Human Reasoning and Cognitive Science

Keith Stenning and Michiel van Lambalgen

A Bradford Book
The MIT Press
Cambridge, Massachusetts
London, England
2008
c Massachusetts Institute of Technology

All rights reserved. No part of this book may be reproduced in any form by any
electronic or mechanical means (including photocopying, recording, or infor-
mation storage and retrieval) without permission in writing from the publisher.

MIT Press books may be purchased at special quantity discounts for business or
sales promotional use. For information, please email special [email protected]
.edu or write to Special Sales Department, The MIT Press, 55 Hayward Street,
Cambridge, MA 02142.

This book was set in latex by the authors. Printed and bound in the United
States of America.

Library of Congress Cataloging-in-Publication Data

Stenning, Keith.
Human reasoning and cognitive science / Keith Stenning and Michiel van
Lambalgen.
p. cm. – (A Bradford book)
Includes bibliographical references and index.
ISBN 978-0-262-19583-6 (hardcover : alk. paper) 1. Cognitive science. 2.
Reasoning. 3. Logic. I. Lambalgen, Michiel van, 1954- II. Title.
BF311.S67773 2008
153.4–dc22
2007046686

10 9 8 7 6 5 4 3 2 1
1 Introduction: Logic and Psychology

The purpose of this book is twofold. Our first aim is to see to what extent the
psychology of reasoning and logic (more generally, semantics) are relevant to
each other. After all, the psychology of reasoning and logic are in a sense about
the same subject, even though in the past century a rift has opened up between
them. Very superficially speaking, logic appears to be normative, whereas the
psychology of reasoning is descriptive and concerned with processing. The first
question then is: what is the relation between these two fields of inquiry?
The psychology of reasoning as a field currently adopts a particular view of
this relation: we propose a quite different one. The book therefore should be
relevant to students of logic who are interested in applications of logic to cogni-
tion. But the book should also be relevant to any psychologist who is interested
in reasoning or communication, or any other cognitive capacity where a cog-
nitive account has to be founded on an informational analysis of a cognitive
capacity. These two groups come to the topic with very different methodolog-
ical equipment. The logic student interested in cognition comes with an un-
derstanding of the level of abstraction that modern logical theories operate at,
but with possibly sparse knowledge of psychological observations of reasoning.
Students of psychology come with knowledge of the experimental literatures,
but those literatures are strongly formed by a different conception of logic – a
conception, current in the nineteenth century,1 in which logic is thought of as a
mechanism for reasoning, and a universal, normatively valid mechanism at that.
This presents us with an educational dilemma. The logical analyses presented
here are couched at a level which is intended to be comprehensible to nonlo-
gicians with sufficient patience to digest logical formulas.2 From experience,
the problems encountered are not so much problems about the technicalities of
the systems (which are often not the main point here) but background assump-
tions about what logic is about and what such systems do and do not attempt
to provide. So our message to the psychology student venturing here would

1. Although already at that time more refined conceptions existed; compare section 1.3 on Husserl.
2. Chapter 2 does duty as an introduction to those aspects of logic most important for our purposes.
4 1 Introduction: Logic and Psychology

be that we promise to show that these analyses can make a real difference to
how empirical investigations are designed, so venturing is well worthwhile.
But understanding requires that many routine assumptions about logic are left
at the door. Modern logical theories provide a conceptual and mathematical
framework for analyzing information systems such as people’s reasoning and
communication. They do not settle mechanisms or processes of reasoning, but
without their conceptualization, it is impossible to know what are empirical and
what are conceptual questions.
To the student venturing from logic our message is that we obviously cannot
provide more than the very bare outlines of a few empirical results, along with
some pointers to the literature. So we need to warn against assuming that the
empirical phenomena are as simple or separable as they are bound to appear.
The immense contribution that psychology has made to understanding the mind
largely consists of bodies of empirical knowledge about what phenomena are
replicable under what range of conditions. Much of this knowledge is implicit
in the literature.
This educational dilemma leads naturally to our second wider aim, to dis-
cuss some of the theories offered in the literature from the point of view of
the philosophy and methodology of science. For instance, both mental models
theory and evolutionary psychology, which take their starting points in obser-
vations about the psychology of reasoning, have become hugely popular ex-
planatory paradigms in psychology. We will see that the experiments claimed
to support these theories are marred by conceptual confusions and attendant
methodological errors, and that the theories themselves show little awareness
of the subtleties of logic. Part of our purpose is therefore to propose a differ-
ent methodology for this field, which takes Marr’s idea of “levels of analysis”
seriously.

1.1 Forms of Rationality

Traditionally, rationality is taken to be a defining characteristic of human na-

ture: “man is a rational animal,” apparently capable of deliberate thought, plan-
ning, problemsolving, scientific theorizing and prediction, moral reasoning, and
so forth. If we ask what “rational” means here, we can read such things as: “In
rational discourse one strives to arrive at justified true belief,” a definition of ra-
tionality from an era oriented toward theory. Our more pragmatically oriented
age has extended this concept of rationality to actions. For instance, in the MIT
Encyclopedia of Cognitive Science, “rational agency” is defined as a coherence
requirement:

[T]he agent must have a means-end competence to fit its actions or decisions, according
to its beliefs or knowledge representations, to its desires or goal-structure.
1.1 Forms of Rationality 5

Without such coherence there is no agent. The onus here is on the term fit
which seems to have a logical component. If an action is performed which is
not part of a plan derived to achieve a given goal, there is no fit. In this sense
checking my horoscope before mounting the bike to go to work is not rational,
and neither is first puncturing the tires.
Philosophy, then, studies the question: are there optimal rules for conducting
such activities? Various logics, scientific methodology, heuristics, probability,
decision theory, all have claims to normative status here, where normativity
means that everybody should obey the rules of these systems in all circum-
stances. As a consequence, there exists an absolute distinction between valid
arguments and fallacies. Judged by these standards, human reasoning in the
laboratory is very poor indeed (as shown by the seminal experiments of Wason
[295] for logic and Kahneman and Tversky [150] for probability), and it has
therefore been said that humans are actually not rational in the sense defined
above.
It is usually assumed that the results obtained in the psychology of reasoning
tell us something about the rationality, or rather the absence thereof, of hu-
man reasoning. The following extended quotation from Peter Wason, one of
the founding fathers of the field whose “selection task” will serve as our en-
trypoint below, exemplifies this attitude to perfection. He writes, concluding
an overview of his selection task paradigm for The Oxford Companion to the
Mind,
Our basic paradigm has the enormous advantage of being artificial and novel; in these
studies we are not interested in everyday thought, but in the kind of thinking which oc-
curs when there is minimal meaning in the things around us. On a much smaller scale,
what do our students’ remarks remind us of in real life? They are like saying “Of course,
the earth is flat,” “Of course, we are descended from Adam and Eve,” “Of course, space
has nothing to do with time.” The old ways of seeing things now look like absurd preju-
dices, but our highly intelligent student volunteers display analogous miniature prejudices
when their premature conclusions are challenged by the facts. As Kuhn has shown, old
paradigms do not die in the face of a few counterexamples. In the same way, our volun-
teers do not often accommodate their thought to new observations, even those governed
by logical necessity, in a deceptive problem situation. They will frequently deny the facts,
or contradict themselves, rather than shift their frame of reference.
Other treatments and interpretations of problem solving could have been cited. For in-
stance, most problems studied by psychologists create a sense of perplexity rather than
a specious answer. But the present interpretation, in terms of the development of dogma
and its resistance to truth, reveals the interest and excitement generated by research in
this area. (Wason [300,p. 644])

What lies behind remarks such as Wason’s is the view that reasoning, whether
logical or probabilistic, can be judged to be rational if certain reasoning rules
from a fixed, given set are followed. If these rules are not followed, dire conse-
quences may result. A good example of this attitude is furnished by Stanovich’s
book Who is Rational? [254]. The following quotation gives some idea of the
6 1 Introduction: Logic and Psychology

passions that infuse this approach. Stanovich considers irrationality to lead to

the occurrence of
wars, economic busts, technological accidents, pyramid sales schemes, telemarketing
fraud, religious fanaticism, psychic scams, environmental degradation, broken marriages,
and savings and loan scandals [254,p. 9]

and believes teaching good reasoning, that is, normatively correct rules, will go
some way toward improving this distressing situation.
Stanovich’s discussion of rules governing reasoning introduces a distinction
between normative, descriptive, and prescriptive rules. We give brief charac-
terizations of the three kinds, followed by representative examples.
• Normative rules: reasoning as it should be, ideally
– Modus tollens: ¬q, p → q/¬p,
P (S|D)P (D)
– Bayes’ theorem: P (D | S) = P (S) .

• Descriptive rules: reasoning as it is actually practiced

– Many people do not endorse modus tollens and believe that from ¬q, p →
q nothing can be derived.
– In doing probabilistic calculations of the probability of a disease given a
cluster of symptoms, even experts sometimes neglect the “base rate” and
put P (D | S) = P (S | D).
• Prescriptive3 rules: these are norms that result from taking into account
our bounded rationality, i.e., computational limitations (due to the computa-
tional complexity of classical logic, and the even higher complexity of prob-
ability theory) and storage limitations (the impossibility of simultaneously
representing all factors relevant to a computation, say, of a plan to achieve a
given goal).
– The classically invalid principle ¬q, p ∧ r → q/¬p ∧ ¬r is correct ac-
cording to closed–world reasoning, which is computationally much less
complex than classical propositional logic, and ameliorates storage prob-
lems.
– Chater and Oaksford’s “heuristic rules” for solving syllogisms. [36]
In terms of these three kinds of rules, Stanovich then distinguishes the following
positions on the relationship between reasoning and rationality [254,pp.4–9]:
• Panglossian. Human reasoning competence and performance is actually nor-
matively correct. What appears to be incorrect reasoning can be explained
3. The term is not very apt, but we will stick to Stanovich’s terminology.
1.1 Forms of Rationality 7

by such maneuvers as different task construal, a different interpretation of

logical terms, etc. (A famous defense of his point of view can be found
in Henlé [122].) As a consequence, no education in “critical thinking” is
necessary.

• Apologist. Actual human performance follows prescriptive rules, but the

latter are in general (and necessarily) subnormal, because of the heavy com-
putational demands of normatively correct reasoning. This point of view was
defended by Oaksford and Chater [205, 36]. As a consequence, education in
“critical thinking” is unlikely to be helpful.

• Meliorist. Actual human reasoning falls short of prescriptive standards,

which are themselves subnormal; there is therefore much room for improve-
ment by suitable education (Stanovich’s own position).

• Eliminativist.4 Reasoning rarely happens in real life, and mainly in insti-

tutional contexts such as schools. By contrast, true rationality is adaptive-
ness: we have developed “fast and frugal algorithms” which allow us to
take quick decisions which are optimal given constraints of time and energy.
This position is defended by evolutionary psychologists such as Cosmides
and Tooby[47, 48] and, in more constructive detail, by Gigerenzer [95].

It will be helpful for the reader if we situate our own position with respect to this
scheme. We are definitely not in the eliminativist camp, since we take the view
that reasoning is everywhere, most prominently in discourse comprehension.
This prime example is often overlooked because of the association of reason-
ing with conscious processing, but this association is wrong: some reasoning
is automatic.5 The same example leads us to think that human reasoning may
not be so flawed after all, since it operates rather competently in this domain.
We are not Panglossians either, although we emphasize that interpretation is
of paramount importance in reasoning. But even if interpretation is important,
and interpretations may differ, people may reason in ways which are inconsis-
tent with their chosen interpretation. From a methodological point of view this
means that if one uses a particular interpretation to explain performance, one
must have evidence for the interpretation which is independent of the perfor-
mance. The apologist and meliorist positions introduce the distinction between
normative and prescriptive rules. Here it becomes clear that Stanovich’s scheme
is predicated on the assumption that reasoning is about following rules from a
fixed, given set, say classical logic, rules which should apply always and every-
where. For if there is no given set of rules which constitutes the norm, and the
norm is instead relative to a “domain,” then the domain may well include the

4. Actually, this category does not occur in [254], but we have added it due to its increased prominence.
5. Chapter 5 has more on this.
8 1 Introduction: Logic and Psychology

cognitive constraints that gave rise to the notion of prescriptive rules, thus pro-
moting the latter to the rank of norm. This is what we will argue for in several
places in the book, in particular in chapter 11.
In the next section we briefly look at the role logic once played in cognitive
science, and the reasons for its demise.

1.2 How Logic and Cognition Got Divorced

The cognitive sciences really got off the ground after they adopted the information-
processing metaphor (Craik [52]):
1. Cognitive explanations must refer to models, conceived of as representa-
tional mechanisms
2. which function “in the same way” as the phenomena being represented
3. and which are capable of generating behavior and thoughts of various kinds.
The role of logic in this scheme was twofold: on the one hand as a formal, sym-
bolic, representation language (which is very expressive!), on the other hand as
an inference mechanism generating behavior and thoughts. It was furthermore
believed that these inference mechanisms are continuous with overt reasoning;
that is, the same processes can be applied both reflectively and automatically.
An extreme form of this attitude is of course Piaget’s “logicism” [216], which
maintains that the acquisition of formal-deductive operations is the crown of
cognitive development. Piaget did the first studies to show that preschool chil-
dren do not yet master simple classical predicate logic; but he also assumed
that everyone gets there in the end. This proved to be the Achilles heel of
this form of logicism. Indeed, Wason’s selection task was inter alia directed
against this assumption, and its apparent outcome – a striking deviation from
classical logical reasoning – seemingly undermined the role of logic as an infer-
ence mechanism. A further criticism concerned the alleged slowness of logical
inference mechanisms, especially when search is involved, for example when
backtracking from a given goal. Thus, Newell and Simon style “production sys-
tems”[199], of which Anderson and Lebiere’s ACT-R [2], is the most famous
example, keep only the inference rule of modus ponens, allowing fast forward
processing, at the cost of considerable complications elsewhere.
Lastly, the advent of neural network theory brought to the fore criticisms of
the symbolic representational format given by logic: it would be tied to brittle,
all-or-none representations, uncharacteristic of actual cognitive representations
with their inherently fuzzy boundaries. As a further consequence of this brit-
tleness, learning symbolic representations would be unrealistically hard. As a
result, from the position of being absolutely central in the cognitive revolution,
which was founded on conceptions of reasoning, computation, and the analysis
1.3 Two Philosophers on the Certainty of Logic: Frege and Husserl 9

of language, the psychology of deduction has gone to being the deadbeat of cog-
nitive psychology, pursued in a ghetto, surrounded by widespread skepticism as
to whether human reasoning really happens outside the academy. “Isn’t what
we really do decision?” we increasingly often hear. Many eminent psychology
departments do not teach courses on reasoning. Imagine such a psychology de-
partment (or indeed any psychology department) not teaching any courses on
perception. Even where they do teach reasoning they are more likely to be fo-
cused on analogical reasoning, thought of as a kind of reasoning at the opposite
end from deduction on some dimension of certainty.
We will argue that logic and reasoning have ended up in this ghetto because of
a series of unwarranted assumptions. One of the tasks of this book is to examine
these assumptions, and show that they do not bear scrutiny. As a prelude, we
consider the vexed issue of the normative status of logic, through some of the
history of present day conceptualizations.

1.3 Two Philosophers on the Certainty of Logic: Frege and Husserl

Famously, Aristotle provided the first rules for reasoning with quantifiers of the
form “All A are B,” “Some A are B,” “No A are B” and “Some A are not
B,” starting centuries of work on how to provide principled explanations for
the validity of some syllogisms, and the invalidity of others. This search for an
explanation turned to the notion of validity itself (überhaupt, we are tempted to
say), and Kant opined in the Critique of Pure Reason that logical laws constitute
the very fabric of thought: thinking which does not proceed according to these
laws is not properly thinking.6
In the nineteenth century, this “transcendental” doctrine of logic was wa-
tered down to a naturalistic version called psychologism, which holds that all of
thinking and knowledge are psychological phenomena and that therefore log-
ical laws are psychological laws. To take an example from John Stuart Mill,
the law of noncontradiction ¬(A ∧ ¬A) represents the impossibility of thinking
contradictory thoughts at the same time. Thus, normative and descriptive rules
coincide. What came after, a strong emphasis on normativity, can to a large
extent be seen as a reaction to this view. Gottlob Frege was the driving force
behind the reaction, and his views still exert their influence on the textbooks.

6. It is impossible to do justice here to Kant’s thinking on logic. It is still common to think of Kant’s logic
as primitive, and its role in the Critique of Pure Reason as an instance of Kant’s architectonic mania. This
is very far from the truth, and Kant’s thinking on, for instance, logical consequence remains relevant to this
day. In fact, our concluding chapter 11 has many affinities with Kant, although it would require another
book to explain why. Béatrice Longuenesse’s Kant and the Capacity to Judge [175] is an excellent guide to
the wider significance of Kant’s logic. The reader may also consult Patricia Kitcher’s Kant’s Transcendental
Psychology [159] for an exposition of Kant’s relevance to cognitive science. Kitcher’s remarks on the simi-
larities between Kant’s first Critique and Marr’s program in cognitive science [183] have influenced chapter
11. We thank Theodora Achourioti for pointing out these connections.
10 1 Introduction: Logic and Psychology

1.3.1 Frege’s Idealism in Logic

Frege did not hesitate to point out the weak empirical basis of psychologism: is
it really true that we cannot simultaneously think contradictory thoughts? Wish
it were so! His chief argument, however, was theoretical, and consisted of two
main reservations about a naturalistic treatment of logic (and mathematics):
1. Psychologism makes logic pertain to ideas only, and as a consequence it
lacks resources to explain why logic is applicable to the real world.
2. Logical and mathematical knowledge are objective, and this objectivity can-
not be safeguarded if logical laws are properties of individual minds.
We now present a few extracts from Frege’s writings to illustrate his views
on psychologism.
As regards 1 we read:
Psychological treatments of logic . . . lead then necessarily to psychological idealism.
Since all knowledge is judgmental, every bridge to the objective is now broken off. (G.
Frege, Nachgelassene Schriften; see [85])

Neither logic nor mathematics has the task of investigating minds and the contents of
consciousness whose bearer is an individual person. (G. Frege, Kleine Schriften; see
[85])

The logicians . . . are too much caught up in psychology . . . Logic is in no way a part of
psychology. The Pythagorean theorem expresses the same thought for all men, while each
person has its own representations, feelings and resolutions that are different from those
of every other person. Thoughts are not psychic structures, and thinking is not an inner
producing and forming, but an apprehension of thoughts which are already objectively
given. (G. Frege, letter to Husserl; see 6, p. 113 of [135])

The last sentence is especially interesting: if “thinking is not an inner producing

and forming,” cognitive science has no business investigating thinking, logical
reasoning in particular. What the psychologist finds interesting in reasoning
is precisely what steps the mind executes in drawing an inference, i.e., in the
process more than the result. The quotation just given suggests that logic itself
has little to contribute to this inquiry, and indeed psychologists have generally
heeded Frege’s message, either by designing logics which are supposedly cog-
nitively more relevant (e.g., Johnson-Laird’s ‘mental models’ [145]. See Chap-
ter 10, especially 10.6.3), or by ignoring the contributions that formal logic
can make to theories of processing. This is a pity, since, as will be shown in
the body of the book, the technical apparatus of logic has much to offer to the
psychologist.
Here is an excerpt relevant to 2:
If we could grasp nothing but what is in ourselves, then a [genuine] conflict of opinions,
[as well as] a reciprocity of understanding, would be impossible, since there would be
1.3 Two Philosophers on the Certainty of Logic: Frege and Husserl 11

no common ground, and no idea in the psychological sense can be such a ground. There
would be no logic that can be appealed to as an arbiter in the conflict of opinions. (G.
Frege, Grundgesetze der Arithmetik; the relevant part is reprinted in [85])

From the last quotation we gather that it is apparently highly desirable that
logic “can be appealed to as an arbiter in the conflict of opinions,” and that
therefore there must be a single, objectively valid logic. The second quotation
(from the letter to Husserl) provides a reason to believe there is one: logic is
as it were the physics of the realm of thought, since it studies the structure of
the “objectively given” thoughts. Psychologism is a threat to this normative
character of logic, and since a logic worthy of the name must give rise to norms
for thinking, psychologism is not a possible theory of logical validity. But,
as we have seen, Frege must invoke an objectively given realm of thought to
buttress the normative pretensions of logic, and this assumption seems hard to
justify. However, if one is skeptical about this objective realm of thought, the
specter of relativism rises again. At first sight, the normativity of logic seems
to be bound up with the uniqueness of logic; and what better way to safeguard
that uniqueness than by positing some underlying reality which the logical laws
describe?
This is indeed a serious problem, and to solve it requires rethinking the sense
in which logic can be considered to be normative. In a nutshell, our answer
will be that norms apply to instances of reasoning only after the interpretation
of the (logical and nonlogical) expressions in the argument has been fixed, and,
furthermore, that there are in general multiple natural options for such inter-
pretations, even for interpreting the logical expressions. Thus, the reasoning
process inevitably involves also steps aimed at fixing an interpretation; once
this has been achieved, the norms governing logical reasoning are also deter-
mined. It will not have escaped the reader’s attention that the view of reasoning
just outlined7 is very different from the one implicitly assumed by the standard
paradigms in the psychology of reasoning. As this book goes to press, a spe-
cial issue on logical views of psychologism covering a range of contemporary
positions is published [171].
One aim of this book is to present a view of reasoning as consisting of rea-
soning to and reasoning from an interpretation, and to apply this view to ex-
perimental studies on reasoning. In philosophy, our precursor here is Husserl,
who is playing Aristotle (the metaphysician, not the logician) to Frege’s Plato.
Husserl’s views on logic never made the logic textbooks (at least explicitly), but
we nonetheless believe that one can find in him the germs of a semantic con-
ception of logic, which comes much closer than Frege’s to how logic functions
in actual reasoning.
7. The outline is very rough indeed. For instance, the phrase “once this has been achieved” suggests that
the two stages are successive. As we will see in the experimental chapters, however, it is much closer to the
truth to view these stages as interactive.
12 1 Introduction: Logic and Psychology

1.3.2 Husserl as a Forerunner of Semantics

It was Frege who converted Husserl to antipsychologism. When we read
Husserl’s criticism of psychologism in Logische Untersuchungen [134], we at
first seem to be on familiar ground. If logical laws were empirical laws about
psychological events, they would have to be approximative and provisional,
like all empirical laws. But logical laws are exact and unassailable, hence they
cannot be empirical.8 Psychologism about logical laws also leads to skeptical
relativism: it is in principle possible that different people reason according to
different logical laws,9 so that what is true for one person may not be true for
another – truth, however, is absolute, not indexed to a person.
So far these arguments are question begging: we may have a strong desire
for logical laws to be exact and unassailable and objective, but we need a justi-
fication for assuming that they are. In trying to provide one, Husserl develops a
strikingly modern view of logic, and one that is much more conducive to play-
ing a role in cognitive science than Frege’s. Husserl’s Logische Untersuchungen
brings the important innovation that logic must be viewed, not as a normative,
but as a theoretical, or as we would now say, mathematical, discipline. Logic
as a theoretical discipline is concerned with “truth,” “judgment,” and similar
concepts. Husserl grounds the normative status of logic via a combination of
the theoretical statement “only such and such arguments preserve truth” and
the normative statement “truth is good,” to conclude: “only such and such argu-
ments are good.” Splitting off the normative from the mathematical component
of logic is potentially beneficial, since it focuses attention on what exactly jus-
tifies the normative statement “truth is good,” and thus opens up space for a
relativized version such as “(this kind of) truth is good (for that purpose).”
In slightly more detail,10 Husserl introduces an essentially modern division
of logic as concerned with

1. “the pure forms of judgments” (i.e., the syntax of a formal language, but
here implying a Kantian delineation of what can be said at all);

2. “the formal categories of meaning” (i.e., the semantic study of concepts such
as “variable,” “reference,” “truth,” “proposition,” “consequence”11 );

8. Husserl remarks correctly that even if logical laws are considered empirical, psychologism is under the
obligation to explain how we can acquire them, and that no account of how logical laws are learned has been
forthcoming.
9. In modern times this is occasionally cheerfully accepted. The logician Dov Gabbay once said in an
interview: “Everybody his own logic!”
10. Here we are much indebted to David Bell’s Husserl [17], in particular pp. 85 –100. The quotes from
Husserl are taken from Bell’s monograph.
11. It is of some interest to observe here that for Frege, semantics, although intuitively given, was not
a proper field of scientific study, since it involves stepping outside the system which is given a semantic
interpretation. See also footnote 12. We agree with Husserl that it is both possible and necessary to reflect
on semantic interpretation.
1.3 Two Philosophers on the Certainty of Logic: Frege and Husserl 13

3. “the formal categories of objects” – that is, what is known as “formal on-
tology,” the study of such concepts as “object,” “state of affairs,” “contin-
uum,” “moment.” This can be read as saying that part of logic must be
the characterization of the structures on which the chosen formal language
is interpreted; the next quotation calls these structures “possible fields of
knowledge”:
The objective correlate of the concept of a possible theory, determined exclusively in
terms of its form, is the concept of a possible field of knowledge over which a the-
ory of this form will preside. [This field] is characterized by the fact that its objects
are capable of certain relations that fall under certain basic laws of such-and-such a
determinate form . . . the objects remain entirely indeterminate as regards their mat-
ter. . . ([17,p. 90-1])

4. Lastly, rational thinking also involves systematization, and therefore pure

logic must also comprise a study of formal theories, not only of propositions
and their inferential relationships; in Husserl’s words
The earlier level of logic had taken for its theme the pure forms of all significant
formations that can occur within a science. Now however, judgment systems in their
entirety become the theme (Formale und transzendentale Logik [17,p. 90-1]).

Modern logic has followed this last injunction, and studies what is known as
“metaproperties” of a logical system such as consistency, the impossibility of
deriving a contradiction in the system. Among the most important metaproper-
ties are metatheorems of the form “only such-and-such argument patterns pre-
serve truth,” which depend on a preliminary characterization of the notion of
truth in the “possible field of knowledge” studied. Normativity comes in only
via a principle of the form “in this particular field of knowledge, truth of such-
and-such a form is good, therefore only such-and-such arguments are good.”
This means that logical laws are unassailable in the sense that they are mathe-
matical consequences of the structure of the domain studied, but by the same
token these laws are relative to that domain. The reader will see in the body of
the book that this view of logic, as not providing absolutely valid norms but as
giving norms valid relative to a particular domain, sheds new light on results
in the psychology of reasoning which have traditionally been taken to show the
incompatibility of logic and actual reasoning.12

12. The preceding paragraphs are not intended as an exegesis of Husserl’s thought; our intention is only
to identify some strands in Husserl which we consider to be fruitful for thinking about logic. The contrast
drawn here between Frege and Husserl is a particular case of the more general distinction, first proposed by
Jean van Heijenoort [281], between “logic as a universal language” and “logic as a calculus.” On the former
conception of logic, whose main champion is Frege, logic is concerned with a single universe of discourse,
and the semantic relation between logical language and that universe is ineffable. On the latter conception
(which ultimately gave rise to the modern “model–theoretic logics” [15]), there are many possible universes
of discourse, logical languages are reinterpretable to fit these universes, and semantics is a legitimate object
of scientific study. Van Heijenoort’s contrast has been called “a fundamental opposition in twentieth century
philosophy” by Hintikka [125] and has been applied to Frege and Husserl in Kusch [165].
14 1 Introduction: Logic and Psychology

For the mathematically inclined reader we include an example of Husserl’s

views as applied to the domain of arithmetic. If this domain is given a clas-
sical, Platonistic interpretation, that is, as concerned with objects which exist
independently of the human mind, then the following is not a logical law:
(1) (†) if A or B is provable (in system S), then A is provable (in system
S) or B is provable (in system S)
because on the one hand Gödel’s incompleteness theorem has shown that there
is a sentence A such that neither A nor ¬A is provable in classical arithmetic,
whereas on the other hand the “law of excluded middle” A ∨ ¬A is a logical
law in classical arithmetic. If, however, the domain of arithmetic is conceptu-
alized as being about particular mental constructions, as mathematicians of the
intuitionistic persuasion claim, then it is a mathematical fact that (†) is a logical
law (and that therefore the law of excluded middle is not). Normative issues
arise, not at the level of logical laws (e.g., the law of excluded middle), but at
the level of what description to choose for the domain of interest. Changing
one’s logical laws then becomes tantamount to changing the description of the
domain.
Husserl’s view has the value of focusing attention on the relation between
mathematics and empirical phenomena in general, as one source of difficulty
in understanding the relation between logic and human reasoning. The relation
between mathematics and empirical phenomena is problematical in any do-
main, but it may be more problematical in this domain than most. Appreciating
the continuity of these problems, and identifying their source is one way for-
ward. Seeing logic as the mathematics of information systems, of which people
are one kind, is quite a good first approximation to the view we develop here.
This view helps in that it makes clear from the start that one’s choice is never
between “doing psychology” and “doing logic.” Understanding reasoning is
always going to require doing both, simply because science does not proceed
far without mathematical, or at least conceptual apparatus.
So we see history turning circle, though not full circle. Like Mill and Husserl,
we see logic and psychology as very closely linked. Frege rejected this view.
Husserl developed a much more sophisticated view of the relation, which fore-
shadows our own in its emphasis on semantics. Later, Frege’s view of logic
foundered on Russell’s paradoxes which showed that logic couldn’t be univer-
sal and homogeneous. In response logic developed the possibility of explicitly
studying semantics, and still later, developed a multiplicity of logics. Much of
the technical development necessary for studying semantics took place in the
context of the foundations of mathematics which took logic very far from psy-
chology. In mid-twentieth century, Montague reapplied the much transformed
technical apparatus of logical semantics to the descriptive analysis of natural
languages. We now apply the availability of a multiplicity of logics back onto
1.4 What the Reader May Expect 15

the subject matter of discourse and psychology. Of course psychology too has
changed out of all recognition since Mill – the whole apparatus of psychologi-
cal experiment postdates Mill’s view. So our “psychologism” is very different
from Mill’s, but the closeness of psychology and logic is something shared.
Our psychologism clearly requires an account of how logic in its modern guise
as mathematical system is related to psychology in its modern guise as experi-
mental science.

1.4 What the Reader May Expect

The remainder of the book is structured as follows. Chapter 2 is a somewhat un-

orthodox introduction to logic, which tries to break the hold of classical logic
by showing it results from contingent assumptions on syntax and semantics.
Systematic variation of these assumptions gives rise to several logics that have
applications in actual human reasoning. This chapter in particular introduces
closed–world reasoning, a form of reasoning that will be very important in part
II of the book. In chapters 3 and 4 we study the Wason selection task from the
vantage point developed in this introduction and in chapter 2: as a task in which
subjects are mostly struggling to impose an interpretation on the experimental
materials instead of engaging with the materials as the experimenter intends
them to do. Chapter 3 contains many examples of tutorial dialogues with sub-
jects which show what interpretational difficulties they experience, and chapter
4 reports on experiments establishing that alleviating these difficulties by modi-
fying the task instructions leads to a vast increase in correct performance, when
measured against the classical competence model.
The selection task has played a major role in debates on evolutionary psy-
chology, with Cosmides [47] claiming that her results on facilitation of the task
with social contract materials show that the only abilities humans have in logical
reasoning are due to an innate module for “cheater detection.” We believe that
this highly influential point of view is mistaken, for two reasons: a faulty view
of logic, and a faulty view of evolution. chapter 5 considers the influence this
faulty logical paradigm has had on the psychology of reasoning, and chapter 6
is a lengthy discussion of the evolution of human cognition. The latter chapter
introduces a hypothesis that will play an important role in part II: that the origin
of the human ability for logical reasoning must be sought in the planning ca-
pacity, and that closed–world reasoning (which governs planning) is therefore
a very fundamental form of reasoning, stretching across many domains.
Chapter 7 applies this idea to the analysis of the suppression task (Byrne
[28]), which is standardly interpreted as providing evidence for “mental mod-
els” and against “mental rules.” We show that the data from this task can be
explained on the assumption that subjects assimilate the task to a discourse–
processing task, using closed–world reasoning, by presenting a rigorously for-
16 1 Introduction: Logic and Psychology

mal model of subjects’ reasoning in a logical system called “logic program-

ming,” which we consider to be the most appealing form of closed–world rea-
soning. We also present data from tutorial dialogues corroborating our inter-
pretation of what subjects are doing. In chapter 8, it is shown that closed–world
reasoning has a revealing neural implementation and that there need not be an
opposition between logical and connectionist modeling. Chapter 9 applies the
ideas of chapters 7 and 8 to autism. We analyse several tasks on which autistic
people are known to fail, such as the false belief task and the box task, and find
that these tasks have a common logical structure which is identical to that of the
suppression task discussed in chapter 7. This leads to a prediction for autistic
people’s behavior on the suppression task, which has been verified. This latter
result is analysed in terms of the neural implementation developed in chapter
8, which then allows us to make a connection to the genetics of autism. Chap-
ter 10 discusses syllogisms. These are, of course, the first reasoning patterns
studied in psychology, but for us their interest lies in the necessity to apply sub-
stantial interpretational theories from linguistics and philosophy to explain the
data. This explains why syllogisms only occur near the end of the book: the
reader must first be familiar with both interpretation processes of reasoning to
an interpretation and derivational processes of reasoning from the interpretation
imposed.
Lastly, chapter 11 makes explicit our view of logic and its relevance to actual
reasoning. The role of logic is to aid in “going beyond the information given”
when processing information. Just as in visual information processing, mathe-
matical structure (edges etc.) must be imposed upon the retinal array, because
this structure is not literally present in the data, so some logical form must first
be imposed on a problem requiring reasoning before the actual reasoning can
take place.
In the end, we aim to convince the reader that using the formal machinery
of modern logic leads to a much more insightful explanation of existing data,
and a much more promising research agenda for generating further data. If we
succeed, then there are general morals to be drawn about what philosophy of
science is appropriate for the psychology of “higher” cognitive functions.
At present, experimental psychology is much influenced by a Popperian phi-
losophy which sees hypothesis testing as the central activity in science. We
will see that Wason himself was much influenced by this account of science in
explaining his subjects’ responses. But Popper’s account, important as it is, in
the intervening years has been shown to be a very partial account. If science is
hypothesis testing, where do hypotheses come from? Why test this one rather
than that? A great deal of science is exploration and observation which don’t
fall easily under the umbrella of hypothesis testing. Highly developed sciences
such as physics, which are overwhelmingly the cases studied by philosophers
of science, have powerful abstract bodies of theory which guide exploration
1.4 What the Reader May Expect 17

and observation, and through them the selection of hypotheses, when the time
comes to test hypotheses. Psychology lacks such bodies of abstract theory, and
so one sees implicit theories playing important roles in choosing what hypothe-
ses to test. “Surprisingness” of a phenomenon as compared to some implicit
theory of that phenomenon is a crucial quality indicating when hypothesis test-
ing is worth the effort. If Wason’s observations of failures of rationality in the
selection task hadn’t been so counterintuitive, then we (and you) would never
have heard of them. We will expend considerable effort in chapter 3 in mak-
ing explicit Wason’s (implicit) background theory against which the results are
so surprising, and in showing how logic can provide explicit background ab-
stractions which change the way hypotheses can be chosen and experiments
designed.
A corollary of the lack of background abstract theories in psychology is the
use of direct operationalization of abstract concepts in experimental procedures:
data categories are assumed to be very closely related to their theoretical cat-
egories. As a consequence, the data observed are supposed to have a direct
bearing on the theoretical hypotheses. In mature sciences this doesn’t happen.
There are always ‘bridging’ inferences required between observation and the-
ory, and an apparent falsification may direct attention to unwarranted auxiliary
assumptions. Especially in chapter 10, though also throughout, we will illus-
trate how logic can open up space for observation and exploration between data
and theory. Young sciences like psychology require lots of observation and ex-
ploration, so a methodology which opens up space for these activities is vital.
In the study of human cognition, this space between data and theory or hy-
pothesis is particularly broad because of our human capacity for multiple inter-
pretation. Indeed, some resistance to taking interpretation seriously in studying
reasoning comes from the belief that this space is too broad to be bridged. Once
we acknowledge the possibility of full human interpretive capacities, then, so
the argument goes, the possibility of science goes out the window. In partic-
ular it is often felt that the scientific study of reasoning is impossible once it
is allowed that the logical expressions are subject to the possibility of multiple
interpretations. Rejecting the possibility of multiple interpretation because it is
held to make science impossible is truly the logic of the drunk searching be-
neath the lamppost who prefers the illuminated circle to the dark space where
he knows he lost his keys. We take the general human capacity for multiple
interpretation to be as close to fact as it is possible to get in cognitive science,
and prefer to follow it where it leads in choosing our methods of investigation.
In fact, one sees this battle fought repeatedly in each area of human cognition.
It remains under control to some extent in perception, because there the exper-
imenter has “stimulus control” – she can twiddle the display, and observe what
the subject reports seeing.13 But in, for example, memory, the problem has been
13. In fact this is something of an illusion, because the subject brings preexisting knowledge to bear in
18 1 Introduction: Logic and Psychology

recognized ever since Ebbinghaus [68] invented nonsense syllables in order to

eliminate the interpretive effects of subjects’ long-term knowledge which is be-
yond experimental control. In memory, the problem leads to a split between
the Ebbinghaus tradition of laboratory experimentation on abstracted materials,
and the Bartlettian tradition of studying, for example, autobiographical mem-
ory [12]. Both have contributions to make, but both require an understanding
of their distinctive approaches to assimilate those contributions.
The problem of treating “content” or “general knowledge” or “experience
prior to the experiment” is endemic in psychology, and the life of the psycho-
logical researcher is much taken up with getting around the barriers it throws
up. The psychology of reasoning’s adoption of “abstract tasks” can be seen as
following in the Ebbinghausian tradition, and reaction against those tasks as
Bartlettian rejection of that tradition. Logic itself is interpreted by psycholo-
gists as the most extreme form of the Ebbinghausian approach in which content
is banished entirely. However, our argument is that this is no longer true in
modern logic. The default logics we present here are actually interpretable as
modeling the relation between a working memory holding the experimental in-
put materials and a long-term memory holding “general knowledge.” So these
logics present a formalization of “content.” They thus attack psychology’s cen-
tral problem head-on. Of course, they do not offer a model of the long-term
memory of some actual adult human being (nor even an idealized adult hu-
man being at standard temperature and pressure). But they do offer precise
formal models of how large databases of default conditionals can control the
interpretation of richly meaningful input texts. Essentially similar computer ar-
chitectures are the basis for implementing real-world useful databases of gen-
eral knowledge in practical applications. Of course the philosophical problems
of “symbol grounding” remain. But nevertheless, here is the first plausible
head-on approach to the formal modeling of content which offers to reconcile
Ebbinghaus with Bartlett.
It should be evident that these issues are close to the heart of problems about
the relations between the humanities and the sciences, and it is entirely fitting
that they should come up when trying to do scientific research into the nature
of the human mind. It is interesting that representatives of more conservative
approaches to both sciences and the humanities have felt it important to try to
defeat the very possibility of cognitive science’s computational model of the
mind. We hope to explain just why that model is felt to be so threatening,
to defuse some of the concerns arising, and to show that one can avoid both
misplaced scientific reductionism and postmodern hyperrelativism by engaging
in a logically based and experimentally informed study of human interpretive
capacities.

making perceptual interpretations, but it is an illusion persistent enough to allow progress.