult process of unifying 
legal knowledge-bases created at different times by different analysts. And it will 
even obstruct the integration of the efforts of a team of analysts engaged on the same 
large legal domain. 

Quite another attitude towards paraphrasing might be justified - why not admit 
different interpretations? Dealing with law in the European Community it may be 
wise to allow the same piece of text to be treated in several different ways in a 
knowledge-base, to take account of national differences of behaviour. Interpretations 
can be localised to individuals or to groups. Unfortunately, this logic has no place to 
record the agent providing the interpretation of the meaning of the text. Meanings, 
it seems, are assumed to belong objectively to the natural language text, and the 
analyst only has to perceive and record them in another, but formal language. 

Paraphrasing problems are explained by the shallowness of the semantic analysis 
required for predicate logic. It is quite easy to shift between different forms when, 
with equal validity, one can accommodate a concept by treating it as a distinct 
individual, as the date is treated in 

x acquires British citizenship by section 1.1 on date y 

x acquires British citizenship at birth by section 1.1 
whilst it can also be incorporated into a prolix predicate, as in 

This illustration demonstrates the looseness of the ontology behind predicate logic. 
What is deemed to exist or not exist is decided at the whim of the analyst! 



Semantics in Legal Expert Systems 229 

One can only draw the conclusion that predicate logic, whilst being wonderfully 
rigorous from a syntactic viewpoint, is sloppy and informal semantically. 

10. Predicate Logic Suits Bureaucratic Systems 
Bureaucratic systems are expected to operate as impartial machinery giving effect to 
policies worked out by the political system where value-judgments are paramount. 
Logical systems devised to serve mathematics may suit the ideal bureaucratic system. 

Legal systems are more than bureaucratic ones. They may give rise to systems of 
routine office work to take care of the commonest cases but the creation of the legal 
norms and their interpretation in the difficult cases involves the examination and re- 
examination of the values which the norms are supposed to embody. Differences of 
value-judgments become exposed as disputes about meanings. A logic for legal 
systems must give semantics pride of place. 

Mathematics, the paradigm of deductive reasoning, despite the claims made on its 
behalf (Susskind 1987, 185), does not necessarily provide the ideal model for 
reasoning in the domain of practical legal and business affairs. A logic inspired by 
mathematics will take advantage of simplifications that have no justification in the 
world of human relationships. 

Despite all the above noted objections to the rule-based expert systems built on the 
foundation of classical logic, they do have a place in the automation of routine 
administrative tasks defined by laws. I have no objection to such systems but I 
would hesitate to call them "legal expert systems" rather than "bureaucratic expert 
systems." 

Why should we expect a logic devised by mathematicians for their work to transfer 
comfortably to the domains of law and other practical social affairs that do not share 
the special properties of timelessness, abstraction, precise formality, independence 
from human judgment, desires and intentional action? It seems to me potentially a 
poor candidate for expressing legal issues but, faufe de mieux, people have embraced 
it too enthusiastically and quite uncritically. 

"What can we put in its place?" you will rightly ask. 

11. Escaping from a Misleading Paradigm 
In aresearch project aimed at discovering ways of modelling organisations as systems 
of social norms,'" my colleagues and I have attempted to escape from the frame,of 
reference within which classical logic was created. Leaving a familiar framework can 
be very difficult, especially if there does not exist a ready-made alternative. 
Sometimes even more daunting is the passionate, irrational opposition with which a 
new framework will probably be greeted, revealing that the underlying metaphysics 
of a formal system owes as much, perhaps, to blind faith as any religious 
fundamentalism! 

This work on developing a legally orientated language, Legol, was conducted at 
the London School of Economics over many years but has now relocated at the 
University of Twente. Several versions of this language were formulated and 
implemented. 

Throughout, the law itself acted as the main source of ideas. Instead of aiming to 
apply logic to the law, we hoped to explicate the logic of the law. Whereas classical 
logic embodies the structures observed in mathematical systems, the law (as ancient 
a discipline as mathematics) might be expected to have evolved very different 
structures. Instead of dealing with timeless abstractions for which no person is 

A strategy which, we are pleased to note, Bob Kowalski's team has now embraced. 



230 Ronald K. Stamper 

responsible, the law concerns a strict time-frame where specific people, as far as 
possible, are held responsible for the concrete events and actions to which its norms 
apply. This led us to incorporate into the formalism structures that require the analyst 
to take account of a number of features of obvious importance in every legal problem 
we had investigated. The result was an informal, partial answer to some of the 
questions about semantics raised in this paper. 

This partial solution" can be illustrated using the same Nationality Act example. 
Our empirical workmade it clear that the analysis would have to penetrate to the level 
of individual words or expressions that function as the semantic units in natural 
language. It was also clear that time had a special role; there was no point in treating 
it as either a class of individual instants or a class of intervals. Also, there had to 
be a place for the authority that determined the existence of every legally significant 
state of affairs. The result, from a semantic point of view, is indicated in the table 
below. 

Entity 

nation 
nation 
person 
citizenship 

citizenship 
citizenship 
territory 
territory 
located in 
act name 
act "1981~61" 
commencement 
fatherhood 

motherhood 

settled in 

Characteristic 
Antecedents 
name 
"Britain" 
name 
"1981~61" 
citizen, nationality 
"1981c61ssl(l)" 
"1981c61ss1(2)" 
name 
"United Kingdom" 

? 

"1981~61~53" 

father child 

mother child 
"1981c61s50" 

? 

? 

Iden tifierslExistence 
StartFinish 

< time > < time .: 
birthdeath 
person, nation 

person, nation 
person, nation 

<time > <time > 
person, territory 

<time > < time > 
act 
person. 1, person. 2 

person.1, person.2 

person, territory 

Most of this table corresponds to a schema (in database terminology) which specifies 
the universals or entity types, that is, the classes of particular instances. In bold type 
are a few particulars. Every particular has associated with it a period of existence and 
the corresponding universal can be annotated to indicate how the start and finish 
times should be selected (birth and death for a person, for example). 

The chnracferistic can be name, in some cases, or it can be a criterion for the existence 
of a particular. For example, citizenship exists by virtue of a number of different 
criteria, giving rise potentiall to a number of different meanings of the concept. In 
the cases of "located in," "&herhood" and "motherhood" we do not know the 
criterion, and this is an indication that the analyst should clarlfy this point in order to 
complete his semantic schema.'* 

Identifiers are the names of the types of individuals being qualified or related by the 
entity. Later these were more accurately described as antecedents when they were 
generalised to be any other entity. The importance of antecedents is that their 

"See Jones et al. (1979), Tagg (1979) and Stamper (1980) for accounts of Legol-2.0 and 
Legol-2.1, the latter having been demonstrated in the 1979 Symposium on Computers and Law 
at Swansea. 
l2 Note how the syntax of this formalism guides the analyst. 



Semantics in Legal Expert Systems 231 

co-existence was a necessary precondition for the existence of an instance of the 
entity, except in the case of what we call “things” which could have an independent 
existence. The antecedents also commonly have role names which can be inserted in 
the definitions, as above for citizen, nationality, father, mother, child. 

This is not the place to explain the Legol language for manipulating these structures 
and for expressing norms. An illustration will be enough. It would not have been 
difficult to provide an infix notation that would have made the first subsection of the 
Nationality Act 1981 read as follows. 

Citizenship @Britain of a person [criterion: 1981c61ssl(l)] starts when 
the birth of that person occurs 
- after commencement the British Nationality Act 1981~61 
while his father is a British citizen 0’ settled in the UK 
or while his mother is a British citizen 0’ settled in the UK 

The words underlined are operators that function as logical operators that in every 
case take account of time. Some words here are just syntactic sugar and would be 
ignored by the interpreter. However, those words that appear in the table above can 
be manipulated as separate concepts (and so are in a sense “understood”) by the 
computer. 

This intermediate solution is full of defects. In particular, it does not function as a 
logic although it does allow the forward and backward chaining of rules that a legal 
expert system requires. In that respect it does not have the theoretical underpinning 
enjoyed by Prolog. Nevertheless, it has the advantage of taking care of some of the 
semantic problems catalogued earlier. 

In particular 
1. fat predicates disappear and the semantic units of the formalism are those of the 

natural language; 
2. semantic analysis no longer depends totally on the analyst’s skills in para- 

phrasing the original text; 
3. indeed, semantic analysis is guided by the quite rich structure awaiting his 

results - each relevant word in the text generates a string of questions that must 
be answered about the start, finish, criterion, antecedents and role names; 

4. accidental synonyms or multiple interpretations cease to be a danger because 
the arbitrary solutions that bedevil most systems specifications are greatly 
reduced, if not entirely eliminated when the level of analysis is reduced to the 
word. 

Despite these advances, this intermediate solution fails to break away from the old, 
defective framework. It still assumes the existence of individuals although it does not 
force one to treat instants or intervals of time among them. It still ado ts the stance 

and what is false. 
Nevertheless, it is loaded with signposts directing us towards a new framework. 

For example, 
- existence is clearly a candidate to replace truth as a primitive concept; 
- the idea of a criterion suggests that each instance of anything existing has to be 

determined by an authority, either in the form of an agent or a norm; 
- there appears to be a unlfying structure that eliminates the distinctions between 

individuals and predicates, or particulars and universals; 
- there seems no reason to treat norms as another kind of entity. 
The signposts begin to make sense once one abandons the old metaphysical 

that we can describe an objective reality using our language and so deci 1 e what is true 

assumptions. 



232 Ronald K. Stamper 

12. A New Framework 
The goal is to create a logical language incorporating a theory of meaning appropriate 
for domains of practical human affairs. The work is not yet complete. At present the 
result is a formalism, capable of mechanical interpretation, and, we believe, capable 
of development into a fully functioning logical system despite its many rather 
unusual features. It has become convenient to employ two formalisms, one, still 
called Legol, for manipulating the knowledge-base and another, called Norma, for 
knowledge representation. Norma is an attempt to capture in a logical syntax what 
we have discovered about the structures underlying social and legal norms. 

The theory has been developed through the analysis of very large numbers of legal 
problems and business information systems. This had led to the evolution of a 
method of analysis which has proved to be effective in designing systems that are 
cheaper to build and far cheaper to maintain than those developed using the 
conventional methodologies. This practical aspect of the work is emphasized in the 
Appendix to this paper where a concrete example is presented. This is also included 
in the hope that it will help the reader to follow some of the more abstract concepts 
introduced below. 

The metaphysical assumptions on which Norma is based are radically different 
from those that seem tacitly to be adopted by builders of rule-based expert systems 
today. I hope that you will agree that they are not difficult to accept, indeed that they 
are less mysterious than the metaphysical assumptions upon which orthodox logics 
are based. In our view there is no need to depend on the notions of individuality, 
identity, truth, and possible worlds, for example. These are all very sophisticated 
concepts of which we are innocent at birth. However, we start our lives with a desire 
to act, and a rudimentary system of values that defines the boundaries between 
states of behaviour that we like and dislike. During our most innocent months of life, 
we live in the here-and-now and we act directly on the world; only later do we 
begin to construct realities distant in space and time, and learn to act indirectly 
through communication. That very simple beginning is also our starting point for 
constructing our new framework. 

A belief in the existence of at least one agent (yourself) and hislher behaviour are 
the most basic ontological assumptions. We need symbols to represent these two 
kinds of things. Two epistemological principles govern the syntactic structure: 

all knowledge entails a knowing agent 
the agent only gains his knowledge through action. 

The agent, an or anism that has acquired (genetically, presumably, though it really 
does not matter i ow) a rich enough structure (an issue that can be examined later 
using the theory to be created) stands at the centre of reality. For the organism, reality 
unfolds through its action (see Gibson 1979; Michaels and Carello 1981). Each of us 
is responsible for the knowledge we have but the social system enables us to share our 
efforts and so live in a much richer reality. The concept of an agent can be extended 
to include groups, or social agents such as committees, teams, companies, nation- 
states even. Their collective behaviour is what we are interested in when we study 
norms.13 

The idea of responsibility is basic. In the crudest sense of having to live with their 
consequences, the agent takes responsibility for its actions; by organising its 
behaviour system well, the agent has a better chance of surviving as an organism in 
the physical world or as a social agent in a social environment. In a more sophisticated 
way, responsibility arises in a human society as a result of the expectations produced 

l 3  The whole social agent is much more than its individual membership because of the addition 
of its norms. Every norm exists in relation to a particular social grouping. 



Semantics in Legal Expert Systems 233 

by the agents’ intentional acts involving one another and the capacity of the social 
system to compel its members to endure the social consequences of their actions. 

The two fundamental principles result in a syntactic structure that mirrors the form 
of every pattern of behaviour in the shape of well-formed-formulas (wff) having two 
components: 

<agent > <behavioral invariant > 

Such a formula we call a “realisation” because it represents a realised or actual state 
of affairs, here and now. 

An agent that attains some behaviourally invariant state may be regarded as a 
modified form of agent which can enter as the agent expression into another realis- 
ation (wff). So we find the early concept of an antecedent in Legol incorporated into 
this language (for example, a state that has created an Act can then perform the 
behaviour it calls “commencement” in the example above). The logical operation of 
building a new pattern of behaviour on the foundation of its antecedents is 
represented by the concatenation of invariants: 

AXY 

where the agent, A, realising x, has become the agent (Ax) and so has been able to 
realise the dependent behaviour y. Because the existence of y can only occur during 
the existence of x, we call y the ontological dependent of x, its ontological antecedent. 

What are these behavioural invariants? The easiest way of recognising them is by 
means of our natural language vocabulary. Most of our words (certainly our nouns, 
verbs, prepositions, adjectives, adverbs) are names of invariants. It may seem 
puzzling to treat the names of objects this way but just consider what you know 
about any ordinary object such as a cup. All your knowledge of a cup comes from 
what you do with it, or what someone else tells you about what he does with a cup. 
To speak of a cup is to speak of a repertoire of behaviour that the cup makes avail- 
able - provided that you have the invariant cup, you have available to you all the 
things you can do with the cup.“ Vocabulary (rather neglected by linguistic theory 
(Aitchison 1987) ) is thus linked to our collective choice of where we draw boundaries 
to mark biologically or socially significant shifts in the value of our behaviour. Words 
are names for zones in our behavioural space, zones where some important features 
are invariant. 

To this, the most basic of the logic structures, ontological dependency, we add a 
number of logical constants to account for our ability to combine behaviour patterns 
at the same level. These are the analogues of the familiar operators and, or, and not of 
propositional logic with the important difference that they do not combine sentences 
but behaviours; one may do one thing while doing another, onvhile doing another, or 
whilenot doing another (notice that we cannot have negative actions, so whilenot is 
dyadic). The agent only exists in the present, but its present behaviour (overt and 
internal) can become extremely complex if it is to accommodate its beliefs about the 

l4 This repertoire of behaviour is associated with any number of amount of cups. Given an 
indeterminate quantity of cups, we can engage in the important and very basic behaviour of 
dividing them up and discarding part. This behaviour can continue until we reach a minimum 
amount of cups, and then the partitioning behaviour will cause our behaviour repertoire to 
change - our one remaining cup is broken! In such a manner we do arrive at the notion of an 
individual as a rather sophisticated pattern of behaviour. 



234 Ronald K. Stamper 

past and the future. These operators play a large part in constructing the necessary 
descriptions. For example:I5 

In which the composite expressions also refer to behavioural invariants having 
existences that are functionally dependent on their components. 

The correlate of material implication in propositional logic is the norm, for example: 

A ((x while y) whilenot (z onohile w)) A((x,y);(z:w)) 

A (y whenever x) 
A (x then y) 

the existence of which is not functionally dependent upon the existence of its 
components x and y. A norm specifies a mechanism, physical or social. This is also an 
invariant. 

Another way of representin a mechanism is to do so without sa ing anything 
about its structure by naming t%e ability to behave in a certain way. &r example: 

In a similar way, we can name two other related behaviours, the beginning and the 
ending of an invariant: 

A walk able Aw* 

A walk begin 
A walk end 

Aw < 
Aw> 

which are OUT names, not necessarily A's names for those behaviours. The agent A 
may not be aware that x and x <  are related. In order for the agent to behave 
intentionally, he will have to recognise the connection. 

So far, although we have introduced the notion of existence into the discussion, 
nothin has been said about time. Time is a rather sophisticated construct that 

represent another. The model of behaviour introduced so far is sufficient to account 
for the appearance of semiological behaviour. (But that is another long story.) All we 
need to recognise is that the agent has the capacity to behave in a way that signifies 
something else: 

A"0x" 
represents A realising a sign, "Bx," meaning Bx. The precise nature of "Bx" is often 
of indifference to us in modelling behaviour, the use of voice, print, picture, 
electronic signals, and so would all be covered by the same expression provided that 
the agent A were using any of them to mean "Bx." One important result of this 
analysis is that we can easily go on to show the communication acts that the agent can 
perform with the sign 

A "Bx" asserts A "Bx" commands A "Bx" suggests 
for example. The interpretation or misinterpretation of a physical signal is not 
considered here but this more detailed level of analysis is readily accommodated. 

Once the agent has acquired the ability to use signs helshe can do more than live 
only in the here-and-now (see Mead 1932). The well-formed formula, Ax, represents 
A bein in the invariant state x. The syntax only allows one to represent the here-and- 
now. l%e start and finish of the invariant cannot be known directly, one is always in 
the past and the other in the future but the agent can here and now use signs to 
represent them. We use these symbols: 

depen % s upon the agent having, first, the ability to use one form of behaviour to 

A walk start 
Aw + 

A walk finish 
AW- 

l5 The abbreviations that we use for the logical operators (e.g., :; for while, onuhile, and whilenot) 
are introduced without other explanation than the presentation of the formula in the two versions. 



Semantics in Legal Expert Systems 235 

Thus we are able, in our current behaviour, to accommodate worlds that are past, 
future and distant. The importance of signs in social and legal behaviour cannot be 
overestimated. We should distinguish, of course, between the internalised norms 
and the signs representing norms, "rules" are what we usually call them. 

The fundamental notion of responsibility has also to be given an explicit place in 
the formalism. Every realisation has an associated authority for its existence, 
symbolised thus 

Ax@ 
This may be an agent, often it is a subagent of a group agent. It may also be a norm 
expressing the combined judgments of several agents. Ultimately, the authority 
always reduces to a set of human judgments. In other words, this logic will always 
point to agents in the social system who are responsible for the use of the constructs 
expressed in the formalism. 

Every realisation has an associated authority. so every compound pattern of 
behaviour has its authority, so the logical combinations that in classical logic we are 
free to assemble at will, and even assemble mechanically, need to be authorised in 
Norma, strictly speaking. From a mathematical point of view this is a dreadful 
restriction but for modelling the social world, the restriction is realistic. In particular, 
although we have accepted such facts and norms as 

Ax 
A (y whenever x) 

Ax 
A(Y - X) 

we do not take it for granted in Norma that we are allowed to make combinations of 
norms and facts, such as 

A(x while (y whenever x)) A(X4Y -4) 
Such an act is not automatically permitted by the logic of the system, as in practice we 
assume that a competent judge will take responsibility for invoking the norm. 
It would be unwise to make the authority, as in classical logic, some universal 
rule permitting the unrestricted combination of expressions recorded in the formal 
system. Of course, we may wish to make use of this kind of mechanical linking 
of formulas but we should not make the mistake of assuming that the laws of logic 
are divinely authorised for all places and all times. They are human constructs 
with human authority behind them. All too often, the dogmatic logician (like the 
fundamentalist believer) insists upon pushing logical rules beyond their limits of 
practical validity. 

This language is very strange from a mathematical point of view. It can never be 
fully represented as a closed formal system. The agent responsible for the formal 
expressions is needed to complete the well-formed formula. (The author of the expert 
system is one factor in its expression.) In Norma, one has no alternative but to 
represent an open system. One is always referred, by the authority, to the containing 
social system. In Norma one cannot deal with a finite system because every behaviour 
it represents has other behaviours necessarily associated with it - the ability, the 
beginning, the ending, the sign for it, its authority - and these, recursively, have 
theirs. No matter how deep his analysis, the knowledge-engineer (dreadful, 
misleading expression!) has ultimately to pin his faith upon the informal system. 
Norma shows where to put the pins. 

13. Semantic Problems Solved? 
The kind of logical language outlined above is potentially capable of capturing the 
complexities of real social behaviour. As ointed out, if you want to simpllfy, you can 
do so. But, if you decide to simpllfy, you K ave no alternative but to take responsibility 



236 Ronald K .  Stamper 

for imposing your simplification. At least we may, in the new framework, choose to 
describe our legal system as a mechanical bureaucratic structure but the new kind of 
logic proposed will not force us to do so. 

The mechanical nature of the classical logic approach, as observed above, intro- 
duces semantic problems. They were classified as either theoretical or as by-products 
of applications to expert systems. Let us look again at those points and see how far the 
proposed new framework might solve them. 

The theoretical problems identified relate to the concepts of truth, individuality, 
identity, extension, and the retreat into syntax. Truth is to be abandoned as a 
primitive concept. Truth, in the context of social behaviour is a very sophisticated 
notion that depends upon some responsible agents reaching a consensus on the 
matter in hand. Existence here and now takes the place of truth as a primitive concept, 
with the benefit of being a concept that can readily be operationalised. Individuality 
and the identification of individuals are also regarded as sophisticated concepts that 
an agent discovers only after a few years of experience which includes playing such 
games as peek-a-boo! In the legal domain it would be foolish to assume that all 
questions of individuality and identity are intrinsically resolved in some objective 
reality, whilst all we have to do is to uncover these facts. Heraclitus's river is the same 
river if all he does is to walk along its banks with his dog or gaze at it from the same 
bridge, but if, employed by the local water authority, he has the job of tracing its 
contents on their way to the sea, his different behaviour repertoire in this different 
context gives the word "river" a different meaning. If the concept of an individual is 
lost, so is the concept of a set of individuals defined extensively by enumerating its 
members. The idea of a behavioural invariant is close to the idea of an intension. 
Finally, we spurn the evasion of semantic issues by retreating into syntactics. All 
these theoretical advantages are justified by their contribution to giving us an 
operational semantics. 

The application problems that commonly arise when predicate logic is used for 
building expert systems were seen to be numerous. They are almost entirely caused 
by the freedom to formulate fat predicates. This liberty leads the analyst to invest a 
private language, introducing complexity, system rigidity, artificially enforced 
uniformity, the illusion that the machine uses natural language, and explains many 
of the difficulties in maintaining and merging systems. The new framework requires 
us to analyse legal norms down to the level of individual words or equivalent 
semantic elements and to place these in an ontological structure that goes a long way 
towards removing the arbitrariness of the analysis. This analysis gives us flexibility, 
a genuine uniformity to which diversity can be added by linking different inter- 
pretations to different agents, and insight into the operational meanings of our 
natural vocabulary. These benefits arise from an insistence on having an operational 
semantics and ruling out-of-order mathematical shortcuts, however elegant they 
may appear. 

14. Jurisprudence in Support of the New Framework 
As a coda, we may ask whether the quest for a radically different kind of logic is 
justified by the arguments of legal philosophers who cast doubt on the notion that the 
law is a system of rules. This task is made easier by Susskind's excellent book 
(Susskind 1987). He has marshalled their principal arguments preparatory to 
dismissing each of them in turn. Susskind rejects them on the grounds that, despite 
their force, they still leave open the possibility of useful practical applications for 
mechanical devices that can simulate deductive reasoning within a system of explicit 
rules. I agree with his thesis but prefer to classlfy this residual set of applications as 



Semantics in Legal Expert Systems 237 

“bureaucratic expert systems.” However, the arguments he dismissed provide 
strong support for pursuing the very different line of thinking advanced here which 
can lead to systems supporting the decision-maker or adviser faced with the semantic 
problems that the bureaucratic expert system leaves untouched. 

The courts seldom concern themselves with disputes about the accuracy of a chain 
of deductive argument, but rather adjudicate on disputes about meaning or the 
assignment of liability either at the substantive level or at the level of procedure. Legal 
problems that go before the courts may certainly entail a measure of deductive 
reasoning but much of the required reasoning can only be carried out by the exercise 
of creative thinking combined with judgment and accountability. This kind of 
reasoning cannot be mechanised. However, we can support the person engaged in 
reasoning about meaning or responsibility if we use a logic that deals with the 
semantic elements of natural language and which does so by clarifying who has the 
authority to convert words into actions either directly (the hangman) or at several 
removes (the clerk of the court). I propose that we explore the line of thinking 
described here as a new framework in order to build systems to support the lawyer 
in his quintessential legal work. 

The names of the arguments cited below are those coined by Susskind and the 
quotations are of his words or from the authors he has quoted. Just two of his 
arguments are considered but the others would also support the position advanced 
here. 

Argument from Act of Will (Susskind 1987, 174). “Hart captures the thrust of the 
argument in the comment that ’rules cannot provide for their own application, and 
even in the clearest case a human being must apply them’.” Susskind also quotes 
Hart saying ”fact situations do not await the judge neatly labelled with the rule 
applicable to them” and ”rules cannot claim their own instances.” 

No self-contained logical system can supply the will that Hart refers to or the 
“judgment” that Kant, an earlier proponent of this view, points out that logic does 
not provide. However, the proposed logic of action, Norma, is being devised to 
account for open systems. It cannot be treated as a closed system and its syntax is 
designed to force the analyst to draw attention to the agents who must supply the 
”will” and the “judgment.” In fact, because it presupposes a responsible authority 
for every realisation, Norma provides for a link reaching from the formal structure 
held by the computer to the informal, human system, for each of the cases cited by 
Hart requiring an agent responsible for: 

1. existence of the norm - 
the agent who successfully began the norm’s existence 

2. expressing the facts in terms employed in the norm - 
the author of the sentences expressing the facts 

3. invoking the norm - 
the agent responsible for conjoining the facts and the norm 

4. communicating the result with due authority - 
the one who is empowered to perform the communication act 

A given body of legal norms (some tax law, for example) may not deal with these 
issues directly, relying on other statutes. The Norma syntax provides a place for this 
information, and it is up to the analyst to use when appropriate. 

The third case has been touched upon already. It was earlier noted what a grave 
assumption is made in the belief that the laws of logic apply categorically in m y  
formalised legal knowledge-base. Should this assumption be made, then someone 
must explicitly take responsibility for it. For example, if you want to use the rule of 
modus ponens then you may, but you will have to incorporate that rule quite explicitly 



238 Ronald K. Stamper 

into the systemanducknowledge your responsibilityfor doing so, together with your indirect 
responsibility for the results it generates. Judges do not indiscriminately link facts and 
rules to draw "the logical conclusions" because the results can be unjust. Logic is a 
mechanical model of language; there is no licence to drive it everywhere except 
perhaps in the Platonic world of mathematics.I6 

It is interesting to note that Susskind's expert system for the Scottish law of divorce 
does have an informal treatment of the notion of authority. This provides a partial 
answer to the question of who supplies the will to make the norm system function. 
I am proposing that a formal treatment of authority should be a necessary component 
of any logic to model social or legal affairs. 

Argument from principles (Susskind 1987, 169). "Dworkin argues that lawyers and 
judges, while solving legal problems, as well as reasoning with rules, often also have 
recourse to non-rule standards that he terms 'principles' ." These principles do not 
lead to necessary conclusions but lend weight or probability to certain possible 
conclusions. Often they are formulated at a far higher level of generality, with 
unlimited exceptions hence incapable of being reduced to rules. "There is no such 
thing as 'the law' as a collection of discrete propositions." Hart goes so far as to say 
that "it remains the judge's duty, even in hard cases, to discover what the rights of 
the parties are, not to invent new rights retrospectively." Principles are essential to 
this task. 

This argument points to an essential openness in legal reasoning. A mechanical, 
deductive system will sooner or later break down. Comparing it with a machine in a 
factory, the operative may have done all he can to make it work, but a point comes 
when he must hand over to the engineer or manager to take actions that have never 
been prescribed in the handbooks. Engineering and managerial expertise, like the 
lawyers' expertise, goes beyond the books to the principles they have learned. A 
solution to the problem has to be found; they cannot opt out. 

If one accepts Dworkin's views, one might attempt to accommodate these 
principles as additional rules together with a host of illustrative exceptions. To do so 
would be to fall into the Artificial Intelligence trap: The ridiculous attempt to usurp 
the roles of people.17 Applying classical logic to expert systems perhaps encourages 
the tendency to fall into the Artificial Intelligence (AI) trap by compelling one to work 
in a closed system. The same computing techniques (the genuine, non-ideological 
contribution of AI) can be employed to offer social groups a medium in which they 
can collaborate more effectively. That is the spirit in which the new framework is 
proposed. It permits any desired degree of automation but it links every relevant 
decision point to the human agent involved and so acknowledges the social context 
of legal and other practical affairs. However, it will not permit closed models to be 
formulated, but always it involves the informal system which sustains the principles 
to which Dworkin has rightly drawn our attention. 

l6 Even in mathematics, there is not a universal agreement about the logical rules which can be 
applied legitimately (Bloor 1976; Haak 1978; see the discussion of Brouwer in Kneale and Kneale 
1978, and Lakatos 1979). 
" Very few AI practitioners among themselves fall into this trap but they are not always 
scrupulous about warning the less sophisticated where to tread. Some are even prepared, in 
their advertising copy, to express themselves in a manner that might mislead the layman into a 
belief that the expert systems they are selling can, in effect, place the relevant expert on the 
user's desk. Susskind himself(1989), sadly, falls into this excess in the opening paragraph of an 
article on his Latent Damage expert system: "Imagine the Chairman of the Law Faculty of 
Oxford University sitting on the desk of every lawyer in the land and always ready for 
consultation. " 



Semantics i n  Legal Expert Systems 239 

As Dworkin indicates, the judges uncover the relevant law that exists within the 
informal legal system. He argues that we cannot make a complete, explicit 
representation of a system of general, probabilistic standards that have unlimited 
exceptions. There is no need to pretend that we can. A logic which enshrines 
openness in its syntax denies the pretence. The openness of the legal system means 
that, no matter how much of the law and the given situation you have reduced to 
explicit formulas and rules, much more remains unsaid. But a responsible agent can 
invoke the unspoken principles within the informal world of law-in-action, and stop 
a deductive chain that would otherwise lead to a miscarriage of justice. 

Argument from Particularity of Facts. Susskind (1987, 181) quotes Hart again: “Fact 
situations do not await us neatly labelled, creased, and folded; nor is their legal 
classification written on them to be simply read off by the judge. Instead, in applying 
legal rules, someone must take the responsibility of deciding what words do or do not 
cover some case in hand, with all the practical consequences involved in this 
decision.” We have already dealt with the responsibility issue but the new issue 
raised by this quotation concerns the problem of choosing descriptive terms in which 
to report the facts of a particular case. 

“The facts of any case can always be characterised in a vast number of different, 
highly specific ways . . .” but the vocabulary we employ will be related to the norms 
invoked. Susskind appears to favour a scheme in which the user and the law relating 
to his case are kept so far apart that they do not interact until the deductive process 
begins. “It is possible, in rinciple, to have a system with such extensive knowledge 

would be so far removed from judgments of law that it would be obtuse to withhold 
the term ‘deductive’ in relation to the process of legal reasoning executed by the 
computer” (Susskind 1987, 181). 

The proposed new paradigm incorporates the principle that meanings are 
relationships between linguistic structures and behaviour, hence we are led to adopt 
the position that the reporting of the facts will justifiably always be coloured by the 
norms. It is obvious that the terms used to characterise a case must take some of their 
meaning from the norms that are considered relevant, simply because the behaviour 
induced by using the chosen descriptors will be mediated by the norms. Children are 
fond of playing this kind of semantic trick: “What’s in my hand? Get it right and I’ll 
give you a penny.” The prankster is safe, the answer “Nothing” will bring the 
riposte “Wrong! I’ve got air in my hand” but the answer “Air“ leads to “No. My 
hand is in the air but nothing is in my hand.” The same kind of semantic manipu- 
lation forms the basis of many a confidence trick and dubious sales technique. Unless 
you know the behaviour that will result from the various formulations from which 
one might reasonably select a description of a case, one does not know the meanings 
of the words to be employed in that socie 

reveal the links between the choice of words and their legal consequences. By 
attempting to lock legal knowledge inside a black box we do the opposite. 
Knowledge, justice and other important social constructs have to be understood and 
constantly rehearsed in a society even to exist. Expertise obscured from critical 
appraisal inside computer packages should not be considered acceptable. 

that any judgments ma B e in relation to questions actually answered by the user 

in that legal context. 
For a computer system to help with this 1 ind of semantic problem-solving it must 

Appendix 
A Simple Example of the Semantic Analysis and Norm Analysis Techniques of MEASUR: 
Canadian Unemployment lnsurance Law - Revised Statutes of Canada (2985), Ch U-1 
To illustrate the method of Semantic Analysis outlined in this paper, here is an 
example taken from Mackaay et al. (1990). 



240 Ronald K .  Stamper 

Work on the formalisation of legal systems to handle the law has tended to neglect 
the problems of methodology that become serious issues when one scales up “desk” 
research to the size of real systems. The LEGOLlNORMA research programme has 
generated a methodology suitable for such problem areas. It is called MEASUR. It 
incorporates a variety of analytical techniques that take one from the articulation of 
somewhat vaguely stated needs (for example, at the earliest stages of legal drafting), 
through the analysis of meanings, to the formulation of norms. The second and third 
phases are illustrated below. Two further stages deal with the elaboration of relevant 
procedures and appropriate sanctions. Of all the techniques included in MEASUR, 
that of Semantic Analysis has proved to be the most important. 

Phase I - Semantic Analysis 
Step I: Generate list of candidate semantic units 
The problem domain has to be reduced to semantically atomic units each of which 
corresponds to an invariant in the behaviour of some responsible agent. The analyst 
begins by listing candidate semantic units for the problem or application domain. 
Here we select the significant items from the small sample of the act covered by this 
example. Great care is not important at this stage. We could select all the words in the 
text not accounted for by functions within the system. Thus: 

Sec.l8.(1) [. . .]the qualifyingperiod of an insured person is the shorter of 
(a) the period of fifty-two weeks that immediately precedes the commencement of a 

benefit period under subsection (1) of section 20, and 
(b) the period that begins on the commencement date of an immediately preceding 

benefit period and ends with the end of the week preceding the commencement of a 
benefit period under subsection (1) of section 20. 

Sec.20. (1) A benefit period begins on the Sunday of the week in which 
(a) the interruption of earnings occurs, or 
(b) the initial claim for benefit is made 

whichever is the later. 

Step 2: Classify the candidates 
This is not a significant task for this example but it helps to look for the following 
classes, among others. 

1. responsible agents, e.g., person 
2. roles indicative of agents (e.g., claimant) 
3. relationships indicative of missing components, e.g., insured which requires two 

4. periods used in formulating conditions, e.g., qualifying period, week, date 
5 .  particular values of measuring or classification frames, called “determinants, ” 

6. measurements or classification frames, called “determiners,” e.g., day (or day of 

Our attention is now drawn to week as a unit of measurement of a duration (a 
determiner applicable to every invariant) 52 weeks may be a determinant for a duration 
of a period. Other questions are suggested. Is week, in the sense of a period named by 
the current state of a calendarlchronometer, different in meaning from week in the 
expression “52 weeks”? We need to check the interpretation of week - does it have 

agents person and stute so we add state to our candidate list 

e.g., Sunday which may be indicative of missing determiners 

week). 



Semantics in Legal Expert Systems 241 

a standard meaning as an interval lasting 7 days of duration starting on a Sunday or 
does it have a different meaning in the context of the act? [We shall treat 52 weeks as 
a determinant.] 

7. semiological behaviour - e.g., claim which is a "speech act" performed by a 
person using a sign, the meaning of which is the benefit he or she wants. 

Step 3: Create an ontology chart 
Time is central to our understanding of meaning in the context of this method. Every 
word or expression that is a semantic unit corresponds to an invariant in the flux of 
events or behaviour of the agents. Thus, to know the meaning of a word entails 
knowing when the corresponding invariant starts and finishes. The existence of each 
invariant depends upon the existence of others, hence we can clanfy meanings by 
constructing an ontology chart to demonstrate these dependencies. See Fig. 1. 

Figure 1: A specific ontology char! for a fragment of legislation 

Notes: 
1. The responsible agents are given in capital letters. 
2. Each element names a universal for which there will be numerous particulars. 
3. Determiners are prefixed by #, their determinants are omitted here just as names of 

individuals are omitted. 
4. Signs, such as "benefit" have the relevant linguistic community as their antecedent. 
5. The invariants inbehaviour are often dependent o n  others (see the discussion in the first few 

paragraphs dealing with the "New Framework" in the main body of the paper, above). 
Thus benefit canonly exist as a joint invariant of PERSON and STATE. It may appear that the 
insurance contract between PERSON and STATE should be an antecedent of benefit but the 
entitlementlobligation referred to as the benefit, though created by the legislation, will not 
necessarily cease to exist if the legislation is repealed, it may be paid subsequently and cease 
to exist then. 

6. Notice that PERSON, STATE and the sign "benefit" have no stated antecedents. As the root 
agent we may assume thecommunity at large, as the custodian of commonsense meanings. 

We may supplement this with other ontology charts to clanfy meanings of certain elements. 
Some such fragments of analysis are very general and belong in a library of semantics. For 
example 



242 Ronald K. S t a m p e r  

payments # 
benefit- 
premium- 

Figure 2: A generic ontology chart for "benefit-payment" and "premium-payment" 

Notes: 
1. The antecedent for money is most probably the STATE but it could be the responsibility of 

another agent (e.g., ECUS in the EEC). 
2. The box under payment lists specific forms including benefit. 
3. Notice that payment is a process taking place during the co-existence of receiving and 

relinquishing acts. 
4. Ownership enters here - the processes of beginning ownership = receiving 

5. Beginning and ending processes are quite different ontologically from start to finish times 
which are signs indicating when, respectively, a beginning and an ending were successfully 
completed. 

ending ownership - relinquishing. 

Phase 11 - Norm Analysis 
Having made it clear what we are talking about by constructing an ontology chart 
showing the relevant words organised as a structure of behavioural invariants, we 
can begin to spec+ the authorities governing behaviour in our problem domain. I n  
the case of an informal system, we should probably only indicate the agents 
responsible for deciding when instances of each type of invariant start and finish their 
existence. In the case of a formal or legal system, many of the authorities will be norms 
or agents empowered (conditionally) to take these decisions. 
In this example we have two legal norms. One defines the start of a benefit period 

and the other, consisting of two logical norms, defines the start and finish of a benefit 
period. In LEGOL, they would be expressed as follows: 

Context insurance (STATE,PERSON) 
Start ofbenefit period [20.(1)] <- day of( # Sunday(day) 

while (week while lust of(finish of earning omhile start of claim)) 

Note that the context operator specifies that the norm is applied to each individual 
case of its operand. This defines the antecedents of benefit period, week, earning, and 
claim. This simplifies the expression of the norm. Also note that all the logical 
operators take account of time. The system has always been implemented with 



Semantics in Legal Expert Systems 243 

one-sided open time intervals, the start being part of the period but not the finish. 
One needs to check the formal interpretation. 

The benefit period is tagged with the reference to the norm defining it. Of course, 
start and finish may have different authorities. The other two norms can be specified 
within the additional context of a single benefit period. 

Context add benefit period 
Finish of qualifying period [18.(1)] <- start ofbenefit period 
Start of qualifying period [1&3.(1)] <- lust of ((finish of quallfylng period minus 
52 weeks) 

orwhile start of prior benefit period) 

The efficiency of evaluation would be improved by defining a function the previous to 
select only last of the prior instances of an invariant type. 
Norms themselves have periods of existence and so do the universals in the 

ontological structure (the analo e a database schema). Thus changes of meaning 
can be handled easily. The onto Y ogical structure, however, tends to be very stable, 
only changing with major paradigm shifts. Hence this method has advantages in 
the design of large administrative systems where restructuring massive files is often 
too great a task to contemplate. Once an unsuitable database structure has been 
implemented one is struck with it and with the resulting inflexibility of the system. 
Using the method of semantic analysis, one obtains a structure that can easily be 
extended in scope or detail, leaving the associated norms unaffected. The amount of 
thought that must go into the preparation of the semantic scheme before the system 
is built is the extra "up-front" investment, but this thought can help to uncover some 
suspected legal problems before drafting errors are made, and so bring immediate 
dividends. 

University of Twente 
Department of Information Management 

P.O. Box 227 
NL-7500 A€ Enschede 

The Netherlands 

References 
Aitchison, Jean. 1987. Words in theMind: An Introduction to theMental Lexicon. Oxford: Blackwell. 
Bloor, David. 1976. Knowledge and Social Imagery. London: Routledge and Kegan Paul. 
---. 1983. Wittgenstein: A Social Theory of Knowledge. London: Macmillan. 
Camap, R. 1942. lntroduction to Semantics. Cambridge, Mass.: Harvard University Press. 
Davis, P. J. and R. Hersh. 1983. The Mathematical Experience. Harmondsworth: Pelican Books. 
Dowty, D. R., R. E. Wall and S. Peters. 1981. Introduction to Montague Semantics. Dordrecht: 

Dreyfus, J. L. and S. E. Dreyfus. 1986. Mind over Machine. Oxford: Blackwell. 
Gibson, James, J. 1979. The Ecological Approach to Visual Perception. Boston: Houghton Mfiflin 

Haak, Susan. 1978. Philosophy of Logics. Cambridge: Cambridge University Press. 
Jones, S., P. Mason andR. K. Stamper. 1979. Legol-2.0: ARelational Specification Language for 

Kneale, W. and M. Kneale. 1975. The D a t e l o p e n t  of Logic. Oxford: Oxford University Press. 
Kowalski, R. 1979. Logic for Problem Solving. Amsterdam: North Holland. 
Lakatos, I. 1976. Pro@ and Refutations. Cambridge: Cambridge University Press. 
Lakoff, G. and M. Johnson. 1980. Metaphors We Live By. Chicago: University of Chicago Press. 
Mackaay, Ejan, Daniel Poulin, Jacques Frbmont, Paul Bratley, and Costant Deniger. 1990. The 

Reidel. 

Company. 

Complex Rules. Information Systems 4. 

Logic of Time in Law and Legal Expert Systems. Ratio Juns 3: 254-71. 



244 Ronald K. Stamper 

Mead, G. H. 1932. The Philosophy of the Present. Chicago: University of Chicago Press. 
Michaels, C. F. and C. Carello. 1981. Direct Perception. Englewood Cliffs: Rentice-Hall. 
Michie, D. and R. Johnston. 1984. The Crentiw Computer. New York: Viking. 
Montague, R. 1974. Formal Philosophy. Ed. R. H. Thomason. New Haven, Conn.: Yale 

Niblett, B., ed. 1980. Computer Science and Law. Cambridge: Cambridge University Press. 
Ortony, A., ed. 1979. Metaphor and Thought. Cambridge: Cambridge University Press. 
Reddy, M. J. 1979. Conduit Metaphor - A Case of Frame Conflict in Our Language About 

Language. In Metaphor and Thought. Ed. A. Ortony. Cambridge: Cambridge University 
Press. 

Ser ot, J. J . ,  F. Sadri, R. Kowalski, F. Kriwawek, P. Hammond and H. T. Co . 1986. The 
gritish Nationality Act as a Logic Programme. Communications of the A.C.M. 2 7  

Stamper, R. K. 1991. Expert Systems - Lawyers Beware. In Law, Decision-Making and 
Microcomputers. Ed. S . S .  Nagel. Westport, Conn.: Quorum Books. 

Stamper, R. K. 1980. Legol: Modelling Legal Rules by Computer. In Computer Science and Law. 
Ed. B. Niblett. Cambridge: Cambridge University Press. 

-. 1985. Information: Mystical Fluid or a Subject for Scientific Enquiry. The Computer 

Susskind, R. E. 1987. Expert Systems in Law: A Jurisprudential Inquiry. Oxford: Oxford University 

University Press. 

]ouml28. 

Press. 
.1989. An Expert Lawyer On Every Desktop. The Independent London. 

Tagg, Clare. 1979. The Impact of Data Modelling on LEGOL. (Internal Report.) London: London 

Tarski, A. 1965. Introduction to Logic. New York: Oxford University Press. 
Twining, W. and D. M. Myers 1976. How to Do Things with Rules. London: Weidenfeld and 

Weber, Max. 1946. Essays in Sociology. Trans. and ed. H. H. Gerth and C. Wright Mills. Oxford: 

Whitehead, A. N. and B. Russell. 191011927. Principia M a t h m t i c a .  Cambridge: Cambridge 

School of Economics. 

Nicolson. 

Oxford University Press. 

University Press.