.‘__. fig. ‘\‘,‘.‘._.‘..1......7.,.-~..:. ‘ A

.-..«~o-—umfi-.\ . .-..

I

.. ..

. “can-..
'; ...'. {3&1
. ““9.

 

 

 

 

 

 

 

 

L.< ... HILT.
~ . ' \

 

 

. ‘.~ m-.. ,.

 

Willi/3W ‘3

I

1” ,/ //I///I

I;

P.

This is to certify that the

dissertation entitled

Argumentative Subsumption:
A Test of a COgnitive Schema For
Argumentative Discourse
presented by

Paul James Mineo

has been accepted towards fulfillment
of the requirements for

Ph.D. , Communication
degreem

 

 

 

Date 4%,":7' la [99/

 

MS U is an Affirmative Action/Equal Opportunity Institution 0- 12771

 

 

LIBRARY
Michigan State
' University

 

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before due due.

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

l

MSU In An Affinnetlve Action/Equal Opportunity Institution
cumulus-gt

 

 

 

 

 

 

 

 

ARGUMENTATIVE SUBSUMPTION:
A TEST OF A COGNITIVE SCHEMA FOR
ARGUMENTATIVE DISCOURSE

By

Paul James Mineo

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Communication

1991

ABSTRACT

ARGUMENTATIVE SUBSUMPTION:
A TEST OF A COGNITIVE SCHEMA FOR
ARGUMENTATIVE DISCOURSE

By

Paul James Mineo

A general methodology, described as functional and psycholinguistic,
is presented for theoretically formulating and empirically validating
cognitive schemata for use in explaining the production and
comprehension of argumentative discourse. The method is contrasted with
previous methods for describing the structure of arguments in ordinary
language. Two studies are reported in which the method is used in an
attempt to empirically validate a proposed schema referred to as
Argumentative Subsumption.

In the first study, the conditions were created by crossing three types
of argument conforming to the general schema (causal reasoning, sign
reasoning and the fallacy of accident), two types of complex argument
structure (coordination and subordination), and two levels of text cues
indicating argument structure (high cues and low cues). Subjects listened
to audio-taped passages of the stimulus arguments spoken in naturalistic
language, and produced reconstructions of the arguments immediately
after hearing them. Conformity to the proposed schema was measured as
proportion of match between proposed schema and subject protocols.

The second study repeated the conditions of the first but subjects were
simply asked to recall the passages and conformity to the proposed schema
was measured as proportion of schema-relevant vs schema-irrelevant

information retained.

The pattern of results were the same for both studies: (1) conformity to
the proposed schema was greatest for sign reasoning; (2) there was no
difference in conformity to schema due to text cues; (3) there was greater
conformity to schema for coordinate arguments than for subordinate

arguments, but only in the case of sign reasoning.

ACKNOWLEDGMENTS

I would like to thank Dr. Michael Burgoon for having the wisdom to
know that the best way to guide and motivate me through this project was to
leave me pretty much alone

——Dr. Judy Burgoon and Dr. Joseph Hanna for the discipline inspired
by the thought of their reading this dissertation

—G.R. Miller for reaffirming my belief that the strength of one’s
convictions gives one the security to be fair-minded with respect to the
convictions of others

—Kirsten Bech, a student volunteer, for her help with this project at
its inception much too long ago

—Sandra Starnaman, both for mundane assistance and
extraordinary inspiration

—my friends Mike Allen, Dave D'Alessio, Mark Hamilton and Brian
Patterson for believing in me, whether they understood me or not

—and especially Maureen, for believing in me despite understanding

me...and for weaving, unweaving and reweaving the tapestry.

iv

TABLE OF CONTENTS

List of Tables .............................................
List of Figures ............................................
Introduction: Problem and Background ........................
A Comprehensive Schema for A Natural Language Organon . . . .
More specific schemata: Argument types ................
Motivation for positing AS ............................

The Extent and Limitations of Current Research ..............

Research on inference processes in general and
psycholinguistic inferences in particular ................

Logic and attitude change research ....................
Argumentation research from a speech-act perspective
Theoretical Implications .................................
Importance for communication theory ..................
Importance for rhetorical theory .......................
Importance for theory of logic .........................
Methodology ...........................................

Independent Variables:
Controlling the Form of Stimulus Arguments .................

Propositional links ..................................
Argument types ....................................
Macrostructures ....................................

Apology ...........................................

00
H

8829.836

Measuring Match Between Subject Protocols and Schema .......
Criteria for match between protocols and schema .........

Coding of subject protocols ............................
Covariates .............................................
Design ................................................
Study I ............................................

Study II ...........................................
Hypotheses ............................................
Study I ............................................

Study II ...........................................
Method ..................................................
Subjects .................................................
Materials ................................................
Procedures ...............................................
Results ..................................................
Study I ...............................................
Study II ..............................................
Discussion ...............................................
Appendix ................................................

List of References ..........................................

vi

3288

SSSSSES‘wtfififiehtfi

838391

57
7O

LIST OF TABLES

Table Page
1. Analysis of Variance, Study I ........................ 55
2 Analysis of Variance, Study II .......................
3. Table of Means, Study I .............................
4 Table of Means, Study II ............................

688%

vii

LIST OF FIGURES

Figure Page

1. Macrostructure Examples .......................... 37

viii

Introduction: Problem and Background

The purpose of this project is to begin a systematic investigation of the
system of argumentation schemata or natural language organon employed
by ordinary language users in producing, comprehending and evaluating
argumentative discourse.

There is no assumption that this organon is the same for all language
users. Rather, it is assumed that the organon for each language user is
learned in the process of acquiring competence in argumentation, and that,
therefore, there would be individual differences in argumentation habits,
just as there are with respect to any acquired set of basic language skills.
Nevertheless, since argumentation is for the most part a cooperative
endeavor, it seems likely that speakers in a language community would
acquire and employ similar organa, that the simplest schemata would be
shared by all language users, with more elaborate schemata being
employed in more formal discourse or by better educated individuals. In
these respects, a natural language organon is parallel with the grammar
of a natural language. It is a set of conventions used by language users
with varying degrees of competence, the most basic conventions being
shared by the greatest number of speakers of the language. Indeed, the
natural language organon is the pragmatics of argumentative discourse.

Neither is the assumption made that the natural language organon is
logically correct, in the sense of a logical system grounded by reference to
philosophical criteria of validity and soundness. It seems likely that
schemata philosophers regard as fallacies would be learned in the same
way as those regarded as logically correct, so that fallacious reasoning

would often be indistinguishable from good reasoning for the untrained

1

2
arguer. The fact that good reasoning and fallacies are often
indistinguishable to the untrained is precisely why fallacies must be
identified in logic textbooks and classrooms. Cultures, subcultures,
ideological groups, cognitive styles, personality types, etc., may be marked
by preferences for types of argument that standard textbooks regard as
fallacious.

An investigation of this kind has never been carried out consistently
and intentionally. The reasons for this, elaborated below, may be stated
briefly as follows: researchers in the fields of communication,
argumentation and logic who are most familiar with the structure of
arguments and argumentative discourse seem not to be interested in
validating the models they propose by means of empirical studies aimed at
determining whether those models reflect cognitive schemata used in
processing arguments. On the other hand, researchers with a cognitive
orientation have typically borrowed the models they have investigated from
the field of logic, that is, models that have been developed to satisfy
prescriptive, epistemological criteria and which are elaborated in the form
of artificial languages (J ohnson-Laird, 1983, ch. 2). Given the aims of
formal logic, it is unreasonable to assume that one might find an easy
match between logical systems and the systems governing mundane
argumentation occurring in natural language.

Researchers using a cognitive approach to argumentation must start
somewhere, however, if not merely by adopting the models provided by
logicians. It should be profitable to begin by proposing a theory of the
functions of argumentative discourse and construct models of
argumentative logic that are capable of fulfilling these functions. In doing

this, one should keep in mind the contributions and, no less important, the

3

limitations of descriptive, naturalistic accounts of argumentation.

OttO‘I‘j ' c-_._fo 1"-1 21'4.-".,'-!0.!

There is an argument schema that appears fundamental to the
functions of argumentation and ubiquitous with respect to descriptions of
argumentative discourse, but which has not been empirically validated
using the methodology proposed here. For convenience this schema will be
referred to as Argumentative Subsumption (AS) and Toulmin’s
terminology (“warrant”, “data” and “claim”) is adopted to refer to the
constituent functional elements of AS.

The schema specifies a relationship that is analogous to (but not
necessarily identical with) that which is identified in Toulmin’s model of
argument by the relationship between warrant and data (“grounds”)
(Toulmin, Rieke & J anik, 1984, p. 45). The schema is also analogous to the
pattern in traditional syllogistic logic most clearly represented in the
Barbara syllogism between major premise and minor premises and in
propositional logic between the premises of modus ponens and in cognitive
psychology as the process in which a general rule is applied to a new
instance. Also worth noting here is the attempt of traditional logicians to
define the so-called “Rule of Syllogism.” H. W. B. Joseph (1916) gives it as
“Anything that satisfies the condition of a rule falls under the rule .”

The claim is not being made that the forms enumerated above are
identical, though it is assumed that they are functionally isomorphic in
important ways and so at least analogous. A different term is used to
distinguish AS from the purely formal relationship which characterizes
those deductively valid forms that are analogous to it as well as from the

pragmatic relationship described by Toulmin and his followers in

4
elaborations on the Toulmin model, because in both those fields the
constructs are deliberately isolated from the psychological context.

It might be necessary also to distinguish AS from a very general
cognitive schema that may be employed in inferring something about a
specific case from prior generalized knowledge. AS may be intimately
related to cognitive inference, but the details of that relationship need to be
worked out. The groundwork for systematically relating the cognitive
process of reasoning and the discourse process of argumentation was
presented in an earlier paper (Mineo, 1983).

However it- is determined that AS may relate to the analogous

relationships referred to, the following definition of the concept is

stipulated:
i 'n ' i hm in h r
H u- 5' I.'I0r’n’-!O‘ 1.-..1'-,=4 -n.-o_1 o ' i
io-'hn .-. ‘I’. or- 0.. .i'q-o hwn-.1. 'n
' o., 11' 0 01! ° ‘ -10 ° 0h: _.;mn- r ' of . mn h
g. .. or. Q‘ . o u_.1o .po . ,gq1V1q.1 . ‘ orrn.°- of . ‘
i ° o1. ' o o: -. a: _o_-. ‘ . _-.t 9- 0' 4911111”. h inivi a1
rrn

To say that AS is We entails that it is “psychologically
real”, i.e., that it underlies the production, comprehension and evaluation
of discourse of a certain type. To say that it is a schema is to indicate: (1)
that it is a hierarchically organized, cognitive framework against which
discourse is compared in order to render it coherent, whether in the
process of constructing discourse or attempting to comprehend it; (2) that
inferences are performed in the linear processes of encoding discourse into

strings of written or spoken text or decoding discourse while reading and

5
hearing, such inferences being employed to maintain coherence between
text elements; and that (3) in the process of encoding, inferences are made
to determine the text elements that need to be produced in order to fulfill the
requirements of the schema, while in the process of decoding, the
inferences made determine whether and how adjacent text elements fit the
schema.

The terms schema and m are used in the psycholinguistic and
cognitive literature as used in (1) and (2), above. Unfortunately, the two
terms are used by authors of conflicting theories concerning the way
discourse is processed. It has been suggested (Brown & Yule, 1983) that the
two approaches need to be combined in any adequate theory of discourse
processing. An attempt to outline how the two approaches might be
integrated (developed in Mineo, 1988) is summarized in (3), above. It has
not clear whether psycholinguistic researchers have followed the
suggestion of Brown and Yule. Lundquist (1986) has suggested a very
similar approach for the study of argumentation.

The term mfegenee may cause some confusion. As used in the
psycholinguistic literature, it refers to a process of linking text elements by
providing information not explicit in the text when attempting to construct
or comprehend discourse (Warren, Nicholas & Trabasso, 1979). When
arguing, for example, reconstructing an unexpressed warrant could
legitimately be referred to as inference — the warrant is “inferred” from the
claim and data. While the term inference may be used to refer to these two
processes, it is not assumed that they are identical. However, inferences
employing AS are assumed to be of the same type as the text-linking
inferences referred to in the psycholinguistic literature. For the purposes

of the present investigation, it is stipulated only that AS is employed in

6

processing argumentative discourse. There certainly are other types of
inferences employed in processing argumentative discourse and there
certainly are other types of psycholinguistic inferences that are not typically
employed in processing argumentative discourse.

WW Arguments may be said to
derive their force from the connection expressed in the warrant (Toulmin et
al., 1984, p. 45). There are apparently many types of connections
represented in the warrants of naturally occurring arguments, i.e.,
whenever arguments are expressed, they will always take a specific form
that is determined by the type of connection represented in the warrant.
These forms are usually discussed when scholars attempt to distinguish
types at reesening, e.g., causal reasoning, sign reasoning, reasoning from
analogy, reasoning from principle, reasoning from example, etc. Given the
presuppositions discussed above, those argument types traditionally
identified as fallacious would have to be included in the list. Each of these
reasoning types has a different type of warrant associated with it and for
each type, different types of data are required to satisfy the condition stated
in the warrant. No argument can be written or spoken which is a pure
argument in the sense that there is no specific type of connection specified
in the warrant.
In view of this, one of the things that needed to be determined for the
purposes of this investigation was the range of argument types that should
be considered for inclusion. The literature on types of reasoning is in no
way consistent, but it seems clear that the study should include more than
one argument type, otherwise it may be argued that the results of the
investigation are valid due to the specific type of reasoning included. The

problem then is to anticipate in advance which of the many types are likely

7
to be most common, easiest to learn, simplest to understand and learned
first in the acquisition of argumentative competence. These are the types
that would most likely be found in an organon for natural languages.

It is also likely that the the types of reasoning are related in the sense
that we must understand how to use some of them before we can
understand others. It can be argued, for example that reasoning from
analogy is enthymematic in that it can be analyzed as a combination of
simpler forms, or that reasoning from representative sample (upon which
the whole elaborate structure of statistical inference rests) is a technical
elaboration of reasoning from example. Again, those that must be
understood first in the order of development of argumentative skills would
be those most likely to be found widely distributed in the population of
arguers of average competence.

These considerations indicated a direction for the preliminary
investigations for this study: a small number of the simplest argument
types should be selected for inclusion in the study. These types should be
simple in the sense of being developmentally more fundamental but diverse
enough so that they serve the function, at least, of enabling the average
language user to construct and comprehend a wide range of arguments
typically found in mundane argumentation.

Some simple examples may clarify the process here proposed as
operating when a bearer/reader attempts to comprehend discourse. If a
text begins with the phrase “Once upon a time...” the reader infers that this
is a setting and so attempts to apply a schema for stories for interpreting
the rest of the text. On the other hand, if the text begins with the words
“Because there is less inflation this election year...” the heater/reader may

infer that the statement is a reason or W and attempts to apply a

8
schema for arguments for interpreting the rest of the text. So far,
processing is guided only by the most general level of the schema (AS). If
the hearer/reader subsequently encounters the phrase “...leads inevitably
to...” the schema is further specified and the rest of the text interpreted as
an instantiation of one of the distinct reasoning types, in this example,
causal reasoning (cf. Lundquist, 1986).

Conceived in this way, the schema AS and its specifications of
different reasoning types provides a language user with a series of
strategies for interpreting argumentative discourse. As such, it is similar,
both in the way it functions and the level at which it functions, to the Given-
New strategy (Clark & Haviland, 1974) or the Order of Events strategy
(Clark & Clark, 1977) proposed by psycholinguistic researchers.
Presumably, AS and its variations, as well as other schemata within the
repertoire of the average language user, would be more general and would
operate at a more fundamental level than schemata used for processing
more elaborate forms of discourse, e.g., the schema hypothesized by
researchers for processing stories (Freedle & Fine, 1982), since arguments
can occur within stories, jokes, etc., as well as within standard
argumentative contexts.

W. In the literature reviewed concerning the
relationship between logic and the psychology of reasoning, researchers
have focussed on the issue of whether humans typically reason logically.
The consensus seems to be that they don’t (Hample, 1979). Some
researchers maintain that reasoning is basically logical but that there are
psychological processes that interfere with the tendency to be logical, while
others have just accepted the idea that humans do not reason logically

(Reynolds & Burgoon, 1983).

9

Both these responses seem unsatisfactory. The willingness to accept
that we reason logically but that there are processes that interfere with
logic seems to be guided by the assumption that the logic people employ, is
“correct” in some sense or at some level. If we do not reason logically all the
time, they seem to assume, we must reason logically sometimes, otherwise
our reasonings will not even be capable of correction. These researchers
have continued to investigate reasoning in order to determine the extent to
which reasoning is logical and to identify the interfering influences. On
the other hand, the willingness to accept that humans do not reason
logically seems to be guided by the assumption that logic plays a very minor
role in human behavior, including argumentative interaction (Jackson,
1985), and researchers adopting this response have been content to ignore
the issue.

Both of these tacks seem to be based on the researchers’ acceptance of
“textbook logic” — logic as it has been defined by philosophers attempting to
explicate what validity is. This is the “logic” referred to in the researchers’
empirical investigations or theoretical critiques. It is easy to see why those
who believe that we must reason logically sometimes would be unwilling to
give up that assumption: they reason that if the philosophers are correct,
then we must at least sometimes reason in such a way that our
conclusions correspond to the world in which we try to survive, or that they
are coherent within the system of beliefs of those with whom we argue.
Likewise, we can understand how others can regard logic as irrelevant to
human behavior: They reason either that the philosophers are wrong about
the structures they have identified, or that true conclusions are irrelevant
to survival or the coherence of belief systems.

In the research concerning the relationship between logic and

10
argumentation, researchers have focussed on the question of whether logic
is necessary in order to explain or understand argumentation practice.
The consensus seems to be that it is not. Some maintain that
argumentation practice is largely illogical and that much of what guides
argumentation is influenced by rules governing argumentation as
interaction (hence extra-logical), but that it would be a good thing if it could
be made logical (van Eemeren, Grootendorst, & Kruiger, 1984). Others
maintain that it is largely irrelevant whether argumentation practice is
logical or not, that the interaction rules are all important (Jackson, 1985)

Again, both these responses seem unsatisfactory. Those who
maintain that argumentation can and should be made logical seem to be
guided by the assumption that logic can best serve the needs of arguers, and
that even if we do not argue logically all the time, we can and should argue
logically more often, if we are to achieve the ends of argumentation. Those
who maintain that logic is irrelevant to argumentation practice seem to be
guided by the assumption that logic cannot help us achieve the ends of
argumentation practice.

Both these tacks seem also to be based on a common assumption, viz.,
that the function of argumentation practice is the resolution of disputes or
disagreements. Those who accept the idea that logic is relevant to
argumentation assume that logic can help lead us to the resolution of
disputes, whereas those who maintain that logic is irrelevant to
argumentation maintain that logic can lead us to the resolution of disputes
only in rare cases.

The alternatives to the above assumptions which motivate the present
project are the following:

There must be a natural language organon that operates in such a

11
way that it leads us, more often than not, to true conclusions about the
world in which we are trying to survive. Whether these rules are
isomorphic with the structures of philosophical logic as presently
formulated is an empirical question. Further, it is assumed that the aim of
argumentation is, at least sometimes, the desire to uncover truths
necessary for survival — what has been called the epistemic function of
argument. Finally it is assumed that even when argumentation is used to
serve other functions, it is largely guided by the system of rules governing
the epistemic function, that is, the epistemic function is regarded as
privileged, though not necessarily favored all the time. If the system of
rules we use to uncover truths necessary for survival is called “logic,” then,
given these assumptions, the claim can be made that argumentation
practice is, for the most part, logical. Whether this claim is true, however,
is also an empirical question insofar as the logic proposed as capable of
fulfilling the stipulated function should be demonstrated as capable of
explaining observed argumentative behavior.

The choice of AS as a logical schema of the sort defined is guided by
the belief that the ability to subsume a new case under a general rule is one
that is useful and fundamental for survival. Aside from these
considerations of human nature and the requirements for human survival,
this choice is guided by the the analogies referred to above. If this is
correct, it is not surprising that logicians have repeatedly identified
structures that are at least analogous to AS.

Furthermore, positing AS seems consistent with observed
argumentation practice. The existence of AS may explain why the Toulmin
model schema of warrant, data and claim appears to be universal in

everyday argumentation, at least in terms of our ability to analyze

12

arguments in terms of the model. There doesn’t seem to be any naturally
occurring argument for which it is impossible to specify “warrant” and
“data.” There are obviously cases in which one of these elements is claimed
to be “implicit” or “unexpressed.” In those cases, when novices with
minimum training are asked to supply the missing elements, they can
usually do so fairly easily in a manner consistent with the model, and
agreement is often high. There are certainly cases in which novices and
even experts disagree on the assignment of the labels. In the case of
novices, one can usually trace the disagreement to ambiguities in the text
or differences in interpretation, and typically, when the ambiguities or
divergent interpretations are pointed out, the disagreement can be resolved.
In the case of experts, the disagreements might all be attributable to
differences in the way “warrant” and “data” are defined.

It is also interesting to note that even with highly artificial
arguments, as long as they are expressed in natural language, or in any
symbol system translatable into a natural language, the warrant/data
analysis works. Admittedly, the translation often results in grammatically
tortuous results (picture recasting every line in a thirty line proof in
symbolic logic in terms of the Toulmin model!) , but the same is true when
one tries to formalize ordinary language arguments. In the first case, one
murders artificial language in order to translate it into ordinary language;
in the second case, one murders a naturally occurring argument in order
to translate it into standardized logical form.

It may be argued that the distinction between data and warrant can be
made but that it is “so abstract as to be meaningless.” Aside from the fact
that people often make the same claim about the basic structures of formal

logic, there are a number of practical advantages to making the distinction

13

for argumentation training, and it seems that these advantages are
exploited even by arguers with no formal training. For example, warrants
and data are typically evaluated using different criteria —— criteria
appropriate to the meaning and function of each. Thus in trained practice,
rendering implicit elements explicit is taught as a skill useful for
exhibiting flaws in an argument. In untrained practice, we often hear
criticisms that can be explained by positing this ability to make
unexpressed elements explicit prior to evaluation, for example in the

following exchange between a five year old (B) and a six year old (A):

A: You can’t go down there!
B: Why? Mommy didn’t say so!
A: So? You’ll still get hurt!

B has interpreted A’s claim as depending on the warrant “I can do
anything unless Mommy said I can’t do it.” A replies, attacking B by
pointing out that the data is only relevant relative to the warrant stated,
which A regards as false, since there are other reasons for not doing
things, like the possibility of injury. (The explanation may be more subtle if
we assume that B intentionally “misinterprets” A’s warrant at first in
order to evade the issue that descending into the cave is dangerous. In
either case, ability to reconstruct unexpressed premises prior to evaluation
is presupposed.)

The project here described was an attempt to demonstrate empirically

that arguers use AS in their processing of argumentative discourse.

14
I] Ell 11... EC lB l

i‘u ,, u infurn ' .oro - '9 “3' ., -.o .0. h0°lin '
W. Since this investigation deals with cognitive

elements that seem at least analogous to elements involved in simple
inference, some of the research into the nature and development of
inference ability may be relevant. The problem is that much of this
research is limited in that it is not clearly linguistic in orientation, e.g.,
problem-solving research (Rice & Kemper, 1984), research using Venn
diagrams (Erikson, 1978), etc.

There has been a significant amount of psycholinguistic and cognitive
research that has concerned itself with the processing of words or
sentences of different logical types, e.g., investigation into childrens’ ability
to use quantifiers, or transitive relations, investigation of certain logical
fallacies like the “conjunction fallacy”, etc. (Galotti, 1989) Some of this
research has specifically involved subjects’ processing of simple arguments
(Hawkins, Pea, Glick & Scribner, 1984; J ohnson-Laird, Oakhill & Bull,
1986; Markovits, 1985). In the research reviewed, however, the logic
employed has been philosophical logic. In order to determine the extent to
which this research is relevant, it would be necessary to isolate those
studies in which the relationships investigated are isomorphic with
natural language logic. Until this task is accomplished, the overall
implications of previous psycholinguistic research remains unclear.

WWW. Some relevant research has been
done by communication researchers interested in whether logic has any
bearing on attitude change (Hample, 1978, 1979a; Miller, 1969; Petty &
Cacioppo, 1981; Steinfatt, Miller & Bettinghaus, 1974).

Here again, though, logic has been operationalized as “textbook logic”

15
or, even worse, it has not been operationalized at all, as in the research
motivated by the elaboration likelihood model of persuasion (Petty &
Cacioppo, 1981), in which “good vs. bad arguments” is used as an
independent variable with only very vaguely defined criteria of “good” and
“bad” having been used to construct the argumentative messages used as
treatments.

Another problem with some persuasion research, noted by Reynolds
and Burgoon (1983), is that researchers have paid insufficient attention to
the question of whether and how the acceptance of premisses leads to the
acceptance of conclusions. Instead, they have focused on the acceptance of
conclusions and merely assumed that this was a result of the acceptance of
the reasons offered.

Until those studies can be isolated in which treatments can be
identified as isomorphic with AS, the implications of this body of research
remain unclear as well.

WWW. There is a
body of research by speech-act theorists that may also be relevant.
Pragmatic models of various aspects of argument have been proposed and
criticized, including, the speech-act model of van Eemeren and
Grootendorst (1984), and the interaction model of Jackson and Jacobs (1982).
The work done on enthymemes would be especially relevant (Jackson &
Jacobs, 1980; van Eemeren & Grootendorst, 1982).

Research along these lines is useful as an important preliminary to
the investigation of the psycholinguistic aspects of argumentation. The
method used by these researchers, however, relies primarily on the use of
the researchers’ own intuitions of the legitimacy of the examples and

counterexamples constructed or selected as evidence. This method is

16
referred to as “analytic induction” (Jackson & Jacobs, 1982).

It was noted above that the Toulmin model as herein modified can be
used to analyze any naturally occurring argument. Anyone who
understands the model can apply it to any unit of discourse that is regarded
as an argument, and the model cannot be used to generate examples that
would not be regarded as arguments. No disconfirming instances can be
found. This is true as far as the author's intuitions go, and it is as far as
one can go with the method of analytic induction.

The fact that AS, or any model of argument can be used by
researchers to analyze any example of naturally occuring discourse,
however, does not demonstrate that the proposed model is used by ordinary
language users to process discourse. Even if it were true that every
naturally occurring argument can be reconstructed in terms of the model,
that doesn't tell us whether the model models something “psychologically
real”—whether it represents a cognitive schema that is habitually
employed by language users in constructing, comprehending and
evaluating arguments. If we do not demonstrate that the schema the
model represents is psychologically real, then the model is of no help in
helping us explain communication processes involving argumentation,
unless all the subjects are taught to use the model and practice using it for
years until it becomes “second nature” to them, that is, until it becomes part
of the habitual processes determining their production of and responses to
argumentative discourse. If, for example, we could teach everyone modern
textbook logic and force them practice it until it becomes habitual, then we
could determine how those habits are involved in their communication
processing. Until that Brave New World arrives, however, a reasonable

alternative is to discover the habits of natural language logic the average

17
person already has, and work with those, both as a pedagogical base and as
a construct to inform communication research.

Moreover, if researchers' ability to apply the model and their inability
to invent or discover counterexamples are the only criteria, as is the case
with analytic induction, it can happen that competing models are equally
supported. This is, in fact, what occurred in the study of grammar where
traditional researchers used a method similar to analytic induction in
which evidence consisted of sentences discovered or invented by
researchers which conformed to the researchers' hypothesized rules of
grammar. Whether the sample sentences counted as confirming instances
of the rule was determined by the researchers' “intuitions of gramaticality”
of the sample sentences, that is, if the sample sentence conformed to the
hypothesized rule and was judged grammatical by the researcher, that was
counted as a confirming instance (Fodor, 1981). It was discovered,
however, that different hypothesized systems of grammatical rules—
competing models—will often work to explain the same data. If one
assumes that the competing models represent psychologically real
cognitive structures, then differences in observable consequences of
positing the competing models can be used to decide between them. This is
precisely the approach taken by psycholinguistic researchers whose work
was able to break the log-jam created by the inability of traditional intuitive
research methods to decide between competing, intuitively based
grammars (Halle, Bresnan & Miller, 1981; Palermo, 1978; Slobin, 1980).

What the assumption of psychological reality provides, then, is a
method of testing hypothesized rules or models that is not limited to
researchers' intuitions. Rather than leave it to researchers to judge

whether a sample unit of discourse is a confirming or disconfirming

18
instance of a rule or model, observational consequences are derived from
the assumption that the rule or model is a psychologically real part of the
cognitive apparatus by which discourse is processed. If competing models
are suggested by different theories, selection of the better model can be
accomplished by precise specification of differences between observational
consequences predicted by the competing models. For the studies reported
in this paper, AS is stipulated as a model reflecting a psychologically real
schema for processing argumentative discourse. The projected empirical
consequences of that assumption are specified in the form of linguistic
evidence extracted from subjects' protocols produced in response to the
stimulus messages.

Aside from these methodological limitations, researchers in the
speech-act tradition have not paid sufficient attention to the logic of
argumentative discourse. Jackson and Jacobs accept the assumption that
logic is irrelevant to the aims of argumentative discourse (Jackson, 1985),
which they define as the resolution of disagreement (Jackson & Jacobs
1982). It may be that little of what they have done is relevant to this project.
They seem, however, to make implicit and extensive use of logical criteria
in their identification of arguments. More specifically, given their
definition of argument as a “disagreement relevant expansion” (Jackson &
Jacobs 1982), it is easy to construct examples in which the only way of
determining whether one utterance is a relevant expansion of another is by
means of the logical relationship between the two utterances. For example,
while rules governing interaction may provide a partial explanation of the
interchange between A and B given above, it was shown that the utterances
chosen by the children bear a logical relationship to one another as well.

Speech act theorists have criticized the Toulmin model for various

19
reasons, and these criticisms may be relevant depending on the extent of
the analogy between AS and Toulmin’s functional model. Important for
my considerations are the criticisms noted by van Eemeren, Grootendorst
and Kruiger (1984). A pair of related criticisms, drawn from early
philosophical criticisms of Toulmin, is that backings can be applied to data
and claims as easily as to warrants, and likewise, that rebuttals and
qualifiers can be applied to warrants and data as easily as to claims.
Backings are defined as “generalizations making explicit the body of
experience relied on to establish the trustworthiness of the ways of arguing
applied in any particular case” (Toulmin, Rieke & J anik, 1984, p. 61);
rebuttals are defined as “extraordinary circumstances that might
undermine the force of the supporting arguments” (Toulmin et al. 1984, p.
95); and qualifiers are defined as phrases that “show what kind or degree of
reliance is to be placed on the conclusions” (Toulmin et al. 1984, p. 85).
Neither one of this pair of criticisms is relevant to the specification of the
functional relationship between data and warrant, however, which is
incorporated in the definition of AS. And in fact, the criticism that
backings are not unique to warrants depends on the observation that the
warrant and data stand in the same functional relation to the claim as
backing to warrant. If this is true, AS could explain the production and
comprehension of those utterances typically identified as backings, though
it would not justify giving the backing a different label.

(It does not seem, however, that backings are merely further
arguments in support of the warrant of an argument as this criticism
suggests. Backings may be characterized as linking the argument to the
appropriate field, i.e., of attempting to make clear the context in which the

argument is to be assessed. This seems a legitimate function given

Z)
Toulmin’s conception of fields and in practice one can find innumerable
examples of expressions that seem to function in this way, e.g., “According
to speech act theory...”, “In the light of the evidence presented...” in
elaborated discourse and more mundane examples like “I seem to
remember from biology class that...” or “In my experience...”. In teaching
argumentation, defining backings in this way makes it easier for students
to pick them out and the practice of attempting to identify fields by looking
for backings is a useful preliminary to argument assessment. Also, in
exploratory research conducted by the author, defining backings in this
way enabled coders analyzing transcripts of naturally occurring
arguments to locate backings more reliably. In any case, whether the
backing-warrant relationship is distinct from or identical to the warrant-
claim relationship is a separate issue from whether the warrant-data-
claim model is analogous to the AS model.)

A much more important criticism, or set of criticisms, depends on the
definitions of “data” and “warrant” and the charge that these are not
adequate to enable us to determine, in many instances, which is the data
and which is the warrant, whereas it is here proposed that AS does
function in precisely this manner. These criticisms assume that in the
Toulmin model (1) the data is always explicit, while the warrant may
remain implicit, (2) the warrant has a general, rule-like, justifying
character and (3) the warrant fulfills a bridging function between data and
claim. Using these criteria, a counterexample is given by van Eemeren et
al. as follows:

Claim: (a) Harry is a British Subject
Datum: (b) A man born in Bermuda is a British Subject

Warrant: (c) Harry was born in Bermuda

21
In Toulmin’s original example, (c) was given as the datum and (b) as

the warrant. van Eemeren et al. claim that their analysis, given above, is
equally legitimate: Assuming that (b) is unknown to the person to whom
the argument addressed, that (c) is known and that the arguer has stated
only (a) and (b), (c) may remain implicit and (b) is a general, rule-like
statement that performs the bridging function between (a) & (c). There is
no reason to assume that data must be explicit and warrants implicit,
however. Toulmin originally claimed that this was only one of the
distinctions that may be drawn between data and warrants, but this is not a
difference in logical function, which was the core of that discussion
(Toulmin, 1958, pp. 99-100), and in later discussions, there is no insistence
implicitness as a defining characteristic of warrants (Toulmin et al. 1984).
The essential difference between data and warrants, then, seems that
warrants have “general rule-like quality” and that data are facts pertaining
to the specific case or cases about which something is claimed, which
differences are stipulated in the definition of AS. Thus, in the above
example (b) is the warrant because of its general rule-like function,
regardless of the fact that it is explicit, and (c) is the data, because it
provides imformation about “Harry,” the subject about whom a claim is
being made.

van Eemeren and Grootendorst have carried the project of developing
a speech—act theory of argumentation further than any other researchers in
the field (van Eemeren & Grootendorst, 1984; van Eemeren, Grootendorst &
Kruiger, 1987). They have argued that this project would require the
specification of argumentation schemata used to identify and evaluate the
logic of arguments. This part of the project has not been carried out. They

have specified and elaborated rules for interaction between arguers, but

22
have acknowledged that the rules of logic must be included in a complete
theory. In their work on enthymemes and fallacies, they have maintained
that some aspects of these phenomena can be understood in terms of
arguers exploiting interaction rules but that others require the rules of
logic. The “logic” that they have consciously (albeit provisionally) adopted
for this work, it seems, is textbook logic. They appeal to Grice’s idea of
conversational implicature, but, nevertheless, in the model they propose by
which speakers “infer” unexpressed premises (and this is, as yet, a totally
unsupported empirical claim), a crucial step in the process is the
assumption that the speaker “intends a valid argument” (van Eemeren &
Grootendorst, 1984, p. 134).

In the model proposed by van Eemeren et al., they define the
relationship between premises (which they call “arguments”) and the
claim, or opinion, that is justified in a simple argument. They also identify
the relationship of mendingtign, in which two or more independent simple
arguments are used to support an opinion, and the relationship of
MM. in which two or more simple arguments are chained in a
series to support an opinion. They do not, however, specify any of the
internal structure for simple arguments. This leaves us with a serious
problem if we are going to regard their model as one that applies to
ordinary language argumentation. The statements comprising arguments
in ordinary language are not always grouped together or marked
linguistically in such a way as to make their relationships transparent. In
order to determine which statements belong to which simple argument and
in order to determine whether simple arguments are related to one another
by coordination or subordination, the internal relationships that link

statements together in a single argument must be specified.

23

Consider, for example, a text consisting of five statements in the
order: A—B-C-D-E. Suppose a bearer determines that A+C forms a simple
argument supporting E and that B+D forms another simple argument
supporting E, and further, that B+D supports neither A or C, and A+C
supports neither B or D. This would be a case of two simple arguments
supporting E coordinately. In the process of interpreting the text, the
hearer would have to construct these groupings, since the groupings are
not given in the surface-structure ordering. The only way these groupings
can be formed is if the hearer looks for relationships that link non-adjacent
statements together into coherent, simple arguments. For example, the
hearer interprets A and C as premises supporting E, despite the fact that B
and D separate them, because the bearer notices that A and C bear a
certain relationship to one another that enables, or legitimizes that
interpretation. The relationship between warrant and data specified by AS
would enable such groupings. No such relationship is specified by the
model proposed by van Eemeren, et al.

van Eemeren and Grootendorst assume that the function of
argumentation is the resolution of disputes and they are content to assume,
for the time being, that textbook logic is very useful for resolving disputes.
This seems to be why they have spent so much of their efforts working on
the interactional aspects of argumentation. Starting from the alternative
assumption that the privileged function of argumentation is uncovering
truths necessary for survival and recognizing that whether philosophical
logic is used to serve that function is an empirical question, the focus of the
project here proposed is on defining and validating the structure of
argumentative logic. The empirical validation of AS was chosen as a first

step in that project for reasons already discussed.

WWW. Communication researchers
interested in message features relevant to persuasion have, for the most
part abandoned the idea that logic is one of these features. The advantage of
assuming that receivers use textbook logic in comprehending and
evaluating persuasive messages was that it provided researchers with
something well defined conceptually and easy to operationalize for use in
research. One of the problems with the assumption that there is a “psycho-
logic” that is relevant to persuasion is that it has provided no systematic
way of identifying that psycho-logic, hence, there was nothing definite for
use in research. What is here proposed is a systematic way of specifying
the nature and limits of a natural language logic.

There are numerous areas of communication research other than
persuasion that might be aided by understanding natural language logic.
It seems likely that the one's definition of communication competence,
would be influenced greatly by understanding norms for argumentative
competence. Differences in argumentative competence may be an
important genetic or limiting factor in conflict management, and may be
related to argumentativeness (Infante, 1987) and thus, to the effects of
argumentativeness (Infante, 1988, pp. 9-10). Furthermore, understanding
natural language logic would certainly aid in our understanding of small
group decision making.

In general, whenever there is reason to believe that the inferences
drawn from a message may mediate the effect of that message as in
attribution processes, person perception, counterarguing, etc., there we

may find the concept of a natural language organon to be a valuable tool for

theory elaboration or integration.

WWW. Validation of a natural language
organon would provide support for rhetorical prescriptions concerning
appropriateness of the level of logical complexity of persuasive messages.
Messages that must be processed during the rapid flow of speech in the
broadcast media, public speaking contexts, small group discussion and
dyadic conversation would presumably demand that receivers employ a
natural language organon, during initial processing at the very least, due
to cognitive limitations on the speed of comprehension in such contexts.
Also, by moving toward fulfillment of Toulmin’s recommendation for
empirical studies in the uses of argument, it would provide a tool for a more
naturalistic, less formalistic basis for analyzing the logic of rhetorical
discourse, precisely the motivation that drew rhetoricians toward the
Toulmin model in the first place.

The approach adopted here assumes that the epistemic function of
argument is privileged due to survival demands. In the practice of
everyday life a child will continually be placed in situations in which
principles previously learned must be applied to new situations. The child
will often discover that such principles are false or incorrectly applied. If
no rule or principle is available to guide action, or if the reason for the
failure of a rule or principle is not understood, the child will turn to others
for help. If a child is asked for justification of behavior, the child learns
that the answer must be in the form of a rule or principle under which the
behavior in question can be subsumed. And when children are asked to
behave in a certain way, or to refrain from so behaving, or when they are
attempting to understand the behavior of another, they soon learn to ask

Why?-questions. In such situations, parents attempt, with varying degrees

%
of success, and varying degrees of concern for the well-being of the child, to
provide a structure by which the child can determine what counts as a
satisfactory answer. It is assumed that in the private and social uses of
language in the above contexts, the child will learn a repertoire of
argument structures and that a stable core of these will be such that they
will lead the child to true conclusions about the world in which the child is
trying to survive. Note, however, that there is considerable opportunity for
the child to learn structures that have been traditionally regarded as
fallacious, also. The privileged status assumed here for the epistemic
function of argument stems from the acceptance of the often quoted notion
of Aristotle’s that “things that are true...are, by their nature, practically
always easier to prove and easier to believe in” (1952, p. 594). This is usually
misinterpreted in modern rhetorical scholarship as a kind of naive,
mystical faith in the Power of Truth (Rieke & Sillars, 1984, pp. 14-17). The
Philosopher, however, states in a previous passage that “the underlying
facts do not lend themselves equally well to the contrary views” (p. 594),
which may be taken to mean that false beliefs are more likely to be falsified
in practice, i.e., in the “experimentation” of everyday living. Thus, while it
is granted that the social function of argumentation is to gain audience
adherence to claims, it is nevertheless assumed that the typical audience
member will be sensitive to argument forms the core of which, at least,
have been selected by the epistemic function.

Immtangefgmhegmflegie. If it turns out that argumentative
logic is similar in many ways to logic as defined by philosophers, they can
certainly feel good about themselves. Beyond this, understanding natural
language logic and how it develops would be a boon to logical pedagogy. For
over a century, logicians have sharply distinguished logic from

27
psychological processes. The result has been the elaboration of logical
systems that people can’t readily use, taught only to a minute fraction of the
population years after the developmental period in which a person most
likely acquires natural logic. There is reason to believe that Piaget’s
investigation into the acquisition of reasoning skills estimated the logical
formative years as being much later than they may in fact be because his
research concentrated on problem-solving tasks and neglected to consider
the high level of logical skills required for language production and
comprehension (Rice & Kemper, 1984). There is no reason that six-year-
olds cannot be taught logic. In their desire to isolate logic from psychology,
it may be that philosophers have neglected the opportunity to raise the
quality of everyday thinking and arguing.

Note, however, that while we often think in terms of logic informing
everyday practice, philosophers are not unaware of the possibility of
everyday practice informing logic. Goodman (1983), for example, while
recognizing that we justify particular inferences by their conformity to valid
deductive rules, also claims that valid deductive rules are justified by their
conformity to particular inferences. He argues that the circularity of this
mutual justification “is a virtuous one....The process of justification is the
delicate one of making mutual adjustments between [valid deductive] rules
and accepted inferences; and in the agreement achieved lies the only

justification needed for either” (p. 64).

Methgdclm
The approach which was followed in specifying the theory for this
project may be called functional-psycholinguistic. This is the approach

used in a previous paper by the author on story coherence (Mineo, 1988).

23
The steps in specifying the theory are the following:

1. _ we; I‘ .0 0.34.01. -1 u o a: u: o 0. 0- _‘ o '1 ‘r‘ in
W. The pragmatic function of AS, as given in the
definition provided above (p. 4), is to apply a general rule or principle to a
new specific case.

2. U: '00- 09-“. 109'... 1% -m.. o - 0.1- or: u'.
The top-down model, at the highest level of generality, is very simple. It
specifies that AS is a schema relating statements to each other as warrant,
data and claim. At the next level down from the top of the hierarchy, it is
proposed that there are a small number of distinct reasoning types in
which the nature of warrant and data (as well as other features, such as
appropriate qualifiers) are further specified.

3. " ' H 1.00 0H'.' "00- o 1' L‘ ' ' _1-. o. .0: .99 ‘0
.0 h ‘ -'o.rin19-. - o '_ 1° of o: x 0. ' o Q'mn.o_ of h
schema. As the bearer/reader processes the text, surface cues trigger
inferences that suggest what schemata might be used in order to interpret
the text as a coherent whole.

An investigation of naturally-occurring argumentative discourse
provided a sample of the linguistic cues employed by arguers in expressing
the types of argument included in the studies reported below. These
linguistic cues were used in the construction of extended examples of
argumentative discourse containing more than one argument. Subjects
were assigned two different tasks in which they listened to the examples
and produced written protocols afterward: in Study I, subjects were
instructed to reconstruct the arguments; in Study II, subjects were simply
asked to recall the messages. It was reasoned that if AS is operating, it

would predict that subjects reconstruct or recall arguments by linking

29
elements of the simple arguments together in a manner consistent with the
AS schema.

It was argued above that speech act theorists—in particular, van
Eemeren, Grootendorst and their colleagues—have failed to identify the
logical aspects of argumentative discourse, though they maintain that this
needs to be done (van Eemeren, et al., 1987 ). It was also argued that the
model they have proposed seems to require the ability to recognize the
internal structure of simple arguments in order to comprehend complex
arguments. In view of these considerations, a crucial empirical question is
whether AS, as a model of the internal structure of arguments, can help
explain how people group statements together in the process of attempting
to interpret argumentative discourse.

The purpose of the studies reported here, then, was to demonstrate
that subjects attempting to comprehend arguments expressed in natural
language will reconstruct the arguments for later recall in accordance
with the AS schema. Arguments thus reconstructed should provide
evidence that subjects related text elements together in a manner that
reflects the underlying argument structure predicted by the schema. More
specifically, these relationships should reflect the structural links predicted
by the schema between functional elements within simple arguments
(warrant-claim; data-claim). Only if subjects reconstruct these simple
links correctly will they be able to recognize the structure (coordination or
subordination) of complex arguments. If the model of processing proposed
above is accurate, the interpreter's ability to process argumentative
discourse accurately depends on cues in the text triggering the AS schema
that is then brought to bear in an attempt to comprehend the text. If

surface cues are minimized (for example, by the elimination of standard

30
argument markers—words like ‘so’ and ‘because’) accuracy of
interpretation should be reduced.

In order to test these assumptions, it was necessary to control the
construction of the stimulus messages included in the studies in terms of
the propositional links within simple arguments, the specific argument
types considered as variations of AS, and the structure of the complex
arguments (referred to below as “macrostructures”). The methods used to
achieve these controls is described in detail below.

It was also necessary to decide on the method of coding subjects'
protocols so that the extent of agreement between their reconstructed
messages and the AS schema could be determined. The method for
achieving this is also described in detail below. Briefly: coders were
provided with definitions of the argument types along with a list of
examples of markers for warrant, data, and claim for coding subject
protocols. Proportion of the match between subject protocols and the
argument structure predicted by the schema was determined by the
number of individual matching links. Links counted as a match whenever
a pair of propositions recalled by the subject was coded as relating the
propositions in the manner predicted by the schema.

It may be argued that this approach does not clearly demonstrate that
interpretation depends on a cognitive schema, rather than on the surface
cues alone. However, to say merely that “surface cues alone” produce the
observed effects provides no explanation at all as to how this occurs. The
fact that different surface cues result in the same perceived structure
means that the cues have been generalized in some way, certain classes of
cues interpreted as indicating the same relationships between text

elements, and such a generalization is what a schema is supposed to be.

31
Furthermore, it was reasoned that if differences in schema produce a
difference in processing beyond that accounted for by cues alone, that
difference in processing could be attributed to differences in schema
applied, rather than differences in surface cues.

Additional support was provided to demonstrate that a schema is
operating by determining whether a second group of subjects could provide
evidence of distinguishing schema-relevant and schema-irrelevant
information. Since the subjects in Study I were to reconstruct the
arguments, they were instructed to reproduce the gist of the stimulus
arguments, whereas the subjects in Study II were assigned the task of
simply recalling the texts. It was assumed that overall recall of schema-
relevant information would be greater than information that is not relevant
to the schema. If argument-irrelevant information was totally irrelevant to
the text, however, it would be regarded as nonsensical, incoherent. For the
arguments constructed for this study, the information included was
relevant to the text but not argument-relevant, and it was assumed that the
latter would be better recalled than the former when subjects were given the
task to recall the text and that this relationship would hold true

independent of the specific argument type and macrostructure.

_u'o'io:o {-1.0 ‘_ H 0 0° 1' 0!!! 0 - 01-- ; am-t ‘
That arguments occurring in natural language exhibit a wide range
of differences is a truism. For the purpose of this study, and any study
which purports to uncover something meaningful about the structure of
arguments occurring in natural language, a limited range of forms must
be selected, so that the results will be clear. This introduces some control

over what is traditionally referred to as message content , considered as a

32
global variable. Though what is being controlled has been referred to
throughout the discussion above as argument We, it is clearly part of
gentent of the message considered globally. Exactly what part of a message
is logical structure, and what part is topic or substantive content is not
clear. This is further complicated by lack of agreement or clarity
concerning what parts of argumentative discourse are motivated by the
relationship between participants, what elements are purely
conversational, what can be classified as ethes and names, etc.
Furthermore, even if one could isolate those elements that are logical, it is a
matter of controversy as to whether there is anything in natural language
that is purely formal and non-substantive in the logical sense. One of the
main motivations of this study is precisely to attempt to clarify the scope of
Legne for naturally occurring argumentative discourse.

It is assumed that dozens of studies will have to be conducted using
this research paradigm, systematically investigating a variety of isolated
forms, before a clear picture of a natural language organon emerges,
including a knowledge of its limits and variations.

From the range of forms possible given variations of argument types
conforming to the AS schema and different ways of arranging coordinate
and subordinate structures, a limited number of argument types and
macrostructures were selected for use as variables in the studies reported
below. Rationale for selection and the outcome of the selection process is
explained in the remainder of this section.

Emsjtjnneliinks. The propositional links used for constructing the
stimulus arguments and coding subject protocols for this study were of
three varieties: Warrant-Claim (W); Data-Claim (D); Subordinate Claim-

Main Claim (S). Each of these produces an ordered pair of propositions

(W<1,2>; D<1,2> and S<1,2>).

The Warrant-Claim and Data-Claim relationships were as defined for
the AS schema. The Subordinate Claim-Main Claim relationship was
defined for this study as linking the claim of an argument included in a text
with a further claim in the same text, where the first claim is used as data
for the second. This structure was chosen because it is probably more
common than that in which an argument is used to support a claim that is
then used as warrant for a further claim, since traditional wisdom
maintains that warrants are more often implicit.

Argnment typee. The forms for the simple stimulus arguments used
in these studies were selected with a number of constraints in mind. It
may be that there are arguments in natural language that do not make use
of subsumption. It may be argued, for example, that there are so-called
inductive arguments in natural language that use a number of premises
about specific cases to establish a general principle. Reasoning from
example (which Aristotle referred to as the “rhetorical form of induction”)
is often described this way. The forms selected for this study should all
make use of subsumption as described in the definition of AS provided
above.

Even if we limit the forms selected to those that conform to AS, it was
noted above that there are an unknown number of specific variations of AS
from which to choose, among which are some of the types of reasoning
discussed in the literature on informal reasoning.

The following criteria were used to select the argument types included
in the study:

(1) The type should be one that has a long history of discussion in the

literature on informal reasoning.

34

(2) The type should be one that has been discussed in the recent
discourse-analytic or speech-act traditions.

(3) Since no assumptions are being made concerning the correctness
of reasoning, those argument types that discussions on informal reasoning
have identified as fallacious should be considered.

(4) Since it is recognized in the literature on informal reasoning that
some argument types can be regarded as analyzable into simpler types,
only the simplest types should be considered because of the greater
likelihood of their occurring. in the average person’s repertoire of natural
language argumentation skills.

In their 1986 article, “Identifying argument schemes,” the speech act
theorists, van Eemeren and Kruiger, discussed three types of argument:
reasoning from sign, reasoning from cause and reasoning from analogy.
These three types also enjoy a long history of discussion in the tradition on
informal reasoning. However, there is good reason to suppose that analogy
is a more complex form of informal reasoning composed of simpler forms.
For this reason, analogy was not included in the study.

Reasoning from authority is a type of reasoning traditionally regarded
as fallacious. More recently, authors of texts on reasoning and
argumentation have become somewhat more liberal concerning the use of
authority appeals, usually redefining the fallacy to include inappropriate
appeals to authority. Since the criteria for distinguishing between
appropriate and inappropriate authority appeals are more complex and,
presumably, less well known to the average person than the general
structure of the appeal, and since appeals to authority abound in ordinary
discourse, that argument type was considered for inclusion in the study.

Upon attempting to construct stimulus arguments that include

35
authority appeals, however, the author found that constraints on the
microstructure made it difficult to create authority appeals that didn’t
sound ludicrous. This is attributed primarily to the demand that the
warrant be explicit. With authority appeals, this requires that the
argument contain statements like “Whatever Michael Burgoon says is
true.” In authority appeals that were found by the author in naturally
occurring arguments, and in typical examples in logic texts, selected from
natural sources or constructed, the warrant is typically left implicit, which
seems to indicate that, though many people are not above using blatant
authority appeals, they tend to make them appear less blatant by leaving the
warrant unstated. In view of these considerations, and to attempt to
conform to the criterion mentioned above concerning the inclusion of
fallacies, other fallacies were considered.

The fallacy that was selected for inclusion in the study was the fallacy
of accident. Since there are divergent definitions of this fallacy in the
literature on informal logic, the following definition was stipulated for this
study: The fallacy of accident consists of application of a generalization to a
case it is not intended to cover. Usually the fallacy occurs by exploiting the
fact that generalizations are often stated with important qualifications
omitted. To state a generalization without proper qualification is itself
sometimes referred to as the fallacy dieteejjnplicitez. In practice, however,
generalizations are often stated with relevant qualification suppressed, and
it is assumed that the hearer knows what the qualifications are (or at least
understands that generalizations are often stated with qualifications
suppressed.) If a generalization with unstated qualifications is understood

as unqualified it can easily be applied to cases that would be excluded by the

qualifications.

36

A simple example of the fallacy might be: “The sign says ‘Parking
reserved for vehicles with M.S.U. identification stickers. Others will be
towed.’ So, clearly, that fire truck [parked in front of the burning
dormitory] is in violation of the ordinance and should be towed.” On signs
such as the one in the example, a general rule (“Parking reserved for
vehicles with M.S.U. identification stickers”) is often posted without all the
qualifications that may apply in special circumstances (such as “except for
emergency vehicles or Safety and Security vehicles and snow removal
vehicles, etc.”). In the example, the speaker exploits the fact that the
qualifications are not stated in order to apply the general rule to a case that
would be excluded if the proper qualifications were understood — a case in
which a fire truck is parked in order to fight a fire in a dormitory.

In the examples of accident found in naturally occurring arguments,
unlike examples of authority , it was usually the case that the warrant (the
simplified generalization or general rule) was stated explicitly. Perhaps
this is because explicit statement of the generalization without qualification
reinforces the likelihood that it will be understood without qualification. In
any case, the fact that the warrant is usually made explicit in naturally
occurring argumentative discourse, qualified the fallacy of accident for
inclusion in the study as well as the fact that it is often included among the
fallacies traditionally identified in logic texts. Though discussion of the
fallacy was not found in the writings of speech-act theorists, the way the
fallacy occurs can easily be understood in terms of Grice’s maxims of
conversational logic, often referenced by speech-act theorists.

Even if the argument types are limited as discussed above, variations
remain that must be further reduced. Functional elements in a single

argument may be implicit. Furthermore, given the AS schema alone,

37
there are no constraints on the number of conditions (implicit or explicit)
included in a warrant, and hence no constraint on the number of data
elements that may be required (explicit or implicit) to substantiate the
claim. For the purpose of this study, these possibilities were limited so that
each stimulus argument would contain an explicit warrant and at least
one data element. These limitations were further modified by
considerations of possible macrostructures.

W. The macrostuctures selected for use in the studies
were the two on the left in Figure 1.

92W in r r W W8 r in r r 2
Warrant1 Datal Dataz Warrant\/ Data Warrant
Sub-claimt

S\/ub—claim (data)

Claim I Sub-claim 2
Main Claim

 

Main Claim

Figure 1. Macrostructure Examples

In the “Coordinate Structure,” the two data items belong to different
simple arguments and the warrant corresponding to Data 2 is implicit. In
“Subordinate Structure 1,” the sub-claim functions as data for the main
claim, and the warrant corresponding to this data item is implicit.

It is important to make some observations concerning these
macrostructures that could have interfered with expected results. First of

all, it was considered desirable to choose structures that have the same

38
number of propositions and the same number and type of microstructural
elements (each form contains two data elements, one warrant element and
one main claim element; both structures contain an implicit warrant). In
limiting the stimulus arguments to those with four propositions and
selecting only two from a number of possible forms in which four
propositions and two simple arguments might be related, a situation is
created that is less than ideal. In “Subordinate Structure 1,” the part of the
argument leading to the sub-claim may, by itself, be considered an example
of coordination, if one defines ‘argument’ as any statement offered in
support of an opinion, as do van Eemeren, et al. (1987). As argued above
(pp. 18-19), van Eemeren, et al. (1987) provide no functional distinctions to
distinguish between types of premises. Thus, they provide no way to
determine whether two statements belong to a single simple argument or
are, instead, two separate arguments supporting the claim in the
relationship they term ‘coordination.’ Using their definitions, then, one
could not say unambiguously that Subordinate Structure 1 was “purely
subordinate.” Designing an argument with four propositions that is avoids
this difficulty, however, would result in double subordination, as in
Subordinate Structure 2 , which would be much more difficult to
comprehend and recall, presumably, than Subordinate Structure 1, an
interesting question for further research.

The second problem, though much less serious, is that simple
arguments like the Coordinate Structure in Figure 1 with two data
elements may be less common than simple arguments with a single data
element.

Agelegy. While the above considerations may seem to greatly reduce

the scope of this study, two points may be made in defense of the apparently

39
restrictive nature of the limitations:

(1) Most of the studies on logical structure have been conducted using
structures that are far more restricted in form. The categorical syllogism
and three term series problems, for example, are much more limited in
scope than the forms used here. Consider, for example, the fact that any
syllogism can be recast in the form of Warrant-Data-Claim and that any
sorite can be recast as a subordinate argument structure of the type used in
this study. Furthermore, since there are many other specific argument
types and fallacies that may share the forms used here, the results should
be much more generalizable than the standard logic experiment.

(2) An effort is made here to use only arguments worded as they
typically are in natural language, rather than to recast arguments for the
study into a linguistic context that is rarely found in ordinary
argumentative discourse. The latter technique introduces more control,

but renders results more difficult to interpret relevant to natural language

argumentation.
M rin A m n B e ' Pr 1 n h In
W In order to

measure the match between subject protocols and the schematic structure
claimed to be underlying the text, the proportion of propositional links
identified by the subject out of all the links included in the stimulus
arguments were determined. In each condition, subjects were presented
with three stimulus arguments to increase the number of total links
available (3 links/argument x 3 arguments = 9 links), and to diminish
somewhat the possible effects of interaction with the content of the

arguments.

40

For the second study, a second set of subjects was assigned to groups
reflecting the same conditions as the first study (three types of argument,
two macrostructures and two levels of cues) with different instructions in
an attempt to offer independent evidence for the schema by demonstrating
more accurate recall of argument-relevant propositions vs. argument-
irrelevant propositions. The criterion here was the proportion of argument-
relevant and argument-irrelevant propositions out of the total propositions
available in the stimulus arguments. Note that here the links between
propositions are not of concern, merely the schema-relevancy of the
individual propositions.

Since proportional data were used in both studies, the data were
transformed using an arcsine transformation (Winer, pp. 399-400).

The criteria used for determining a match between protocols and
schema for Study I were kept deliberately very conservative. This was
considered necessary in order to minimize dependency on coder intuitions
of a match. It was argued above that AS can be widely, if not universally,
applied to naturally occurring arguments, but that the ability of trained
coders to apply a model does not by itself provide sufficient evidence that the
model represents a cognitive schema employed by ordinary language users
in producing or comprehending discourse of that type. That additional
evidence is provided by precisely specifying observational criteria for coding
subject responses.

For these reasons, specific surface cues were used as criteria for the
coding scheme in Study I and the decision to use such criteria resulted in
the conservative coding scheme. In that study, subjects were instructed to
reconstruct the arguments in their own words, as if speaking the

argument themselves, eliminating material that they judged extraneous to

41
the arguments (Appendix, p. 70).

We. In Study I, subjects were instructed to
reconstruct the arguments in their own words, as if speaking the
argument themselves, eliminating material that they judged extraneous to
the arguments (Appendix, p. 70). As indicated below (p. 49), subject
protocols for Study I were considered uncodable if the coders judged that the
subjects were merely trying to reproduce the texts verbatim or if subjects
gave a third person report of the argument. These responses were
considered uncodable because in both cases the subjects were regarded as
not having accepted the task of reconstructing the arguments. In pre-tests
(reported below, p. 50), this decision resulted in 74% codable responses. For
these codable protocols, the general rule for coders was that, in order for a
pair of propositions in the protocols to count as a matching link, there had
to be surface cues in the protocols (for the reasons explained above) that
indicated the same relationship between pairs of propositions (Warrant-
Claim; Data-Claim or Subordinate Claim-Main Claim) as that predicted by
the schema.

The clearest surface cues used to verify a match were standard
argument function markers — words such as ‘therefore,’ ‘because,’ ‘so,’
and their synonyms — occuring in the proper place between propositions.
Note that the markers ‘because’ and ‘so’ were used in the example given to
subjects in their instructions (Appendix, p. 70). To illustrate: if the schema
specified that A was a datum or warrant and B was the claim, and the
subject wrote “A so B,” that would be coded as a match, whereas “B so A”
would not.

The more difficult cases to code were those in which there were no

standard argument function markers. Intuitively, the sentence pair “You

42
can't vote. You're not even twenty-one” would count as an argument
consistent with the AS schema. The problem with this kind of construction
is that there are no obvious surface cues to support the interpretation that
the first sentence is a claim and the second data offered in support of that
claim. Thus, if a subject produces such a sentence pair there is no evidence
in the form of surface cues that the subject intended the pair as a claim and
a warrant. To maintain the conservative nature of the coding scheme, it
was decided that such cases should be coded as a non-match.

The only evidence of a match permitted without the presence of
argument function markers was the use of grammatical conjunction
within sentences linking premise and claim or grouping pairs of premises.
This was permitted, first, because of the frequent use of such forms in
naturally occuring arguments (for example, “You can't vote — you're not
even twenty-onel”) and second, because they contain surface cues
(punctuation, in the example given) as clues to intended argument
structure. In general usage, such grammatical conjunction within
sentences indicates that the second clause is intended as an explanation or
elaboration of the first, and grammarians recognize giving a reason as a
specific subclass of this type of conjunction. Since the subjects in Study I
were instructed to look for arguments, it was not considered too liberal to
consider such structures in their protocols as indicating that the second
clause (“you're not even twenty-one”) was a reason offered in support of the
first clause (“you can't vote”). On the other hand, if subjects merely
juxtapose two separate sentences, there is no clear surface evidence
concerning which they intended as claim and which as supporting reason.
To illustrate using the above example: the argument “You can't vote

because you're not even twenty-one” could also be expressed as “You can't

43
vote. You're not even twenty-one” or “You're not even twenty-one. You
can't vote.” However, another argument in which the individual clauses
function in the opposite manner, viz., “You can't vote, therefore you're not
twenty-one,” can also be expressed as either “You can't vote. You're not
even twenty-one” or “You're not even twenty-one. You can't vote.” Since
both arguments can be expressed in the same ways as a juxtaposition of two
separate sentences, the occurrence of such a structure in a subject's
protocol provides no evidence concerning which argument the subject
intended.

Coding in Study II was less problematic because coders were required
to match single propositions rather than pairs of propositions conjoined
logically. In that study, coders were given a list of the individual
propositions contained in the stimulus messages, including alternative
forms of expressing the propositions. Each proposition was identified as
schema-relevant or schema-irrelevant and coders merely had to count the
number of schema-relevant propositions. For example, for the following

stimulus message

Don't worry! You're always such a worry wort. It's gonna be a
success! If you're gonna have a really good party, you gotta limit
it—y'know?—to just friends. And you sent invitations, right?
'Course, they're not exactly the same people I would invite—just
kidding! And you're good at keeping them mixing, too, y'know—
ha, hal—that interpersonal stuff.

coders were provided with the following list:

44

1. Don't worry! (There's no need to worry) — irrelevant.

2. You're always such a worry wort. — irrelevant '

3. It's gonna be a success! (Your party will be a success.) —
relevant.

4. If you're gonna have a really good party, you gotta limit it to
just friends. (If a party is to be a success, you should invite only
close friends) — relevant.

(Etc)

It was understood that the list of alternative forms for each
proposition was not necessarily complete, and that there may be other

forms in which subjects would express the same proposition.

(leverietee

Based on Galotti’s (1989) excellent review of the research on formal
and everyday reasoning, reasoning ability does not seem to be related to
intelligence, but, for everyday reasoning, it does seem to be related to
general world-knowledge. Since the general world-knowledge of the
subjects chosen (see below) was presumed to be about the same, no effort
was made to control for intelligence or world knowledge in this study.

However, while all subjects were presumed to have been exposed to
the same general pressures for learning natural language argumentation
skills, it was thought that they might differ with regard to education in
logic, informal reasoning, argumentation or debate. Thus, the self-
reported level of prior logical training was used to achieve some statistical
control over that variable.

Also, though subjects may all have been exposed to the same

45
pressures to learn argumentation skills, subjects with more elaborate
general linguistic skills will be assumed to be better able to decipher the
underlying structure of any text. Consequently, subjects’ scores on the
language portion of their A.C.T. tests were used as a way of controlling for

differences in general linguistic skills.

Design

SLndLI. In study one 3 x 2 x 2 AN OVA was used with three types of
argument (cause, sign and accident), two types of macrostructure
(coordinate and subordinate), and two levels of cue elaboration (high and
low).

Study II. The second study duplicated the conditions of the first, with
different instructions. Since every subject was exposed to both types of
proposition (schema-relevant and schema-irrelevant), proportion of recall
for each type of proposition was compared using repeated measures, with
type of proposition nested within the cells of the 3 x 2 x 2 design described

above.

Mathew

SHAM-

(1) If a natural language argumentation schema is operating, it is
expected that subjects will reconstruct the arguments with a higher
proportion of accuracy when text cues are more elaborate.

(2) Since subordinate argument macrostructure is more complex
than coordinate argument macrostructure, it is expected that subjects will
reconstruct the arguments with a higher proportion of accuracy with

coordinate arguments.

46

(3) There is no reason to assume that any of the three types of
arguments is more likely to be familiar to the average subject than the
others. In fact the types were selected for the likelihood of their familiarity.
Consequently, it is not expected that subjects will reconstruct the
arguments with a higher proportion of accuracy for any of the three
argument types.

W.

(4) If a natural language argumentation schema is operating, it is
expected that subjects will recall a greater proportion of schema-relevant

propositions than schema-irrelevant propositions.

Method

Subiesta

There were 263 participants in Study I and 193 in Study II. All were
student volunteers from general education courses at a midwestern
university. More subjects were used than is typical for reasoning
experiments, as the artificiality of the language used in those studies

introduces more control than was assumed here.

Materials

Each argument constructed for pretesting contained four propositions
with three propositional links (see Figure 1.) All the arguments also
contained two or three propositions that were relevant to the text but
irrelevant to the argument structure. Schema-irrelevant propositions that

were nevertheless relevant to the text included statements clarifying

47
schema-relevant propositions, statements providing context, relational
comments, etc.

Transcripts of naturally occurring argumentation were examined in
an attempt to isolate surface cues used to distinguish between the
argument types used in this study. Several arguments of each type were
found in the transcripts examined, but there did not seem to be any
consistent linguistic markers to distinguish them. In view of this,
research assistants who were asked to construct arguments were provided
with a definition, a schematic form of the argument type and two or three
examples of each type.

The goal was to produce thirty-six arguments t3 be pretested in
printed form, six for each argument type, half of the arguments of each
type having coordinate macrostructure, half having subordinate
macrostructure. High and low cue-elaboration versions of the arguments
were then constructed by deleting argument markers (words like ‘so’,
‘because’, etc.) to form the low cue-elaboration version or adding them to
create the high cue-elaboration version, depending on which version was
created originally. Minor adjustments were often made in the wording of
the modified versions to accommodate the changes in cue-elaboration if
simply deleting or adding the cues made the texts sound unnatural in the
judgement of the author and research assistants.

The arguments were pretested in typed form for subject perceptions of
coherence, as measured using an instrument developed by Hamilton and
Burgoon (1984). (For the pretested arguments used in this study, coefficient
alpha reliability for the 14 semantic differential items was .84). The
arguments were also tested for “naturalness”, using three Likert-type

items. Though all the arguments were fabricated, subjects were told that

48
some of the passages they would be reading were made up by the author
and research assistants and some were transcripts of discourse recorded in
real-life contexts and that the purpose of the survey was to determine
whether they could tell the difference between the fabricated and natural
texts. It was hoped that by challenging them in this way they would be
more consciously critical, thereby giving us a better picture concerning the
naturalness of the texts. Subjects were later informed that all the texts
were fictitious and reasons for their distinctions were recorded and
discussed. Of the texts rated by them as below the mean on “naturalness,”
the most common reason given (34%) was “language too elaborate for the
context;” the second most frequent reason given (21%) was “topic not typical
for the speaker.”

Of the thirty-six original arguments, those falling below the mean for
coherence or naturalness were eliminated. For some cells in the design,
this left less than the three arguments required. Twice as many texts as
were needed to fill in the gaps were fabricated, using all the leads provided
by previous pre-testing, and this process continued until three arguments
for each condition remained that were above the mean for coherence and
naturalness established in the measurement of the original thirty-six
arguments.

The process of constructing and pre-testing the stimulus arguments
was time-consuming and required a large pool of subjects. It was deemed
necessary, however, in order to insure, as far as possible for the typed
versions of the arguments, that subjects’ perceptions of incoherence or non-
authenticity would not interfere with their ability to recall the structure of
the arguments in the main studies.

After the first pre-test of subject instructions for the main studies, it

49
was found that only 63% of the subjects were producing codable responses.
Codable response were considered those in which the subjects had
reproduced the arguments in their own words, as if speaking them
themselves. If it was clear to the coders that the subject was merely
attempting to reproduce the arguments word-for-word, or if the subject
gave a third-person report of the argument (The speaker said, “...” then he
said, “...”) those protocols were considered uncodable since the processes
required to produce such responses most likely do not require
comprehension.

The first set of instructions pre-tested were printed on the cover of the
subjects’ packets. The subjects were only told that the study concerned the
manner in which people recalled arguments, then they were instructed to
read the directions on the cover sheet. The number of codable responses
increased somewhat in the pre-test of the second set of instructions (to 67%)
when the author or research assistants read the directions exactly as
printed on the subjects’ cover sheets and a second worked example was
included.

Two additional pre-tests of instructions were conducted. Another
minor change that had some effect on producing codable responses was
elimination of the term “paraphrase” from the instructions. Apparently,
there is wide diversity among subjects concerning what that term means.
For the final set of instructions, the subjects were told that the researchers
had a way of scoring how much of the original passages they recalled and
that the person who scored the highest out of all the subjects involved would
get $5.00. They were told it was very likely that several subjects would be
tied with “perfect” scores and that in that event, all tying subjects would get
$5.00. (Only seven subjects had scores reflecting 67% accuracy. They were

50
told by their instructors to pick up their prizes in the department office at
the university where the studies were conducted. One prize remains
unclaimed.) In this last pre-test, 74% of the subjects were producing
codable responses. Due to the expectation of lost information in the main
studies, it was decided that a larger number of subjects would have to be
used than originally planned: 263 subjects for Study I (150 originally
planned) and 193 for Study II (120 originally planned).

While checking the pretest arguments to determine whether subjects
were producing codable responses, it was observed that of the four
arguments presented to each subject for the pretest, the first seemed to
produce considerably less codable responses than any of the other three,
regardless of the order in which the four arguments were presented.
Subjects responses to the first argument were badly garbled or seriously
truncated. Apparently, due to subjects being unclear concerning what was
expected of them, there seemed to be a practice effect differentiating their
responses to the first and subsequent stimulus arguments. In order to
diminish this effect in the main studies, an uncoded argument, the same
for all conditions in both studies, was used as the first argument in the set
presented to the subjects. The argument used differed in both type and
macrostructure from those used in any of the conditions in both studies.
This argument was not coded for accuracy in the first study, nor for recall
of relevant versus irrelevant information in the second study, and was
merely used as a warm-up exercise to familiarize the subjects with the
task. N 0 further effort was made to control for practice effects, since it was

not confounded with any of the variables of interest in either study.

51
Pr r

Subjects in both studies were presented with four audio-taped
arguments. Subjects in Study I were instructed to reconstruct the
arguments after hearing each one, while subjects in Study II were simply
asked to recall them. (See Appendix p. 70 for instructions used in Study I, p.
71 for instructions used in Study II.)

In two studies on subjects' ability to recall argument lines, Marcus
(1982) had subjects listen to five recorded arguments, write out the
arguments as they recalled them and then do the same with five more
arguments. The structure of the arguments for that study, though
presented somewhat more artificially than those to be used here, were
considerably more elaborate when analyzed formally. Marcus was able to
detect differences in the way different structures were processed when
subjects listened to five arguments in sequence before attempting to recall
any, so, given the simpler task in the present studies, it was assumed that
the differences in the structures of arguments used in these studies would
be detectable by the subjects.

Since recall data used in psycholinguistic studies of the type
conducted here presuppose that the pressure of reduced processing time
provides a more accurate picture of processing, during pretests the author
and research assistants made head-counts at 15 second intervals of subjects
who appeared to be finished writing their responses. It was determined
that 45 seconds was adequate time for subjects to write out their responses.
If a full minute was permitted, some subjects were observed to go back and
cross out their responses and rewrite them, and in some cases even went
back to previous pages of their booklet to modify their responses, or were left

sitting idle long enough for them to scribble comments or questions about

52
the study. If only 30 seconds were provided, about a third of the subjects
were still writing when the taped warning to go on to the next page was
heard. For Study 11, however, for which the task was presumed to require
less difficult processing than that required in Study 1, subjects were given
only 30 seconds to recall the arguments.

After all the data were collected, the author visited the classes from
which subjects were drawn and debriefed the subjects concerning the
purpose of the study.

Subject protocols for both studies were coded by two research
assistants, both of whom audited the portion of the author’s argumentation
class in which the Toulmin model was presented. The research assistants
were given a sample of arguments constructed by the author along with the
definitions of the relevant components and were asked to take them home
and analyze them. The author discussed the result of their analysis with
the assistants and the exercise was repeated with another sample of
arguments. Total training time, including assistants’ report of time spent
on “homework” and the time spent auditing the author’s class, was
approximately eight hours. Assistants could probably be trained in less
time, however, since that part of the training that occurred in the
classroom (nearly four hours) would have been reduced if the assistants
were tutored privately. Assistants gained further experience coding the
protocols from the pre-tests.

The coders were not aware of the hypotheses of either study. Due to
the possibility of the coders assuming that “accuracy” and “recall of
relevant information” were preferred results, it was stressed that no such
preferences were to be assumed, since the purpose of the study was to test a

model, and that accuracy and relevancy were defined only in terms of the

53
model so that there was no implication that conformity to the model was
necessarily a better way for the subjects to process the texts.

This extensive training and experience resulted in very high coder
agreement for both studies. The coder agreement for Study I was 95.3%.
For Study II, coder agreement for relevant information was 90.1%; for
irrelevant information agreement was 95.7%. (Though the difference
between coder agreement on relevant versus irrelevant information is not
large, it is curious. The research assistants who did the coding suggested
that the reason their agreement was greater on irrelevant items was
probably due to the author’s greater care in defining what constitutes
schema-irrelevant information and his overconfidence concerning the
likelihood of agreement on what subject interpretations were to be

considered relevant.)

Results

Snidxl

The data for Study I were analyzed using a 3 x 2 x 2 design with type of
argument, macrostructure and level of cues as factors and ACT language
score as a covariate. Analysis was performed using the SPSS-PC AN OVA
routine.

The covariates considered for the study were subject's ACT language
scores and prior experience in courses in which the structure of arguments
was included as a topic. The range of experience was rarely more than one
course in high school or college. Since the data on prior experience was

categorical and since the categories differed widely form the point of view of

54
the focus of the course (public speaking, introduction to logic, practical
reasoning, etc.) subjects were simply classified as having prior academic
experience or not. The means for accuracy were .89 for no prior experience
and .78 with prior experience. A t-test was performed which revealed that
this difference was not significant, 1, (261) = 1.52, p = .129. On the basis of
these considerations, prior experience was not included as a covariate.

The correlation between ACT language scores and accuracy was .09.
Though this amounts to only 1% of the variance of the scores, the
superiority of the measurement compared to that for prior experience, and
lack of guidelines for anticipating magnitude of effects due to type,
macrostructure and cues, general language skill was retained as a
covariate. After the main analysis was performed, it was observed that the
contribution of the covariate was not significant, 13(1, 250) = 2.74, p > .05.

The first hypothesis identified above was that if a natural language
argumentation schema was operating, subjects would reconstruct the
arguments with a higher proportion of accuracy when text cues are more
elaborate. The results, however, show no main effect for cues, F(1, 250) =
0.297.

While the argument types were chosen because it was assumed that
they would all be equally familiar to the average person, the analysis
revealed that there was a remarkable effect for argument type E(2, 250) =
37.73, p < .001. Subjects found sign reasoning far easier to comprehend
than either causal reasoning or the fallacy of accident. The means for the
three argument types were .30 for causal reasoning, .59 for sign reasoning
and .38 for the fallacy of accident.

The mean accuracy for coordinate macrostructure (.44) was higher

than that for subordinate macrostructure (.40), but this difference was not

55

significant F_(1, 250) = 2.61, p > .05. Note that the argument type by
macrostucture interaction was significant, however, E(2, 250) = 5.84, p >
.05. In view of this, each of the argument type conditions was examined to
determine whether a simple main effect for macrostructure was present.
For subjects in the sign reasoning condition, the simple main effect for
macrostructure was significant, E(1, 84) = 16.74, p < .001, whereas this was
not the case for causal reasoning, E(1, 9)1 = 0.22, p = .64, nor accident, E(1,
82) = 0.15, p = .71. These tests of simple main effects were computed using
the error terms from only those cases included in each condition rather
than the error term from the entire sample as a test of homogeneity of
variance was not performed. Neither did the analyses include the
covariate, which contributed little to the results of the main analysis.

The remaining two-way interactions (type x cues, macrostructure x

cues) and the three-way interaction were not significant.

Table 1

Analysis of Variance, Study I

f ri i n f i ifi n f F
Covaritate; ACT .148 1 .148 2.735 .099
Argument Type 4.079 2 2.040 37.727 .000*
Macrostructure .141 1 .141 2.607 .108
Cues .016 1 .016 .297 .586
Type x Macro. .632 2 .316 5.842 .003
Type x Cues .122 2 .061 1.127 .326
Macro. x Cues .030 1 .030 .554 .457
Type x Macro x Cues .131 2 .066 1.213 .299
Explained 5.232 12 .436 8.064 .000
Residual 13.516 250 .054
Total 18.748 262 .072

StmbLII

The data for Study II were analyzed using 3 x 2 x 2 x 2 design with
type of argument, macrostructure and level of cues as between-subjects
factors and schema relevance (relevant and irrelevant information) as the
within-subjects factor. Analysis was performed using the SPSS-PC
MAN OVA routine.

The fourth hypothesis, that subjects would recall a greater proportion
of schema-relevant propositions than schema-irrelevant propositions, was
supported with E(2, 179) = 3.67, p = .027. (3’5 reported in this section are
those calculated from Pillai’s Trace statistic. See Table 2, below.)

The remaining results for this study parallel those of Study I. There
was a main effect for argument type, E(4, 360) = 7.97, p < .001, and a
significant type x macrostructure interaction, E(4, 360) = 4.52, p = .001. As
in Study I, cues had no effect, E(2, 179) = 0.931, p = 0.396, and the remaining

two-way interaction and the three-way interaction were not significant.

Table2
Analysis of Variance, Study II.
o__ ‘0 4.10; " i’ intro; 0.100.-. _-_l0_0 unit on '0

Within subjects effect:

Relevance .03941 3.672 2 179 .027
Argument Type .16273 7.971 4 360 .000
Macrostructure .00721 0.650 2 179 .523
Cues .01030 0.931 2 179 .396
Type x Macro. .09566 4.521 4 360 .001
Type x Cues .03440 1.575 4 36) .180
Macro. x Cues .02147 1.964 2 179 .143

Type x Macro. x Cues .02446 1.114 4 360 .349

Discussion

While only one of the four hypotheses in the two studies were
supported, the results indicate that what happens when people attempt to
comprehend natural language argumentative discourse can be illuminated
by the theory of argument schemata.

In developing the story about what happens when people attempt to
comprehend arguments it was claimed (above, p. 29) that “Only if subjects
reconstruct these [links within simple arguments] correctly will they be
able to recognize the structure (coordination or subordination) of complex
arguments.” It was not intended that this implication of the theory be
tested, since an attempt was made to select types of reasoning that the
author thought would be equal with respect to comprehension. However,
the fact that sign reasoning turned out to be far less difficult for the subjects
to comprehend suggests an explanation for the type x macrostructure
interaction. When subjects were confronted with an example of sign
reasoning, the three internal propositional links fell into place for them
more easily so that the macrostructure was more clear. That overall
comprehension was better was reflected in the higher accuracy scores for
coordinate macrostructure as opposed to subordinate macrostructure when
the type of reasoning in the simple arguments was familiar to the subjects,
specifically, in the case of sign reasoning. As reported above, for subjects in
the sign reasoning condition, the simple main effect for macrostructure
was significant whereas this was not the case for causal reasoning nor
accident. This supports the claim that comprehension of the
macrostructure of complex arguments is contingent upon familiarity with

the types of reasoning used in the simple arguments. Since the pattern of

57

58

significant results was identical in both studies, the overall outcome of the
two studies may be interpreted as supporting the idea that, for argument
types with which the average person is familiar, a schema is operating that
influences the information listeners retain accurately when exposed to
spoken argumentative discourse and that comprehension of the structure
of complex subordinate or coordinate arguments is contingent upon
familiarity with the type of simple arguments making up those structures.

Though the author assumed that causal reasoning and the fallacy of
accident are included in the average persons repertoire of familiar
argument types, this is apparently not the case. Though pe0ple may use
the fallacy of accident themselves, and though they may fall victim to it
occasionally, perhaps when listening to arguments under the conditions of
these studies their focused attempt to comprehend the arguments caused
them to notice “something fishy” about the accident arguments which
interfered with their comprehension. If, indeed, this is what is happening,
then from the point of view of prescriptive logic, this is a happy result.
More importantly in terms of this investigation, if the fallacy of accident is
not in the average person's repertoire of argument forms, that would be
consistent with the theory developed in the introduction, viz., that there
must be a natural language logic that leads us, more often than not, to true
conclusions about the world in which we are trying to survive. As defined
above, the fallacy of accident amounts to over-applying a generalization,
which habit would not seem to have survival value.

It may be disturbing to contemplate subjects' poor performance with
causal reasoning, which seems to indicate that this form of reasoning is
somewhat too sophisticated to be included in the average persons

repertoire. It should be pointed out, however, that this finding may also be

59
consistent with the survival-value assumption. For getting through life,
being able to detect “signs” indicating the presence of absence of beneficial
or harmful phenomena may be quite enough, and being able to reason
about the causes of those phenomena may very well be a sophisticated
embellishment not developed to a significant degree in the average person,
despite Virgil's dictum “Happy is the man who knows the causes of
things.”

In a study of argument recognition, van Eemeren and his colleagues
(1985) discovered that the only one that had any effect on subjects’ ability to
recognize arguments was the inclusion of premise or claim markers —
words like “so,” “because,” “therefore,” etc. For this reason, it was thought
that such cues would aid comprehension of argument structure in the
present studies. While these cues may help people recognize arguments,
they don’t seem to make a difference in subjects’ ability to comprehend
argument structure. Contrary to hypothesis 1 for Study I, surface cues —
at least those used to indicate argument microstructure — seem to have
little effect on helping subjects reconstruct argument structure accurately.
The failure of these studies to detect a main effect for cues, however, is an
ambiguous outcome with respect to the theory that there are argument
schemata implicated in the production and comprehension of
argumentative discourse.

It was mentioned above (p. 31) that if differences in schema produce a
difference in Processing We.
that difference could be attributed to differences in schema applied rather
than differences in surface cues. Since no differences were found for cues,
the theory suggests that the schema accounted for the differences observed.

Nevertheless, since the theory posits cooperation of text cues and schema,

a)

an alternative explanation is required for the observed differences in
interpretation despite lack of observed differences for surface cues. It may
be argued that these studies simply failed to isolate the relevant surface
cues, and that some obvious marker, or set of markers, does indeed exist as
the surface cue that influences subject responses. A less ad hoc, more
cognitive explanation, however, is that subjects’ comprehension of the
meaning of the propositional content of the statements in the text is the
primary cue that determines what schema will be applied, with or without
the help of surface cues.

Consider the author’s following favorite example used to demonstrate

the operation of implicit schemata to students of argumentation classes:

A: John left for the grocery store. He had noticed that it was
raining heavily outside.

B: John’s no champ. He hadn’t fought everybody.

Prior to discussion of argument structure, students in introductory courses
typically identify example A as a narrative, whereas B is typically identified
as an argument. There are no obvious surface cues to distinguish the two
examples, however. The only cues available in each case seem to be the
juxtaposition of two sentences and the anaphoric link between ‘John’ in the
first sentence and ‘he’ in the second. Stenning (1978) argues convincingly
that success in attempting to make sense out of pairs of sentences linked in
this way does not depend on surface cues, but on the hearer/reader
supplying a cognitive interpretation — a context — that renders the pairs
coherent. The enormous number of contingent rules that would be

necessary to render such texts coherent if success depended on surface

61
cues alone makes the alternative theory far more attractive. What seems to
be happening here is that when two sentences are grouped together, pe0ple
assume that they are intended as a single coherent text and attempt to
interpret them that way, applying various schemata with no help from
surface cues to indicate what schema is appropriate. The two sentences in
A can be interpreted as a coherent unit if the reader interprets them as a
narrative. Whereas, if they try to apply argument schemata, coherence is
more difficult to obtain. E.g., “John noticed it was raining heavily outside,
therefore he went to the grocery store” can only be assumed coherent if we
add the assumption that “John always goes to the grocery store when it
rains heavily.” Since this assumption seems peculiar to most people, they
are reluctant to interpret the text as an argument. By contrast, they are
apparently not reluctant to supply the warrant “A champ has got to fight
everybody,” which renders B coherent as an argument.

The prediction that surface cues would make a difference was based
on a concession to the literature (van Eemeren, et al., 1985). The failure of
these studies to support that prediction may be attributed to differences in
task (merely recognizing arguments versus recalling argument structure)
rather than to a weakness in the composite theory. That theory asserts that
successful comprehension depends on both schema and surface cues and
thus that there should be an effect for schema beyond that accountable for
surface cues alone. The pattern of effects observed in these studies supports
that claim.

A discovery made by a research assistant in the process of coding the
subjects’ protocols may indicate that the errors that led to inaccuracy were
somewhat systematic and, in fact, indicated a straining toward accuracy.

What was discovered was that subjects seemed to exhibit distinct

62
processing biases. In many cases where the subjects were coded as
inaccurate because their responses failed to accurately match linked
proposition pairs, their responses seemed to exhibit specific patterns, i.e.,
their failure to match the original structure seemed due to the fact that they
were looking for something else. In terms consistent with the theoretical
foundation for these studies, a different schema was operating, or the
schema was applied using a different strategy. In other words, the subjects
exhibited processing biases. They are referred to as “processing biases”
rather than simply “processing errors” because it was often the case that
the same subject committed the same type of error in more than one of the
three arguments processed, and different subjects would often exhibit the
same type of error. If their responses were simply random, to say that
subjects were employing a different schema would be an ad hoc
modification of the theory, i.e., the theory would be unfalsifiable if we were
permitted to say merely that “another schema was operating” any time the
results were insignificant. Since the possibility of processing biases was
not considered beforehand so that no attempt was made to control for that
variable, all that can be said for the present studies was that inaccurate
responses represented random individual differences. The'discovery of
possible processing biases suggests, however, that in future studies of this
type, error can be reduced by being aware of some of the processing biases
(alternate schemata or strategies) that subjects can employ.

The processing biases exhibited are interesting in that it is relatively
easy to classify them in a manner that relates them to the hypothesized
schema. One such bias may be called “extracting a warrant.” This
consisted of reducing the entire complex argument to a single warrant. In

these cases, the subjects simply reported the warrant that was explicit in

63

the stimulus argument. For example, given the following stimulus

argument

Henry the Eighth probably died of scurvy. He had wild mood
swings and a collapsed nose — typical symptoms of scurvy.
Also note that like most of the upper-class of Tudor England, he
treated fruits and vegetables as unfit to eat.

the subject exhibiting this kind of bias produces a reconstruction that
reduces the entire argument to a single rule-like statement, such as “Mood
swings and collapsed nose are symptoms of scurvy.”

An explanation for this effect might be that the subject approaches the
task with the assumption that the most important element to grasp in any
argument is the general rule which justifies the inference from data to
claim. Obviously, the warrant is pivotal in the Toulmin model, and
grasping the function of a warrant may be an early step in the development
of a person’s ability to comprehend arguments. This bias represents a lack
of pragmatic sophistication, however. It is not pragmatically correct to
assume that the warrant is the most important element to grasp in all
arguments, since often the source assumes that the warrant is well known
to the receiver so that providing the data is source’s primary concern.

If this is the nature of the error that occurs in these cases, it is
consistent with a kind of error that is commonly observed in
psycholinguistic studies of the acquisition of grammatical rules. When
children learn a rule of grammar, they tend to over-apply it at first, and
only later learn to limit application according to exceptions to the rule. For

example, children are observed to make accurate tense distinctions with

64
irregular verbs the uses of which they have apparently learned separately
first, as in “I go outside” versus “I went outside.” Later, the child learns the
rule for the formation of past participles and the formerly correct
constructions are replaced with incorrect ones that appear to represent
over-application of the past participle rule — “I went outside” is replaced
with “I goed outside.” Finally, the child learns the exceptions to the rule
(irregular verbs) and correct construction returns. The suggested analogy
to the habit of “extracting a warrant” is that people first learn of the
importance of grasping the warrant to comprehending arguments and only
later learn that sources often intend the data as the rhetorically crucial
information to be grasped in their arguments.

Another processing bias observed might be called “vertical
simplification” (to distinguish it from “horizontal simplification,” described
below). This consists of failing to reconstruct intermediate steps in
subordinate arguments, the subjects’ protocols indicating links between the
data or warrant of the first argument in the chain and the main claim,
without mention of the intermediate claim). For example, given the

following stimulus argument

Lots of times people try to follow those textbook formulas for
impression management and come off appearing stiff — unreal,
you know? Jimmy does that — Jimmy from French class — he
really believes that stuff, okay? So people get that kind of
impression of him. I mean, like, “Where’s this guy comin’
from,” y’know? So, they naturally don’t like him very much at
first.

65
a subject might produce a protocol that reads “When people use textbook
formulas for their image they appear stiff and unreal. That’s why they
don’t like Jimmy at first.” The intermediate claim, “People get the
impression that Jimmy is stiff and unreal,” is left out of the subject’s
reconstruction.

It is difficult to describe this bias as an error at all, since the main
claim is correctly identified and the direction of support is retained. Also
note that the same pressures that motivate the production of enthymemes
may be used to explain this bias.

Finally, some subjects exhibited “horizontal simplification.” In this
situation, subjects collapsed coordinate structure (two simple arguments
supporting a claim independently) to a simple argument, i.e., they phrased
their protocols in a manner that indicated all premises supporting the
main claim were premises of a single argument. For example, given the
stimulus argument used previously concerning Henry the Eighth, the
subject might write, “Henry the Eighth probably died of scurvy because he
had wild mood swings, a collapsed nose and didn’t eat fruits and
vegetables.” In this case, the fact that two warrants are required for the
argument to fit the schema is not reflected in the subject’s reconstruction
and the data from the coordinate arguments are presented in such a way as
to suggest that the subject regarded all the data items as belonging to a
single simple argument (wild mood swings, a collapsed nose and not eating
vegetables are regarded as data relative to a single simple argument.)

This last bias posed a coding problem due to the fact that one of the
hypotheses predicted greater comprehension for coordinate arguments.
The original coding instructions defined links between propositions only.

Which propositions are linked incidentally distinguishes simple

66
arguments from subordinate arguments (the two data elements are linked
to different claims), whereas it doesn’t distinguish simple arguments from
coordinate arguments (the data elements from the two arguments are
linked to the same claim). Thus there was no way, given the original
coding instructions, for identifying protocols like the one in the last
example (where a subject gives no indication of recognizing coordinate
structure) as inaccurate. This could have resulted in a distortion in favor of
accuracy in coding coordinate arguments, since examples of the type just
illustrated would have all been coded as accurate though the protocols
reveal no evidence that the subject recognized the coordination. Once this
bias was brought to the attention of the author, however, a computer-aided
search of all those cases coded as accurate in the coordinate argument
conditions were reviewed. If there was no indication that the subject
recognized the coordinate structure of the stimulus arguments, their
responses were recoded as inaccurate. The remaining uncoded protocols
were then coded in the same way.

The necessity of defining criteria for determining whether the
subjects recognized coordination led to another discovery that may be of
some interest. In both the arguments constructed for the study and the
subjects’ protocols, there were some identifiable surface cues that appeared
to be used to indicate coordination — i.e., distinct phrases that indicated
that there two different arguments supporting the same claim. The phrase
“....Also note” in the Henry the Eighth example seems to serve this
function. The phrase (plus the fact that it begins a separate sentence from
that in which the data from the other argument is presented) seems to be
designed to indicate that the following data is distinct in some way, perhaps

because it belongs to a separate, coordinate argument. Other such

67
coordination markers were “....Besides,” “....And, besides,” etc.
“Furthermore,” “In addition,” “Not only that, but,” may be pragmatically
synonymous expressions, though they did not appear in the stimulus
arguments nor in subjects’ protocols.

Note that each of the processing biases discussed above, far from
indicating mere randomness in the “incorrect” responses, indicate that,
with refinements, the proposed schema could account for even more of the
variance than in the present studies. The biases discussed above can be
integrated into the theory without great difficulty. For example, the
tendency to “extract a warrant” indicates that the subject habitually seeks
out the general rule when attempting to comprehend an argument. This is
encouraging considering the importance of the warrant in making the
argument work logically. It can be dysfunctional, however, because, as
mentioned above, in those arguments where the speaker assumes the
hearer knows the general rule, the pragmatically crucial element is the
new data provided. Thus, a processing bias that favored extracting the
warrant would mislead the bearer in many cases, because the crucial
information from the speaker’s point of view might not be comprehended by
the hearer who, focusing exclusively on the warrant already taken for
granted (“given” information) by both of them, might accuse the speaker of
stating the obvious, while missing the pragmatically crucial data (the
“new” information).

These processing biases could easily be made the subject of future
studies. Familiarity with the kinds of processing biases for arguments can
be used as guides for the construction of arguments in such a way as the
compensate for these differences, if, for examme, there are certain modes of

presentation that overcome these biases.

68

One outcome of these studies that calls for comment is the overall low
level of accuracy in subject responses. In Study 1, the untransformed mean
percent of accuracy across all conditions was only 21.4%. The means for
the three argument types were 33.1% for Sign, 18.6% for Accident and
13.1% for Cause.

These results may be interpreted as bringing into serious question the
impact of the logos dimension of messages, and may seem to invite the
conclusion that people are not logical. It is important to keep in mind,
however, that the immediate purpose of this investigation is to describe the
logic that people use, not to evaluate their performance relative to
philosophical logic. Consequently, “accuracy” as used in these studies
refers to the degree of agreement between subjects' observed performance
and the model of a schema that is hypothesized as underlying that
performance. Accuracy, then, is not necessarily a measure of their logic in
a philosophical sense, rather it is a measure of the extent to which their
behavior can be accounted for by the model.

Furthermore, it must be emphasized that most of the stimulus
arguments were imbedded in conversational, highly colloquial speech, with
either relatively simple argument markers, as discussed above, or the
absence of markers in the low-cues conditions. Note also that up to half the
information contained in the texts was deliberately generated as schema-
irrelevant, no attempt was made to ensure the clarity of argument
structure and the topics of the arguments were relatively innocuous so that
there was little likelihood of high involvement. (See Appendix, pp. 5-9.)
Taking these conditions into consideration, together with the possibility of
processing biases, the accuracy scores need not be interpreted as low. AS

does not purport to model everything that happens when people attempt to

@
comprehend arguments and it is as important to understand the
limitations of logos as it is to understand the form that logos takes in
natural language. It is expected that in future studies a clearer picture
will emerge concerning the variety of forms constitute the logic of natural
language, as well as a clearer picture of the variables that limit that logic.

Note also that the very conditions referred to above as possible
explanations of low accuracy scores—colloquial speech, simple argument
markers, and inclusion of schema-irrelevant information—suggest
methods that may be used to enhance the comprehension of the logical
structure of messages and thereby improve argument practice. If
arguments are presented in a more elaborated code, with clearer argument
markers, with schema-irrelevant information kept more clearly separate,
it may be the case that comprehension increases a great deal. All of these
suggestions may be empirically tested in order to determine their
effectiveness in contributing to logical comprehension.

What the studies reported here indicate confirms the assessment
stated in the introduction, viz., a natural language logic would have to be
complex enough to be account for our ability to construct and comprehend
the diversity of arguments as they are presented in everyday life, but not so
complex as to render it unlearnable by the average person and incapable of
operating at the rapid processing speed required for conversational
argument. The patterns of responses exhibited by subjects in these studies
provide enough support for the idea that there are identifiable regularities
in argument processing to warrant further investigation, and provide

valuable leads concerning the direction such investigations may take.

APPENDIX

Appendix

In i

In this study, we are trying to find out what kinds of conversational
messages are best understood.

There is a way to score your results to see how well you understand
the messages. The person who does best of the subjects selected for this
study will get $5.00. In case of a tie, both of you will get $5.00. Note that I
have asked you to put your student number on the front page. This is so
that I can locate the winner and so I can obtain information your prior
education in language skills. The results of this study will be held in
strictest confidence, as is standard with research, and will not become part
of your University record.

I have taken 4 exerpts from actual transcribed conversations and have
had them performed by an actor so that they would sound clearer. In each
case the speaker is making an argument, i.e., stating his or her position
and supporting it. The passages may contain some extraneous material.
Your task is to listen carefully to each passage and then, after you have
heard the entire passage rewrite it in your own words, writing down only
the essentials—just the gist—of the speaker's arguments as if you were
speaking them yourself. Don't try to memorize the passage word for word.
You may leave out anything that you think is unnecessary to the argument.

For example, if the speaker says:

You know Dr. Smith from Social Science? Well, I think he's the best
professor at this school. He teaches History mostly, I think. He's good
because he almost never gives F's and at least a third of the people in his
classes get A's. So I think you should take him.

You might rewrite the passage as follows, as if stating the argument
yourself:

"Dr. Smith is the best professor at this school, because he almost
never gives F's and at least a third of the pe0ple in his classes get A's. So, I
think you should take him."

Repeat: listen carefully to each passage and then, after you have
heard the entire passage rewrite it, writing down only the essentials of the
speaker's arguments as if you were speaking them yourself.

You will be given a little less than a minute to rewrite after each

passage, then you will hear my voice on the tape instucting you to go on to
the next page.

70

71
mm

In this study, we are trying to find out what kinds of conversational
messages are best remembered.

There is a way to score your results to see how well you remembered
the messages. The person who does best of the subjects selected for this
study will get $5.00. In case of a tie, both of you will get $5.00. Note that I
have asked you to put your student number on the front page. This is so
that I can locate the winner and so I can obtain information your prior
education in language skills. The results of this study will be held in
strictest confidence, as is standard with research, and will not become part
of your University record.

I have taken 4 exerpts from actual transcribed conversations and have
had them performed by an actor so that they would sound clearer. Your
task is to listen carefully to each passage and then, after you have heard the
entire passage write the passage out on the form provided as well as you
can recall it.

You will be given a half-minute to rewrite after each passage, then
you will hear my voice on the tape instucting you to go on to the next page.

72
Table of Means, Study I

 

 

 

 

 

 

Macrostructure Coordinate Subordinate
Cues Lo Hi Lo Hi

Cause .35 .21 .30 .32

(26) (22) (21 ) (24)
Argument

Type Sign .64 .72 .50 .50
(21) (21) (19) (25)

Accident .38 .35 .39 .39
(18) (20) (23) (23)

 

 

 

 

 

 

Table of Means, Study II (collapsed across variable Cues)

 

 

 

 

 

 

Macrostructure Coordinate Subordinate

Relevance Relevant lrrel Relevant lrrel

Cause 1 .38 .45 1 .07 .58
(34) (34) (35) (351

Argument

Type Sign 1 .43 .38 1.62 .44
(29) (29) (32) (32)

Accident 1 .27 .72 1.38 .71
(30) (30) (33) (33)

 

 

 

 

 

 

 

 

W

The following is a list of all stimulus messages used in both studies.

(Material in brackets indicates items that were considered schema-
irrelevant for Study II.)

C
1.

rin w

[Don't worry!] [You're always such a worry wort.] It's gonna be a
success! If you're gonna have a really good party, you gotta limit it—
y'know?—to just friends. And you sent invitations, right? ['Course,
they're not exactly the same people I would invite—just kidding!] And
your're good at keeping them mixing, too, [y'know—ha, ha!—that
interpersonal stuff]

[Thank God it's Friday!] [I don't have a whole helluva lotta cash.] But,
whenever you really wanna relax, I think it takes less to get you drunk.
An' I tell you, if ever there was a weekend I could use some relaxing,
this is it. [Have you finished that term project?] 'Course, I haven't had
much to eat today, either. It's not gonna take me much to get a good
buzz on.

[He doesn't speak their language, you know what I mean?] Like, he
doesn't know what their concerns are. N o, I don't think he'll win.
[Not much of a chance] If people are gonna vote for you, [you gotta you
gotta know where they're comin from]—you gotta understand their
concerns. Besides, he doesn't act like a winner.

MW

Lots of times people try to follow, y'know, those textbook formulas for
impression management and come off appearing stiff—unreal, you
know? [They go right down the checklist] Jimmy does that—[he really
believes that stuff, okay?] People usually get that bogus impression of
him. [I mean, like, "Where's this guy commin' from," y'know?] And
naturally they don't like him very much at first.

[You hear what I'm sayin' girlfriend?] If your gonna forget about it,
you gotta keep telling yourself that it wasn't right for you, otherwise it
lingers. But you don't tell yourself that; [you just hang around, waitin'
for it to fade away] Of course it lingers! And then you got that bad
attitude with the next dude, like he's gonna disappoint you, too. [You
never just END it!]

[Look at this stuff comin' down.] Maybe school'll be closed tomorrow.
This is gonna be an ice storm. [We had one last year, remember?] [Or
maybe that was back home] Anyway, when it's freezing down below,
this here drizzle builds up on whatever it touches, an' we've had
freezing weather for—what?—a week? [Hey, I could use a day off]

74
i ri

7. [Did you see Jerry last night?—Jerry from nextdoor?] He's got those
ikky red spots all over, an' he's tired all the time. Aren't those signs of
AIDS? An' you know he'll sleep with anybody, [really, what a sleaze] I
bet he's got AIDS. [Not that I'd wish it on anybody]

8. Henry the Eighth probably died of scurvy. [Do they still call it scurvy?]
Anyway, he had wild mood swings and a collapsed nose—typical
symptoms of scurvy. Also, [like most of the upper—class of Tudor
England,] he treated fruits and vegetables as unfit to eat. [Imagine a
king dying of malnutrition!]

9. If a person thinks that all who disagree with him are all equally wrong,
he's what they call a dogmatic personality. That's just what you get
with racists. [You know the type?] Another thing is the way they feel
an overwhelming need to spread the word. [Mostly I think that's to
convince themselves] I would say that racists are mostly dogmatics,
yes.

Sin/ ri

10. [Yeah, right, he may be cute, but,] face it, he's a jerk, if you ask me. If
all a guy does is talk about himself and his sexual prowess, that makes
him a jerk in my book. [I can't stand when they do that!] And who do
we know talks that stuff? Like ALL the time. You know who! I'd say
you better dump him quick. [Just be brave!]

11. I know they'll let us spend the night together. [I don't think we have a
thing to worry about] They think of you as perfect for me. I told them
all about you, about how ambitious you are—[W ell, you arell—and how
you treat me right. That's what makes a perfect husband in my
parents eyes.

12. [They announced this here scholarship, y'know, for star students, I
guess, an' Jeff says he's gonna apply.] I don't think he's exactly what
you'd call a star student. [I just don't see it.] You gotta study and it
helps if you do your own work. And he never cracks a book and he's

always asking to see other people's homeworks. Ain't no way he'll get
that scholarship.

A'i [C l' 1 C

13. [No, he lives in Colorado, not Arizona] Anyway, he hasn't called me in
four days. Now, when people are in love they want to communicate,
right? And, really, when he calls he never talks about us much, [he
just talks about his job and school an' like that] [I mean who cares?]
Him sayin he loves me, an' all? I don't buy it. [It's just talk.]

75

14. This is a democracy—everybody's supposed to be part of the decision

15.

A

16.

17.

18.

making process. [How come only one person gets to decide what our
grade is?] The whole class should should do the evaluation. [Think of
the professor as someone whose job it is to just TEACH] [That makes
sense to me] That's what they did in Russia, I believe, during the
revolution.

[See this here in the paper?] The University has no right to say a
fraternity can't sell their calendar on campus, y'know, the swimsuit
thing they tried to sell in M0 Hall? This is a free country right? I mean,
freedom of speech? You can't be havin' no censorship in a free country.
[Who they think they are?] And they ain't our parents or guardians,
either.

in rin

[You know who she is? The attractive blonde two doors down?] Yes,
well, she gives her kid wine every evening with supper. [And she's
involved in a custody suitl] No one under twenty-one is supposed to
drink in this state—she's in violation of the law! I'd say that makes her
irresponsible as a parent, wouldn't you?

There are a lot more gay people around then you might think. Well,
sexual activities with members of the same sex means you're gay,
right? And it says right here that almost everybody tries that when
they're growing up. [J eez, how do they get up the nerve to ask people
that stuff? Imagine if you had to do that for a living—go around and do
those interviews?] Like I said, we live in very confused times when it
comes to sex.

[Mark, the fool, just went right down and signed up, y'know?] You
have to snort cocaine! Cocaine, man, that stuff is addictive. [Professor
what's-his-name?—down in social science?—is running the thing] [I
don't care if it is a psych experiment]—you could become an addict!
Hey, there's no way I'm signing up. [Mark can do it if he wants]

D [C l' 11.] C

1.

[Don't worry!] [You're always such a worry wort.] It's gonna be a
success. Because, if you're gonna have a really good party, you gotta
limit it—y'know?—to just friends. And you sent invitations, right?
['Course, they're not exactly the same people I would invite—just

, kidding!] And your're good at keeping them mixing, too, [y'know—ha,

hal—that interpersonal stuff]

[Thank God it's Friday!] [I don't have a whole helluva lotta cash.] But,
whenever you really wanna relax, I think it takes less to get you drunk.
An' I tell you, if ever there was a weekend I could use some relaxing,

this is it. [Have you finished that term project?] 'Course, I haven't had

76

much to eat today, either. So it's not gonna take me much to get a good
buzz on.

[He doesn't speak their language, you know what I mean?] Like, he
doesn't know what their concerns are. So, no, I don't think he'll win.
[Not much of a chance] If people are gonna vote for you, [you gotta you
gotta know where they're comin from]—you gotta understand their
concerns. Besides, he doesn't act like a winner.

0 IS] 1. 11°10

4. Lots of times people try to follow, y'know, those textbook formulas for
impression management and come off appearing stiff—unreal, you
know? Jimmy does that—[he really believes that stuff, okay?] So,
people usually get that bogus impression of him. [I mean, like,
"Where's this guy commin' from," y'know?] So naturally they don't
like him very much at first.

5. [You hear what I’m sayin' girlfi'iend?] If your gonna forget about it,
you gotta keep telling yourself that it wasn't right for you, otherwise it
lingers. But you don't tell yourself that; [you just hang around, waitin'
for it to fade away] So, of course it lingers! So then you got that bad
attitude with the next dude, like he's gonna disappoint you, too. [You
never just END it!]

6. [Look at this stuff comin' down.] Maybe school'll be closed tomorrow.
Because, this is gonna be an ice storm. [We had one last year,
remember?] [Or maybe that was back home] Anyway, when it's
freezing down below, this here drizzle builds up on whatever it touches,

an' we've had freezing weather for—what?—a week? [Hey, I could use
a day off]

Si [C l' III'lC

7. [Did you see Jerry last night?—Jerry from nextdoor?] He's got those
ikky red spots all over, an' he's tired all the time. Aren't those signs of
AIDS? An' you know he'll sleep with anybody, [really, what a sleaze]
So, I'll bet he's got AIDS. [Not that I'd wish it on anybody]

8. Henry the Eighth probably died of scurvy. [Do they still call it scurvy?]
We know because, he had wild mood swings and a collapsed nose—
typical symptoms of scurvy. Also, [like most of the upper-class of Tudor
England,] he treated fruits and vegetables as unfit to eat. [Imagine a
king dying of malnutrition!]

9. If a person thinks that all who disagree with him are all equally wrong,
he's what they call a dogmatic personality. That's just what you get
with racists. [You know the type?] Another thing is the way they feel
an overwhelming need to spread the word. [Mostly I think that's to
convince themselves] So, I would say that racists are mostly
dogmatics, yes.

5' CS] l' '11” Q

10. [Yeah, right, he may be cute, but,] face it, he's a jerk, if you ask me If
all a guy does is talk about himself and his sexual prowess, that makes
him a jerk in my book. [I can't stand when they do that!] And who do

78

we know talks that stuff? Like ALL the time. You know who! So, I'd
say you better dump him quick. [Just be brave!]

11. I know they'll let us spend the night together. [I don't think we have a
thing to worry about,] because they think of you as perfect for me. I told
them all about you, about how ambitious you are—[W ell, you are!]—and
how you treat me right. That's what makes a perfect husband in my
parents eyes.

12. [They announced this here scholarship, y'know, for star students, I
guess, an' Jeff says he's gonna apply.] I don't think he's exactly what
you'd call a star student. [I just don't see it.] You gotta study and it
helps if you do your own work. And he never cracks a book and he's
always asking to see other people's homeworks. So there ain't no way
he'll get that scholarship.

Ain ri i

13. [No, he lives in Colorado, not Arizona] Anyway, he hasn't called me in
four days. Now, when people are in love they want to communicate,
right? And, really, when he calls he never talks about us much, [he
just talks about his job and school an' like that] [I mean who cares?]
So, him sayin' he loves me, an' all? I don't buy it. [It's just talk.]

14. This is a democracy—everybody's supposed to be part of the decision
making process. [How come only one person gets to decide what our
grade is?] So, the whole class should should do the evaluation. [Think
of the professor as someone whose job it is to just TEACH.] [That makes
sense to me] That's what they did in Russia, I believe, during the
revolution.

15. [See this here in the paper?] The University has no right to say a
fraternity can't sell their calendar on campus, y'know, the swimsuit
thing they tried to sell in M0 Hall? Because this is a free country right?
I mean, freedom of speech? You can't be havin' no censorship in a free
country. [Who they think they are?] And they ain't our parents or
guardians, either.

Ai r' i

16. [You know who she is? The attractive blonde two doors down?] Yes,
well, she gives her kid wine every evening with supper. [And she's
involved in a custody suit!] No one under twenty-one is supposed to
drink in this state—so, she's in violation of the law! I'd say that makes
her irresponsible as a parent, wouldn't you?

17. There are a lot more gay people around then you might think. Well,
because sexual activities with members of the same sex means you're
gay, right? And it says right here that almost everybody tries that when
they're growing up. [J eez, how do they get up the nerve to ask people

18.

79

that stuff? Imagine if you had to do that for a living—go around and do
those interviews?] So I say, we live in very confused times when it
comes to sex.

[Mark, the fool, just went right down and signed up, y'know?] You
have to snort cocaine! Cocaine, man, that stuff is addictive. [Professor
what's-his-name?—down in social science?—is running the thing] [I
don't care if it is a psych experiment]—you could become an addict!
Hey, so there's no way I'm signing up. [Mark can do it if he wants.]

LIST OF REFERENCES

List of References

Aristotle. (1952). Rhetoric. (W. Rhys Roberts, Trans), In R. M.
Hutchins (Ed.), The great begke 9f the fleetem mgrld. (Vol. 9, pp. 585-675).

Chicago: Encyclopaedia Britannica, Inc.

Brown, G. and Yule, G. (1983). W. Cambridge:
Cambridge University Press.

Clark, H. H. and Clark, E. V. (1977). W
W. New York: Harcourt Brace J ovanov1tch

Clark, H. H. and Haviland, S. E. (1974). Psychological processes as

linguistic explanation. In D. Cohen (ed. ), Explaining lingni eti 19
nhenemen na. Washington: Hemisphere Press.

Erickson, J. R. (1978) Research on syllogistic reasoning. In R. Revlin
& R. E. Mayer (eds), W. Washington: Winston, pp. 39-50.

Freedle, R. & Fine, J. (1982). Prose comprehension 1n natural and
experimental settings: The theory1 and its practical implications. In S.
Rosenberg (Ed.,) H n h lin .Hillsdale:
Lawrence Erlbaum,1982.k

Galotti, K. M. (1989). Approaches to studying formal and everyday
reasoning. h l ' l l ' , _1_0_5, 331-351.

Halle, M., Bresnan, J. & Miller, G. (1978). iLngni iiet 9th egg an nd
W. Cambridge: MIT Press.

Hamilton, M. & Burgoon, M. (1984, May). Observers perceptions of
ingratiating discourse. Paper presented at the annual meeting of the
International Communication Association. San Francisco.

Hample, D. (19793). Predicting belief change using a cognitive theory
of argument and evidence. Communication Monographs, 49, 142-146.

Hample, D. (1979b, November). A review of empirical literature on
logical processes. Paper presented at the annual meeting of the Speech
Communication Association.

Hample, D. (1978). Predicting immediate belief change and
adherence to argument claims. Communication Monographs, 45, 219-228.

Hawkins, T., Pea, R. D., Glick, J ., & Scribner, S. (1984). "Merds that
laugh don't like mushrooms": Evidence for deductive reasoning by

preschoolers. Dmlonmentaifimhoiggx. 20. 584-594.

80

81
Infante, D. (1988). W Prospect Heights, IL:

Waveland Press.
Infante, D. (1987). Aggressiveness InJ. C. McCroskey andJ. A. Daly

(Eds). W (pp 157-192)
Newbury Park, CA: Sage Publications.

Jackson, S. A. (1985). What can speech acts do for argumentation?
InJ. R. Cox, M. O. Sillars and G. B. Walker (Eds) AW
91‘- 1"”. ‘oin f f0 A,._01r‘n'o
argnmentatien. Annandale: Speech Communication Association.

Jackson, S. A. &Jacobs, C. S. (1982). Conversational argument: A
discourse analytic approach. In J. R. Cox and C. A. Willard (Eds)

Andya ngee in argumentati en theem and teeeayeh. Carbondale, IL.

Southern Illinois University Press.

Jackson, S. A. & Jacobs, C. S. (1980). Structure of conversational

argument: Pragmatic basis for the enthymeme. Quarterly slenmal ef
529.2211. 5.6. 251-265.

Johnson-Laird, P. N. (1983). MentaLMedele. Cambridge: Cambridge

University Press.

Johnson- Laird, P. N. Oakhill, J., &Bull, D. (1986). Children's

syllogisnc reasoning WWW
38A, 35- 58.

Joseph, H. W. B. (1916). WWW. London: Oxford

University Press.

Lundquist, L. (1986). Towards a procedural analysis of
argumentative operators in texts. In F. H. van Eemeren, R. Grootendorst,

J. A. Blair and C. A. Willard (Eds), Atgnmentatien; Pereneetiyee and
appygachee. Dordrecht: Foris Publications.

Marcus, S. L. (1982). Recall of logical argument lines. Met
Verbal Learning and Veybal Behafler, 21, 549-562.

Markovits, H. (1985). Incorrect conditional reasoning among adults:

Competence or performance? WM, 16, 241-247.

Miller, G. R. (1969). Some factors influencing judgements of the
logical validity of arguments. WM, 55, 276-286.

Mineo, P. (1983). Toward a psycholinguistic perspective on reasoning
and argumentation. Paper presented at the Social Psychology Conference
of the British Psychological Association, Bristol.

82

Mineo, P. (1988). An integration of inference and schema approaches
to discourse coherence. Paper submitted in partial fulfillment of the
requirements for the Ph.D., Michigan State University.

Palermo, D. S. (1978). Eeyghgjeagflangnage. Glenview: Scott,

Foresman and Company.

Petty.R E4 &Cacioppo, J. T. (1981) Attitudesandmnam
. Dubuque, IA: Wm. C. Brown Co.

Reynolds, R. A., & Burgoon, M. (1983). Belief processing, reasoning

and evidence. In R. Bostrom (Ed.), (lemmnnigatien yearbeek 2. Beverly
Hills: Sage Publications.

Rice, M. and Kemper, S. (1984). ledjflguageandmgmtton.

Baltimore: University Park Press.

Rieke. R. & Sillars, M. (1984). Armmentafionandthcjeamn
making preeeee. Glenview: Scott, Foresman and Company.

Slobin, D. I. (1979). Psygheljnglnetiee. Glenview: Scott, Foresman
and Company.

Steinfatt, T. M. Miller, G. R., & Bettinghaus, E. P. (1971). The
concept of logical ambiguity and judgements of syllogistic validity.
(29mmnni eati 19;; Men negyhap e, 41, 317- 328.

Stenning, K. (1978). Anaphora as an approach to pragmatics. In M.
Halle, J. Bresnan, and G. A. Miller (Eds), Lingnietie theery and
neygholegicalfiality. Cambridge: The MIT Press.

Toulmin, S. (1958). The neee Qt aygnment. Cambridge: Cambridge

University Press.

Toulmin, S., Rieke, R. & Janik, A. (1984). W
reaeening. New York: Macmillan Publishing Co., Inc.

van Eemeren, F. H. and Grootendorst, R. (1982a). Unexpected

pggmisses: Part1 JonmalnflthcAmeficarLEorensieAssofiafion 12. 97-

van Eemeren, F. H. and Grootendorst, R. (1982b). Unexpressed

Eggmisses Part II JoumaLofiheArneficasEmsiLAssndation .12. 215-

van Eemeren, F. H. and Grootendorst, R. (1984). Speecheetem
WW Dordrecht: Foris Publications.

83

van Eemeren, F. H, Grootendorst, R. and Kruiger, T. (1987).
MW. Providence: Foris Publications, Inc.

van Eemeren, F. H., Grootendorst, R. and Meuffels, B. (1985).
Gedifferentieerde replicaties van identificatie- onderzoek. __'_j_deg_h1jfl;_yoey
taa lbeheereing, 1, 241- 257. (Unpublished translation provided by Laure

Galjart, graduate student of the Department of Speech Communication,
University of Amsterdam.)

van Eemeren, F. H. and Kruiger, R. (1986). Identifying argument
schemes. In F. H. van Eemeren, R. Grootendorst, J. A. Blair and C. A.

Willard (Eds.,) WWW. Dordrecht:
Foris Publications.

Warren, W.H., Nicholas, D.W. & Trabasso, T. Event chains and
inferences in understanding narratives. In R. Freedle (Ed.), N331
Direetiene in Dieeenree Pyeeeeeing. Norwood, N.J.: Ablex, 1979.