From: AAAI Technical Report SS-95-07. Compilation copyright © 1995, AAAI (www.aaai.org). All rights reserved.
Some Requirements
on an Action
(Position
Language for Legal Discourse
Paper)
L. Thorne
McCarty
Computer Science
Department
Rutgers University
New Brunswick,
NJ 08903, USA
mccarty@cs, rutgers, edu
Myinterest in theories of action arises from mywork
on AI and Legal Reasoning. This work dates back a
number of years [McCarty, 1977], and it includes both a
practical and a theoretical component.
On the practical side, the goal is (eventually) to construct an "intelligent legal information system" [McCarry, 1983a]. Such a system would include at least the
following modules:
1. A Legal Analysis System. This is what most people have in mind when they talk about legal expert systems. Given a representation of the rules
and regulations in effect in a particular area of the
law, the system would provide an analysis of the legal consequences of a given set of facts. There are
a number of such systems in existence today, e.g.
[Michaelsen, 1984; Susskind and Capper, 1988], but
they are based on very primitive (rule-based) technology.
2. A Legal Planning System. Clients are best served,
not by litigation after the fact, but by careful
planning beforehand to avoid adverse legal consequences. Think of this (abstractly)
as a search
through a space of possible transaction patterns to
identify a course of action which, according to the
rules of the legal analysis system, best achieves some
given set of goals. For a discussion of howwe might
automate this search in one important legal area
- estate planning - see [Schlobohm and McCarty,
1989].
3. A Conceptual Legal Retrieval System. Legal retrieval today is based on full-text keywordsearch
systems, using technology that has not changed substantially since the 1950’s. As an alternative, various researchers have proposed the development of
"conceptual" legal retrieval systems [Hafner, 1981],
which could be integrated with the legal analysis
and planning systems mentioned above.
What are the obstacles to building such an integrated
system? I have argued for several years that the main
136
obstacle is the knowledge representation problem [McCarry, 1990]. So far, all attempts to advance beyond the
most primitive rule-based expert systems in law have
floundered on the difficulties of representing the "common sense categories" that appear everywhere in legal
discourse. Accordingly, one goal of my current work is
to contribute to the solution of this problem, and to
establish a better technical foundation for the legal information systems of the future.
More challenging still is a theoretical goal: Wewould
like to understand legal reasoning and legal argument
in computational terms. In earlier work [McCarty and
Sridharan, 1981], I proposed a theory in which the argument for a particular legal result was explained in terms
of a prototypical case plus a carefully chosen sequence of
transformations of this case into certain less prototypical
exemplars. Again, despite some attractive properties,
the theory floundered on the knowledge representation
problem. (This, of course, was also the fate of most of
the ambitious AI theories of the early 1980’s.) To see
the problem, it is sufficient to note that a transformation is a syntactic operation, and for such an operation
to be meaningful it must correspond to the important
semantic relationships in the domain in which it is applied (here: corporate law and corporate tax law). However, the AI languages of the early 1980’s were very poor
in representing commonsense semantic categories. My
response was to rework the semantic foundations, developing what I now call a Language for Legal Discourse
(LLD) [McCarty, 1989]. I am currently engaged in a
new effort to implement the earlier theory - using LLD
as the representation language - and the initial results
are encouraging. In addition, although the specification
of LLDwas driven by a crisis in theory, I believe it will
also provide a foundation for the development of practical legal information systems, as discussed above.
What does this have to do with actions? My Language for Legal Discourse includes a number of diverse
features: it is an order-sorted logic, with count terms and
mass terms, several kinds of negation, several modes of
definition and a logic programmingstyle of computation.
But it is also fundamentally a language of time and ac-
tion, both of which are ubiquitous in legal domains. In
particular, I have found the following features of the action language to be absolutely essential:
1. Indefinite Temporal References. Explicit times and
dates are certainly necessary (e.g., March27, 1994),
but we also need to refer to times that are known
only by their relative positions in a temporal order. Often these references are to temporal intervals, which may be overlapping and which may have
indefinite end points. Thus the expressive power of
Allen’s temporal logic [Allen, 1984] seems necessary.
2. Multiple Agents and Concurrent Events. If the
world had only a single agent, a legal system would
not be necessary. Thus the representation of multiple agents in a legal knowledgebase is the norm,
rather than the exception. Furthermore, the actions
of these agents cannot be constrained to a sequentim order, by such artificial devices as interleaving.
Full concurrency is necessary.
3. Complex Actions. Most legal categories consist of
complex actions defined in terms of more primitive
actions. Myearliest work on corporate tax law [MeCarry, 1977] includes manyexamples: the definition
of a "stock distribution"; the definition of a "Section
368(a)(1)(B) reorganization";
etc. A minimal
quirement for a knowledge representation language
in a legal domain is the ability to represent these
kinds of definitions.
4. Repetitive Actions. Manyof the complex actions in
a legal domainare repetitive actions. That is, if A
is itself an action, either primitive or complex, we
often need to define the action "doing A some finite
number of times". An example, chosen again from
corporate tax law, is the concept of a "creeping acquisition". The tax code attaches certain legal consequences to the "acquisition ...of control", which
is defined as the purchase of a certain amount of
voting stock in the acquired corporation. But this
stock can be purchased in one step, or two steps, or
any finite numberof steps, over an indefinite interval of time, and the precise manner in which it is
purchased may itself have legal consequences. We
clearly need to be able to represent - and reason
about - such actions.
5. Modalities Over Actions. The most familiar modalities over actions are deontic: permitted, forbidden,
obligatory. Since these modalities appear frequently
in legal domains, I have devoted some effort to the
study of their semantic and computational properties, dating back at least a decade [McCarty, 1983b].
Often, the mere statement of the facts of a case
requires the use of deontic modalities, as in the description of the "trust" at issue in manyestate plan-
ning cases [Schlobohmand McCarty, 1989]. This requirement for the representation of deontic concepts
in a legal knowledge base is now widely recognized
[Jones and Sergot, 1992].
6. Natural Language Semantics. Finally, it is important to note that the language of legal reasoning
and legal argument is natural language: English,
German, French, etc. This does not mean that an
intelligent legal information system must be able
to cope with natural language text (although a restricted natural language interface would be very
useful). Rather, it means that whenever we reach
a choice point in our design of a knowledge representation language, we should opt for those features
that seem to be most congruent with the underlying
natural language semantics. I have found this to be
a reliable methodological guide, often pointing the
way to distinctions that turned out later to be justifiable on more pragmatic grounds. One example is
the representation of "free choice permission" in my
work on the deontic modalities [McCarty, 1983b].
For the technical details of the action language used in
LLD,the reader should consult mypapers in K1~’92[MeCarry and van der Meyden, 1992] and KR’94 [McCarty,
1994]. The main idea in [McCarty and van der Meyden, 1992] is to treat actions as predicates over a linear
temporal order, and to draw a distinction between basic
actions and defined actions. Defined actions are represented by Horn clauses with (optional) linear order constraints, and the defined predicates are circumscribed to
capture the intended interpretation of the Horn clauses
as definitions. With this device, it is easy to represent
concurrent events and repetitive actions. It is also easy
to select an appropriate temporal ontology for a particular application, since discrete time and continuous
time are just minor variations of a single logical language. These ideas are extended in [McCarty, 1994] to
include a model theory for the deontic concepts: permitted, forbidden, obligatory. Two subsequent papers,
not yet published, will present a (partial) deontic proof
theory, and an analysis of negation-as-failure as applied
to the deontic modalities. The techniques for answering queries in all of these languages are based on ideas
originally presented in [McCarty and van der Meyden,
1991].
This paper has argued that a Languagefor Legal Discourse requires an action language with certain expressive features. I have not, so far, made any overt claims
about application areas outside of the law. However,
there maybe other applications with similar properties:
e.g., an information system in a bureaucratic environment, or an automated system for commercial transactions. (We might call these "quasi-legal" applications.)
Somewhatfurther afield is Shoham’sproposal for "Agent
Oriented Programming" [Shoham, 1992]. Could these
application areas benefit from a study of the legal domain? Would LLD be useful here as well? These might
be interesting questions for discussion at the AAAISymposium.
References
[Allen, 1984] J.F. Allen. Towards a general theory of
action and time. Artificial Intelligence, 23(2):123-154,
1984.
[Hafner, 1981] C.D. Hafner. An Information Retrieval
System Based on a Computer Model of Legal Knowledge. UMIResearch Press, Ann Arbor, 1981.
[Jones and Sergot, 1992] A.J.I. Jones and M. Sergot.
Deontic logic in the representation of law: Towardsa
methodology. Artificial Intelligence and Law, 1(1):4564, 1992.
[McCarty and Sridharan, 1981] L.T. McCarty and N.S.
Sridharan. The representation of an evolving system
of legal concepts: II. Prototypes and deformations. In
Proceedings of the Seventh International Joint Conference on Artificial Intelligence, pages 246-53, 1981.
[McCarty, 1990] L.T. McCarty. Artificial intelligence
and law: Howto get there from here. Ratio Juris,
3(2):189-200, 1990.
[McCarty, 1994] L.T. McCarty. Modalities over actions.
In Principles of Knowledge Representation and Reasoning: Proceedings of the Fourth International Conference (KRY4), pages 437-448. Morgan Kaufmann,
1994.
[Michaelsen, 1984] R.H. Michaelsen. An expert system
for federal tax planning. Expert Systems, 1(2):149167, 1984.
[Schlobohm and McCarty, 1989] D.A. Schlobohm and
L.T. McCarty. EPS-Ih Estate planning with prototypes. In Proceedings of the Second International Conference on Artificial Intelligence and Law, pages 1-10.
ACMPress, June 1989.
[Shoham, 1992] Y. Shoham. Agent oriented programming. Artificial Intelligence, 60:51-92, 1992.
[Susskind and Capper, 1988] R.E. Susskind and P. Capper. Latent Damage Law: The Expert System. Butterworths, London, 1988.
[McCarty and van der Meyden, 1991] L.T.
McCarty and R. van der Meyden. Indefinite reasoning
with definite rules. In Proceedings of the Twelfth International Joint Conferenceon Artificial Intelligence,
pages 890-896, 1991.
[McCarty and van der Meyden, 1992] L.T.
McCarty and 1~. van der Meyden. Reasoning about
indefinite actions. In Principles of KnowledgeRepresentation and Reasoning: Proceedings of the Third
International Conference (KR92), pages 59-70. Morgan Kaufmann, 1992.
[McCarty, 1977] L.T. McCarty. Reflections on TAXMAN:An experiment in artificial intelligence and legal reasoning. Harvard Law Review, 90:837-93, 1977.
[McCarty, 1983a] L.T. McCarty. Intelligent
legal information systems: Problems and prospects. Rutgets Computer and Technology Law Journal, 9(2):265294, 1983. Also published in C. Campbell, editor,
Data Processing and the Law, pp. 125-151 (Sweet and
Maxwell, 1984).
[McCarty, 1983b] L.T. McCarty. Permissions and obligations. In Proceedings of the Eighth International
Joint Conferenceon Artificial Intelligence, pages 287294, 1983.
[McCarty, 1989] L.T. McCarty. A language for legal discourse. I. Basic features. In Proceedings of the Second
International Conference on Artificial Intelligence and
Law, pages 180-189. ACMPress, June 1989.
i38