A Framework for Self-Explaining Legal Documents

L. Karl Branting
Department of Computer Science
University of Wyoming
Box 3682
Laramie, WY 82071
karl@index.uwyo.edu
James C. Lester
Department of Computer Science
North Carolina State University
Box 8206
Raleigh, NC 27606
lester@adm.csc.ncsu.edu

Abstract
Legal document drafting is an essential professional skill for attorneys and
judges. To maintain stylistic and substantive consistency and decrease drafting time,
new documents are often created by modifying previous documents. This paper
proposes a framework for document reuse based on an explicit representation of the
illocutionary and rhetorical structure underlying documents. Explicit representation
of this structure facilitates (1) interpretation of previous documents by enabling
them to explain themselves, (2) construction of documents by enabling document
drafters to issue goal-based specifications and rapidly retrieve documents with
similar intentional structure, and (3) maintenance of multi-generation documents.
The applicability of this framework to a representative class of judicial orders
jurisdictional show-cause ordersis demonstrated.
1 Introduction
Legal problem solving subsumes a number of distinct tasks, including analysis of the legal
consequences of actual or hypothetical sequences of actions, argumentation, advising clients,
planning transactions, and drafting legal documents. Legal document drafting is an essential
professional skill for attorneys and judges. In the U.S., a significant portion of attorneys
workload consists of drafting documents intended to precisely stipulate legal relationships,
such as wills, contracts, and leases, and persuasive documents arising from litigation, such as
pleadings, motions, and briefs. Judges routinely draft performative documents, such as orders
and decisions.
Complex legal documents are typically created by modifying previous documents.
Adaptation and reuse of previous documents is an almost universal practice in U.S. law firms.
Document reuse is beneficial because it reduces drafting time and promotes stylistic and
substantive consistency. However, document reuse requires access to the original intentions
underlying the document, which may not be readily apparent from the documents surface
text. For example, when both parties to a contract agree that modifications should be made,

Jurix 96: Branting and Lester

the assumptions behind the original contract must be reconstructed to determine the precise
textual changes required.
To address the problem of recovering the intentions underlying documents to facilitate
their reuse, we propose a self-explaining documents framework. We say that a document
containing a given discourse is self-explaining if it contains an explicit representation of the
illocutionary and rhetorical1 structure underlying the discourse. Because self-explaining
documents record intentional knowledge, they offer significant potential for the interpretation
and maintenance of complex, multi-generation documents. In particular, they can explain why
a particular clause was included, suggest how an existing document should be modified to
apply to slightly different circumstances, and present arguments for the pros and cons of
alternative clauses.
This paper presents a framework for the use of justification structures for document
reuse. Section 2 describes a representative class of legal documentsappellate jurisdictional
show cause ordersand illustrates how the self-explaining document approach would assist
drafting in this application. Section 3 proposes a dual justification structure that combines
illocutionary and rhetorical structures to represent document intent. Section 4 illustrates the
envisioned behavior of the system when applied to the domain of show-cause order drafting.
Section 5 describes future implementation plans for this work, and Section 6 outlines related
work.
2 Document Reuse in Judicial Drafting
In the Anglo-American legal system, the most prominent judicial documents are appellate
decisions. Appellate opinions typically contain a summary of the facts of the case,
identification of the issues of law raised in arguments by counsel for each of the parties,
pronouncement of the legal propositions supported by the controlling authorities, and
declaration of a decision that resolves the issues by applying the legal propositions to the facts
of the case. The complexity and individuality of appellate opinions makes automated
assistance for such documents unfeasible. However, courts produce a number of other more
routine documents having considerable stylistic and substantive consistency, including
various types of orders issued in response to motions or sua sponte. Jurisdictional show-cause
orders are typical of such orders.
Jurisdictional show-cause orders are generally issued during jurisdictional screening, a
process of determining whether the requirements for an appeal have been satisfied.
Jurisdictional screening is typically performed at the earliest possible stage of an appeal to
permit cases with jurisdictional defects to be recognized as soon as possible. This minimizes
unnecessary consumption of limited judicial resources.
In this paper, discussion of appellate jurisdictional screening will focus on the Colorado
Court of Appeals, where one of the authors, Karl Branting, worked for several years as a staff
attorney. The Colorado Court of Appeals receives on average over 100 new cases per month.
Screening these appeals is too complex for clerical personnel, but must instead be performed
by a staff attorney. The staff attorney examines the case file to determine whether the subjectmatter, finality, and timeliness requirements for appellate jurisdiction have been met. If there
1

An illocutionary operator is a speech act such as informing, requesting, warning, or promising. A

rhetorical operator is a discourse or coherence relation, such as exemplification, generalization, sequence,
or elaboration. See Allen [1987] for a more detailed discussion.

A Framework for Self-Explaining Legal Documents

appears to be a jurisdictional defect, the staff attorney drafts a show-cause order that sets forth
the apparent defect and orders the appellant to rebut the defect within a fixed time period or
face dismissal of the appeal.
--------------------------------------------------------------------Colorado Court of Appeals
Order
No. 90CA0274
Tr. Ct. No. 79DR221
--------------------------------------------------------------------In re the Marriage of
SUSAN W. KIRKPATRICK
Appellant,
and
JOHN B. KNEZOVICH
Appellee.
--------------------------------------------------------------------To: Susan W. Kirkpatrick, appellant, and her attorney, Susan M. Lach
From the notice of appeal filed by appellant and the register of
actions submitted by the clerk of the district court, it appears that
this appeal is from an order denying a motion to change venue or for
an order declining jurisdiction as an inconvenient forum under Section
14-13-108, C.R.S. It further appears that this is not a final judgment
because it does not end "the particular action in which it is entered,
leaving nothing further for the court pronouncing it to do in order to
completely determine the rights of the parties involved in the
proceeding." Harding Glass Co. v. Jones, 640 P.2d 1123 (Colo. 1982);
D.H. v. People, 192 Colo. 542, 561 P.2d 337 (1965).
IT IS THEREFORE ORDERED that the appellant shall show cause, if any
she has, in writing on or before April 17, 1990, why this appeal
should not be partially dismissed without prejudice for failure to
file a final appealable order.
BY THE COURT
Dated: April 3, 1990
Copies to: Counsel of Record

Figure 1: A typical show-cause order.

Figure 1 shows a typical show cause order. This order identifies an apparent defectlack
of finalityand orders the appellant to show cause with 14 days why the appeal should not
therefore be dismissed. Show-cause orders typify legal documents that are produced in
relatively high volume (several hundred per year), are complex enough to require drafting by
an attorney, but have a sufficient degree of stylistic and substantive consistency to facilitate
reuse.

Jurix 96: Branting and Lester

Currently, staff attorneys at the Colorado Court of Appeals draft show cause orders
manually. One approach is for the staff attorney to reuse only those boilerplate text blocks
that are common to all show-cause orders (e.g., From the notice of appeal filed by
appellant). Alternatively, a staff attorney may refer to a collection of previous showcause orders. The attorney can search this collection for a previous order that involved defects
similar to those in the current case. The reusable language from the previous case may then be
transcribed into the current case, or the staff attorney may photocopy the previous order, cross
out the portions inapplicable to the new case, and write in portions specific to the new case.
Manual drafting of show-cause orders has several clear disadvantages. First, the process
is very time-consuming and laborious. The first approach, which reuses only the most general
boilerplate language, entails repeated replication of drafting effort and creates a high
likelihood of inconsistent language. The second approach, reuse of similar orders, depends on
the staff attorneys ability to find and appropriately modify previous orders. This in turn
depends on the attorneys ability to understand the relevant similarities and differences
between the goals that must be achieved by the current show-cause order and the intentions
underlying previous orders.
The difficulty of finding and adapting previous documents is exacerbated by the frequent
personnel changes. For example, at the Colorado Court of Appeals few staff attorneys are
willing to do jurisdictional screening for more than six months, and many do screening for as
little as three months. As a result, jurisdictional screening is typically performed by attorneys
familiar with only the most recent show-cause orders. Staff attorneys may therefore have
difficulty reconstructing the intent behind particular clauses, impeding interpretation and
modification.
One approach to providing automated assistance in the drafting of show-cause orders
would be to design a set of templates for various show-cause orders, e.g., Word Perfect
macros. However, the wide variety of possible jurisdictional defects and the even wider
variety of factual situations that can give rise to jurisdictional defects make it unlikely that a
adequate collection of macros could be devised. Moreover, even if a sufficient set of macros
could be devised, these macros would present the user with an overwhelming number of
choices, making selection of the most appropriate macro unlikely.
We argue that the goals of accuracy, efficiency, and stylistic consistency are best served
by an approach to drafting routine legal documents, such as show-cause orders, based on
retrieval and modification of previous documents. The task of drafting by document reuse
can be summarized as follows:
Given:
A set of goals to be accomplished by the document to be drafted.
A library of existing documents.
Do:
Retrieval. Find the existing document(s) (or combination of document components)
that best satisfy the current goals.
Analysis. Display the goals achieved by the retrieved document or document
component
Comparison. Display the differences, if any, between, the goals achieved by the
retrieved document(s) and the current goals
Adaptation. Remove the portions of text whose only purpose is to satisfy goals that
arent present in the current situation (excision), and add text to satisfy any of the
current goals not satisfied by the retrieved text (augmentation).

A Framework for Self-Explaining Legal Documents

A system for legal drafting by document reuse must therefore enable users to pose at a
minimum three classes of queries:
Goal Exemplification: What are some examples of archival documents or document
segments that achieve a given goal or set of goals? Goal exemplification is necessary
for the retrieval stage of drafting.
Goal Identification: Why was a given segment included in a document? Goal
identification is necessary for the analysis stage of drafting.
Goal Comparison: To what segments of archival documents is a given document
segment most similar? What are the intentional similarities and differences between
two documents (at the most specific abstraction level at which they differ)? Goal
comparison is necessary for the comparison and adaptation stages of drafting.
In addition, the system must provide the functionalities required for systematically
substituting case-specific data from the current case for case-specific data in the previous
case, e.g., names of parties, dates, etc., and compile the result into a new document.
3 Representational Requirements for Self-Explaining Legal Documents
Legal documents can serve a variety of illocutionary goals, including eliciting information,
persuading, memorializing events such as reciprocal communications, or accomplishing
performative goals, such as creating or revoking legal relationships. For example, the goals of
show-cause orders include to make-findings about relevant facts such as the apparent
judgment and filing dates, and make-conclusions concerning the existence of a jurisdictional
defect, and inform the appellant of the apparent defect, time limit for response, and sanction if
no adequate response is filed. Self-explaining legal documents therefore require a vocabulary
of illocutionary goals sufficiently rich to permit accurate document retrieval, analysis, and
adaptation.
However, the illocutionary goal structure of a document is not per se sufficient to
completely determine the documents surface text. In general, the illocutionary goal structure
leaves unspecified rhetorical features such as (1) the order of the textual elements that satisfy
various illocutionary subgoals, and (2) textual elements and stylistic constraints imposed by
the particular genre of the text. Accordingly, self-explaining documents must include the
documents rhetorical structure as well as their illocutionary structure.
The minimum representational requirements for self-explaining documents therefore
include the following:
A taxonomy of illocutionary goals sufficiently expressive to permit retrieval of
documents, comparison of documents, and explanation of document components. The
necessary granularity of the leaf nodes of this taxonomy depends on the requirements
of the particular document genre and the pragmatics of the users application.
A taxonomy of rhetorical goals.
A representation of templates at a level of granularity corresponding to the leaf nodes
of the illocutionary goals. The substitutable elements of text templates should be
tagged with a data-type so that entire documents, or document components, can be
viewed either as uninstantiated templates or as fully instantiated texts.
A set of link annotations, e.g. , annotations providing the legal authority under which a
given legal goal is satisfied by a given set of performative subgoals, and annotations
explaining why a given rhetorical goal is satisfied by a particular set of subgoals in a
given document genre.

Jurix 96: Branting and Lester

Our model of the intentional structure of jurisdictional show-cause orders includes five
illocutionary operators: establish, inform, find, rule, and order. These operators are
necessity because of the nature of performative judicial documents (i.e., judicial documents
intended to create or alter legal relationships). Typically, such documents must make findings
of relevant facts, rulings of the applicable law, and order some change in legal status. In
addition, there may be intermediate reasoning steps that must established, and the recipient of
the document may need to be informed of facts which are neither findings, rulings, or orders,
such as the applicable legal authority.
To illustrate, a simplified representation of the illocutionary structure of the show-cause
order shown in Figure 1 is set forth in Figure 2. The root illocutionary goal is to establish
(est) the prerequisites for dismissal. Achieving this goal requires the operators displayed as
children of the root: set forth the preamble (inform); establish a jurisdictional defect; and
order a response. The goal to establish a jurisdictional defect, in turn, is achieved by
establishing that the order being appealed is non-final. This goal, in turn is established by
finding that the order is a motion to change venue, ruling that a motion to change venue is
not a final order, and informing the appellant of the authority that supports the ruling. The
subgoals of the order requirement are informing the appellant of the date a response is due,
the sanction for failing to respond, and the defect. Each leaf illocutionary goal is connected to
a text segment intended to achieve that goal.
A simplified representation of the rhetorical structure of the show-cause order is shown
in Figure 3. Unlike the illocutionary structure, the rhetorical structure is closely connected to
the surface text of the document. The top-level node of the rhetorical structure, show-cause
order, indicates the genre of the document. The children of the root node consist of a set of
elements sufficient for a document of that genre. Together, the annotated illocutionary and
rhetorical goal structures constitute the justification structure of a document.

4 Example: Use of Self-Explaining Show-Cause Orders

This section illustrates through an example the manner in which the justification structure
described in the previous section could be used to assist in drafting new show-cause orders.
Suppose that a relatively inexperienced staff attorney, Jones, is screening an appeal in the case
In re the Marriage of Herbert W. Smythe and Catherine Smythe and discovers an apparent
jurisdictional defect. Under the Colorado Appellate Rule 4(a), an appeal to the Court of
Appeals is timely only if a notice of appeal is filed within 45 days of (1) the date of entry
of judgment, if the parties are present when the judgment is announced, or (2) the date of
mailing of the notice, if notice of the judgment is transmitted by mail. However, if the
appellant files a motion for post-trial relief the due date for the notice of appeal is extended
until the motion is ruled upon or expires.

A Framework for Self-Explaining Legal Documents

----------------------------------Colorado Court of Appeals
No. 90CA0274

Order
Tr. Ct. No. 79DR221

----------------------------------In re the Marriage of

SUSAN W. KIRKPATRICK
Appellant

est(preq-for-dismissal)

and

inform(preamble)

JOHN B. KNEZOVICH

est(jurisdictional-defect)

-----------------------------------

Appellee

est(nonfinal-order)

To: Susan W. Kirkpatrick, appellant, and her attorney,

find(venue-motion)

Susan M. Lach

rule(nonfinal
(venue-motion))

From the notice of appeal filed by appellant and the register

inform(authority)

appears that this appeal is from an order denying a motion

of actions submitted by the clerk of the district court, it

to change venue or for an order declining jurisdiction as an

Order(response)

inconvenient forum under Section 14-13-108, C.R.S. It further

inform(response-date)

appears that this is not a final judgment because it does not

inform(sanction)

end "the particular action in which it is entered, leaving

nothing further for the court pronouncing it to do in order to

inform(defect)
completely deterinine the rights of the parties involved in
the proceeding." Harding Glass Co. v. Jones, 640 P.2d 1123
(Colo. 1982); D.H. v. People, 192 Colo. 542, 561 P.2d 337
(1965).

IT IS THEREFORE ORDERED that the appellant shall show

cause, if any she has, in writing on or before April 17, 1990,
why this appeal should not be partially dismissed without
prejudice for failure to file a final appealable order.
BY THE COURT
Date: April 3, 1990
Copies to: Counsel of Record

Figure 2: A simplified representation of the illocutionary structure of the show-cause

order in Kirkpatrick.

Jurix 96: Branting and Lester

----------------------------------Colorado Court of Appeals
No. 90CA0274

show-cause order

Order
Tr. Ct. No. 79DR221

----------------------------------In re the Marriage of

caption

SUSAN W. KIRKPATRICK
Appellant
and

ordered-party

JOHN B. KNEZOVICH
Appellee
-----------------------------------

procedural
history

To: Susan W. Kirkpatrick, appellant, and her attorney,

Susan M. Lach
From the notice of appeal filed by appellant and the register
of actions submitted by the clerk of the district court, it

order

appears that this appeal is from an order denying a motion

to change venue or for an order declining jurisdiction as an
inconvenient forum under Section 14-13-108, C.R.S. It further

signature

appears that this is not a final judgment because it does not

end "the particular action in which it is entered, leaving
nothing further for the court pronouncing it to do in order to

date

completely deterinine the rights of the parties involved in

the proceeding." Harding Glass Co. v. Jones, 640 P.2d 1123
(Colo. 1982); D.H. v. People, 192 Colo. 542, 561 P.2d 337

recipients

(1965).

IT IS THEREFORE ORDERED that the appellant shall show

cause, if any she has, in writing on or before April 17, 1990,
why this appeal should not be partially dismissed without
prejudice for failure to file a final appealable order.
BY THE COURT
Date: April 3, 1990
Copies to: Counsel of Record

Figure 3: The rhetorical structure of the show-cause order in Kirkpatrick.

Suppose that the Smythe file shows that notice of the judgment was mailed on September
22, 1995, but the notice of appeal was not filed until November 7, 1995 (more than 45 days
later). Jones therefore wishes to draft a show-cause order having to do with an untimely notice
of appeal. Representation of the justification structure of existing documents could assist
Jones in the following steps:
Initial Document Retrieval
User: Initiates the document planning session by selecting Construct new document
from pull-down menus.
Document Planner: Presents user with dialog form listing various types of orders and
notices. Among his choices is Show-Cause Order.
User: Selects Show-Cause Order.
Document Planner: Presents user taxonomic hierarchies of illocutionary and rhetorical
goals and substantive legal areas.

A Framework for Self-Explaining Legal Documents

User: Selects marriage from the taxonomy of substantive legal areas.
Document Planner: Asks user if he would like to see a typical show cause order
pertaining to marriage.
User: Answers affirmatively.
Document Planner: Retrieval system indexes into a partition of the case library
containing show-cause documents pertaining to marriage. It retrieves the prototypical
show-cause order for marriage, Kirkpatrick
User: Studies Kirkpatrick, considers how it relates to his current task, and highlights a
region of the document that seems different than the one he would like to have in the
current: It further appears that this is not a final judgment because it does not end the
particular action in which it is entered, leaving nothing further for the court
pronouncing it to do in order to completely determine the rights of the parties involved
in the proceeding.
Document Planner: Asks user if he would like to know why this section was included.
User: Answers affirmatively.
Document Planner: Uses the illocutionary structure to generate an explanation of the
communicative goals behind the document. First it locates the substructure of the
illocutionary structure that covers the region:
find (non-final-order-type(motion-to-change-venue))
It then ascends the supergoals to find
est (jurisdictional-defect(no-finality))
From this it generates the explanation The text you have highlighted was included to
establish that there was a lack of a final appealable order. It then poses the question, Is
this true in the case for which you are drafting an order?
User: The user selects No.

Document Planner: The planner inspects its operator library to determine other

reasons for establishing jurisdictional defects. It then finds all of the operators that
achieve the goal of establishing jurisdictional defects and asks the user about the
applicability to the current case. After asking about two others to which the user
responds negatively, it encounters
est (jurisdictional-defect(failure-to-file-timely-noa))
<
inform (defective-order)),
est(untimely-NOA(failure-to-file-timely-NOA)).
and asks the user, Is the jurisdictional defect in your case a failure to file a timely notice
of appeal?
User: Responds affirmatively.
Document Planner: Asks the user if he would like it to search for show-cause orders in
which the jurisdictional defect is a failure to file a timely notice of appeal.
User: Responds affirmatively.
Document Planner: The planner again locates the sub-partition of the document
library pertaining to documents whose illocutionary structure contains an instantiated
Establish operator with the following pattern
est (jurisdictional-defect(failure-to-file-timely-NOA))
It locates Cohen, whose illocutionary structure has
estjurisdictional-defect(failure-to-file-timely-NOA))

Jurix 96: Branting and Lester

<
est(untimely-NOA(failure-to-file-timely-NOA))
It then asks the user if he would like to use Cohen as a model.
User: Responds affirmatively
Interactive Document Compilation
Document Planner: The planner then begins to construct an illocutionary structure for
the users show-cause order by traversing the illocutionary structure of Cohen,
performing a structure mapping, and querying the user for information. As it traverses
the tree, it encounters
est (time-limit-begins(cohen))
<
find(parties-present-at-judgment(cohen))
and so it asks if both parties were present at the Smythe judgment.
User: Responds in the negative.
Interactive Document Adaptation and Constraint Propagation
Document Planner: As a result of the interaction, the planner has established that the
Cohen case is not a precise analogue of the Smythe case. It can now provide the user
with a good deal of assistance by exploiting its knowledge of the illocutionary
structures of other documents. Its first step is to determine alternative means of
achieving the goal of establishing when the time limit begins. It searches the operator
library and discovers
est(commencement-of-time-limit)
find(mailed-noticed-of-judgment).
so it asks the user if, in the Smythe case, the judgment notice was mailed to the parties.
User: Responds affirmatively.
Document Planner: Next it asks if the user if he would like to see an excerpt of a show
cause order dealing with a mailed judgment.
User: Responds affirmatively.
Document Planner: Indexes into the document library and examines the illocutionary
structures. It is unable to locate a show cause order pertaining to marriage in which the
judgment notice was mailed. It then searches other types of show cause orders and it
finds a document with an illocutionary structure including the instantiated operator
est (time-limit-begins(canada))
<
find(mailed-noticed-of-judgment(canada))
It then climbs the illocutionary structure of Canada up two levels to obtain the context
in which the decision was made. Here it finds the instantiated goal
est(duration-of-time-limit(45 days, C.A.R. 4(a)))
Next, it locates all text in the Canada document that was generated by this goal and all
its progeny goals. It then spawns a document viewer, and displays this text to the
user:

10

A Framework for Self-Explaining Legal Documents

However, it appears that the trial courts order granting summary judgment
was entered on February 9, 1987 and mailed to counsel of record on February
10, 1987, and that no C.R.C.P. motion for post-trial relief was filed.
It then asks the user if Smythe presents a similar situation.
User: Responds affirmatively.
Document Planner: It now structure-maps the Canada illocutionary structure for the
mailing segment over to the Smythe case by interacting with the user. It first asks
when the judgment was mailed.
User: Enters September 22, 1995.
Document Planner: This triggers a constraint propagation that calculates a new value
for the ?DUE-DATE, which is 45 days after the mailing date. The planner explains the
calculation to the user, using the authority clause (?AUTHORITY = C.A.R. 4(a)) to
point the user to the justification. The planner then requests the user to confirm the
due date.
User: Confirms date.
Document Planner: Encounters the goal in Canada
find (no-post-trial-relief-motion(canada))
and asks if in Smythe, there was a post trial relief motion.
User: The user responds affirmatively.
Document Planner: The planner then integrates the sub-structure of the illocutionary
structure that it analogized from Canada into the super-structure that it is in the process
of building from Cohen.

Completing Document Compilation

The planner passes the newly constructed illocutionary structure for the Smythe case to
the document compiler, which performs the following actions. It first checks if any internal
constraints have been violated (which may happen as a result of adaptive interactions). In this
case, no violations have occurred. It next passes the illocutionary structure to the document
generator, which takes all inform operators and realizes them in text. Finally, it locates the
rhetorical structure associated with documents of the given class (in this case, Show-Cause
orders). The rhetorical structure is used to (1) order the text resulting from the inform
operators and (2) format the document. The resulting document is then displayed in the
document studio and saved in the archive.

5 Implementation Plans
We are currently applying the self-explaining document approach to Colorado Court of
Appeals show-cause orders. As argued above, show-cause orders appear to typify legal
documents that are produced in relatively high volume (several hundred per year), are
complex enough to require drafting by an attorney, but have sufficient stylistic and
substantive consistency to facilitate reuse.
We have developed an initial implementation of illocutionary rules for a set of
representative show-cause orders and are current engaged in developing rhetorical rules for
these documents. To supply a uniform representation for all of the knowledge structures, all
illocutionary structures, rhetorical structures, and planning operators arel be implemented in a

11

Jurix 96: Branting and Lester

unification-based constraint system [Elhadad, 1991]. Use of this single representation permits
all adaptation procedures (document compilation and consistency checking) to be performed
with a single mechanism: constraint propagation. Although this brings with it an initial
development overhead, we expect to witness significant software engineering benefits in the
form of rapid extensions to the domain theory. Once the system for our first class of
regulatory documents is complete, we will run extensive empirical evaluations to measure the
gains in efficiency and correctness.
6 Related Work
Our approach to self-explaining documents draws on four different lines of research:
discourse structure analysis; the theory of argumentation; explanation generation; and
automated document drafting. The primary focus of research in discourse structure has been
accounting for the coherence of expository or other communicative text through hierarchical
structures of rhetorical and other discourse relations, e.g. Grosz et al. [1986] and Hobbs
[1979]. The formalization of inter-sentential discourse relations is a key requirement for the
development of self-explaining documents.
The most directly relevant portion of research in discourse structure is speech act theory.
Initiated by J.L. Austin, who was primarily concerned with explicit performatives [Austin,
1962], speech act theory addresses the illocutionary content of discourse, that is, the goals
that a speaker intends to accomplish through that discourse [Grice,1975; Searle, 1969].
The theory of argumentation addresses texts intended to persuade, establish, or prove.
For example, Toulmin [1958] analyzed argumentative texts in terms of the concepts of
warrant, ground, conclusion, backing, and qualification. This model has been widely applied
to the analysis [Marshall,1989; Zeleznikow et al., 1995] and creation [Bench-Capon et al.,
1995] of legal documents. Argument structure, like other forms of illocutionary goal structure
but unlike rhetorical structure, does not directly address the surface form of texts. This line
of research is particularly relevant to the analysis of the illocutionary structure of persuasive
or dispositive documents such legal briefs and judicial decisions [Branting, 1993a].
The explanation community has extensively studied the process of planning and realizing
text given a set of discourse specifications. Over the past decade, their work research on
discourse planning [McKeown, 1985; Paris, 1988; Hovey, 1990; Hovey, 1993; Cawsey,
1992; Suthers, 1993; Moore, 1995; Mittal, 1993; Lester et al., 1996] has produced a variety of
techniques for determining the content and organization of many genres of text. Perhaps
because of the necessity of coping with the myriad underlying rhetorical, illocutionary, and
argument structures in discourse generation, this work has yielded a variety of mechanisms for
determining the content and organization of multi-sentential text, a key capability of selfexplaining documents.
Automated document drafting research is the fourth relevant research area. Two
important areas of automated document drafting research are automated legal drafting and
automated report generation. A large number of automated legal drafting systems have been
developed in recent years, but most involve creation of text templates that are then
instantiated to create particular documents [Lauritsen, 1992]. Some progress has been made in
exploiting explicit representations of the relationship between generic documents and
document instances and of constraints among document components [Daskalopulu et al.,
1995]. However, there is a growing recognition in the Law and AI community that a
declarative representation of the knowledge underlying the selection and configuration of

12

A Framework for Self-Explaining Legal Documents

textual elements is essential for the development of tools that embody the expertise of legal
drafting experts [Gordon, 1989; Lauritsen, 1993].
The automated report generation community has addressed another form of text
production from an underlying domain structure: the derivation of technical documentation
from program traces generated during software development or use [McKeown et al., 1995;
Korelsky et al., 1993]
7 Summary
Document drafting is a critical task for attorneys and judges. To cope with the complexities
of constructing and maintaining complicated, multi-generation documents, we have proposed
the self-explaining document framework. We are currently instantiating the framework in a
system for drafting appellate jurisdictional show-cause orders. It is our hypothesis that, by
equipping documents with a justification structure that includes both illocutionary and
rhetorical components, self-explaining document systems can assist document designers with
constructing, maintaining, and interpreting complex documents.
Our initial focus has been on legal documents in the common law system. However, there is
nothing about the self-explaining document approach that is limited to the common-law
system or, indeed, legal documents. This approach is applicable to any documents that
accomplish well-defined illocutionary goals and are subject to consistent rhetorical
constraints.
Acknowledgments
This research is supported in part by grants from the National Center for Automated
Information Research and by NSF Faculty Early Career Development Grant IRI-9502152.
References
J. Allen. Natural Languare Understanding.Benjamin/Cummings, Menlo Park, CA, 1987.
J. Austin. How to do things with words . Oxford U. Press, New York, 1962.
T. Bench-Capon and G. Staniford. PLAID - proactive legal assistance. In Proceedings of
the Fifth International Conference on Artificial Intelligence and Law, pages 81--88, 1995.
L. K. Branting. An issue-oriented approach to judicial document assembly. In Proceedings of
the Fourth International Conference on Artificial Intelligence and Law, pages 228--235,
Amsterdam, The Netherlands, June 15--18, 1993. ACM Press.
L. K. Branting. A reduction-graph model of ratio decidendi. In Proceedings of the Fourth
International Conference on Artificial Intelligence and Law, pages 40--49, Amsterdam, The
Netherlands, June 15--18, 1993. ACM Press.
A. Cawsey. Explanation and Interaction: The Computer Generation of Explanatory
Dialogues. MIT Press, 1992.

13

Jurix 96: Branting and Lester

A. Daskalopulu and M. Sergot. A constraint-driven system for contract assembly. In

Proceedings of the Fifth International Conference on Artificial Intelligence and Law, pages
62--70, 1995.
M. Elhadad. FUF: The universal unifier user manual version 5.0. Technical Report CUCS038-91, Department of Computer Science, Columbia University, 1991.
T. Gordon. A theory construction approach to legal document assembly. In Pre-Proceedings
of the Third International Conference on Logic, Informatics, and Law, pages 485--498,
Florence, 1989.
H. Grice. Logic and conversation. In P. Cole and J. Morgan, editors, Syntax and Semantics
2: Speech Acts, pages 41--58. Academic Press, New York, NY, 1975.
B. Grosz and C. Sidner. Attention, intention, and the structure of discourse. Computational
Linguistics, 12(3), 1986.
J. Hobbs. Coherence and co-reference. Cognitive Science, 3(1):67--82, 1979.
E. H. Hovy. Pragmatics and natural language generation. Artificial Intelligence, 43:153-197, 1990.
E. H. Hovy. Automated discourse generation using discourse structure relations. Artificial
Intelligence, 63:341--385, 1993.
T. Korelsky, D. McCullough, and O. Rambow. Knowledge requirements for the automatic
generation of project management reports. In Proceedings of the Eighth KnowledgeEngineering Conference. IEEE Computer Society Press, September 20--23 1993.
M. Lauritsen. Technology report: Building legal practice systems with today's commercial
authoring tools. Law and Artificial Intelligence, 1(1), 1992.
M. Lauritsen. Knowing documents. In Fourth International Conference on Artificial
Intelligence and Law, pages 185--191, Amsterdam, 1993. ACM Press.
J. C. Lester and B. W. Porter. Scaling up explanation generation: Large-scale knowledge
bases and empirical studies. In Proceedings of the Thirteenth National Conference on
Artificial Intelligence, Portland, Oregon, to appear 1996.
C. Marshall. Representing the structure of a legal argument. In Proceedings of the Second
International Conference on Artificial Intelligence and Law, pages 121--127, Vancouver,
B.C., June13-16 1989.
K. R. McKeown. Text Generation: Using Discourse Strategies and Focus Constraints to
Generate Natural Language Text. Cambridge University Press, 1985.

14

A Framework for Self-Explaining Legal Documents

V. Mittal. Generating Natural Language Descriptions with Integrated Text and Examples.
PhD thesis, University of Southern California, September 1993.
J. D. Moore. Participating in Explanatory Dialogues. MIT Press, 1995.
K. McKeown, J. Robin, and K. Kukick. Generating concise natural language summaries.
Information Processing and Management, 1995. Special Issue on Summarization.
C.Paris. Tailoring object descriptions to a user's level of expertise. Computational
Linguistics, 14(3):64--78, September 1988.
J. Searle. Speech Acts: An Essay in the Philosophy of Language. Cambridge University
Press, Cambridge, 1969.
D. D. Suthers. An Analysis of Explanation and Its Implications for the Design of Explanation
Planners. PhD thesis, University of Massachusetts, February 1993.
S. E. Toulmin. The Uses of Argument. Cambridge University Press, 1958.
J. Zeleznikow and A. Stranieri. The split-up system: Integrating neural networks and rulebased reasoning in the legal domain. In Proceedings of the Fifth International Conference on
Artificial Intelligence and Law, pages 185--194, 1995.

15