A Modest Proposal for Annotating the Dialectical State of a Dispute
Ronald P. Loui
Department of Computer Science and Engineering
Washington University
St. Louis, MO
Apology
This conference is about visualizing evidence in legal settings, which is
a much broader topic than I am addressing in this paper. Normally no
comment would be required for the specialization of an author's concerns,
except that as the academic descendant of the inductivists in philosophy
of science, I might have had very different things to say about
visualizing inductive and evidential arguments, arguments based on
probabilistic connections and default assumptions, were I not already
invested in the visualization of qualitative defeasible arguments.
My starting point is the Toulmin diagram for visualizing the structure of
defeasible arguments. That is what I have written about in the past and
it is what I am addressing today. But I do so with some reluctance
because I believe that this Toulmin kind of argument-diagramming is of
limited relevance to those who have practical persuasive legal needs. The
dialectical diagramming that my colleagues and I discuss in the theory of
argumentation, after Toulmin, is mainly a device for elucidating logical
structure. Our uses of these diagrams are for theorizing, not
persuading.
Nevertheless, Peter Tillers and Henry Prakken seem to think that
a paper on this subject is required from me, so here it is.
Toulmin
In an earlier paper with many co-authors, I referred to Toulmin diagram's
as "a spaghetti of boxes and arrows." Many people in the CSCW
(computer-supported collaborative work) community had been writing on the
user interfaces that they had created based on Toulmin's box-and-arrow
diagram, and there seemed in the early 90's a plethora of programs
available to support the manipulation of such box-and-arrow diagrams. Of
course, one draws a Toulmin domino when one wants to connect the
supporting claim to the claim it supports, and there is nothing wrong with
one Toulmin domino (three or four boxes and two or three arrows) by
itself. The problem is when many Toulmin dominoes are connected. Since
there is no prescription for the layout of the various boxes and arrows,
one must rely on the discipline of the user to create a visually
understandable diagram. As boxes and arrows proliferate, it is not clear
whether even the most disciplined arrangements of Toulmin are
cogent, or even comprehensible.
(from http://www.csdl.tamu.edu/~marshall/ht91.pdf)
Meanwhile, we had been diagramming arguments as trees, where the main
claim of the argument sat at the root of the tree, and the unsupported
claims sat at the leaves of the tree. (Trees were drawn with the root at
the top of the page, as computer scientists prefer.) A defeasible
connection between the children of a node and the node was made
implicitly. Toulmin instead would make a box that explicitly stated this
defeasible connection. By convention, if a node labeled p had children
labeled a, b, and c, then the required logical connection was "if a, b,
and c, then defeasibly p." This was perhaps the most common diagramming
of arguments in the literature in the late 90's.
As several arguments might particpate in a dialectical structure,
where one argument countered another, and a third countered it in turn,
seeking to reinstate the first argument, we would draw a second tree
showing the relationships between the arguments. Some would call this the
meta-argument tree, or the inter-argument tree, because it showed how the
arguments related rather than how the arguments were constituted.
(from http://citeseer.ist.psu.edu/cache/papers/cs/11317/http:zSzzSzwww.cs.wustl.eduzSz~louizSzsurvey.pdf/carlos00logical.pdf)
My own preference was to combine the two diagrams, so that dialectical
relevance had a left-to-right discipline, and support was top-down.
Alternate arguments would also be listed top-down, but would not be
confused with support since each alternate argument would be its own tree,
connected within itself, but not connected to the other tree.
If one insisted on comparing this to a Toulmin diagram, there were several
differences, all working in our favor and against Toulmin. First, the
largely redundant warrant and backing boxes would not be explicitly
shown. If an attack were made against a warrant, then the warrant could
be added at that time, or a counterattack would simply be aimed at a link
instead of a node. Nothing prevented putting the names of cases or
statutes on links, for example, and warrants are often citation references
rather than textual statements. In any case, it was useful not to double
the number of boxes unnecessarily.
Second, the arrangement into columns was a required part of our regimen.
Alternate columns would represent pro and con, so that any argument in the
second and fourth columns would presumably be adversary to the position
taken in the first column. Reading left-to-right would reconstruct the
dialectical dialogue in a logically useful way.
Third, there was no explicit requirement that a counterargument link to
the exact node that was its logically contrary proposition or claim.
This could be done, but indeed would promote spaghetti. It was
sufficient, it seemed, to link the argument to its preceding column,
wherein one would find the object of the argument's attack. Since the
number of arguments in a column and the number of claims in an argument
presumably would not be too large, the connection could be made easily
simply by visually scanning for a negation. Even without locating this
contrary proposition in the prior column, it was knowable simply by
negating the claim at the root of the counterargument tree. Indeed, the
heavy dotted line served as much to indicate a new, alternate attack, as
to indicate, redundantly, that its relevance depended on something claimed
in the prior column.
While this style of recording dialectical structure was intuitive, it
actually clashed with the conventions of Rescher and the dialogue
logicians, who preferred dialogical time to run top to bottom, and to
intertwine support, challenge for support, and rebuttal. The logical
exchange under Rescher would appear as:
| PRO | CON |
| P | why P? |
| A,B, C, and D | why D? |
| D2 | why D2? |
| E and F | not-D |
| why not D? | G |
| why G? | H |
| not-G | why not-G? |
| H, J, and K | not-B |
| why not B? | H and L |
It is compact and can be written without graphical artifacts, but it does
not display the argumentative structure, nor even indicate whether a move
is sufficient. Rescher-form is in some sense the textual locutionary
form that the Toulmin diagram and all subsequent diagrams attempt to
compile into a more visually scannable, pictorial form.
Encapsulation
With the advent of HTML, and inspired by the encapsulation of dialogues
into separate windows in computer GUI's, we proposed simplifying the
diagrams further by removing all of the arrows. Support relations would
be implicit in the narrowest containments. All containments after that
would be of dialectical significance: left-to-right for countering,
rebutting, or undercutting; top-to-bottom for alternate attacks of the
same argument.
We even proposed several views of arguments in this encapsulated form,
where one view gave the authorities of the claims, another gave a paraphrase,
and yet another gave a simplified logical form.
(http://www.cs.wustl.edu/~room5/new/ladue/antigone.out3.html)
But we were crestfallen to realize that this form derives much of its
intuitiveness from the simple two-column outline form that is familiar
to high school debate participants.
For example,
could be written as
P
A
B
C
D
D2 not-D2
E G
F H
not-G
H
J
K
|
with no loss of visual clarity.
This two column form, "debate-form," is not entirely satisfying, as here it
permits the lines for E and F's support of D2 to collide with the lines
for G and H's support of not-D2. E and G share lines, but their sharing
has a different semantics from D2 and not-D2 sharing a line. Where would
a counterargument against E be placed in this scheme? Moreover, the
semantic overloading of the line is exacerbated if we permit the third
column and make no provision for overlapping lines (E bears no relation to
not-G, and they share a line):
P
A
B
C
D
D2 not-D2
E G not-G
F H H
J
K
|
But the reality is that the semantics of boxes within boxes is not so
transparent, and strict adherence to too many encapsulation rules can
actually decrease comprehension. The main advantage of the boxes is that
they permit a rotation from top-down to left-right for lists of supporting
propositions.
For the diagram,
the equivalent naive outline,
P
A
B
C
D
D2 not-D2
E G not-G
F H H
J
K
not-D2
H
L
|
is really equally useful, if a bit ad hoc. And any attempt to clarify the
semantics of a line by imposing a push-and-pop spatial grammar (e.g., the
left-hand-side contiguity always defers to the right-hand-side, during a
counterargument) simply ruins the visual appeal. The proper rendering,
while preserving line semantics, decreases density and reduces
juxtaposition within arguments, thus increasing scan effort:
P
A
B
C
D
D2 not-D2
G not-G
H
J
K
H
not-D2
H
L
E
F
|
These are exactly the considerations that first led to our adoption of
encapsulation.
Web Markup, Dynamics, and Animation
We are now all familiar with the font control and dynamics of HTML, and
the possibilities for rendering argument are dizzying. Not only did we
use color to differentiate PRO's and CON's positions above, but we
could have used size and style, face and font-weight as well. All of the
original CSCW argument systems were precursors of hypertext systems, so
their authors envisioned actions of hiding and unhiding upon user action.
For example, the outline form above seems not so sparse when counterarguments
can be hidden or restored from hiding. Thus,
might display the main argument, whereupon clicking on the box next to
D2 exposes the argument for not-D2, and further options for exploring
counterarguments.
Interface dynamics might include a mouseover or pop-up window for viewing
longer forms of the claim, or viewing the authorities of the defeasible
rules when mousing over the "links".
In the extreme, animations could be used, which would be difficult to
print in a paper such as this, even with screen captures. Animations
could show not only the historical development of the dialogue or its
dialectical structure; animations could also be used to show the
differing views of the state of the dispute, especially as participants
were willing to accept certain assertions, presumptions, or defeat relations
among competing arguments.
TO DO: an animation of multiple views
TO DO: non-animation where font corresponds to criticality of claim.
A Plea for Structure over Style
No doubt, the enterprising software interface designer has already thought
of connecting Microsoft Word functions and Adobe Photoshop tools to the
layout of arguments. Why not check the spelling and grammar? How about
smart quotes and a minimum font size?
There is a fine line between aiding the thought process that underlies the
formation of rational beliefs, and the markup of text for its own sake, or
for mainly stylistic purposes. Of course, persuasion may indeed at times
require style. It is simply good practice to be as clear as possible, and
style helps. Perhaps the rhetorical desire to produce multi-column
aligned outlines would eventually be satisfied by authors of PowerPoint
providing some kind of columnar text support for PRO/CON presentations.
Our real interest in diagramming arguments derives from a desire to
understand structure as sophisticates. It seems our natural obligation to
offer symbolic regimes that sometimes go far beyond the work-a-day
visualizations. It would be folly to impose complex structure on all
argumentative forms. Even when the argument could be diagrammed so as to
expose inherent specific structure, the academic must recognize that the
practitioner need not do so; even when the user could use a template with
more numerous well-formedness constraints, the programmer must recognize
her right to use a simpler template. Still, our job is to invent those
templates.
Here are two forms that derive from the analysis of legal reasoning
from precedent. These are simple diagrams, but have had no real hearing
in the community of those who like diagramming.
The first is the idea of Kevin Ashley, in HYPO. The idea begins simply
enough: a present case is related to a prior case by the enumeration of
its similarities and dissimilarities.
To argue P on the basis of a past case exhibiting A, B, C, and D, one need
only remark that the current case has A and B in common, has C as a
distinction, did not report on D, and has an added fact of possible
relevance, E, that was not reported in the precedent. We can find this
basic idea in Raz, complete with letters A through E.
PrecedentCase CurrentCase
P P
A A
B B
C not-C
D
E
|
Without going into too much detail, Ashley proposes an improvement of this
model for the restricted domain of reasoning about trade secrets
violations. He manually determines the dimensions along which precedent
cases and current cases may be compared for similarity. He identifies
each dimension as inherently pro-plaintiff or pro-defendant so that
reasoning can be augmented by dimensions, or factors, that are not present
in past cases.
The result is a "claim lattice," which is intended to show the most on-point
cases for plaintiff and defendant, based on subset-relations among factors
and the inclusion of additional pro-plaintiff factors by the plaintiff,
the omission of additional pro-defendant factors by the plaintiff, the
inclusion of additional pro-defendant factors by the defendant, or the
omission of additional pro-plaintiff factors by the defendant.
The second template, or argument form, is the generalization and possible
improvement of Ashley's analysis, according to my co-author, Jeff Norman,
and myself. (This in turn may have been improved by Prakken and Sartor,
depending on what you think of analogies that are based on less similarity
than could even be used to effect the past case's main argument!)
The main idea here is to record the actual disputation tree from the past
case. An attempt to make use of a past case's argumentation is not just
a listing of similarities, but a recall, re-enactment or re-citation of
the past arguments, to the extent that they can be made in the current
context. The synthetic contribution of the prior case is the judicial
decision that one argument defeated another in the past case; hence, that
same argument should defeat the same counterargument in the current case,
all things being equal. However, if the past case's main argument
survived a counterargument in the presence of a reinstating argument, and
if that reinstating argument is not available to the arguers of the
current case, then the authority of the past case is dubious.
What can be visualized in this scheme are the past relationships among
arguments, and the current context's ability to recall each of those
arguments, in part or in whole.
Here, Precedent1 and Precedent2 appear to conflict over the conclusion of
Q in the current case. But the relevant similarities are not "A B C D H
J" versus "A B C E H M P", which is an indeterminate contest. The
relevance of Precedent1 is that the "A B C for Q" argument can be made.
The counterargument that was disfavored in Precedent1 is irrelevant in the
current case because C is not a current issue. Meanwhile, Precedent2
contains three arguments, all of which can be made in the current fact
situation, and the argument for Q in Precedent2 is actually stronger than
the argument for Q in Precedent1. Nevertheless, at least in the context
of the disrupting argument "P for not-F", the "A B C E H for not-Q"
argument was preferred in Precedent2. Thus, Precedent2's decision is
authoritative here.
Perhaps not so many case-based analogical arguments in the real world can
benefit from this kind of analysis, but I believe this kind of argument
diagramming, and the ambition for a calculus of rationality, should not be
lost as we see stylistic markup proliferate among GUI tools for
argumentation and cooperative problem solving.
No less a figure than Cass Sunstein pronounced HYPO to be inadequate as an
account of general legal analogy, at a conference sponsored by Chicago Law
Review students about a decade ago. However, all of his concerns appear
to my eyes to be addressed in this deeper, argument-based model of the
case. I do look forward to legal scholars' reactions to this subsequent
modeling of the logic of analogical reasoning. It has been almost fifteen
years since we proposed it, and HYPO itself is now over twenty years old. It
is a bit unfair to be looking at first generation models when judging the
entire AI and Law field.
Jonathan Swift's Modest Proposal
The title of this talk contains the phrase "Modest Proposal," and there is
a certain obligation to live up to Jonathan Swift's legacy. If there were
something as outrageous as the famine of Irish families in the world of
argument visualization, I would not hesitate to stryke a satyrical tone.
Perhaps if a generation of lawyers had grown too enamored of Toulmin
spaghetti diagrams, it would be high time to suggest that we feed them a
carb-free diet of old-fashioned prosodic meat.
Truth be told, I was simply in a modest mood, in a humble place, when I
gave Henry Prakken a title for this talk.
I had in mind the unambitious annotation of an argument with the
warrant-status of all hidden arguments: red for a defeating counter
argument, pink for an interfering counterargument, pale pink-gray for a
defeated putative counterargument, black for an effective reinstater, and
gray for an ineffective putative reinstater. Parentheses could be used to
group the dialectical chains. D2 below has a putative interferer that is
defeated by a reinstater, but also has an actual interferer with no
rebuttal.
I have indicated the warrant status of P with a gray surrounding box
instead of a black surrounding box because of the existence of the
unrebutted interferer. Visually, one can see immediately that this
is an argument supported on four main points, with one deep (potentially
weak) chain, where a counterargument has been rebutted and another
counterargument has not been.
But thinking further, and a bit lightly, about how arguments could be
annotated based on their warrant status and apparent strength, I am
happy to suggest the following markup.
Here is how you would indicate that an argument is little tested, but
inherently sound, has unusually wide appeal, and is probably an argument
that you would not want to have to defeat.
P
A
B
C
D
D2
E
F
|
|
Here is an argument you have been waiting to make for some time, which you
expect would be a very good argument, that is, if there isn't a better
argument to be found; and it is a familiar argument that has survived many
pointed attacks in the past.
P
A
D
D2
E
F
|
|
And here is an argument that you just don't want to be making, an argument
that many may have made in the past with good intentions, an argument that
derives from a very strong precedent, but an argument that has far too
many points that are now simply indefensible.
not-P
C
D
F
|
|
Finally, this is a fantasy view of how the dialectical state of
longstanding and well-understood arguments might be visualized,
taking mainly into account the potential strength of each.
Postscript
I want to add two brief comments after attending the conference.
First, this conference turned out not to be a celebration of the graphical
forms and templates developed from logics of reasoning. The presentations
that made the greatest impression were the ones that questioned or even
mocked the persuasiveness of ostensibly rational well-formed argument.
Far from the well known rhetorical idea that persuasiveness differs from
rationality (because for example, the audience is easily swayed by
irrational considerations, and therefore, rhetoric ought to be brought
closer to logic), the locus of dissatisfaction actually uncovered a deep
problem in logical method. The innocent question was whether the mere
serialization of premises and conclusion conferred a logical force that
might be mere appearance. Someone might just draw a right-arrow or a
summation line, and claim to be lone rational party in a disagreement,
when in fact there is no real logic being used. The use of logic might be
mere prestidigitation, or worse, pure palaver. No one is suggesting
outright irrationality, the adherence to no reasoning norms whatsoever.
But it is fair to ask, does the mere appearance of being logical, the
adoption of a specific form of the use of language, bring with it
conclusive force? Does this adoption of form, by itself, carry the weight
of an entire intellectual history, culture, or institution, with respect
to belief formation on the issue at hand, among the interested parties to
a specific dispute? Is it sufficient to sound logical in order to be
compulsory regarding a specific conclusion? If so, why? And what culture
does it bring to bear on probative determinations? And which logical form
or forms are so privileged as to carry this weight?
These are questions that arise in logic, usually when two logicians are
disagreeing about the correct logic. "My logic confers validity; your
logic does not." "Right back at you." Who is right? There are usually
no good answers. One answer is that the more complex the logical
constraints, the more difficult it is to speak in that logical way.
Therefore, it would seem, the party that is able to conform to that
complexity must have the superior point of view. Hogwash! Another answer
is that induction has shown that specific ways of drawing conclusions from
premises attach more probably to correct, or at least warranted,
conclusions. Unfortunately, it is a simple matter to corrupt the
empirical data by concocting myriad examples of the invalidity of logical
rules of inference. "p, therefore 'p or q'" is inductively valid? Here
are a thousand examples, where "or" is taken to be "xor", and there is
no longer an inductive justification. Sometimes it is a bit harder to
make examples, such as the dilution of 'p and q' to the conclusion p.
In these cases, the conventionalist and the semanticist have some
satisfaction that a proper speaker of the language should conform to some
rules.
The best answer I could find I reported in Journal of Philosophy, under
the title "How a Formal Theory Can Be Normative" (1990). The principle of
charity and the indeterminacy of tranlation make it difficult to assess
the validity of another speaker's derivations; it is too hard to pin down
the meanings of symbols and too easy to shift blame around the web of
belief, from disputing conclusions to disputing premises. Hence,
formality of language, including logical form, is never compulsory in
social belief formation (unless so mandated by a social authority, such as
a court). Agreement based on a derivation has force in the degree to
which two people agree on the attachment of symbols to the world, and on
the uncontested basis that forms the premises of the argument. To
disagree on logic is as easy as disagreeing on meanings. I am reminded of
Bill Clinton's clever phrase, "It depends on what the meaning of the word
'is' is." To agree on logic is to share a lot of epistemological and
semantical basis. In the extreme, a logical form or a formal theory has
considerable force for the person who imposes it as discipline on the
self. Within the single individual, especially as time goes by, there may
be incommensurability of meaning and plasticity of predicates, but there
can also be a determination to keep meanings fixed, a will to
epistemological constancy. This is especially pertinent to people who
are prescriptive with formal systems, including those who build artificial
intelligence programs. So the Western canon of logical rhetorical form
fails as the ultimate arbiter of valid argument, but it does lead to an
industry of self-help "How To" books, as well as the possibility of
building automated reasoners. The latter might be unable to attach
meanings to their syntax, but they do derive with enviable
self-discipline.
My second comment on this conference has to do with the formal study of
persuasive visual aids. I had the web design course this semester,
in which half the students are interested in marketing, not javascript.
When conference speakers wrestled with the proposition that writing a name
first, or a name larger, or a name in a lighter color, had hidden
rhetorical force, I had no choice but to agree. In fact, I wondered where
was the "Manual of Visual Inference" that described the most fundamental
and systematic ways that information could be manipulated in graphical
presentation to appeal to visual bias. Of course, there are people such
as Edward Tufte who have taken questions like this seriously, and the
conference included some of his disciples. But in AI, we often ask for
the most basic manual that one could give to a four year old, or to a
moron, or to an Army private (because it is such a manual expressed at
that level of detail and explicitness that eventually can become the
program for an autonomous agent). What is the minimal kind of
intelligence that is needed to perform visual inference-biasing tasks?
One might even imagine a more constrained domain, such as the
specification of type for nespaper headlines. "Rule Ten: The more words,
the less the force; Example: 'Bush Administration Strategy Fails To Reach
Milestones' versus simply 'Bush Fails'." In any case, I have never seen
such a manual, and I would like to see one. Instead of inventing another
two hundred logics this year, perhaps AI researchers should try to
formulate two hundred rules for the manipulation of inference in graphical
objects. "One premise is deemed more important than another if it is in a
larger font." There is no reason why such a basic manual cannot be written.
The students arrive in my web design course knowing that powerpoint slides
should use fonts at least 20pt (with exceptions of course). I try to
convince them that a professional looking site has levels of encapsulation
that reach depths of seven or more, or that blue on white signals that a
corporation takes itself seriously, and dark backgrounds are favored by
prurient adolescents and pastels appeal to aging baby boomers. This is a
far cry from knowing how to manipulate video into a simulatneous 3x3 array
to suggest a Hollywood Squares repetition of a corporate executive's
repeated avoidance of a question under oath, but it is a start. Tillers
and his people really have identified an area where the gap between
formalists and practitioners can and should be closed.