Mental Imagery and Human Memory PDF
Mental Imagery and Human Memory PDF
Mental Imagery and Human Memory PDF
JOHN T. E. RICHARDSON
Lecturer in Psyclwlogy, Brunei University
© John T. E. Richardson 1g8o
Softcover reprint of the hardcover 1st edition 1980
This book is sold subject to the standard conditions of the Net Book Agreement
Dedicated to Sonia
Contents
Priface lX
1 Introduction
5 Pictorial Memory
5.1 Memory for Pictures
5.2 Faces: A Special Case?
5·3 Pictures and Words
54 Pictures and Mental Imagery
5·5 Conclusions
VII
Vlll Contents
10 Conclusions 1 43
References
Subject Index
Preface
Over the last eight years, I have carried out a variety of experimental
investigations of the role of mental imagery in human memory. In common with
many other research workers, I have found mental imagery to be a fascinating,
yet extremely elusive phenomenon to study in the psychological laboratory.
Under many circumstances, it is possible to demonstrate substantial empirical
effects which may be attributed to the operation of mental imagery, as I shall
explain in the central chapters of this book.
Be that as it may, our understanding of the significance of mental imagery for
remembering in everyday life is still very limited. The obvious experiments, the
easy, well-controlled studies of remembering under artificial conditions have
now been done; the much more difficult problem of saying anything meaningful
about the cognitive tasks which confront ordinary people in their everyday
activities has yet to be faced. Nevertheless, the considerable amount of
laboratory research which has been carried out to date may be seen as an
important preliminary to the latter activity. In this book, I have tried to give a
perspicuous summary of this research within a reasonable and reasoned
theoretical context.
The experimentation on mental imagery has been heavily influenced by the
behaviourist tradition in experimental psychology. To understand the research
which has been carried out, it is necessary to understand that tradition, and to
assess its conceptual adequacy as a scientific methodology for the investigation of
personal experience. I hope that I have given a fair and balanced account of the
behaviourist tradition, even though I believe its effects upon the development of
cognitive psychology to have been largely detrimental. More generally, I hope
that it will become clear to the reader that advances in human experimental
psychology are likely to result from the combined use of empirical investigation
and of conceptual analysis.
To the extent that this book is able to make any contribution to the
development of our knowledge and understanding of the role of mental imagery
in human memory, it is because I have been able to focus my training and
research experience both in human experimental psychology and in philosophy.
Accordingly, I am pleased to be able to acknowledge my gratitude to those who
have been instrumental in determining such skills as I have in these areas. In
particular, I would like to thank Alan Baddeley for his guidance and example
IX
X Preface
John T. E. Richardson
Department of Psyclwlogy
Brunei University
Uxbridge, Middlesex
April 1979
I
Introduction
Until a few years ago, research on the nature and function of mental imagery was
one of the success stories of modern experimental psychology. A considerable
amount of experimental work was carried out during the rg6os, and this
appeared to implicate mental imagery not only as an empirical phenomenon of
considerable predictive importance, but also at the theoretical level as a major
representational system underlying human cognitive behaviour. Especially in
North America, this research was widely hailed as reflecting a major change in
the direction of experimental psychology, and a reaction against the excesses of
the behaviourist tradition. Indeed, it was said that this change represented a shift
in the scientific paradigm employed by experimental psychologists which made
it possible for them to carry out a rigorous empirical investigation of an aspect of
human experience which they had long neglected (see, for example, the papers
by Holt, 1964; Kessel, 1972; and Neisser, 1972a, 1972b).
However, for the last six or seven years, the literature has manifested a
perceptible disillusionment, and this has been engendered by several different
factors. First, some of the experimental research has not demonstrated clear and
unambiguous effects in accordance with the predictions of the prevalent
theoretical approach. Moreover, some of the effects which were clear and
unambiguous appeared to be open to alternative interpretations. Second, a
variety of methodological procedures have been devised for investigating mental
imagery in a laboratory setting, but they are so many and so diverse that it has
become difficult to integrate the findings within a single framework. Third, the
original success of the theories which identified mental imagery as a distinct
cognitive mode was partly due to the patent inadequacy of the alternatives,
which were based upon purely verbal representations in human memory.
Nevertheless, several researchers began to develop more sophisticated, pro-
positional theories of memory which were not subject to these limitations, and
which seemed to encompass most of the empirical phenomena which had been
explained by the imagery approach. Finally, these theoretical alternatives tend
to be supported not so much by further empirical evidence, as by conceptual
objections to mental imagery as a coherent format for mnemonic representation.
In the late 1970s, then, the interested though sceptical inquirer could perhaps
be forgiven for thinking that research on mental imagery was empirically
unsound, methodologically fragmented, theoretically outdated, and conceptu-
2 Mental Imagery and Human Memory
ally confused. It is certainly true that sections of the literature leave one 'half
distressed and half confused', as Newell (1973) described his own reaction.
Despite this, however, the field has continued to develop and expand at a
remarkable pace. The percentage of publications in psychology which is devoted
to mental imagery has increased in an exponential fashion (Paivio, 1975d), and
in 1977 a Journal of Mental Imagery was established. There is clearly an urgent
need to integrate and organise the research findings which are being produced.
Indeed, this need was identified by Newell in 1973 and has recently been
emphasised by Sheehan (1978). In this book, I hope to go some way towards
satisfying this need. Specifically, my intentions are as follows: to spell out the
major theoretical alternatives in understanding and explaining empirical
findings on mental imagery; to consider the conceptual problems inherent in
previous psychological research; to assess the methodological adequacy of the
experimental procedures which have been employed; and to review the
available empirical evidence in a manner which enables it to be integrated
within a single theoretical framework.
Some remarks are in order concerning the areas of research which are not
covered in this book. First, it will be concerned with mental imagery insofar as it
contributes to human memory. It will not therefore be concerned with forms of
mental imagery other than those which are involved in remembering. An
obvious reason for this is that by far the greatest proportion of experimental
research has been carried out in connection with human memory, and that,
however fragmented this research might be, its links with studies ofother forms of
mental imagery are much more tenuous. An interesting account of after-
imagery, dreaming imagery, hallucinatory imagery, and other forms has been
given by Richardson (1969, chaps 2, 5). Second, this book will be concerned with
human memory insofar as it is illuminated by the study of mental imagery. It will
not therefore be concerned with other approaches to understanding human
memory. Concepts such as primary and secondary memory, and the role of
organisation in memory will be mentioned from time to time, but excellent
discussions of these alternative approaches are readily available, and it would be
redundant for me to discuss them in any detail.
I shall begin by giving a brief review of the attitudes which have been
expressed by experimental psychologists since the beginning of this century
towards the possibility of carrying out a scientific study of mental imagery. This
will mainly involve a discussion of American behaviourism, and of the
procedures adopted by the behaviourists for investigating apparently 'men-
talistic' concepts. Chapter 2 will continue by contrasting the dual coding theory
ofimagery and verbal processes with the common coding theory of propositional
representations. The experimental methods which have been employed in
laboratory research were devised in order to handle the behaviourists' misgivings
concerning the ordinary-language concept of mental imagery. Chapter 3 will
consider the philosophical presuppositions of the classical behaviourist attitude;
it will be argued that these were confused and ill-founded, and that they imposed
Introduction 3
4
Psychological Attitudes to Mental Imagery 5
go back upon the definition: never to use the terms consciousness, mental
states, mind, content, will, imagery, and the like ....
In the psychological revolution, the second phase is just now getting under
way. The first banished thought, imagery, volition, attention, and other such
Psychological Attitudes to Mental Imagery 7
seditious notions. The sedition of one period, however, may be the good sense
of another. These notions relate to a vital problem in the understanding of
man, and it is the task of the second phase to bring them back ....
The behaviorist taboos have been broken, and the mind suddenly seems worth
studying after all. Ideas and images are once again discussed in respectable
journals .... Because of this paradigm shift, contemporary psychologists feel
able to attack crucial and long-neglected problems and have a new array of
concepts with which to do so.
However, Neisser seems to have heavily overstated this point, smce recent
8 Mental Imagery and Human Merrwry
The problem of defining images is not essentially different from the problems
commonly faced by psychologists in defining many of their working concepts
like intelligenee, habit, drive, personality, or cognitive dissonance. The general
approach to definition of such hypothetical constructs or concepts is that of
operationism which, in the words of Hull, tries to 'anchor' such assumed
mechanisms, devices, states, or processes at both ends of an S-R sequence--by
spelling out the procedures and measurements applied to the independent and
dependent variables under study. In practice, the operational procedure
specifies certain conditions under which some alleged phenomenon, process,
or state can be postulated to exist.
However, many psychologists feel that they have to go beyond the specification
ofoperational procedures, because these identify the function of mental imagery,
rather than its nature (cf. Neisser, I 972a). Many suggested definitions emphasise
the relationship between mental imagery and perception. Thus, Neisser (I972b)
claimed that 'a subject is imaging whenever he employs some of the same
cognitive processes that he would use in perceiving, but when the stimulus input
that would normally give rise to such perception is absent'. Unfortunately, this
suggestion is not really helpful, since it does not distinguish between voluntary
imagery and hallucinations. (It also seems to imply that after-images are not
examples of mental imagery, since they are normally produced by a specifiable
'stimulus input'.) Moreover, what is meant by the phrase 'employs some of the
same cognitive processes' is not explained, and so it is difficult to know how in
principle one would verify, from this definition, that someone were imaging.
(The question of the precise relationship between mental imagery and
perception is a difficult one, but not one which need cause undue concern here.
At least three different sorts of suggestion have been made. First, that mental
images are like perceptions. This stems from the classical position that both
imaging and perceiving involve the apprehension of ideas, and that there are no
intrinsic properties which distinguish the two sorts of idea. Second, that mental
images are based upon perceptions. This occurs in the traditional empiricist
theory that mental images are representations produced by perceptual ex-
perience: see Yuille and Catchpole, I 977. Clearly, it cannot explain the creative
Psychological Attitudes to Mental Imagery
capacity of mental imagery. Third, that mental imagery and perception are
functionally overlapping. This idea is contained in Neisser's 'definition' of
mental imagery, and has produced much interesting research on the possibility
of using one sort of process to interfere with the other: see chapters 4, 6, 7 and 8.
Neisser's more recent writings have involved the more complex idea that mental
images are anticipatory phases of perceptual activity: Neisser, I976, pp. I30,
I 7<r- 1. It does seem to be the case that the domain of possible objects of
perception is coextensive with the domain of possible objects of mental imagery.
Nevertheless, Neisser's 'definition' appears to be based upon the unwarranted
conclusion that the processes of apprehending the two sorts of object are
overlapping. The issue of the functional relationship between perception and
mental imagery is to be resolved as much through conceptual analysis as through
empirical investigation.)
It is important to ask whether it is legitimate to try to fill out a hypothetical
construct by suggesting theoretical properties which are not clearly tied to
operational procedures. Paivio (I 975d) has claimed that it is legitimate, and that
no one has really believed that the meaning of any theoretical concept is
exhausted by defining operations: 'The investigator must have some kind of
theoretical definition of a concept in mind or he would have no basis for selecting
'l particular indicator and no motivation to look for one. Pure operationism is
accordingly impossible.' However, it is obvious that operational procedures
constitute the only basis for carrying out an empirical investigation of a
theoretical construct. Thus, any hypothesis which relates to properties of the
theoretical construct which have not been 'unpacked' or translated into
operational procedures will be untestable. It is therefore the duty of the theorist
to ensure that his theoretical constructs are unpacked thoroughly, and are
exhaustively specified by defining operations. This involves making unambigu-
ous predictions concerning the empirical effects (both simple and interactive) of
the various operational procedures upon performance. Paivio's own account of
mental imagery appears to be inadequate in just this respect, since he conceded
that he was unable to make specific predictions concerning possible interactions
among the operational procedures in terms of their effects upon performance (for
example, Paivio, 1971c, p. 502).
Another major assumption is that the two systems are taken to be different in
the way in which their units are organised into higher-order structures. The
imaginal system is assumed to represent information in a synchronous or
spatially parallel manner, so that different components of a complex thing or
scene are simultaneously available. On the other hand, the verbal system is taken
to employ the sequential organisation which is characteristic of linguistic
utterances. Similarly, the imaginal system is assumed to be capable of
transformations along spatial dimensions, such as size, shape and orientation,
whereas the verbal system is taken to allow transformations on a sequential
frame, such as additions, deletions and changes in sequential ordering.
One of the major purposes ofPaivio's research was to argue for the inherent
inadequacy of theories of memory based upon the assumption of a single, verbal
code. Such theories were popular up to the I96os, but research on the role of
mental imagery was generally taken to present grave difficulties for such
accounts. Accordingly, dual coding theory was very successful, and was at first
accepted by most psychologists investigating mental imagery as providing
essentially the correct framework for subsequent research (for example, Bower,
1970a, 1972). However, in the last decade various theories have been put
forward to try to handle the results of this research within a single hypothetical
system, in a manner which was not subject to the limitations of the traditional
verbal coding model.
This approach appears to stem from some ideas expressed by Bower (1972) at a
conference in 1969. Although at that time he generally seems to have accepted
the assumptions of dual coding theory, Bower specifically disagreed with Paivio
over the organisational properties of the two hypothesised systems. On the basis
of his own experiments on the effects of imagery instructions upon memory
performance (see chapter 6), Bower argued that roughly the same general
principles of relational organisation applied to imaginal and verbal represen-
tations. Indeed, he speculated that 'we may have a common generative
grammar that underlies our verbal production of sentences and our imaginal
production of visualized scenes or of hand-drawn pictures'. In a subsequent
paper, Bower (I 970b) made a more radical proposal. He suggested that this
relational generating system 'may be a "conceptual deep structure" ... into
which sentences and perceptual experiences are translated for storage and out of
which either surface sentences, imagery, or drawings may be generated,
depending on the material and the task demands'.
This idea has become an important part of several theories of long-term
memory which assume that information is represented in memory at an abstract
level by networks of propositions (for example, Anderson and Bower, 1973;
Pylyshyn, 1973; Rumelhart, Lindsay, and Norman, 1972; Yuille and Catchpole,
Mental Imagery and Human Merrwry
I 977). These theories were originally devised to handle the results of experiments
on the comprehension and retention of narrative by postulating abstract
semantic representations (propositions) which were related in complex struc-
tures or networks (see chapter g). However, it soon became obvious that it would
be parsimonious to represent perceptual and linguistic information in a common
propositional base which would be neutral with respect to the source of that
information. That is, all cognitive and mnemonic processes should be explained
by a single system of abstract propositional representations. Mental imagery
should not therefore be regarded as implicating a qualitatively distinct form of
mental representation. (Not all propositional theorists have made this step. In
particular, Kintsch has assumed that mental images and propositions have
underlying representations which are structurally distinct and functionally
independent: see Jorgensen and Kintsch, I973; Kintsch, I974·)
Superficially, at least, dual coding theory and common coding theory appear
to be quite different explanatory frameworks, and one might expect that it would
be entirely possible to decide between them on purely empirical grounds. In fact,
the situation is much more complicated, and a considerable amount of discussion
has taken place concerning the appropriate experimental tests of the two
positions. However, propositional theorists have also attempted to attack dual
coding theory on conceptual grounds, by arguing that mental imagery is not a
theoretically adequate form of mental representation. Before discussing the
evidence for each of these theoretical positions, therefore, it is important to
consider these arguments in some detail.
It is clear that Pylyshyn's approach is much less restricted than the standard
behaviourist position: he accepts the existence and importance of mental
imagery, and regards mental imagery both as a possible source of scientific
evidence and as a possible object of scientific investigation. The main question
which he wishes to consider is 'whether the concept of image can be used as a
Psychological Attitudes to Mental Imagery
imagery agree with Pylyshyn that mental images come already interpreted and
organised, and should not be assimilated to 'raw', uninterpreted patterns of
sensory activity. If mental images are taken to be composed of relatively large,
interpreted, perceptual 'chunks', then the amount of storage capacity required
may be much reduced. Second, in any case we simply do not have any accurate
idea of either the storage capacity of the brain or the amount of information
contained in a mental image. There is thus no sound basis for claiming that the
representation of mental images would make unreasonable demands upon the
storage facilities of the human brain (see also Bugelski, 1977). Finally, the
objection appears to apply with equal force to the major alternative theories of
cognitive representation. It is difficult to make any clear comparison between
dual coding theory and propositional theories, since there is no criterion for
deciding which of an indefinite number of sets of propositions best represents the
scene or event depicted by a given mental image. Nevertheless, as far as specific
proposals by the proponents of common coding theory are concerned, vast
numbers of propositions seem to be necessary to represent even relatively simple
configurations of objects. As Anderson ( 1978) concluded: 'The simple fact may
be that there is a great deal of information in an image and any representation of
an image will have to acknowledge this fact.'
One of Pylyshyn's most important arguments is that dual coding theory is
incoherent without the assumption of a third code or representation which is
amodal and propositional in nature:
Similar suggestions have been made by Anderson and Bower (1973, p. 453),
Clark and Chase (1972) and Moscovitch (1973). However, Anderson (1978)
pointed out that Pylyshyn's argument leads to an infinite regress, since it may be
applied with equal force to the problem of translation between either of the two
original representations and the postulated third code. Kosslyn and Pomerantz
(1977) suggested that, however complex the transformation rules might be for
carrying out translations between imaginal and verbal representations, it would
be a mistake to suppose that such processes or routines constituted an additional,
intermediate form of representation. Finally,Janssen (1976, p. 18) argued that,
even if a propositional representation were necessary on a priori grounds, and
even if it could be shown to be structurally isomorphic to any imaginal
representation, this would not prove that the latter was redundant, since the two
representations might have quite different functional properties.
20 Mental Imagery and Human Memory
Anderson's paper contains a formal proof of this latter assertion, and a concrete
Psychological Attitudes to Mental Imagery 21
and that instead they would adopt a form of epistemological realism, according to
which many hypothetical entities are believed to exist, and some may be shown
to exist. The purpose of psychology is to investigate human faculties, and the role
of psychological theories is to specify putative mechanisms which underlie those
faculties. This will involve ascribing properties to those structures which are
causally responsible for the behaviour being studied, in particular the structures
of the human brain. Any adequate theory must therefore postulate hypothetical
entities which must be considered at least as candidates for objective existence
among the physiological structures of the central nervous system. This entails
that physiological data and neuropsychological research will be of immense
value in discriminating among alternative theories. Moreover, since these
theories are offering putative descriptions of the real mechanisms responsible for
behaviour, the truth or falsehood of those descriptions must in principle be
demonstrable. (Further discussion of these points is contained in chapter 3·)
This approach is actually supported by Anderson's account of an experiment
by Patterson and Bradshaw ( 1975) which he himself described as 'particularly
interesting'. This st·_:dy employed a face-recognition task to test the conventional
idea that different representations are used in the two cerebral hemispheres (see
chapter g). Anderson interpreted the results to mean that the two hemispheres
differ in their ability to perform certain types of operation upon a common
representation of visual information. This is exactly the sort of conclusion which
is necessary to specify the functions of the two cerebral hemispheres and the
nature of cognitive representation, but it is the sort of conclusion which can only
be made if neuropsychological evidence is relevant to deciding among different
theories of cognitive representation.
2.7 CONCLUSIONS
vary in how closely they are tied to an operationist approach. At one end of the
spectrum, there is the neobehaviourist position of Bugelski (I 97 I, I 977) who
adopts a rigorous operationism, and who defines mental imagery as a covert
reaction which is 'at no time ... equated with phenomenal or "mental"
experience'. At the other end, there is the position of Alan Richardson (I96g),
who emphasises the phenomenal aspects of mental imagery, but who fails to
specifY an explicit, coherent methodology for experimental research (Bugelski,
I97I). However, Neisser (I972a) has pointed out that Richardson's book is
written in the classical, introspectionist mould, and does not reflect the current
state of research into mental imagery. In between these two extremes, there is the
neomentalism of Paivio (I97Ic, I975d), which is claimed to combine the
introspectionist and behaviourist traditions 'in that it embraces mental pheno-
mena as its subject matter and behavioural approaches as the method of study'.
(It should also be pointed out that this book is concerned almost entirely with
research carried out in English-speaking countries. In European countries,
psychologists have never departed to the same extent from the introspective
method. However, although American and British psychologists are beginning
to consider the implications ofEuropean phenomenological research, it has so far
had little impact on the specific field of mental imagery.)
There are two further, related problems which should be mentioned briefly.
In contemporary research, the theoretical construct of mental imagery is defined
by reference to various operational procedures, of which the three most
important were described earlier in this chapter. However, each of these
procedures involves the use of the term 'imagery' in the instructions given to the
experimental subjects. Investigations of mental imagery as a mnemonic aid
include instructions to the subjects to use mental imagery in their learning; and
research on both stimulus attributes and individual differences includes
questionnaires in which the subjects judge the extent to which various stimuli
evoke mental imagery. It is surely circular to offer an operational definition of a
theoretical term if the specified operations involve the use of that term (cf.
Bugelski, I 97 I). Even if the manoeuvre is not actually circular, it still begs the
question of how the subjects know what the term means, and the question of
whether their interpretation of this term corresponds to the psychologist's.
Presumably, the subjects in an experiment have learned the meaning of the
expression 'mental imagery' in everyday discourse with all its 'mental and
phenomenological aspects'. But in that case the behaviourist's operational
definition of mental imagery is still contaminated by a surreptitious reference to
'private' experience. This semantic point has certain psychological implications.
Earlier in this chapter, it was mentioned that whether mental imagery is a
conscious experience or not is usually taken to be irrelevant to its adequacy as a
theoretical construct. In fact, as Marks (I977) has pointed out, the operational
procedures which are taken to define that construct actually assume a conscious
awareness on the part of the subjects of the mental imagery evoked by the
stimulus material, and of the strategies employed in remembering that material.
Mental Imagery and Human Memory
The second problem concerns the psychologist's motivation for using the tertn
'mental imagery' in the first place. If a term is rigorously defined by certain
operations, and if it is introduced to refer to a hypothetical construct within a
scientific theory, it is not dear why one should employ a term which is already in
common currency with an informal, unscientific meaning. On the contrary, one
might expect this to be a rich source of confusion. Since the operations already
refer to mental imagery, it might have been thought reasonable to use the term in
theorising ('Mental imagery is what people use when told to use mental
imagery'). However, it is more likely that an individual psychologist considered
'mental imagery' to be an appropriate label for his theoretical construct before
he had determined particular operational indicators for that construct (cf.
Paivio, 1975d). Presumably, the label was appropriate because the theoretical
construct bore certain similarities to the everyday concept of mental imagery. In
this case, it is important to ask how the two concepts differ, and whether an
ordinary-language concept can function as a hypothetical construct within a
psychological theory. In order to answer these questions, it is necessary to attain
a clear idea of the concept of mental imagery in ordinary language.
3
A Conceptual Analysis of Mental Imagery
The philosopher Ludwig Wittgenstein (1953, p. 232) once made the following
remarks concerning the state of psychology in the 194os:
The behaviorist, having made a clean sweep of all the rubbish called
consciousness, comes back to you: 'Prove to me,' he says, 'that you have
auditory images, visual images, or any other kinds of disembodied processes.
So far I have only your unverified and unsupported word that you have them.'
Science must have objective evidence to base its theories upon.
But is it really the case that the sort of evidence available to other people
concerning a person's mental states is not conclusive? Baker (1974) has pointed
out that the notion of conclusive evidence is typically ambiguous:
1. In one sense, one might say that evidence is conclusive if it is not logically
A Conceptual Ana!Jsis if Mental Imagery
I shall now try to spell out a more positive argument for this latter conclusion by
giving a conceptual analysis of mental imagery which is based upon the theory of
criteria to be found in Wittgenstein's ( I953, I958, I967) later philosophical
writings. The idea of a criterion is an important device in his philosophy of
language (Baker, I974; Richardson, I976c), and is central to his philosophy of
mind (Hacker, I972). It is a difficult concept, over which there has been much
philosophical controversy, but it is important to get to grips with the concept of a
criterion if one is to achieve a proper appreciation of the status of mental
imagery.
Wittgenstein ( I958, p. 24) suggested that, to obtain a systematic explanation
of a concept, one must be able to specify what must be the case if that concept is
correctly applied. The basis on which such an ascription is made he called a
Mental Imagery and Human Memory
criterion. The criteria for the use of a word or phrase determine its meaning or
sense. (They also specify what are to count as legitimate knowledge-claims: one
can claim to know that a concept applies to a given object only if the criteria of
that concept are satisfied.) Wittgenstein explained what kind of thing he had in
mind in the following way:
When we learnt the use of the phrase 'so-and-so has toothache' we were
pointed out certain kinds of behaviour of those who were said to have
toothache. As an instance of these kinds of behaviour let us take holding your
cheek. Suppose that by observation I found that in certain instances whenever
these first criteria told me a person had toothache, a red patch appeared on the
person's cheek. Supposing I now said to someone 'I see A has toothache, he's
got a red patch on his cheek'. He may ask me 'How do you know A has
toothache when you see a red patch?' I shall then point out that certain
phenomena had always coincided with the appearance of the red patch.
Now one may go on and ask: 'How do you know that he has got toothache
when he holds his cheek?' The answer to this might be, 'I say, he has toothache
when he holds his cheek because I hold my cheek when I have toothache'. But
what if we went on asking:-'And why do you suppose that toothache
corresponds to his holding his cheek just because your toothache corresponds
to your holding your cheek?' You will be at a loss to answer this question, and
find that here we strike rock bottom, that is we have come down to
conventions.
Thus, there are two kinds of evidence which might be used to justify the
ascription of toothache to another person. The evidential value of criteria is
guaranteed by linguistic convention: they are specified as the defining properties
of a concept when it is taught. However, the evidential value of the other kinds of
evidence which might support the application of a concept is guaranteed merely
by their empirical correlation with its criteria. (Wittgenstein referred to this
latter sort of evidence as the symptoms of the concept.)
It is characteristic of many concepts, according to Wittgenstein (I 958, pp. 5 I,
I 43-4), that there are many different criteria for their application which might
be employed under different circumstances. It follows that there is not
necessarily one thing in common to all the objects to which the concept may be
ascribed; rather, they may be related to one another in many different ways, just
like the members of a family (Wittgenstein, I 953, §§66-7, I 64). Thus, he rejected
the philosophical position known as essentialism (Richardson, I976c, pp. 85-7),
which holds that the meaning of an expression is given by what is common to all
of the things to which it applies, their essence. He also rejected the idea that a
criterion for ascribing a concept to an object must constitute a logically necessary
condition for the truth of the proposition that the object falls under Lhe concept.
It is perhaps more important that the notion of a criterion is also used to show
that an item of evidence need not constitute a logically sufficient condition for the
A Conceptual Anarysis if Mental Imagery
After the more general philosophical preliminaries of the two previous sections, I
shall now apply the framework of criteria! semantics to the concepts of mental
imagery and of memory. In each case the appropriate strategy will be to locate
the concept within a class of similar concepts, and then to consider in a very
schematic way the criteria which operate in the ascription of the particular
concept of interest. It should perhaps be mentioned that an extended discussion
ofboth mental imagery and memory is to be found in Wittgenstein's (I 967) ,Zettel
(the former in sections 62 I-49, and the latter in sections 65c:r-6g). Unfortunately,
34 Mental Imagery and Human Memory
(and typically does) continue to have that ability even when it is not being
exercised. At any time, under suitable circumstances, one may verify that the
person continues to have that ability by a spot check. Moreover, like all
capacities and dispositions, the ability to remember does not exhibit an
epistemological asymmetry between its first-person present-tense use and its
third-person present-tense use, and the self-ascription of this ability is corrigible.
Consciousness and self-consciousness are not defining characteristics of the
ability to remember: a person may manifest that ability without realising or
otherwise being aware of doing so, and, on the other hand, he may falsely believe
that he can remember something.
The faculty of memory is a capacity of a certain kind of object: namely, a
conscious, living body, or what one would ordinarily refer to as a person. That
body exhibits or embodies the capacity by virtue of certain properties of its
physical structure. This distinction between a capacity and its vehicle is a perfectly
general one, applying equally to human beings and to inanimate objects or
systems. (It can be argued that there are no major logical differences between the
two sorts of capacity; cf. Wittgenstein, 1953, §182.) For example, Kenny ( 1972)
gave the example of the ability of a thermostat to regulate a room's temperature
and the structure by virtue of which the thermostat has that ability. Elsewhere,
he gave the example of the capacity of whisky to intoxicate and the ingredient of
alcohol which is responsible for that capacity (Kenny, 1975, p. 10). These
examples indicate that one explains the possession of a capacity by specifYing the
physical properties of its vehicle which are causally responsible for that capacity.
Naturally, limitations in one's knowledge may mean that one can give only a
putative, sketchy and functional account of the physical structure of interest, in
other words a theoretical account rather than a definitive description. For
example, one might explain apparent constraints upon the forms of human
natural languages by the theoretical assertion that the brain is 'pre-
programmed' in some way (cf. Kenny, 1972). However, it is important to realise
that the hypothetical states and processes which are ascribed to the vehicle which
underlies a human capacity are the putative physical states and processes of a
physical structure (the central nervous system), and not the mental states and
processes of the person who possesses that capacity.
More generally, one must bear in mind two important distinctions. The first is
that between psychological states and human dispositions, which was mentioned
earlier. The second is that between a capacity and its vehicle, which has just been
explained. Both distinctions reflect differences of conceptual category; to violate
or to ignore these categorial differences will lead to profound conceptual
confusion (Kenny, 1972; 1975, pp. 1o-1; Wittgenstein, 1953, p. 181).
Nevertheless, contemporary research in cognitive psychology commits exactly
this error. The source of the error lies in taking expressions which have an
established employment as the names of mental states, and using them in
theoretical descriptions of the vehicles underlying human dispositions. This
move is especially obvious in research on mental imagery, which employs a term
A Conceptual Anarysis if Mental Imagery 37
only be able to define the parts of the structure in terms of their function,
without having any understanding of the nature of the material embodiment
of the structure; just as one might well know that the body, since it regulates its
own temperature, must contain a thermostat, and yet have no idea of the
material structure of the thermostat.
In the previous section, it was mentioned that mental images are representations
(in the usual sense of that word). An important logical feature of all
representations is their intentionality: the object of a representation (what it is a
representation of) is determined by how it is intended, not by what it resembles.
This point, and its implications for understanding mental imagery have been
discussed by Fodor ( I975, chap. 4). On the one hand, an image does not have to
resemble what it represents. Malcolm ( I977, p. 2 I6) has given the examples of a
cartoon caricature and a piece of modern sculpture. Of course, if an image did
have to resemble what it represented, one would have to inspect one's images to
find out what they represented (cf. Wittgenstein, I 967, §62 I). On the other
hand, even if an image does resemble what it represents, any pictorial
representation can be interpreted in various ways. Fodor ( I975, pp. I82-3) gave
the example of a hexagon with its major diagonals, which can be seen as either a
plane figure or a three-dimensional cube viewed from one of its corners (see also
Malcolm, I977, pp. I47-53; Wittgenstein, I953, p. I39)· In general, two
pictures can represent the same thing (for example, an isosceles triangle and a
scalene triangle can both represent the notion of triangularity), while one picture
can represent two things (for example, an ambiguous figure can be interpreted as
a duck or as a rabbit) (Fodor, I975, p. I92).
Thus, the object of a mental image is not defined by any of the pictorial
properties of the image itself, but is carried by the description under which the
image is intended. This is similar to the idea expressed by Pylyshyn (I973) that
mental images come to mind already interpreted. It has already been mentioned
that whether another person has a mental image and which image he has is
Mental Imagery and Human Memory
determined by his verbal utterances. This last point means that, for the imager
himself, the identity of, his image is determined by the description which he
attaches to it. Fodor (1975, p. 191) explained this as follows:
Suppose that what one visualizes in imaging a tiger might be anything from a
full-scale tiger portrait (in the case of the ideticist) to a sort of transient stick
figure (in the case of poor imagers like me). What makes my stick figure an
image of a tiger is not that it looks much like one (my drawings of tigers don't
look much like tigers either) but rather that it's my image, so I'm the one who
gets to say what it's an image of. My images (and my drawings) connect with
my intentions in a certain way; I take them as tiger-pictures for purposes of
whatever task I happen to have in hand.
This suggests that the functional origin of mental imagery lies in the
descriptions in accordance with which images are constructed. Of course, there
may be an indefinite number of equivalent descriptions of one and the same state
of affairs, and any one of these possible descriptions would be accepted by the
imager as a true account of his intended representation. Thus, one must regard
the identity of a mental image as consisting not in any particular verbal
statement, but in the abstract proposition which underlies the set of equivalent
de!!criptions. In other words, the functional origin of mental imagery lies in a
system of knowledge which is 'essentially conceptual and propositional, rather than
sensory or pictorial, in nature' (Pylyshyn, 1973; see also Yuille and Catchpole,
1977). It should be added that exactly the same argument can be carried
through even when it is difficult or impossible for the imager to give a verbal
account ofhis mental image (for example, in contemplating the look or smell of a
thing). In such cases the image will allow the same ambiguity or indeterminacy
without the interpretation or intention of the imager (cf. Fodor, 1975, pp. 193-
4).
This is not to say that mental imagery is without functional utility. This must
be established on the basis of experimental investigation, but a priori one can
point out two respects in which the construction of mental images might be
helpful (Fodor, 1975, p. 191). First, if mental images represent spatial properties
and relationships in a quasi-pictorial form, then they may be more efficient for
carrying out certain tasks than discursive, verbal representations. An example of
such a task would be successive crossmodal similarity judgements (Posner, Boies,
Eichelman and Taylor, 196g). Second, a mental image may be determinate
under several different descriptions other than that on the basis of which it was
constructed. The image might therefore manifest emergent properties which
could not be readily computed from the original description (Kosslyn and
Pomerantz, 1977). An example of this would be counting the number ofwindows
in one's house by 'reading off' the information from a mental image (Shepard,
1966).
This last point would seem to block a fundamental and otherwise devastating
A Conceptual Ana!Jsis if Mental Imagery
criticism of dual coding theory which might be taken to follow from the previous
discussion. The intentionality of mental imagery appears to imply that an image
cannot contain any new information which was not already available when the
image was constructed; for, otherwise, the imager would have to inspect the
image to determine whether it was in accordance with his intentions (cf.
Wittgenstein, 1953, §38g, p. 177). Consequently, mental imagery would be
merely an epiphenomenon, and could not be used as a means to recalling
information. (In personal discussion, Peter Hacker has suggested that mental
imagery might be seen rather as a manifestation of the ability to recall that
information, of the same logical status as producing a spoken or written response.
Cf. Wittgenstein (1967, §6so): 'My memory-image is not evidence for that past
situation, like a photograph which was taken then and convinces me now that
this was how things were then. The memory-image and the memory-words stand
on the same level.') Nevertheless, the intentional description according to which
an image is produced does not exhaustively specify all of the properties of the
image; on the contrary, any mental image will have emergent properties which
are not included in that description (though they might be deductively entailed
by the properties which are contained in the description). In those situations
which make use of the emergent properties of images, therefore, mental imagery
will deserve a role in psychological investigation as a distinctive mode of thinking
(cf. Kosslyn and Pomerantz, 1977). In the following chapters, special attention
will be paid to whether the functional elements in cognition and memory are the
emergent properties of a mental image, or properties contained within the
propositional description under which the mental image is interpreted.
3.6 CONCLUSIONS
In the previous chapter, I gave a conceptual analysis which implied that one
should distinguish carefully between the phenomenal experience of mental
imagery and a form of mnemonic representation which might be produced by
the use of mental imagery in learning. I would now like to extend this idea by
proposing a theoretical distinction between what may be called the constructive
and the elaborative uses of mental imagery. First, mental images are symbolic
representations which may be maintained over a definite period of time and
which may be manipulated in various ways. It is therefore reasonable to suggest
that mental imagery constitutes a non-verbal, short-term, working memory in
which information may be pictorially represented and spatially transformed.
Second, mental images are symbolic representations which may be evoked by
the presentation of verbal information to be remembered over an indefinite
period of time. In this case, the use of mental imagery may be regarded as a way
of elaborating or qualitatively transforming the material to be learned, and it is
therefore reasonable to suggest that mental imagery constitutes an elaborative
form of coding in long-term memory. Most of the original research on the nature
and function of mental imagery was concerned with its elaborative role in long-
term memory, but more recently a 'second generation' of experimental research
has been concerned with operations on mental images in immediate memory
(Anderson, 1978).
Other researchers have made similar distinctions. For example, Marks ( 1972)
observed that experiments on mental imagery could be divided into two groups.
One group, concerned with the literal function of mental imagery, considers the
performance of subjects who are asked to form an exact mental image of a
stimulus pattern. The other group, concerned with the associative function of
mental imagery, considers the performance of subjects who are asked to form an
association between two different stimuli by the use of mental imagery.
Similarly, Janssen (1976) argued for the existence of two components in visual
imagery: a spatial component, relating to the location and orientation of a
mental image; and a non-spatial component, relating to the identity and internal
properties of the mental image (cf. Baddeley, 1976, p. 229). Other writers have
suggested a functional distinction between 'dynamic' or 'kinetic' imagery and
'static' imagery (Janssen, 1976, chap. 6; White and Ashton, 1977). Paivio
(1971c, p. 509) made a rather different distinction between the use of mental
43
44 Mental Imagery and Human Memory
imagery as a form of short-term memory for concrete events, and the use of
mental imagery as a non-verbal representation in long-term memory. Finally, a
distinction between 'maintaining' and 'elaborative' activities has been made in
considering the nature of the verbal rehearsal of stimulus material (Craik and
Watkins, 1973).
In this chapter, I shall discuss the constructive or manipulative use of mental
imagery in immediate memory. In subsequent chapters, I shall discuss the
elaborative use of mental imagery in long-term memory.
One might suppose that mental imagery offers an efficient means of representing,
maintaining and manipulating information under many different circum-
stances. Paivio ( 1975h) suggested that it should be especially helpful in tasks that
involve the spatial organisation of informational units or the retrieval of spatial
information from long-term memory. Pinker and Kosslyn (1978) have made
more detailed proposals:
One desirable property of imagery would be that one could 'move' one part or
portion [of a pattern] and all of the spatial relations between that part and the
others would 'emerge', that is, would become evident to the mind's eye
without specifically being calculated .... Such a property of images would be
especially useful if images occurred in a three-dimensional structure, a kind of
'work-space'. The space which we perceive and in which we move about is
three-dimensional, and it surely would be useful to have an internal three-
dimensional 'model' of space that we can manipulate mentally and in which
the consequences of various contemplated actions can be visualized ....
who are asked this question report that they have to imagine moving around the
house, visualising and counting the windows (Janssen, I976, p. 59). Meudell
( I97 I) found that the time taken to answer the question varied in a linear fashion
with the number ofwindows counted. Another task which has been studied is one
where the subjects are asked to name the states, counties, or cities of a particular
country. This is a form of category generation, which typically shows a negatively
accelerated function between the cumulative number of items named and the
time spent in naming them (Bousfield and Sedgewick, I944). However, if the
subjects are asked to name the items by working in a given direction (from north
to south, say), they produce a linear function comparable to that obtained by
reading the items from an actual map (Indow and Togano, I97o; cf. Berlyne,
Ig6s, P· 142).
Naturally, mental imagery should be especially useful in tasks which require
the short-term retention of spatial-order information. Milner ( I97 I) described a
task devised by P. Corsi, in which the subject has to recall a sequence of spatial
locations on a display. The 'spatial memory span' measured by this test is quite
independent of a subject's verbal digit span (J. T. E. Richardson, I977a).
Another task is one devised by Bower (I972), in which the subject memorises a
sequence of compass directions by imagining a sequence of steps taken in the
appropriate directions. Healy (I 97 5, I 978) carried out an extensive investig-
ation of the memory codes used in the short-term recall of sequences of letters.
When the spatial order in which items occur is completely correlated with their
temporal order, it appears that spatial-order information is not retained (Hitch
and Morton, I975)· However, when the two orders are varied independently,
Healy found that the recall of the spatial ordering was based upon a complex
encoding of the temporal-spatial pattern which had been presented. Similarly,
Anderson (I 976) found evidence for the independent coding of spatial and
temporal information in short-term memory. She found that pictorial displays
tended to be recalled better in terms of their spatial structure, whereas verbal
presentation emphasised their temporal structure (cf. Fodor, I975, pp. I86-7).
Finally, a rather different situation in which mental imagery seems to be
important is the short-term recognition offaces (Nielsen and Smith, I973) and of
random visual patterns (Phillips and Christie, I977a).
The experiments which have been described thus far have provided some
support for the idea that mental imagery is a literal representation of spatial
information. That is, the spatial relationships between the objects in an imaged
array appear to correspond to the relationships which would hold between the
same objects in an actual array. A further question is whether the way in which
imaged objects are manipulated corresponds to the way in which actual objects
might be manipulated (Shepard and Podgorny, I979).
Mental Imagery and Human Memory
Most of the experimental research in this area has been carried out by Shepard
and his collaborators. The original study by Shepard and Metzler (I 97 I) showed
that the time taken to match two different views of the same three-dimensional
object was linearly related to the angle between the two views. This indicated
that the subjects were mentally rotating one or both of the objects at a constant
rate until they had the same orientation. All of their subjects reported using
mental imagery in order to carry out this process of mental rotation. The
question arises whether similar patterns of performance will be obtained when
subjects have to carry out a whole sequence of distinct manipulations. Shepard
and Feng (I972) investigated this possibility by presenting patterns of six
connected squares which result when the faces of a cube are unfolded onto a flat
surface. The subjects were asked to decide whether two arrows marked on edges
of different squares would meet if the squares were folded back up into the cube.
The results indicated that reaction time varied in a linear fashion with the sum of
the number of squares that would be involved in each fold, if the folds were
actually performed physically. Once again, all of the subjects reported using
mental imagery to carry out the task: 'Some Ss described this imagery process as
being primarily visual; others spoke, as well, of a strong kinesthetic component in
which they imagined folding up the cube with their own hands .... '
A more complex task was devised by Cooper and Shepard (I 973), in which the
subjects were required to judge whether common alphanumeric characters were
presented in their normal form or as mirror-image reversals. On each trial, the
subject was instructed to construct a mental image of the appropriate character
in one of six different orientations; the stimulus was then presented in that
orientation or in one of the other five. The results showed that reaction time
increased in a linear fashion with the angular discrepancy between the
anticipated orientation and the actual orientation of the stimulus. On the basis of
this relationship, it was suggested that the subjects rotated their visual images at
a constant rate until they were at the same orientation as the test stimulus. This
interpretation of the results was borne out by the subjects' own reports
concerning their use of mental imagery; however, as Sheehan ( I978) has pointed
out, some subjects indicated that what they were mentally rotating was very
schematic and that their phenomenal experience was as much kinaesthetic as
visual. Essentially the same findings were obtained by Cooper ( I975) using
nonsense patterns as stimuli, and further discussion of these experiments is
contained in papers by Metzler and Shepard ( I974) and by Cooper and
Shepard ( I979).
1974). Since the effect appears to result from mental comparisons made along
any ordered dimension whatsoever, it will not be informative as to the specific
strategy or process used by subjects in order to carry out certain particular sorts of
mental comparison.
Perhaps I should spell out this latter point more clearly. Most researchers are
agreed that the finding of a symbolic distance effect in mental comparisons
entails that the relevant information is represented in a continuous, analogue
form (Kerst and Howard, 1977; Moyer, 1973; Paivio, 1975b, 1975e, 1976b,
1978d). Paivio (1975b, 1975e) went further to argue that the symbolic distance
effect was consistent with his account in terms of mental imagery, but inconsist-
ent with 'any model which assumes that the size attribute of animal names is
represented in the form ofdiscrete semantic features of propositions'. The finding
of a symbolic distance effect with abstract properties of concrete things (Banks
and Flora, 1977; Kerst and Howard, 1977) creates some difficulty for this view,
but Paivio (1978b) argued that, insofar as these properties were attributes of
things, it was reasonable to suppose that mental images of those things might be
involved. Nevertheless, this does not resolve the problem caused by Friedman's
( 1978) finding of a symbolic distance effect with abstract properties of abstract
concepts.
Paivio's arguments were well-motivated and accurate so far as specific
proposals by propositional theorists went at the time (see, for example, Anderson
and Bower, 1973, p. 461). However, Friedman ( 1978) has recently pointed out
that the assumption that propositional structures can only represent discrete
information rests on a confusion between structure and content. Thus, the
symbolic distance effect does not even count as evidence against propositional
theories (see also Anderson, 1978). Indeed, in his most recent papers, Paivio
( 1978b, 1978d) has accepted that other analogical representations might be
conceptualised beyond mental imagery. As Kerst and Howard ( 1977) concluded
in their investigation:
The symbolic distance effect ... cannot be taken as a unique property of the
imagery system which sets it apart from the verbal system. Rather the case for
- the role of visual imagery in the comparison process must rest on evidence
other than the symbolic distance effect.
(A finding which does seem to create difficulty for propositional theories is that
reaction times are similar when comparisons are made between conceptual
categories to when comparisons are made within a conceptual category: see
Paivio, 1975e.)
A major alternative source of evidence concerning the nature of the
representation used in mental comparisons has been the investigation of
performance with pictures and words as stimuli. A basic assumption ofPaivio's
(1971c) theoretical position is that mental imagery should be evoked more
readily by pictures of objects than by the names of those objects. If mental
so Mental Imagery and Human Memory
comparisons are based upon the use of mental imagery, it follows that such
comparisons should be faster with pictures than with words as stimuli (even if the
pictures do not faithfully represent size information). The first investigation of
this idea was carried out by Paivio (I975e), with concrete objects being
compared in terms of their physical size. Both pictures and words produced a
symbolic distance effect, and, as predicted, reaction times were substantially
faster with pictures than with words. Similar results were obtained by Paivio
(I978a) with comparisons of mental clocks, but in this case the pictorial stimuli
directly represented the relevant information (the angular separation of the
hands), and so the pictorial comparison was reduced to a perceptual comparison.
Paivio ( I975t) also investigated performance with pictorial presentation when
the pictured sizes of the two objects were incongruent with their actual size.
Thus, the subjects had to respond 'larger' to the physically smaller picture.
Under these circumstances, reaction times were slower than with congruent
pictorial representations. Moreover, pairs which are incongruent with respect to
size are congruent with respect to apparent distance. In accordance with this
idea, such pairs produced faster reaction times when the subjects made
judgements of apparent distance. However, although these experiments show
that mental comparisons may be speeded or slowed by congruent or incongruent
perceptual information, they do not show that such comparisons are made on the
basis of representations which preserve such information in an analogical
fashion.
The superiority of pictorial presentation over verbal presentation is consistent
with the idea that mental imagery is involved in mental comparisons between
concrete objects on the basis of physical attributes. However, Banks and Flora
(I 977) considered performance on words and pictures in their study of
comparisons between animals in terms oftheir intelligence. They also found that
the comparison time was faster with pictures than with words; since the relevant
dimension was abstract rather than concrete, they took the result to be evidence
against Paivio's ( I975e) position. Paivio ( I978d) himself compared pictorial and
verbal presentation in judgements of pleasantness and monetary value. In both
cases, mental comparisons were faster with pictures than with words, and for
judgements of pleasantness mental comparisons were faster with concrete nouns
than with abstract nouns. Paivio suggested that properties such as intelligence,
pleasantness and value should be regarded as attributes of things rather than
words, and that it is necessary to produce mental images ofsuch things in order to
make comparisons with respect to these properties (see also Paivio, I978b).
Support for this idea has been provided by comparing performance on
pictures and words in judgements of non-semantic properties of words. For
example, Paivio (I 975e) showed that mental comparisons of the pronounce-
ability of object names were slower when the objects were pictured than when
their names were visually presented. Similar results were obtained by Paivio
( I978b) when subjects compared the relative frequency with which object names
Mental Imagery in Immediate Memory
differed In terms of their overall reaction times. The results suggested that the
mental comparisons were based on mental images, and that these were more
readily produced by subjects of high spatial ability (Paivio, 1978h).
Evidence on individual differences for mental comparisons involving abstract
attributes was provided by Paivio ( 1978d). For comparisons of both pleasantness
and monetary value, the reaction times of subjects of high spatial ability were
significantly faster than those of subjects of low spatial ability. In neither case
was the effect of verbal ability significant. These results support Paivio's idea that
properties such as intelligence, pleasantness and value are attributes of things
rather than words, and that even mental comparisons on abstract dimensions
such as these are made on the basis of mental images of the named objects.
Finally, Paivio (1976h, 1978d) referred to an unpublished study which found no
sign of an effect of spatial ability upon mental comparisons with respect to word
familiarity. This supports the converse suggestion that mental comparisons on
non-semantic dimensions of words are not made with the help of mental
imagery.
In order to check on the compatibility of these conclusions with the subjects'
phenomenal experience, unpublished experiments on mental comparisons
carried out at Brune! University have included questionnaires on the strategies
employed to carry out these tasks. One study employed pairs of animal names, as
in the original experiment by Moyer ( 1 973); the subjects reported using mental
imagery 73% of the time in makingjudgements along concrete dimensions (size
and angularity), 79% of the time in making judgements along abstract
dimensions (ferocity and intelligence), and 13% of the time in making
judgements along non-semantic dimensions (frequency and pronounceability).
A further study employed homogeneous pairs of concrete and abstract nouns
with judgements along the Evaluative and Potency Dimensions of the Semantic
Differential (Osgood et al., 1958); the subjects reported using mental imagery
68% of the time with concrete pairs, but only 24% of the time with abstract
pairs.
These results are quite consistent with Paivio's account of mental com-
parisons, according to which concrete objects are judged along both concrete
and abstract dimensions by a comparison of mental images. However, there is a
crucial conceptual difficulty with this account which has been pointed out by
Banks (1977):
If images are to be compared to compute the correct response in, say, a size
judgement, it is necessary to make the images the right size to begin with.
Kosslyn's (1975) research shows that mental images are flexible as to relative
size, and it would seem necessary to retrieve size information along with shape
information in constructing the images. (Our long-term memory for im-
ageable things cannot be little snapshots that are all just the right size.) Thus,
size information must be available, and used, before the image is constructed,
and the imagery process itself hardly seems to be necessary.
Mental Imagery in Immediate Memory 53
This point is connected with the property of intentionality which was discussed
in the previous chapter. Not only is the identity of a mental image determined by
the description under which the image is intended; the apparent or relative size
of the imaged object is also carried by the intentional description. To the extent
that the words to be compared evoke a mental image in which the appropriate
objects are represented with the correct relative sizes, that image only serves to
manifest knowledge which must have been independently available.
Banks' conceptual argument was supported empirically by the results of a
study by Holyoak ( 1977). The subjects in these experiments made judgements of
relative magnitude of pairs of objects whose names were presented successively.
The subjects were instructed to prepare themselves with an image of the first
object in each pair at its normal size, with an image that was abnormally big or
small, or with no image at all. Independent of this, the subjects either were
specifically instructed to use mental imagery to make the comparison of the two
objects, or were not so instructed. The different preparatory instructions only
influenced the reaction times when the subjects were specifically instructed to
use mental imagery in the comparison itself. In addition, when the subjects were
asked to maintain irrelevant images of digits at the same time as making the
mental comparison, this had a greater effect upon performance when the
subjects were told to use mental imagery in the comparison itself. These results
suggested that subjects normally compare the sizes of named objects not using
mental imagery, but using more abstract information. Moreover, they do not
even show that subjects can use mental imagery to make such comparisons when
specifically instructed to do so. Holyoak's results indicate merely that mental
comparisons are slowed by maintaining two mental images (either a congruent
image and an incongruent image, or a relevant image and an irrelevant image)
instead of one.
Banks (1977) interpreted his conceptual criticism and Holyoak's results to
mean that 'imagery is not frequently or importantly involved in symbolic
comparative judgements'. Nevertheless, it is unlikely that his point applies to all
or even to most of the dimensions that have been used in mental comparison
studies.lt is apposite precisely because the apparent size ofan imaged object is an
intentional property which is incorporated into the description under which the
image is constructed. However, most of the other attributes of concrete objects
which have been employed (angularity, intelligence, ferocity, angular separ-
ation of the hands of a clock, pleasantness, monetary value, and so on) are not
intentional properties, and could only be computed from the intentional
properties with the utmost difficulty, if at all. Rather, they are emergent
properties which are only manifest in mental images of the relevant objects. So,
although there are good conceptual reasons for rejecting Paivio's account in the
case ofjudgements of relative size, and although the various sorts of evidence on
mental comparisons might individually be open to alternative interpretations,
the explanation which is most in accord with the findings to date is that which
identifies mental imagery as the cognitive representation employed in making
54 Mental Imagery and Human Memory
judgements among pictured and named concrete objects along physical and
semantic dimensions. As Paivio (I 978d) remarked: 'The combined influence of
symbolic distance, picture superiority, and imagery ability in the present tasks
strongly suggest that the comparison is based on information that is analog and
continuous, as well as specifically nonverbal and imagistic in nature.'
Other researchers have used somewhat different psychophysical techniques to
argue for an isomorphism between perceptual and cognitive representations.
Shepard and Chipman (I 970) used multidimensional scaling to analyse
judgements of shape similarity between pairs of states of the U.S.A., presented
either as cut-out maps or as their names. Fillenbaum and Rapoport (I 97 I) used
similar methods to compare the memory structures underlying colour names and
the perceptual structures of the colours themselves. Kerst and Howard (I 978)
demonstrated that similar psychophysical functions were produced from
estimates of geographical areas or distances whether these were made from
studying a map or from memory of a map. The latter experiment produced
intriguing evidence on the question of the functional similarity between
judgements based upon perception and 'internal psychophysical judgements'
based upon memory representations. The psychophysical function obtained in
perceptual estimation is typically a power function whose exponent may deviate
from unity in either direction. Kerst and Howard found that estimates made
from memory were also well fitted by power functions, and that the exponent of
such a fimction was close to the square of that produced in perceptual estimation.
They argued that this is exactly the. relationship which would be expected if
memory judgements of perceptual continua result from a 're-perceptual' process
that operates upon stored perceptual representations (see also Moyer and
Dumais, I 978).
A rather different sequence of experimental research has been concerned with the
possibility of finding specific sorts of experimental task which might selectively
interfere with a subject's ability to create and use mental images. This research is
based upon the general idea that the psychological mechanisms underlying
perception and mental imagery are functionally overlapping. In chapter 2, it
was suggested that there were no adequate grounds for this proposal.
Nevertheless, one has the intuitive idea that 'looking at one thing and visualizing
another at the same time is as difficult as trying to look at two things at once'
(Marks, I977; see also Neisser, I976, p. I46), and this indicates the possibility of
using perceptual tasks to disrupt the use of mental imagery. Indeed, Neisser
(I 972b) argued that this sort of experimental paradigm could provide an
'operational interpretation' of his 'perceptual definition' of mental imagery as
the use of the cognitive processes involved in perception in the absence of
adequate stimulus input. Stronger forms of the assumption that perception and
Mental Imagery in Immediate Memory 55
Thus, Brooks' ( 1968) results appear to show that the processing of spatial
Mental Imagery in Immediate Memory 57
The results were fortunately very clear. The auditory tracking caused far
greater impairment on the Brooks spatial task than on the equivalent verbal
task, while for the brightness judgement no such difference occurred. In short,
it appears that, for the Brooks tasks at least, disruption is spatial rather than
visual.
sensitivity is roughly twice as great when the signal and the mental image are in
the same sensory modality. Thus, auditory imagery interferes more with the
detection of auditory signals, but visual imagery interferes more with the
detection of visual signals (Segal and Fusella, 1970, 1971). So, although there is a
generalised effect of mental imagery upon perceptual sensitivity, there is also a
clear modality-specific effect as well. This latter effect might be taken to support
the idea of a functional overlap between mental imagery and perception, but
Bower ( 1972) suggested that it might merely reflect peripheral effects of mental
imagery upon modality-specific attentional mechanisms; for example, visual
imagery might reduce visual sensitivity by producing pupil dilation and
misfocusing.
Unfortunately, recent research by Phillips and Christie ( 1977b) has tended to
undermine the conventional interpretation of Brooks' (1967, 1968) original
experiments in a radical manner. They observed that Brooks had failed to
include adequate control conditions, and so could not legitimately claim to have
demonstrated interference specificity. Experiments which incorporated such
controls have clearly shown that verbalisation can affect visualisation (Kelly and
Martin, 1974; Yuille and Ternes, 1975). They proposed that one should analyse
various complex intervening tasks to determine precisely which aspects or
components are responsible for the resulting interference.
In previous research, Phillips and Christie ( 1977a) had identified a recency
effect in the short-term recognition of novel abstract patterns. This effect was
confined to the last item in a sequence and was attributed to a visual short-term
memory, which they called visualisation. Accordingly, they carried out a series
of experiments to determine which interpolated tasks would remove or interfere
with this component. It was found that visualisation was disrupted by mental
addition, but not by reading; that the disruption caused by mental addition was
independent of the presentation modality of the digits to be added; and that
interference from similar stimulus patterns depended upon whether the subjects
had to maintain visual representations that outlived iconic storage. Contrary to
the usual interpretation of Brooks' findings, that visualisation and perception
compete for special purpose visual processing resources, Phillips and Christie
took their results to indicate that visualisation requires general purpose
resources, and that interference between visualisation and perception might be
due to competition for these resources. However, their experiments suggested
that visualisation might be disrupted by any concurrent activity which is not pre-
programmed and over-learned, but which might compete for the resources of a
central executive system or working memory. Previous results may be readily
interpreted in these terms. For instance, the experiments by Healy (1975),
mentioned earlier, showed that the recall of spatial order would be disrupted by
a relatively novel, unpractised task, such as naming the spatial locations of a
sequence of digits, but not by a relatively over-learned task, such as naming a
sequence of digits. One may conclude, however, that research which attempts to
disrupt mental imagery by the specific use of perceptual concurrent tasks is
Menta/Imagery in Immediate Memory 59
4·5 CONCLUSIONS
This chapter has been concerned with a very limited number of rather
specialised areas of experimental research. However, generally speaking, the
results of that research indicate that mental imagery may be employed in the
representation, preservation and manipulation of spatial and pictorial inform-
ation. In particular, it may operate as a form of short-term working memory in a
variety of experimental tasks to enable the subject to grasp emergent properties
of a stimulus array which could not be readily computed or deduced purely on
the basis of a propositional description of that array. This use of mental imagery
may be disrupted by a concurrent cognitive task involving the processing of
spatial information, and possibly by any concurrent task which is not pre-
programmed but makes demands upon a central executive system.
5
Pictorial Memory
It has been well established that the normal adult's ability to recognise pictures is
remarkably good. The first systematic investigation of this ability was carried out
by Shepard (I967). He compared his subjects' recognition memory for words,
sentences and pictures. In each case, the subjects attempted to memorise a series
of approximately 6oo stimuli, and were then tested on their ability to
discriminate these from other, 'new' stimuli using a forced-choice procedure. On
immediate testing, the median percentage of correct responses for each of the
three kinds of material was go.o, 88.2 and g8.5. As Shepard concluded:
Robinson, 1968; Nelson and Kosslyn, 1976; Reese, 1975; Robinson, 1970; Ward
and Legant, 1971). Thus, although it might eventually turn out to be necessary
to allow that image superiority contributes to the fact that memory performance
is better with pictorial material than with verbal material, it seems essential to
incorporate the possibility of redundant and simultaneous codes into an
adequate account of the effect, and the coding redundancy hypothesis tends to be
accepted by most psychologists who accept the existence of a qualitatively
distinct imaginal representation (for example, Bower, 197oa; Paivio, 1971c,
1972). Perhaps more important, it is difficult to see how the evidence on the
effects of labelling can be handled by the opposing, common coding theory of
imagery and language, which assumes that pictorial and verbal material are
encoded into a common mnemonic representation.
In connection with the effects oflabelling upon memory performance, it has
long been known that, if subjects are asked to recall ambiguous geometrical
figures by drawing them, supplying verbal labels may lead to distortion of the
figures in the direction of a prototypical instance of the relevant concept (Bruner,
Busiek and Mintura, 1952; Carmichael, Hogan and Walter, 1932; Hall, 1936;
Herman, Lawless and Marshall, 1957). However, such distortion also occurs if
the labels are provided only at the time of recall (Hanawalt and Demarest,
1939), and it does not occur in tests of recognition memory (Prentice, 1954). This
suggests that labelling has an effect upon the processes of reconstruction, but not
upon the actual storage of the labelled pictures (Reese, 1977b).
Quite a different experiment creates further problems for the common coding
hypothesis. Jenkins (described by Bower, 1972) required his subjects to learn
paired associates consisting of pictures and words. He found that their learning
performance was actually increased if the perspective of the object depicted in
the picture was varied from one trial to the next, compared to a control condition
in which the same picture was used on each trial. This indicated that the subjects
were integrating the successive perspective views into a single three-dimensional
representation which was a superior mnemonic cue to the constant two-
dimensional representation presumably evoked in the control condition. As
Paivio ( 1975b) remarked, the mnemonic representation produced by mental
imagery is a mental construction rather than a product ofany specific perceptual
experience.
Whether the subject's retention is tested by recall or recognition, it is usually
suggested that the contribution of the hypothesised pictorial representation
should be more important if a verbal description of what is represented is too
complex or is simply not available (Sheehan, 1972). Thus, complex, abstract
geometrical patterns for which there exists no obvious label can plausibly only be
remembered by means of some pictorial, non-verbal representation (Owens and
Richardson, in press). A more interesting class of stimulus is the human face, and
a considerable amount of research has been carried out on the specific problem of
how one recognises faces.
64 Mental Imagery and Human Memory
One of the critical assumptions of the dual coding hypothesis is that the verbal
and non-verbal representations taken to operate in human memory are
functionally independent (Paivio, 1975b, 1978b). To test this assumption, it is
necessary to show that the two sorts of mnemonic code have additive effects upon
performance in particular experimental situations. Of course, since both codes
may in principle be used to remember a given class of stimulus material, the
demonstration of additive effects is no easy matter. However, there is some
evidence for such effects in the recall of pictures and words.
Paivio and Csapo ( 1973) reported a series of experiments which used
incidental recall and different orienting tasks to induce one code or the other
individually, or both simultaneously. They based their research on the fact that
repeated presentations of a stimulus in the same modality are typically not
additive in their effects upon recall performance. This is consistent with the idea
that repeated presentations serve to strengthen a single mnemonic represen-
tation, and inconsistent with the idea that each presentation creates a new,
separate representation in memory. Conversely, they argued, additive, statisti-
cally independent effects of repeated presentations upon memory performance
are evidence for separate, functionally independent representations in memory.
Not surprisingly, the experiments demonstrated that repeated presentations of
a stimulus produced increased recall performance, and pictorial presentations
produced better performance than verbal presentations (that is, names of objects
were recalled less well than pictures of the same objects). Picture-picture and
word-word repetitions produced poorer performance than would be expected
on the basis of stochastically independent effects upon recall performance, but
repetitions which pictured previously named objects or which named previously
pictured objects were additive in their effects. This supported the hypothesis that
pictorial and verbal mnemonic codes are functionally independent.
As Paivio (1975a) observed, the empirical observation of stochastic indepen-
dence cannot be handled by common coding theory without the addition ofsome
purely post hoc device. For example, it might be assumed that knowledge of the
original presentation modality is preserved by attaching differentiating and
independent 'tags' to a 'core' representation in memory. However,
verbal distractor task interfered more with the subsequent recall of the figures
than with their subsequent recognition, but that a visual distractor task had the
opposite effect. Moreover, the subjects' ability to describe the original stimuli
was correlated with their recall performance, but not with their recognition
performance. Baddeley ( I976, pp. 2 I6-I 7) summarised the implications of these
results as follows:
It appears then that the recall of visual material is relatively poor, shows rapid
forgetting ... , and tends to rely on verbal coding, whereas visual recognition
shows relatively little forgetting after the first few seconds and is not
apparently affected by verbal factors.
While there seems to be good evidence that the pictorial presentation of material
directly arouses a distinctive, nonverbal representation in human memory, the
critical proposition for present purposes is that the same representation is evoked
when subjects construct mental images in learning material which is presented in
a verbal form. What experimental evidence is there for this idea?
First, if pictorial presentation directly arouses such a mnemonic represen-
tation, then presumably the construction of mental images should have little or
no effect upon performance. Since the idea of instructing subjects to make up
mental images of what is clearly before them is rather bizarre, there is little
evidence on this matter. Nevertheless, Robinson ( I970) showed that instructions
68 Mental Imagery and Human Memory
to use mental imagery had little effect upon recognition memory for line
drawings, whereas instructions to label the drawings had the usual beneficial
effect upon performance. An important exception to the idea that mental
imagery should have little effect upon pictorial memory is that it should be
helpful if it serves to integrate stimuli which are presented as separate pictures.
Kerst ( 1976) used an incidental paired-associate learning experiment, and found
that instructions to generate such interactive images produced a substantial
increase in performance, equivalent in magnitude to that produced by actually
presenting the pictured objects interacting in some way (see also Wollen and
Lowry, 1974). As Paivio ( 1975d) pointed out, the parallel results obtained from
the use of interactive pictures and from the use of instructions to make up
interactive mental images 'make it reasonable to infer that the instructional sets
and the pictures arouse similar representations'. However, the more detailed
discussion of the effects in instructions to use mental imagery in learning verbal
material contained in chapter 6 will throw considerable doubt upon the idea that
such instructions implicate a qualitatively different sort of mnemonic
representation.
Second, a verbal stimulus which is reliably associated with some (possibly
arbitrary) pictorial stimulus should come to evoke the representation of that
stimulus in non-verbal long-term memory. Bower (I 972) used this idea to
explain why subjects performed better in a paired-associate learning experiment
when the stimuli were the names of their personal friends than when they were
the names of historical characters whose exact facial appearance was unknown,
or when the stimuli were the names of totally unknown people randomly selected
from a telephone directory. Indeed, the association with pictorial or actual
experience is usually taken to be the basis of the effect of stimulus imageability or
concreteness upon memory performance (see chapter 7). But can the operation
of such associations be demonstrated more clearly? An experiment by
Philipchalk (described by Paivio, 1971b; 197Ic, p. 72) used nonsense syllables as
the stimulus terms in a paired-associate learning task, where the response terms
were pictures, concrete words or abstract words. In the second phase of the
experiment, the same nonsense syllables were paired with randomly chosen,
meaningful words. The syllables which had previously been paired with pictures
produced better performance than those which had been paired with concrete
words, and these in turn produced better performance than those which had
been paired with abstract words. However, this variation in performance
occurred only when the subjects were instructed to use their previous associations
in the subsequent learning task. Philipchalk's results suggested that pictorial
associations might subsequently be employed as mental images to mediate new
learning, although this mediation might need to be primed or prompted by
appropriate instructions. This was confirmed by the use of post-learning
questionnaires, in which the subjects reported the greatest use of mental imagery
in learning syllables which had previously been paired with pictures, and the
least use of mental imagery in learning syllables which had previously been
Pictorial Memory 69
paired with abstract words. However, such images must have been constructed
under determinate intentional descriptions, and Philipchalk's results may be
taken to show merely that these propositional mediators enhance memory
performance.
Finally, can one point to similarities between pictorial presentation and the
use of mental imagery in learning in terms of the situations in which they are
most effective? The results of Bower (I 972) and Kerst (I 976) show that both
procedures are more effective when they encourage the integration or organis-
ation of the material in a single, interactive representation, than when they
promote the use of a separate, unrelated representations. However, in the
following chapter, it will be noted that the superiority of interactive encoding over
separative encoding applies equally to verbal representations in memory.
Reference has already been made to the work ofJones (I 978), who showed that
pictorial presentation was especially helpful in retaining location information.
Jones demonstrated a similar conclusion in analysing the results of experiments
by Anderson and Bower ( I973, pp. 305-I9), on the effects of instructions to use
mental imagery. These ·studies investigated memory for sentences of the form,
'The Subject Transitive- Verb-ed the Object who Intransitive- Verb-ed in the Location'
(for example, 'The hippie touched the debutante who sang in the park'). In one
experiment the subjects received standard learning instructions, and in the other
the subjects received additional instructions 'strongly urging them to form vivid
visual images of the sentences to be memorized'. In both cases, the subjects were
cued with one, two or three of the five content words and had to produce the
entire sentence. Anderson and Bower had concluded that the effect of
instructions to use mental imagery upon sentence recall was simply to raise the
overall level of performance, but Jones demonstrated that the improvement was
relatively specific to the grouping, 'the Object who Intransitive- Verb-ed in the
Location' (for example, ' ... the debutante who sang in the park'). As Jones
concluded: 'This may be because the coding of information set in a locational
context is particularly improved by the imaging operation; this explanation is
consistent with the finding reported earlier that recall involving location is
particularly improved by pictorial rather than verbal presentation.' Once again,
however, since the mental images employed by Anderson and Bower's subjects
must have been constructed according to intentional descriptions, the results
might be taken to indicate that both pictorial presentation and imagery
mnemonic instructions encourage propositional representations which emphas-
ise location information.
5·5 CONCLUSIONS
at least with respect to the specific proposals which have been made by the
proponents of the respective theories.
The well-established superiority of pictorial presentation in recognition and
recall does not discriminate between the alternative positions, but the effects of
labelling upon the recall of pictures and the effects of supplementing verbal
material with relevant pictures are difficult to handle in terms of a theory based
upon a common propositional representation in long-term memory. The same
conclusion applies to Paivio and Csapo's (I 973) demonstration of the stochastic
independence of pictorial and verbal presentations in incidental recall. On a
priori grounds, moreover, it is difficult not to assume a basically pictorial
representation in memory underlying in the retention of complex visual displays
which cannot be readily described or labelled, such as abstract geometrical
patterns, or the faces of unknown people.
Thus, the empirical evidence favours the theory of dual coding, and within
this framework gives specific support to the coding redundancy hypothesis,
according to which pictures to be remembered may receive both pictorial and
verbal representations in long-term memory. This account is supported further
by the effects of selective interference from irrelevant spatial tasks interpolated
between presentation and recall. Although, as Anderson (I 978) has proved, it is
possible in principle to specify versions of common coding theory which are
equivalent to dual coding (at least with respect to its predictions for pictorial
memory), the various adjustments needed to ensure this equivalence are unlikely
to constitute well motivated extensions of the current versions of common coding
theory. The dual coding model may therefore be regarded as the most promising
and most coherent theoretical approach for investigating pictorial memory.
For the purpose of this book, however, the principal interest in pictorial
memory lies in its value as a means of investigating mental imagery. There is
nevertheless little direct evidence to permit an identification of the non-verbal
representation aroused by the presentation of pictures and the mnemonic code
produced by the use of mental imagery. Where experimental investigations have
suggested that the two representations have structural or functional properties in
common, the question of the intentionality of mental imagery suggests that the
relevant information might be adequately represented at a propositional level.
Moreover, in subsequent chapters, empirical findings will indicate that any
similarities between the two representations are also shared with verbal coding in
long-term memory. The crucial assumption of dual coding theory that pictorial
presentation gives rise to an imagistic representation in memory may turn out to
be empirically testable only by a consideration of the relevant neuropsycholo-
gical evidence, and this must wait until chapter 9·
6
Imagery Mnemonic Instructions
One of the reasons for the revival of interest in mental imagery was the
experimental demonstration of the effectiveness of instructions to subjects to use
mental imagery in their learning. The efficacy of such instructions had in fact
been known more generally for a very long time. Various techniques prescribed
for orators in Greek and Roman times included the use of mental imagery
explicitly (see Paivio, 1971 c, chap. 6; Yates, 1966), and these techniques survive
essentially unchanged in the present day in courses of memory improvement. In
this chapter, I shall discuss experimental evidence on the effects of imagery
mnemonic instructions and its implications for theories of human memory.
(Bower, I972; Groninger, I97I; but cf. Hasher, Riebman and Wren, I976),
indicating that mental imagery not only improves memory performance, but
also makes the stored material less vulnerable to forgetting. Moreover, Beech (in
press) has suggested that the adequate use of mental imagery depends upon an
adequate delay between presentation and recall, so that subjects have time to
visualise the material to be remembered. It is interesting that he found that
immediate recall produced a specific reduction in the accurate recall oflocation
information.
Whether these effects are specifically due to the use of mental imagery and not
to other factors must be determined by examination of the experimental
evidence. The most plausible alternative explanation is that instructions to use
mental imagery simply enhance the subjects' motivation to learn. This was
discussed at length by Bower ( I972) and by Paivio ( I96g), and rejected fairly
conclusively. The most crucial difficulty for a motivational account is that the
effects of imagery instructions are undiminished under incidental as compared to
intentional learning conditions (Paivio, I 972). Indeed, incidental and inten-
tional learners operating under imagery instructions may produce comparable
levels of performance (Bower, I972). Further evidence comes from the use of
post-learning questionnaires in which the subjects indicate the learning
strategies they have employed. These indicate quite clearly that subjects are able
to increase their use of mental imagery in response to specific instructions to do so
(Paivio and Yuille, I 967, I 969; cf. Bugelski, I 97 I). However, they also show that
subjects frequently use mental imagery spontaneously, especially in learning
relatively concrete material, and this tendency increases over trials even when
the subjects have been given specific instructions to use alternative mnemonic
strategies (see Paivio, I97Ic, pp. 36I-6, for discussion of this point). The
implication is that the substantial observed effects of imagery mnemonic
instructions are nevertheless likely to be underestimates of the effectiveness of
mental imagery as a mnemonic strategy (Anderson and Kulhavy, I972; Bower,
I972).
One of the traditional elements of techniques for the improvement of memory
is the prescription that the learner should try to produce mental images that are
in some way bizarre or peculiar. Instructions to produce bizarre mental images
have been found to produce improved recall performance when compared with
standard learning instructions (Perensky and Senter, I 970). However, studies
which have attempted to separate the effects of bizarreness and of imaging have
failed to find any additional effect of constructing bizarre images over and above
the benefit of using mental imagery (Hauck, Walsh and Kroll, I976; Senter and
Hoffman, I976; Wollen, Weber and Lowry, I972; Wood, I967). Indeed, bizarre
mental images take longer to construct and may result in poorer performance
(Nappe and Wollen, I973); similarly, sentences describing conventional or
plausible situations may be more effective mediators in paired-associate learning
than sentences describing bizarre or implausible situations (Collyer,Jonides and
Bevan, I972; Kulhavy and Heinen, I974)· Neisser (I976, p. I40) has expressed
Imagery Mnemonic Instructions 73
certain reservations about accepting these negative findings, and there are
suggestions that bizarreness may be of benefit after a delay of 24 hours or more
(Andreoffand Yarmey, 1976; Delin, 1968; cf. Webber and Marshall, 1978), or if
the subjects rate the bizarreness of their mental images at the time of presentation
(Merry and Graham, 1978). However, the safest conclusion at present is that
bizarreness is not a factor determining the value of mental imagery in improving
recall.
The most common explanation of the effectiveness of imagery mnemonic
instructions is the coding redundanry hypothesis mentioned in the previous chapter,
according to which memory performance varies with the number of alternative
memory codes available for an item. This is naturally the explanation favoured
by adherents of the dual coding theory (for example, Bower, 197oa, 1972; Paivio,
197Ic, chap. 11). The specific account runs as follows (Paivio, 197Ic, p. 389):
While it has generally been established that instructions to use mental imagery in
learning may lead to substantial improvements in recall and recognition
performance, there is one important exception to this generalisation which was
mentioned briefly in the previous chapter. It is crucial that the mental imagery
used by the subjects in such experiments serves to increase the organisation and
cohesion of the material to be remembered. Accordingly, instructions to produce
separate mental images corresponding to the individual stimulus items may have
no effect upon performance at all, and may even lead to a reduction in recall.
This point was first identified by Bower ( 1972), who compared the incidental
learning performance of a group of subjects receiving interaction instructions (to
construct a mental image depicting two objects interacting in some way) with
that of another group of subjects receiving separation instructions (to construct a
mental image depicting two objects 'separated in their imaginal space, like two
pictures on opposite walls of a room'). The first group demonstrated the usual
marked superiority in recall when cued with one of the words in each pair,
whereas the second group performed at the level expected when subjects were
merely instructed to use rote repetition. As Bower concluded: 'Instructions to
image the terms per se have relatively little effect on associative learning. The
74 Mental Imagery and Human Memory
results seem to imply that mental imagery is spatial in nature, and not tied to any
specific sensory modality (Baddeley, 1976, p. 231). However, their subjects'
reports indicated that they had to make the 'concealed' relationships picturable
in order to carry out the task. In the majority of cases, the subjects reported
making the concealed object visible directly, imagining being able to 'see
through' the concealing surface, or imagining some movement or change of
position which brought the object into view (Neisser, 1972b). Thus, Neisser and
Kerr's experiment cannot be accepted as a valid comparison of 'pictorial' and
'concealed' imagery, although their results supported the general idea of the
mnemonic effectiveness of interactive imagery.
Bower (I 970b) argued that the effects of interactive and separative imagery
permitted one to discriminate between two general classes of explanation for the
efficacy of mental imagery in human memory: 'One class of explanations
supposes that the imagery effect is due to some benefit regarding the
differentiation of the individual elements in the pairs; the other class of
explanations attributes the effect to the element-to-element association process
itself.' He suggested that the poor performance of subjects who used separative
imagery showed that the benefit of mental imagery was in the formation of
relational associations rather than in the encoding of individual items. Bower
also found that interactive imagery instructions enhanced paired-associate
learning without increasing the subjects' performance on a test of stimulus
recognition, which supported his conclusion. (Of course, other experiments have
found effects of imagery mnemonic instructions in tests of recognition memory,
as was mentioned above. These findings indicate either that interactive imagery
instructions also affect the encoding of individual stimulus items, or that
associative retrieval processes operate in tests ofrecognition memory.)
It is of course well known that subjective organisation is important in verbal
learning, and that procedures to increase that organisation which are merely
based upon verbal categorisation may also lead to substantial improvements in
recall performance (for example, Mandler, 1967). Consequently, Bower's
argument tends to undermine the idea, central to dual coding theory, that the
use of mental imagery leads to a qualitatively different mnemonic represen-
tation. Bower ( 197ob, 1972) continued this line by pointing to parallels between
the findings obtained in comparing interactive and separative imagery, and
those obtained in the recall of pictures and in the recall of pairs of words. In the
previous chapter, I mentioned experiments which demonstrated that two
pictured objects are more easily remembered if they are shown in some sort of
interaction. It is also known that the recall of noun pairs is facilitated if they are
embedded in a meaningful sentence and connected by a verb or preposition; this
facilitation does not occur if the sentence is anomalous (in which case
performance may actually be reduced) or if the two nouns are merely connected
by a conjunction (Bobrow and Bower, 196g; Epstein, Rock and Zuckerman,
1960; Rohwer, 1966). On the basis of the similarity among these experimental
Mental Imagery and Human Memory
results, Bower concluded that 'this recall pattern with pictures, images, and
words is probably being produced by the same relational generating system' (see
also Anderson and Bower, I973, p. 457).
More recently, this approach has been developed by Begg (I 978), who studied
cued recall, free recall and recognition memory for pairs or triads of nouns
learned under interaction or separation instructions. Although cued recall
performance was substantially better with interactive imagery than with
separative imagery, this was not so in the case of recognition memory, and was
not so in the case of free recall unless the mnemonic instructions encouraged
interunit organisation as well as intraunit organisation. Begg concluded that the
results merely reflect general principles of mnemonic organisation, rather than
any specific properties of mental imagery. Begg andYoung (I 977) compared the
effects of interactive imagery and type of connective in the retention of pairs of
nouns. As mentioned above, verbs enhance organisation and recall more than
conjunctions. (Begg and Young demonstrated further that prepositions which
imply spatial contiguity, such as in or on, produce this facilitation, whereas
prepositions which merely suggest spatial proximity, such as near or by, do not.)
Interactive imagery instructions were found to remove thisform class effect by
selectively increasing performance on pairs connected by conjunctions; on the
other hand, separative imagery instructions were found to remove the effect by
selectively reducing performance on pairs connected by verbs. Once again, Begg
and Young took the attitude that imagery mnemonic instructions were merely
one way of manipulating the likelihood that pairs of items will be jointly encoded
in memory.
The results which have just been described are not formally inconsistent with
dual coding theory. As Paivio (I 972) pointed out, this position entails that
mnemonic instructions may lead to increased performance either by providing an
alternative memory code (the coding redundancy hypothesis) or by increasing
the organisation of the material to be remembered within one or both
representations. Nevertheless, if the effects of imagery instructions could be
explained by a theory which posits only a single form of coding in memory, then
on the grounds of parsimony this latter position should be adopted in preference
to one which posits two or more representations. Conversely, more evidence on
the effects of imagery mnemonic instructions is required if dual coding theory is
to be given a sound empirical basis.
One source of evidence involves the comparison of imagery instructions and
other mnemonic devices which do not explicitly mention the use of mental
imagery. Just as the presentation of a pair of nouns in a meaningful phrase or
sentence may lead to improved recall, substantial improvements in performance
are also obtained when the subjects themselves are asked to make up a sentence
Imagery Mnemonic Instructions 77
linking two nouns (Bower, I972). Instructions to generate such verbal mediators
may be just as effective in increasing recall as instructions to use interactive
mental imagery (Janssen, I976, p. 42). These findings might be interpreted as
further evidence for the idea that constructing mental images and generating
sentences are two forms of cognitive processing operating on a common
representational system in memory. The stronger idea that this system might be
verbal or propositional in nature was considered by Bower (I97oa):
However, Bower himself rejected this 'strict verbal hypothesis'. On the one hand,
his own experiments had shown that the effects of instructions to use interactive
mental imagery are sometimes greater than those of instructions to generate
linking sentences; however, he agreed that the differences are typically not very
large, and so there may still be considerable overlap between the cognitive
processes aroused by the two procedures. On the other hand, it could be argued
that the overlap arises not because imagery instructions enhance the organis-
ation of material within a common, propositional representation, but because
sentence generation itself depends upon the use of mental imagery (cf. Paivio,
I97Ib). In support of this idea, Bower (I972) mentioned that his subjects
spontaneously reported the experience of imaging the scene described by a
generated sentence (but cf. Bugelski, I 97 I).
Another source of evidence is an extension of the original approach used by
Paivioand Csapo ( I973) which was mentioned in the previous chapter. It will be
recalled that these experimenters attempted to determine whether repeated
presentations of pictures and words created new, independent representations in
memory, or whether they merely served to strengthen a single, original
pr.esentation. Paivio and Csapo carried out further experiments using homogen-
eous lists of either pictures or words, where the subjects were either to construct a
mental image of the object pictured or named, or to pronounce its name to
themselves. Particular items could be presented once or twice, and they could be
imaged or pronounced at each presentation. An unanticipated recall test
produced an analogous pattern of results to that obtained in the original
experiment. Not surprisingly, repeated presentations produced an increase in
recall performance, and imaging produced better performance than pronounc-
ing. Repeated imaging and repeated pronouncing produced poorer perform-
ance than would be expected on the basis of stochastically independent effects,
but imaging and pronouncing the same item conformed with the assumption of
additive effects. These findings were replicated and extended by Paivio (I975a),
and were taken to support the hypothesis of functionally independent imaginal
Mental Imagery and Human Memory
These are two different ways of investigating mental imagery, and there is no
guarantee that they will be affected by a concurrent interfering task in the same
manner. Atwood's experiment provides no way of testing the specific hypothesis
that a concurrent task will reduce the beneficial effect of imagery mnemonic
instructions. (The relationship between such tasks and the effect of stimulus
concreteness will be considered in the following chapter.) Second, the experi-
mental material in Atwood's study was somewhat bizarre {for example, 'nudist
devouring a bird'). Third, and most crucial, attempts to replicate Atwood's
experiment have continually proved unsuccessful (Baddeley, Grant, Wight and
Thomson, 1974; Bower, Munoz and Arnold, cited by Anderson and Bower,
1973, p. 459; Brooks, cited by Paivio, 1971c, p. 374; Quinn, cited by Baddeley,
1976, p. 229).
A series of experiments was reported by Janssen ( 1976) in an attempt to clarify
the situation. (In this chapter, only the results bearing upon the effects of
imagery mnemonic instructions will be discussed.) The first experiment
concerned the intentional learning of paired associates under imagery medi-
ation, verbal mediation and rote rehearsal instructions with Atwood's auditory
and visual tasks and a control condition. The second experiment compared the
incidental and intentional learning of paired associates under imagery and
verbal mediation instructions with the auditory and visual concurrent tasks. The
third experiment concerned the free recall of lists of nouns under separative
imagery, interactive imagery, separative verbal and standard free-recall
instructions with the auditory and verbal tasks and a control condition with no
concurrent activity. A further experiment used paired-associate learning under
imagery and verbal mediation instructions with Atwood's auditory and visual
tasks and a spatial tracking task. In three of these experiments there was a
significant main effect of interference type, but in none of the experiments did the
interaction between type of instructions and type of concurrent interference even
approach statistical significance. Janssen reported one experiment which
investigated the effect of a concurrent task upon the retrieval of paired associates.
This did demonstrate a significant interaction in the expected direction, such
that subjects using imaginal mediators were more disrupted by a visual
concurrent task, whereas subjects using verbal mediators were more disrupted
by an auditory concurrent task. This result is quite different from those of the
other experiments, and clearly stands in need of replication. Given the excellent
agreement amongJanssen's other four studies and previous attempts to replicate
Atwood's (1971) findings, the safest conclusion at the present time seems to be
that concurrent visual or spatial tasks do not normally have any selective effect
upon the use of mental imagery in human learning.
In chapter 4, I mentioned some unpublished experiments by Baddeley and
Lieberman on the disruptive effects of concurrent tasks in immediate memory.
The same series of experiments included studies of the effects of concurrent
pursuit tracking upon performance in long-term memory (Baddeley, 1976,
pp. 230-2). In the first experiment, subjects who were iilstructed to use rote
So Mental Imagery and Human Memory
rehearsal were compared with those who were taught the 'one-bun' mnemonic.
This technique consists of a nonsense rhyme, 'one is a bun, two is a shoe', and so
on, and is used to learn a sequence of items by constructing a compound image
relating the first object and a bun, another image relating the second object and a
shoe, and so on. Baddeley and Lieberman found that the advantage of using the
mnemonic was slightly reduced when the subjects carried out the concurrent
pursuit tracking task. However, a further experiment compared subjects using
rote rehearsal with those who were taught a location mnemonic. In this
technique, the subject learns a sequence of items by constructing an image
locating each object at a particular location on a well known route (for example,
a walk around the university campus). Baddeley and Lieberman found that the
usefulness of this mnemonic was drastically reduced when the subjects carried
out the concurrent pursuit tracking task.
On the basis of these results, Baddeley ( 1976, p. 232) concluded that 'it is
possible to interfere with the control process involved in forming, manipulating,
and utilizing images'. However, this does not explain why concurrent tracking
did not greatly disrupt the use of the 'one-bun' mnemonic, and it ignores
Janssen's ( 1976) continual failure to show any selective interference with the use
of mental imagery in learning, either when compared with the use of verbal
mediators, or when compared with the use of rote repetition. These latter results
indicate that concurrent spatial tasks do not interfere with the primary function
of mental imagery which has been identified in this chapter, the elaboration and
strengthening of associative connections in long-term memory. On the other
hand, it is reasonable to suppose that such concurrent tasks will interfere with
other constructive, manipulative or spatial tasks in short-term, non-verbal,
working memory. The location mnemonic includes a component which seems to
fall within this latter category, namely the manipulation of a pictured
geographical representation, and it is therefore not surprising that it is disrupted
by concurrent tracking. However, the 'one-bun' system merely involves the use
of a well learned set of phonemic associations, and certainly does not place any
demands upon the form of working memory defined above. It is thus equally
unsurprising that concurrent spatial tasks do not interfere drastically with the use
of this mnemonic, nor with the purely associative strategies which were used by
Janssen's ( 1976) subjects. It is of course an implication of this analysis that the
constructive and elaborative uses of mental imagery which were described in
chapter 4 are functionally independent of one another.
The Method has such a circumscribed range of utility that it is useless for all
practical purposes oflearning and remembering. Appropriate task conditions
are essential. The presented items must be translatable into imaged objects;
the rate of presentation must be slow, not faster than about one item every
three seconds; and the Method breaks down if we try to comprehend
relationships between items that are not strictly contiguous. These task
conditions rarely arise in real life ....
Hunter went on to quote with approval Francis Bacon's remark that the
mnemonic is 'not dextrous to be applied to the serious use of business and
occasions'. Similarly, Morris ( 1977) suggested that imagery mnemonics are
really only of value in learning lists of unrelated, concrete words. Of course, even
if mnemonic techniques based upon mental imagery are entirely lacking in
general practical utility, this cannot be blamed upon the efforts of psychologists.
Such techniques have been devised by ordinary human beings for use in specific
mundane activities. Their limitations merely reflect the superficial understand-
Mental imagery and Human Memory
ing which ordinary folk have of their own cognitive faculties. Nevertheless,
psychologists can be blamed for having only recently paid attention to the
possible implications of mnemonic techniques for psychological theories of those
faculties, and for having so far failed to offer more effective devices for improving
memory in real-life situations.
6.6 CONCLUSIONS
While it may be true that mnemonic techniques based upon mental imagery are
of only limited value in assisting learning in everyday life, it does not follow that
the study of mental imagery will only be of limited value in understanding
memory function. This latter point has to be decided by an informed and critical
appreciation of the available experimental research.
The evidence which I have reviewed in this chapter indicates that under
laboratory conditions instructions to use mental imagery may lead to substantial
improvements in memory performance. The comparison of interaction and
separation instructions shows that these effects come about by means ofincreased
organisation of the material to be remembered. More generally, the effects of
imagery mnemonic instructions are entirely analogous to those of verbal
mediation instructions, and the resulting increases in recall performance are
typically comparable in the two cases. The study of repetition effects in
incidental recall does not permit one to specify the nature of the mnemonic
representation produced by mental imagery, and the effects of concurrent spatial
tasks upon memory performance are largely independent of the effects of
imagery mnemonic instructions, unless the mnemonic technique itself contains a
spatial, manipulative component.
The empirical findings on the effects of imagery mnemonic instructions do not
permit one to distinguish between dual coding theory and common coding
theory. Indeed, both approaches tend to interpret the effects in terms of
enhanced relational organisation. However, precisely because both accounts can
readily handle the available evidence, considerations of parsimony must favour
common coding theory: it is simply unnecessary to postulate an additional,
imaginal system of mnemonic representations. A propositional system of
representations is necessary to handle the independent corpus of evidence on
sentential mediators and the form class effect, and this approach can readily
encompass the effects of instructions to use mental imagery without serious
modification.
Nevertheless, once again, the property of intentionality compels one to
reconsider the implications of the empirical evidence. A subject who constructs a
mental image in response to the appropriate mnemonic instructions does so
under an intentional description which must be conceived as an abstract
propositional representation of the content or object of the image. Such
propositions also underlie other mediating devices which might be employed,
Imagery Mnemonic Instructions
An experimenter is at perfect liberty to define the ease with which different words
evoke mental imagery on a purely personal, intuitive basis, but if he wishes to
ensure that his deijnition has any generality, and that his intuitions are similar to
those of his experimental subjects, he must consult them in order to validate his
selection of experimental material. In contemporary research, this is usually
carried out with the assistance of questionnaires administered to large groups of
subjects, in which evoked mental imagery is scored on a seven-point rating scale.
The first large-scale study of this nature was reported by Paivio, Yuille and
Madigan (rg68), who collected ratings on 925 English nouns. The instructions
given to their subjects were as follows:
One set of evidence on this question comes from the experiment by N eisser and
Kerr (I973), which was described in detail in the previous chapter. It will be
remembered that this study compared the relative effectiveness of separative
imagery, 'pictorial' interactive imagery and 'concealed' interactive imagery
(where the relevant interactive relationship could not have been represented in
any ordinary two-dimensional picture). The experiment was an incidental
learning task in which the orienting task required the subjects to rate the
vividness of the mental imagery evoked by each sentence on a seven-point scale.
The vividness ratings indicated that the 'pictorial' images were better than the
other two types, which were essentially alike: Nevertheless, as was mentioned in
chapter 6, the 'concealed' images produced better recall performance than the
separative images, and were not significantly different from the 'pictorial'
images. Thus, the usefulness of mental imagery appears to be determined by its
relational organisation, and not by its vividness.
Further evidence comes from two experiments which Richardson (in
press) carried out as extensions of a study mentioned above by Bower
(I 972). In each case the subjects rated the imageability of the stimulus material
in response to instructions similar to those ofPaivio, Yuille and Madigan (I 968),
and then received an unanticipated retention test. When the subjects rated
individual nouns, and the test was one of free recall, the average imageability
rating given to each item did not show a significant correlation with the number
of subjects who recalled that item (r = +o.28). However, when the subjects
rated pairs of nouns on the ease with which they evoked an interactive mental
image, and one member of each pair was given as the cue to the recall of the other
member of that pair, there was substantial correlation between the average
imageability rating given to each pair and the number of subjects who recalled
that pair (r = +o.67). These experiments suggest that it is the ease with which a
stimulus item evokes an integrative mental image which is the critical property
determining how easily it will be remembered.
The question arises as to whether it is really the case that material which is of
high imageability is easily remembered because it evokes mental imagery when
presented for learning. This is by no means a trivial question: the fact that verbal
material may display considerable variation both in ratings of evoked mental
imagery and in the ease with which it can be recalled in no way guarantees that
evoked mental imagery is causally responsible for improved retention.
Additional evidence is necessary to substantiate this latter conclusion (cf.
Anderson, I978; Neisser, I972b)
The most important source of evidence is the reports given by the subjects
themselves as to the sorts of devices or mediators used in learning experiments.
First, the frequency of reported use of mental imagery is substantially greater in
the case of material which is of high umageability (Paivio, Smythe and Yuille,
I968; Richardson, I978e; cf. Neisser, 1972b). Second, the vividness of reported
imagery both at presentation and recall is greater in the case of material of high
imageability, and the effect at the time of recall is greater if the retention test is
Remembering Individual Words
This idea is not really a new one. As Marks (I972) mentioned, Bartlett (I932)
attributed most of the inventions and importations found in repeated reproduc-
tions of prose passages to visual imagery. However, it is only recently that there
has been any sound evidence for the proposal. Bower (I 972) compared subjects
using interactive imagery instructions and subjects using rote repetition in
learning a paired-associate list containing unique concrete nouns and another
paired-associate list containing synonym pairs. The subjects using mental
imagery produced poorer performance on the synonym list than on the unique
list, but the subjects using rote rehearsal showed no appreciate drop in
performance. Moreover, 3I% of the errors made by the subjects using mental
imagery were synonym intrusions, whereas the corresponding figure for the
subjects using rote rehearsal was only 2%.
Similar results have been found in studies of the effects of stimulus
imageability. Thus, material of high imageability produces more intrusion errors
and fewer omission errors than material oflow imageability (Davies and Proctor,
I 976; Postman and Burns, I 973). Yuille (I 973) showed that the increase in
synonym intrusions in recall with material of high imageability or following
instructions to use mental imagery was not due to a more rapid degradation of
the relevant mediator, and therefore must be attributed to decoding errors
(Paivio, I975b). On the other hand, Bower (I970a) used a similar basis for
arguing that subjects may spontaneously use mental imagery even to remember
items oflow imageability. His evidence for this was that his subjects sometimes
made semantic intrusion errors with such material which took the form of
associated concrete words (for example, church instead of religion). He argued:
'Such errors are entirely understandable in terms of the person's forgetting the
linguistic tag on his concrete image, which tag was supposed to remind him to
name the abstract concept, not the concrete exemplar.' However, the mere
existence of such intrusion errors is at best suggestive of this sort of process.
One would prefer to have quantitative data showing that the proportion of
concrete intrusion errors given for abstract items was greater than the proportion
of abstract intrusion errors given for concrete items. Even this would be far from
conclusive, since such results are also consistent with the idea that the use of
mental imagery involves a 'deeper' level of coding or processing in long-term
memory, and leads to a higher incidence of synonym intrusions merely because it
makes more likely the registration of information in a semantic or propositional
Remembering Individual Words 89
would be the effective stimulus attribute even with material of high imageability.
An experiment was carried out to test these predictions, using a test offree recall
with individual nouns with imageability and concreteness ratings greater than
6 .o (Richardson, I978a). The partial correlation between imageability and
performance (even with this restricted range) was +o.4I, but the partial
correlation between concreteness and performance was -o.I3.
These results clearly demonstrated that the imageability of stimulus material,
rather than its concreteness, is the effective attribute determining how easily it
can be remembered. They are thus quite consistent with the conventional
position on the relationship between concreteness and imageability, but are
clearly inconsistent with the proposals of Richardson's ( I975a) earlier paper.
Accordingly, while it may be of interest to distinguish between concreteness and
imageability, both conceptually and theoretically, this recent experiment
suggests that conventional accounts, which identify stimulus imageability as the
primary determiner of recall performance, are likely to be essentially correct.
A rather different problem arises over the possibility that certain words have a
complex internal structure, and that this dimension of linguistic complexity is
correlated and thus confounded with the image-evoking potential of individual
words. This problem was first raised by Kintsch ( I972a, I972b), who pointed out
that many words commonly referred to as 'abstract' may also be regarded as
syntactically complex, whereas 'concrete' items may be regarded as relatively
simple. The complexity with which Kintsch was concerned is that which has
been ascribed to derived nouns, such as refusal, destruction and eagerness. At one
time, many linguists regarded these nouns as resulting from the application of
nominalising transformations to their stems or base forms, refuse, destroy and eager
(Chomsky, I965, pp. I84--92; Fowler, I97I, pp. I29-3I; Katz and Postal, I964,
pp. I23-4; Lees, I96o, pp. 64--9). (More accurately, sentences containing such
nouns were considered to result from deep structures containing the correspond-
ing base forms, together with some nominalising morpheme.) The idea that
derived nouns are more complex than their stems is intuitively plausible, and
suggests that they should be more difficult to deal with in a wide variety of
psychological tasks than simple nouns which are not produced in this manner.
Kintsch (1972a) carried out three experiments to examine the relative
importance ofimageability and lexical complexity in paired-associate learning.
The results indicated that the two variables had roughly equal effects, although
one experiment produced a complex interaction with word frequency. However,
Richardson ( 1975a, 1975') suggested that there were various respects in which
Kintsch's experiments were unsatisfactory, and he carried out further studies in
which the variables of imageability, concreteness and lexical complexity were
manipulated in an orthogonal fashion in free recall. These showed that lexical
94 Mental Imagery and Human Memory
complexity had no effect upon recall performance when the imageability and the
concreteness of the stimulus material were both controlled. That is, derived
nouns are not more difficult to remember than simple nouns, when one allows for
the fact that they may be more abstract and less imageable.
These results are consistent with the more recent linguistic account of lexical
derivation, according to which derived nouns and their stems are merely
alternative surface forms of the same, neutral lexical entry (Chomsky, 1970); this
would not predict that derived nouns were more difficult to remember.
However, the results are also consistent with the view that derived nouns and
their stems are merely distinct lexical items, andj. T. E. Richardson ( 1977h) has
argued for this conclusion on the basis of an extensive series of experimental
studies. On this account the intuition of a relationship between a derived noun
and its base form should be ascribed to diachronic considerations (pertaining to
the historical development of the language), rather than to synchronic
considerations (pertaining to a correct account of linguistic competence).
Accordingly, the distinction between simple and derived nouns is not relevant to
an understanding of the processes operating in learning tasks.
material, but not in the case of concrete material, and instructions to motivate
the subjects to learn accentuate this effect (Sheehan, I 972). The clear
discrepancy between these results and the effects of imagery mnemonic
instructions constitutes further, excellent evidence that such instructions do not
merely affect performance by increasing the subjects' motivation (see chapter 6).
However, if one considers experiments in which interactive imagery instruc-
tions are compared with verbal mediation instructions, the findings are rather
different. The two types of instructions lead to similar improvements in
performance in the case of concrete items, and have no effect upon the recall of
abstract items. (Equivalently, the effect of stimulus imageability upon recall
performance is roughly the same under both imagery and verbal mediation
instructions.) This has been found principally in paired-associate learning
(Janssen, I976, pp. 42-3, 48, 53; Paivio and Yuille, I967; Wood, I967; Yuille
and Paivio, I968), but also in free recall (Janssen, I976, pp. S<r-I). However, the
different mnemonic instructions seem to be only partially effective in influencing
the subjects' learning strategies. Both subjects receiving imagery instructions and
those receiving verbal instructions report a greater use of mental imagery with
concrete material than with abstract material, and a greater use of mental
imagery with repeated presentations of the same paired-associate list (Paivio and
Yuille, I96g). These results led Paivio (I97IC, p. 362) to conclude 'that
associative strategies are only partly controlled by experimental sets and that,
over trials, subjects increasingly revert to associative habits aroused by the
semantic characteristics of the to-be-learned items'.
In order to resolve these difficulties, Paivio and F oth (I 970) compared the
effects of imagery and verbal mnemonic instructions with concrete and abstract
material when the subjects were forced to specifY the mediator employed in
learning each paired associate. Different groups of subjects learned concrete or
abstract paired associates, but each subject used both types of mediators with
different items. Under instructions to use mental imagery, the subjects were
required to construct an interactive mental image and to give a sketch drawing of
the scene or event depicted. Under instructions to use verbal mediators, they
were required to generate a linking phrase or sentence and to write it down. The
results showed a significant interaction such that imagery instructions produced
better performance than verbal instructions in the case of concrete pairs, but
worse performance in the case of abstract pairs. (Equivalently, the effect of
stimulus imageability upon recall performance was greater with interactive
imagery instructions than with verbal mediation instructions.) These results are
of course entirely consistent with dual coding theory, and with the theoretical
analysis of the effect of imagery mnemonic instructions which was given by
Paivio ( I97 I c) in the passage quoted above. Nevertheless, the procedure used by
Paivio and Foth confounded the mnemonic efficacy of constructing an
interactive mental image with that of drawing an interactive picture (Janssen,
I976, p. 42); since the subjects instructed to use mental imagery were also
required to generate their own pictorial presentation of the material to be
Remembering Individual Words 97
Once again, one may examine the forms of coding or representation used in
remembering concrete and abstract words by considering which sorts of
concurrent tasks will interfere with recall performance. In this case, the
particular problem of interest is that of discovering concurrent tasks which
interfere in a selective manner with the beneficial effect of stimulus imageability.
Such a discovery would count as evidence for the functional independence of the
two mnemonic representations postulated by dual coding theory (Paivio,
1975b).
As was mentioned in chapter 6, Paivio ( 1975d) has pointed out that, in a sense,
increasing the presentation rate may be regarded as a way of selectively
interfering with cognitive processing. The fact that the effect of stimulus
imageability upon performance is greatly attentuated with very fast rates of
98 Mental Imagery and Human Memory
presentation (Paivio and Csapo, I96g; see also Paivio, I975b) may therefore be
taken as evidence that the presentation of concrete material at slower rates of
presentation gives rise to additional forms of representation in memory (see
Paivio, I972, for discussion). So, the interaction of the effect of presentation rate
and the effect of stimulus imageability upon recall performance provides (very
limited) evidence for dual coding theory.
The first empirical study to investigate this problem by requiring the subjects to
carry out a clearly defined concurrent task was that ofAtwood (I 97 I), which was
described in detail in the previous chapter. This study found that a concurrent
visual task interfered more with the retention ofconcrete material under imagery
mnemonic instructions than with the retention of abstract material under verbal
instructions, but that a concurrent auditory task had the opposite effect. On the
other hand, there were various problems with this study, the most important of
which was the complete confounding of the manipulation of mnemonic
instructions and the manipulation of stimulus imageability (Janssen, I976,
pp. 3I-2). Attempts to replicate Atwood's experiment have been unsuccessful
(Bower, Munoz and Arnold, cited by Anderson and Bower, I973, p. 459; Brooks,
cited by Paivio, I97IC, p. 374; Quinn, cited by Baddeley, I976, p. 229), and
several studies which have looked more carefully at the effect of stimulus
imageability have failed to find any interaction with the effect of a spatial
concurrent task (Baddeley, Grant, Wight and Thomson, I974; Byrne, I974;
Elliott, I973; Warren, I977)· Although Janssen (I976, PP· 4I-3, 47, 5I-7) did
find evidence for the predicted interaction, most authorities consider that
concurrent visuo-spatial tasks do not have a selective effect upon the recall of
concrete material as opposed to that of abstract material (Anderson and Bower,
I973, p. 459; Baddeley, I976, pp. 229-32).
This conclusion entails that the cognitive processes which are responsible for
the effect of stimulus imageability upon recall performance are functionally
independent of those which are disrupted by a concurrent visuo-spatial task.
Janssen ( I976, pp. 33-4) and Baddeley ( I976, pp. 230, 232; Baddeleyet al., I974)
have attempted to articulate this point in a more specific manner. Janssen
argued that whether two tasks interfere with one another depended upon
whether they both employ mental imagery in a spatial or a non-spatial manner.
He suggested that the concurrent tasks which are typically employed have a
strong spatial component, whereas the use of mental imagery in long-term
memory does not. However, the experiment by Neisser and Kerr ( I973) shows
that effective relational associations in long-term memory may indeed be based
upon spatial organisation.
Baddeley's argument was that 'it is possible to interfere with the control
process involved in forming, manipulating, and utilizing images, though not
with the more basic semantic characteristics which cause subjects to rate a word
as being concrete or imageable'. He distinguished between the optional strategy
or control process of mental imagery, which 'represents an activity of the subject
and may thus be disrupted by a concurrent visuo-spatial task', and the
Remembering Individual Words 99
concreteness of the stimulus material, which 'depends on the way in which the
material has already been registered and thus is not dependent on the current
activities of the subject'. Baddeley's discussion implies that these two cognitive
modes are functionally independent. However, it does not explain why the effect
of a concurrent visuo-spatial task typically does not interact with that of imagery
mnemonic instructions (which presumably implicate the 'control process' of
mental imagery); nor does it explain why the effect of stimulus imageability
typically does interact with that of imagery mnemonic instructions.
The interpretation of stimulus imageability which is suggested by the research
to date is that of a measure of the ease with which experimental material evokes
integrated, relationally organised images. The research on selective interference
suggests that concurrent spatial tasks do not interfere with the elaboration and
strengthening of such associative connections in long-term memory. The
theoretical dichotomy which Janssen and Baddeley were (unsuccessfully)
attempting to delimit appears to be that which was defined in chapter 4 between
the constructive and elaborative uses of mental imagery. Once again, the
implication of the empirical evidence on selective interference and stimulus
imageability is that these two sorts of use are functionally independent of one
another.
7.6 CONCLUSIONS
The research which I described in the previous chapter indicated that stimulus
imageability was an important factor determining a person's ability to
remember individual, unrelated words. However, it is important to remember
that·human beings rarely encounter lists of unrelated words outside the confines
of the psychological laboratory. If this research is to have any relevance to
remembering in everyday life, similar findings must be demonstrated in more
realistic situations. More generally, the role of mental imagery in retaining
connected material must be carefully determined. Although this is the most
important task in any area of experimental research, the specific question of the
relevance of mental imagery to everyday remembering has really received no
clearly articulated answer. Nevertheless, in this chapter, I shall try at least to
sketch out an answer by considering evidence on the function of mental imagery
in the retention of narrative. To do this, I shall consider findings obtained with
simple noun phrases, with individual sentences, and finally with connected
discourse.
In recent years there has been a considerable amount of research on how people
remember individual sentences, and the results can be summarised fairly easily
Mental Imagery and Human Memory
(see, for example, Johnson-Laird, I970, I974). First, under most circumstances,
subjects demonstrate a retention of the precise wording of a passage only for the last
sentence presented, and possibly only for the last clause presented Uarvella, I970;
Jarvella and Herman, I972; Sachs, I967; but cf. Kintsch and Bates, I977). Even
this ability may be limited in tests of incidental learning Uohnson-Laird
and Stevenson, I970). Second, subjects' retention of the meaning of a passage is
typically good, and may be maintained despite considerable interpolated
information (Begg, I97I; Sachs, I967). A corollary of this is that the incidence of
semantic confusions is relatively high in both recognition and recall (Bransford,
Barclay and Franks, I972; Fillenbaum, I973; Johnson-Laird and Stevenson,
I 970; Sachs, I 967). These findings imply that the verbatim expression of a
sentence is lost once a semantic interpretation has been achieved, and must be
reconstructed on the basis of that interpretation at the time of recall (Bregman
and Strasberg, I968; Sachs, I967).
The most important immediate question is whether mental imagery is
involved in the storage and retrieval of such semantic interpretations. One sort of
evidence comes from studies of cued recall analogous to those mentioned in the
previous section. Several studies have found that the efficacy of a word in a
sentence as a prompt for the recall of the entire sentence appears to depend upon
the syntactic function of the word within that sentence. For example, the final
noun in a passive sentence was found to be a more effective cue when it
functioned as the logical subject of the sentence (as in Gloves were made by tailors),
than when it was merely contained in an adverbial phrase (as in Gloves were made
by hand) (Blumenthal, I967); an adjective which modified a whole sentence or
proposition embedded within the sentence to be remembered (as in John is easy to
please) was found to be a more effective cue than an adjective which merely
modified a noun within that sentence (as in John is eager to please) (Blumental and
Boakes, I 967); and a noun which functioned as the grammatical subject of a
sentence was found to be a more effective cue than a noun which functioned as
the grammatical object of the sentence (Horowitz and Prytulak, I969).
However, it now appears likely that these effects were produced by a
confounding of stimulus imageability: as in the case of connected phrases, when
the imagcability of a word in a sentence is controlled, its grammatical status
seems to be unimportant in determining performance (James, I972).
The critical pr_edictions to be made on the basis of dual coding theory depend
upon the mnemonic representations which are assumed to be available for
encoding concrete and abstract sentences. Paivio ( 197 I b; see also 197 Ic, pp. 45o-
I) stated his position in the following way:
Extended to connected discourse, the dual process model implies that concrete
phrases or sentences, like concrete words, can be coded and stored in memory,
not only verbally, but also in the form of nonverbal imagery .... Although
high imagery words or paired associates are generally easier to learn and
remember than low imagery ones, the model does not imply that high imagery
Remembering Connected Narrative
The first two points made by Moeser seem to be largely correct, and constitute
a firm rebuttal ofBegg and Paivio's position. Nevertheless, her third suggestion is
Remembering Connected Narrative I07
not sound. Her 'analog semantic coding model' is sketched out only in a very
vague manner, and it is not possible to see clearly how it differs from standard
propositional models. Moreover, even when her paper was written, some
propositional theorists had already incorporated into their models the explicit
assumption of a more favourable mnemonic representation for imageable
material (Anderson and Bower, I973, p. 452). This idea is typically handled
either by assuming that concrete sentences lead to more distinctive represen-
tations or that they are encoded in more closely interconnected sets of
propositions. Thus, it must be concluded that Moeser's results are entirely
consistent even with the specific proposals which had been made by pro-
positional theorists. Conversely, however, her results provide no clear support for
the dual coding position, and are actually inconsistent with the interpretation of
that position offered by Begg and Paivio.
Further evidence on the memory representations available for concrete and
abstract sentences comes from the use of interference tasks designed to interfere
with a particular form of coding in a relatively specific manner. In previous
chapters, it has been observed that Atwood's (I97I) original findings obtained
with concrete and abstract phrases have not proved to be replicable, and that
visuo-spatial tasks usually have equal effects upon the recall of both sorts of
material. A series of experiments carried out by Sasson ( I97 I; Sasson and Fraisse,
I 972) investigated the effect of interpolated learning upon the recall of concrete
and abstract sentences. The first two experiments required subjects to recall
concrete sentences under conditions of negative transfer or retroactive inter-
ference (from the learning of unrelated pictures or concrete sentences inter-
polated between the presentation and recall of the critical material) and under
conditions of positive transfer (from the learning of interpolated pictures and
sentences . which respectively depicted and were identical to the critical
material). Interpolated pictures and concrete sentences were found to produce
similar amounts of positive and negative transfer, and Sasson concluded that
sentences were stored and retrieved primarily in the form of visual images. In a
third experiment, the subjects were required to recall abstract sentences under
conditions of retroactive interference from unrelated pictures, unrelated con-
crete sentences or unrelated abstract sentences. The interpolated abstract
sentences produced substantial negative transfer, but the interpolated pictures
and concrete sentences did not. Sasson and Fraisse concluded that abstract
sentences were stored and retrieved primarily in a verbal form.
However, there are two major difficulties which prevent one from accepting
Sasson's conclusions. First, all of the pictures employed in his experiments
depicted events or scenes which could be readily described even under the rapid
presentation conditions used. In fact, the pictures were outline drawings
produced 'such that each drawing would exactly depict the event described' by
one of the concrete sentences. That is, each picture was a symbolic representation
constructed by an artist under an intentional description which was exactly
contained within one of the simple affirmative concrete sentences. The finding
ro8 Mental Imagery and Human Memory
that pictures and concrete sentences have equal effects upon the recall of
concrete and abstract sentences shows at most that pictures and concrete
sentences have similar mnemonic representations; it does not specify whether
those representations are pictorial or propositional in nature, and so does not
distinguish between dual coding theory and common coding theory.
The second problem is that the abstract sentences employed in the third
experiment were rather obscure or bizarre, and had to be interpreted
metaphorically (for example, 'Religion teaches charity'). Under the presen-
tation conditions which Sasson and Fraisse employed, it is possible that most of
the time subjects simply did not grasp the meanings of the sentences to be
learned. The finding that interpolated abstract sentences impair the recall of the
original abstract sentences, whereas interpolated pictures or concrete sentences
do not, shows at most that pictures and concrete sentences have a different
mnemonic representation from abstract sentences; it does not specify whether
those representations are pictorial and propositional, pictorial and verbatim, or
propositional and verbatim. Two features of their results give strong support to
the idea tha the abstract sentences were only remembered in terms of their
verbatim expression. On the one hand, interpolated concrete sentences had no
effect upon the recall of abstract sentences, and yet they had presumably been
encoded in terms of meanings, and this should have interfered with any retention
of the semantic content of the abstract sentences. On the other hand, the
retention of the abstract sentences was essentially non-existent after a delay of
two days. (The performance in delayed recall was less than s% even in a control
condition with no interpolated learning.)
Further experiments on selective interference were reported by Davies and
Proctor ( 1 976), who used a dis tractor task which involved verbal coding
(counting backwards) and one which involved perceptual coding (visual
tracking). The recall of abstract sentences was more impaired by the verbal task,
whereas the recall of concrete sentences was more impaired by the perpetual
task. Errors with the abstract sentences tended to be omissions, whereas errors
with the concrete sentences tended to be intrusions. Moreover, a greater
proportion of the intrusion errors obtained with concrete sentences involved a
change of meaning from the original following the perceptual task. These results
indicated that qualitatively different encoding processes operate in the retention
of concrete and abstract sentences. However, the fact that a visual tracking task
was found to have a selective effect upon the recall of concrete material is
inconsistent with most of the other research which has been carried out on
selective interference effects using phrases and unrelated words (chapter 7).
Inspection of their abstract material suggests that these sentences were rather
obscure or opaque, and so a failure to control the intelligibility of the stimulus
material might have been a factor influencing the results.
Remembering Connected Narrative 109
surely agree that many abstract sentences can be understood and remembered
without recourse to mental imagery (Johnson-Laird, I974), and yet one would
be extremely unhappy about describing that understanding merely as the
arousal of an 'implicit auditory-motor representation' of such a sentence. Davies
and Proctor ( I976) suggested that even concrete sentences may not arouse
mental imagery if they are tautologies or merely make a statement about the
general nature of the world (for instance, The shoes are a product of the skin of the
cow). Although subjects do report using mental imagery when carrying out
sentence comprehension tasks, Garrod and Trabasso ( I973) found that this was
considerably attenuated when the sentences were presented visually, pre-
sumably a selective interference effect of the sort discussed in chapter 4 (see also
Klee and Eysenck, I973). Moreover, Holmes and Langford ( I976) argued that
response latencies in such tasks are too short to permit mental imagery to have
any effective role. Bugelski (I 977) has claimed that 'to argue that all of meaning
cannot be accounted for by imagery is to attack a straw man', since no one has
ever denied this. But the point is that no dual coding theorist has ever given a
clear account of the meaning of abstract sentences which is at all plausible to
someone who does not subscribe to a rigorous form ofbehaviourism. My personal
inclination is to regard the account given by propositional theorists as essentially
correct as a solution to this latter problem, without prejudice to the question of
the status or the functional utility of mental imagery. But, even in their most
recent publications, dual coding theorists continue to regard propositional
theory as inherently contradictory to their own position, and not as a possible
extension of that position (Bugelski, I 977; Paivio, I 978b).
Nevertheless, if one wishes to relate these positions to the nature of sentential
meaning, ohe must ask: What is it to understand a sentence? There are several
different answers which have been given to this question, and they are associated
with different methods for measuring whether the process of understanding a
sentence is affected by its imageability (cf. Moeser, I 974). These were mentioned
in discussing the experiment by Begg and Paivio (I969) earlier in this chapter,
but it will be useful to consider them once more.
I. One might say that a person had understood a sentence when he had
imposed some interpretation on it. This is the sort of idea which seems to have
been behind Paivio and Begg's (I 97 I) use of a simple latency measure to
investigate comprehensibility. It will be recalled that they found no difference
between concrete and abstract sentences (cf. Ernest and Paivio, I97Ih; Yuille
and Paivio, I 967), though Holmes and Langford (I 976) criticised their measure
as crude and insensitive, and their conclusion as erroneous.
2. One might say that a person had understood a sentence when he had made
the absolute judgement that it was meaningful, and had discriminated it from
other sentences in terms of their meaningfulness. This seems to have been
involved in the study by Johnson et al. (I972), in which the suQjects rated the
sentences on a comprehension scale; and also in the experiments by Klee and
Eysenck (I973) and by Holmes and Langford (I976), which measured the
Remembering Connected Narrative Ill
The most co~mon situation in which a person attempts to retain and to integrate
the information contained in a series of sentences is where he tries to learn a story.
The first systematic investigation of story recall was carried out in the 192os by
Bartlett (1932). He was concerned with understanding the extent to which the
interpretation placed upon a story reflected the reader's social and cultural
background. His best known study employed a North American folk-tale called
The War of the Ghosts. Bartlett expected that the unusual social conventions
embodied in the story and the apparently disconnected narrative would create
considerable difficulties for his British subjects; their attempts at recalling the
story would therefore shed light upon the processes of interpretation which had
been used. Each subject was asked to recall the story as accurately as possible on
several occasions over many months. The version recalled by any given subject
showed considerable omissions and distortions, but it remained relatively fixed
from one test to the next. It remained a single passage, and was more coherent
than the original in terms of the subject's own culturally induced expectations.
Bartlett suggested the term, schema, to refer to the abstract mnemonic
representation of a narrative, distorted so as to be consistent with already existing
schemata, acquired on the basis of past cultural experience.
Although Bartlett's ideas are generally recognised as being very important, his
empirical investigations were not followed up for a considerable time. In the
II2 Mental Imagery and Human Memory
I96os, some psychologists turned their attention to the role of thematic structure
in remembering narrative. One of the original findings in this area was that both
the total verbal recall and the incidence ofsemantic confusions in the retention of
connected discourse are greater when its thematic structure is obvious than when
it is obscure (Pompi and Lachman, I 967). This suggested that such material is
encoded in the form of surrogate processes, general ideas or themes from which the
narrative itself is reconstructed at the time of recall. It was subsequently shown
that the comprehension and recall of an obscure or vague passage would be
increased by supplying an appropriate and relevant picture or title (Bransford
and Johnson, I972; Dooling and Lachman, I97I). Yuille and Paivio (I969)
showed that thematic organisation had a greater effect upon the recall of
concrete passages than upon the recall of abstract passages. On the other hand,
Pezdek and Royer ( I974) showed that the provision of relevant context
paragraphs enhanced the ability to detect changes of meaning in abstract
sentences more than in concrete sentences. These two results are entirely
consistent with one another, if one assumes (I) that the thematic structure of
concrete material is typically obvious, promotes semantic encoding, and does not
stand in need of support from the context; but (2) that the thematic structure of
abstract material is typically obscure, promotes literal encoding, and benefits
from the provision of a context. These results might be taken to be consistent with
the idea that thematic structure in the case of concrete passages permits the
construction of interactive, integrative mental imagery, from which the verbal
content may be reconstructed or redintegrated at the time of recall (Paivio,
I97Ib). However, they are in no way inconsistent with the view that concrete
and abstract passages receive exactly the same sort of semantic representation.
Indeed, these findings are entirely analogous to those obtained with individual
sentences, which were described earlier in this chapter.
These experiments had the disadvantage that the notion of thematic structure
was poorly defined, and there was little experimental control over how it was
introduced. This problem was handled by Bransford and Franks (I97I), who
introduced a novel experimental technique to demonstrate the operation of
Bartlett's schemata. The subjects were asked to listen to component sentences
which together defined a complex idea. For example, the sentences, The ants were
in the kitchen, The jelly was on the table, The jelly was sweet' and The ants ate the jelly,
each connote a single proposition, and jointly define the complex idea, The ants in
the kitchen ate the sweet jelly which was on the table. The acquisition sentences in the
original experiment contained one, two or three of the components in a given
idea. The subjects then received a recognition test with sentences containing one,
two, three or four components. The subjects' confidence that a particular
sentence had been presented for learning varied directly with the number of
components which it contained; in particular, the sentences containing four
components were judged as 'old' with the highest confidence, even though they
had never been presented. A subsequent experiment showed that the subjects'
confidence ratings in such a test were predicted by the number of components
Remembering Connected Narrative I I 3
which a sentence contained, but not at all by whether the sentence had actually
been presented. Bransford and Franks concluded that their subjects had
integrated the information communicated by sets of individual sentences to
construct holistic semantic ideas. These ideas 'need not be communicated by single
sentences. They may result from the integration of information expressed by
many different sentences experienced successively and often nonconsecutively in
time.' Bransford and Franks explicitly related their ideas to the research carried
out by Bartlett, and suggested that their methodology might lend some precision
to the concept of a schema as what is learned when subjects are exposed to
connected discourse.
These original findings were replicated both by Bransford and Franks (I972)
and by other researchers (Cofer, I973; Singer, I973)· Bransford, Barclay and
Franks (I972) demonstrated similar integrative processes operating among the
clauses within a single sentence, and Bransford and Franks (I972) produced
similar findings on the integration of information expressed by sentences within
connected paragraphs. It has been shown that the basic pattern of results may
break down under certain circumstances (Flagg, I976; Pezdek, I978), and they
may also be open to alternative interpretations, at least when material other than
sentences is used (Reitman and Bower, I973). Nevertheless, it is obvious that the
experimental tasks employed by Bartlett and by Bransford and Franks are
considerably more typical of remembering in everyday life than experiments on
the learning of individual sentences, phrases or words; and some process of
abstraction and integration such as is postulated by these writers must be
assumed to operate in the comprehension and retention of connected narrative.
The material which was employed in the original experiments was typically
concrete and easily evoked mental imagery (Franks and Bransford, I972).
Indeed, the experimental subjects spontaneously commented that they had
constructed mental images of the situations described by the sentences to be
remembered (Bransford et al., I972; cf. Anderson and Bower, I973, p. 3I4). It is
therefore interesting to consider whether a holistic semantic idea might consist,
in whole or in part, in a composite, interactive mental image (Neisser, I972b).
Bartlett (I932, pp. 64-5, 22o--4, 303-4) himself considered and rejected the
suggestion that mental imagery might be of importance in the retention of
narrative. Similarly, Bransford et al., (I 972) commented that their 'constructive'
approach to sentence memory was based upon the construction of semantic
descriptions, and not upon the concretisation of information in the form of
mental images.
However, Franks and Bransford (I972) noted a clear implication of the
research ofBegg and Paivio (I96g) on the retention of individual sentences. If
concrete sentences are stored as mental images, while abstract sentences are
stored in a superficial verbal form, the integration of information in the form of
holistic semantic ideas should be much more difficult in the case of abstract
sentences, and so the subjects should be better able to remember the particular
sentences which were presented for learning. Franks and Bransford repeated
I 14 Mental Imagery and Human Memory
their earlier experiments using abstract ideas, and obtained essentially the same
results. The confidence ratings given to a sentence in a test of recognition
memory were directly related to the number of components which it contained,
and not at all to whether it had· actually been presented. This indicated that, for
both concrete and abstract ideas, the subjects were integrating the information
contained in the individual sentences presented for learning, and storing holistic
representations of the complete ideas in memory. Franks and Bransford did not
consider that their results constituted a strict refutation of Begg and Paivio's
position, but they did demonstrate an important functional similarity between
the mnemonic representations of concrete and abstract ideas.
The studies of Bransford and Franks demonstrated virtually complete
integration of both concrete and abstract material. They could therefore be
regarded as suffering from a ceiling effect if one wished to consider the specific
hypothesis that concrete material is more easily integrated than abstract
material. Unpublished experiments carried out at Brunei University have tried
to remedy this by severely attenuating the acquisition conditions so that only
partial integration was achieved. That is, when the number of acquisition
sentences was reduced to a minimum, the subjects were able to discriminate
between these sentences and 'new' sentences which related to the same complex
ideas. The recognition test also included sentences relating to entirely different
ideas, so that the subjects were actually being tested both on their ability to
remember the particular sentences used in the acquisition phase, and on their
ability to remember the general ideas described by those sentences. Not
surprisingly, concrete ideas were remembered better than abstract ideas.
However, the results also showed quite clearly that the subjects could
discriminate between 'old' and 'new' sentences relating to concrete ideas just as
well as they could discriminate between 'old' and 'new' sentences relating to
abstract ideas; that is, concrete material showed no more integration than
abstract material. Another way of describing these results is to say that the
subjects were more accurate in detecting changes in meaning but no less accurate
in detecting changed in wording in concrete sentences than in abstract sentences.
It will be recalled that similar findings were obtained by Moeser (1974) in
studying the retention of individual, unrelated sentences.
This research also included conditions where the subjects were instructed to
use mental imagery to help them to remember the sentences. These instructions
did not affect the subjects' ability to remember the general ideas, nor did they
affect their discrimination of 'old' and 'new' sentences in the case of concrete
ideas. However, they produced poorer discrimination of 'old' and 'new'
sentences in the case of abstract ideas. This suggests that the use of mental
imagery may produce better integration of the material to be remembered, at
least in the case of abstract material, provided that the subjects receive explicit
instructions to this effect. However, when the subjects do receive such
instructions, their discrimination of'old' and 'new' sentences relating to concrete
ideas is actually better than of those relating to abstract ideas; that is, they
Remembering Connected Narrative 115
showed less integration and a superior ability to detect changes in wording in the
case of concrete material than in the case of abstract material. This seems quite
inconsistent with the proposals of Begg and Paivio. Moreover, since in the
absence of imagery mnemonic instructions concrete material produced no more
integration than abstract material, it is unlikely that mental imagery normally
plays any important role in the construction of holistic semantic ideas.
It would be of considerable interest to know whether this system of mnemonic
representations were distinct from whatever form of storage is employed in re-
membering pictures, or totally encompassed that form of storage. An important
study along these lines was carried out by Baggett (1975), who presented simple
stories consisting of four pictures, and tested her subjects either immediately or
after a delay. In her first experiment, the subjects were presented with individual
pictures and had to indicate whether each was consistent with the original story.
On immediate testing, the responses to 'old' pictures were faster than those to
'new' pictures which were consistent with the story; this suggested that the
responses were made partly on the basis of an immediate memory for the specific
pictures presented. After a delay of72 hours, 'old' pictures no longer showed this
superiority, indicating that the responses were made on the basis of a conceptual
representation of the story which was neutral with respect to the original pictures
presented. In the second experiment, the subjects had to indicate whether each
specific picture had been one of those originally presented. They were able to
reject 'new' pictures with perfect accuracy, even when they were consistent with
the original story, and even when the test followed a delay of 72 hours. This
implied that a mnemonic representation of the original pictures was available
even after 72 hours, and could be accessed by the subjects if the task demanded it
(but cf. Pezdek, 1978). Finally, the third experiment required the subjects to
respond to written yes/no questions on the basis of whether they were consistent
with the original story. The response latencies were independent of whether the
answer had to be inferred from the events actually depicted in the original
pictures. This suggested, once again, that the responses were made on the basis of
a conceptual representation of the story.
Thus, Baggett's results appear to support the idea of a long-term memory for
specific pictorial presentations, and a separate conceptual, propositional,
nonpictorial representation. She concluded her paper in the following manner:
Our study has focused on the memory representations arising when a viewer
bridges the gaps in the continuity of action in a picture story in order to
integrate the surface structure into a meaningful succession of incidents
related over time. We believe that the viewer, at least with our simple stories,
makes inferences at the time of viewing to aid his comprehension; that these
inferences, when stored in memory, are essentially nonpictorial in nature; and
that 3 days later the subject can still separate the inferences from the pictures
he saw, if the task requires it.
116 Mental Imagery and Human Memory
These conclusions have two important corollaries. First, any theory of human
memory which attempts to model all cognitive processes in terms of a single
system of propositional representations, and which includes no provision for a
specific pictorial memory (for example, Anderson and Bower, 1973; Pylyshyn,
1973) is inherently inadequate. Although the propositional system may be able
to represent most of the important knowledge which is employed in cognitive
tasks, it cannot handle the excellent retention of pictures demonstrated in
Baggett's study (cf. the discussion in chapter 5). Second, the comprehension and
retention of picture stories does not give rise to any additional pictorial
representations in the form of mental imagery. Otherwise, the subjects would be
unable to distinguish efficiently between the original pictures and 'new' but
appropriate pictures which happened to have been imaged. This in turn suggests
one of two possibilities. Either mental imagery does not give rise to a pictorial
representation in memory; or mental imagery does give rise to such a
representation, but is not employed in the retention of picture stories.
8.5 CONCLUSIONS
We come now to one of the most complex and least conclusive of the areas of
experimental investigation concerning mental imagery: the study of differences
between individual subjects in their ability to use mental imagery and in their
preference for using mental imagery in various cognitive tasks. Of course, the
study of how individuals differ in their cognitive strategies and abilities has
always formed a significant part of psychological research. It is therefore not
surprising that the investigation of individual differences is one of the oldest
approaches to the understanding of mental imagery.
In chapter 3 it was argued that mental imagery was a typical example of the
class of ordinary-language concepts relating to psychological states and
processes, in that 'having a mental image' exhibits an asymmetry between its
first-person present-tense use and its third-person present-tense use. The exact
nature of the asymmetry is a matter of some philosophical controversy, but one
can certainly agree that each person has a peculiar authority for asserting that he
himself has a mental image, an authority which others lack. Whether he is
having a mental image is not something about which he can be mistaken. It
would therefore be reasonable to start an investigation of individual differences
in the use of mental imagery by comparing people on the basis of what they are
inclined to say about their own images. (In contemporary psychology, this is seen
as a 'subjective' approach to the study of individual differences; however, for
reasons explained previously, I regard this as a misnomer.)
Accordingly, this chapter will consider first the comparison of individual
subjects in terms of their reports concerning the vividness or manipulability of
their experienced mental imagery. This is the oldest method used by psychol-
ogists for studying individual differences in mental imagery, but it will be argued
on both conceptual and empirical grounds that it does not constitute a useful
approach for psychological research. Nevertheless, certain recent experiments
have been concerned with subjects' reports on the strategies or mediators used in
verbal learning tasks. This approach has produced quite clear and reliable
findings, and it seems to constitute a potentially important way of studying
individual differences in the use of mental imagery.
Many psychologists have been dissatisfied with these 'introspective' tech-
niques for comparing individual subjects, and have sought instead for objective
measures of performance which can be taken as indices of the ability to employ
117
118 Mental Imagery and Human Memory
If a person's account of his own mental imagery enjoys the sort of epistemological
priority which was discussed in chapter 3, then it seems that he should be in the
best position to specify not only when he is experiencing mental imagery, and what
is depicted in his mental imagery, but also how vivid, distinct and efficacious are
his mental images. Psychologists usually interpret these judgements as intro-
spective evaluations of the subject's ability to produce concrete mental images
(for instance, Marks, I972), and formal questionnaires on the vividness of
experienced mental imagery have elsewhere been referred to as 'tests of
subjective imagery ability' (Richardson, I 978c). The use of such questionnaires
has a long history in psychology, going back to the work of Galton (I 883). Recent
reviews of this research have been given by A. Richardson ( I977) and by White,
Sheehan and Ashton (I 977). Research using these questionnaires has typically
demonstrated a considerable variation among individual subjects in terms of
reports concerning the vividness of their experienced imagery. Since other
experimental approaches have indicated that the use of mental imagery is a
highly effective strategy for acquiring new information (chapters 6 and 7), it
would be reasonable to expect that the vividness of a person's mental imagery
should predict how easily such information can be retained.
The most detailed technique for evaluating the subjective vividness of a
person's experienced mental imagery is possibly the Questionnaire upon Mental
Imagery (QMI), which was developed by Betts (I909) from Galton's original
procedure. This included I 50 items, and studied seven major modalities: visual,
Individual Dif}erences
to the incidental recall of block designs, such that vivid imagers were superior to
non-vivid imagers. This was not replicated in a further study using concrete and
abstract nouns as stimulus material (Sheehan, I972), but a similar relationship
between reported vividness and incidental recall has been found in at least two
other investigations (Janssen, I976, p. 48; Morris and Gale, I974). Nevertheless,
the most appropriate conclusion at present seems to be that the QMI has little
predictive validity in the study of human memory.
The negative findings obtained by Sheehan and Neisser ( I969) led Neisser
(I 970) to conclude that the vividness of experienced mental imagery was quite
unrelated to its usefulness or accuracy as a memory code. However, this
conclusion was challenged by Marks (I972, I973), who ascribed the negative
results to two basic shortcomings ofSheehan's research. First, he suggested that it
would not be helpful to evaluate the vividness of each sul;>ject's evoked mental
imagery by averaging across all seven sensory modalities (see also Durndell and
Wetherick, I976h). Rather, 'it would seem more appropriate to select subjects
according to their scores in the imagery modality most likely to function in the
experimental task to be performed.' In experiments on the recognition and
reconstruction of visual patterns and designs, the relevant modality would be
vision;-indeed, whatever the material, it could be argued that visual imagery
is the most easily aroused and of most importance in determining memory
performance (Putnoky, I975; White, Ashton and Brown, I977)· Therefore,
Marks devised a Vividness of Visual Imagery Questionnaire (VVIQ), contain-
ing sixteen items to be rated in terms of evoked visual imagery along five-point
rating scales. The internal consistency of this test is good, the test-retest reliability
is high, and factor analysis yields a single underlying dimension (Dowling,
described by White, Sheehan and Ashton, I977; Marks, I972, I973; McKelvie
and Gingras, I974)·
Marks' second objection was that Sheehan's experiments had employed
abstract geometrical patterns of little inherent interest or meaning of the
subjects. He suggested that 'it is not unreasonable to assume that vividness will be
related to the interest, affect, meaning, and overall level of arousal evoked by a
stimulus.' Accordingly, he carried out three experiments in which the stimuli were
coloured photographs of objects or scent:s, and in which a forced-choice
recognition test subsequently required the subjects to recall details of the pictures
presented. In each case, the 'good visualisers' produced better performance than
the 'poor visualisers'. Subsequent applications of the VVIQ have, however,
produced less convincing results. Marks ( I977) found a weak positive correlation
with recognition memory for pictures, but a weak negative correlation with
recognition memory for words (though presumably vivid visual imagery should
be useful in both cases). McKellar, Marks and Barron (reported by Marks, I972)
found no relationship between VVIQ scores and performance in the serial
learning oflists of words, nor between VVI Qscores and the benefit gained from
the location mnemonic (see chapter 6). Nevertheless, some interesting findings
were obtained in an experiment by Gur and Hilgard (I 975) in which the subjects
Individual Dij]erences I2I
had to detect differences between pairs of pictures. The subjects were classified as
'good' or 'poor' imagers according to their scores on the VVIQ, and the two
stimuli in each pair were presented either simultaneously or separated by an
interval of 20 seconds. The response latenCies showed a highly significant
interaction, such that the poor imagers were worse with successive presentation,
but the good imagers were not. Conversely, the good imagers responded faster
than the poor imagers, but only under conditions of successive presentation. This
is of course exactly the pattern to be expected on the assumption that
introspective judgements of the vividness of evoked visual imagery are a measure
of the efficiency of pictorial memory.
However, there is a serious problem to be faced by any method for
investigating mental imagery which is based upon introspective reports, and this
is the extent to which the results might be influenced by the subjects'
expectations. The idea that subjects might vary their responses according to their
task expectations was first suggested by Angell ( I910). More recently, Holt
(I964) pointed out that the experimental results in any experiment on mental
imagery will be influenced by expectations set up in the subjects 'by the specific
experimental conditions, by their general reading, rumor, and other sources of
notions about what to anticipate, and by the attitude towards imagery that
prevails in their particular subculture'. The first experimental investigation of
this problem was carried out by Di Vesta, Ingersoll and Sunshine (I 97 I), who
found a significant positive correlation between scores on the QMI and scores on
the Marlowe-Crowne Social Desirability Scale (Crowne and Marlowe, I964),
measuririgthe extent to which an individual is dependent upon the recognition
and approval of others (see Durndell and Wetherick, I975)· Di Vesta et al.
carried out factor analyses on performance in a large number of tests of
individual ability, and found that introspective questionnaires on mental
imagery contributed only to a factor defined by the Social Desirability Scale.
This correlation between reported vividness of experienced imagery and social
desirability has not always been replicated (Durndell and Wetherick, I975;
White, Sheehan and Ashton, I 977); for example, A. Richardson (I 977) found a
clear relationship in the case of male subjects, but not in the case of female
subjects. Nevertheless, it remains possible that under certain circumstances
introspective reports may be contaminated by a disposition to respond in a
socially desirable manner. For instance, this might explain the finding of a study
by Huckabee (I974) that introverts produce higher ratings of evoked mental
imagery than extraverts.
These sorts of expectations are probably not fixed, but may well be influenced
by the subjects' perceptions of the psychology experiment as a social situation.
For example, as Bower (I 97oa) suggested: 'The normal person's introspections
are frequently neither very discriminating nor particularly valid, and they are
easily influenced by loaded or leading questions.' More generally, one may
consider whether introspective reports might be influenced by the manner in
which the experimental instructions are presented. Durndell and Wetherick
122 Mental Imagery and Human Memory
Berger and Gaunitz (1977) repeated Gur and Hilgard's (1975) investigation of
the relationship between scores on the VVIQ and visual memory, but they
identified their experimental groups after the memory task had been carried out.
They found no sign of any difference in performance between 'good' imagers and
'poor' imagers. Similarly, Richardson ( 1978c) found no correlation between
scores on the QMI and performance in the free recall of lists of common nouns
when the experimental groups were identified after the criterion task.
Because of the abundance of negative findings on the relationship between
rated vividness of experienced mental imagery and memory performance, and
because of the serious possibility of contamination by experimenter effects, some
psychologists have que~tioned whether introspective reports are a valid measure
of individual differences in cognitive ability at all. For example, in the light of
their finding that introspective reports (as measured by several tests, including
the QMI) loaded only on a factor of social desirability, Di Vesta et al. (1971)
concluded that such reports did not possess construct validity as measures of the
use of mental imagery. However, there are good reasons for not accepting this
sort of conclusion. The first is the conceptual point made earlier in discussing the
problem of validation. Statements about a person's mental imagery are
ultimately to be justified and verified on the basis of his verbal utterances.
Introspective questionnaires such as the QMI consist of central examples of such
utterances, albeit distorted slightly so as to produce a numerical measure of the
vividness of experienced imagery. Any alternative notion of mental imagery, any
alternative measure that might be proposed, would only be contingently related
to what is indicated by the ordinary-language concept. It would therefore have
to be validated (in the sense used by Di Vesta et al.) by demonstrating that it was
empirically correlated with the subjects' introspective reports.
A second reason for rejecting the idea that introspective reports do not possess
construct validity is that relatively similar reports have proved very successful in
predicting performance in a wide variety of learning tasks when interpreted as
measures of the imageability of the stimulus material. In particular, the
procedure ofPaivio, Yuille and Madigan ( rg68) computes the average rating of
evoked mental imagery across groups of subjects, and the resulting measure
shows a high corr-elation with memory performance on individual items or
samples of items in a wide variety of tasks (chapter 7). However, if introspective
reports are to be rejected for the purpose of comparing individual subjects because
they lack construct validity as measures of the vividness of experienced imagery,
then to be consistent they should also be rejected for the purpose of comparing
individual stimulus items. This would clearly remove an important source of
experimental evidence. Conversely, if introspective reports are methodologically
acceptable for studying differences among stimulus items in the vividness of
evoked mental imagery, then additional justification must be given for refusing
to accept them for the purpose of studying individual differences among the
experimental subjects.
Nevertheless, the problem remains: even if introspective ratings are metho-
124 Mental Imagery and Human Memory
In particular, it is likely that subjects lack a definite origin and a definite unit
of measurement for making these subjective judgements, and that in
consequence their appreciation of the vividness of mental imagery can be
taken to define at best an ordinal scale.lt is probable that such judgements are
adequate to ensure that relatively gross comparisons among different stimulus
items predict variations in the memorability of those items. Nevertheless, since
there is no unique mapping of this ordinal subjective scale of vividness onto the
seven-point rating scale (which may be variously interpreted as representing
either an interval scale or a ratio scale), the absolute value assigned by a
Individual Differences 125
An analogy from the area of perception may help to clarify this point. Under
normal conditions, when an observer views a distant object, he can easily give an
accurate estimate of the object's size. However, one may eliminate any
information about the distance of the object in the following manner. The object
is rendered luminous and is presented in a totally dark room. The observer views
the object with one eye through a pinhole in a screen, with his head held
stationary. Under these conditions, the only information available to the subject
is the visual angle which the object subtends, and he is unable to ascribe a unique
objective size to the object. However, if the observer then looks through another
pinhole at a different object, he is able to judge, quite accurately, which looks
larger, that is, which object subtends the larger visual angle (Epstein and
Landauer, 1969; Hastorff and Way, 1952; Rock, 1975, p. 37; Rock and
McDermott, 1964).
The subject who is presented with a questionnaire on the vividness of his
experienced mental imagery seems to be in an analogous situation. He can judge
whether he has a mental image or not, just as the observer in the above
experiment can report the presence or absence of a perceptual object. He can
judge whether one item evokes a more vivid mental image than another,just as
the observer is able to say whether one perceptual object looks larger than
another. However, he is unable to make an absolute judgement of the vividness
of a single image,just as the perceiver is unable to say in absolute terms how large
an object looks. If the argument given above is sound, it follows that comparisons
among experimental subjects in terms of their ratings of evoked mental imagery
are neither valid nor meaningful, and it is quite unsurprising that they should fail
to predict performance in learning tasks.
(An interesting question is whether a subject can make comparative
judgements across modalities, for example, comparing the vividness of a visual
image with that of an auditory image. It could be argued that mental imagery in
one modality is qualitatively different from that in another. So, just as it makes
no sense for a subject to ascribe a unique vividness rating to a mental image,
neither does it make any sense for him to judge whether a given visual image is
more or less vivid than a given auditory image. On this view, the high
correlations which have been found among introspective ratings in different
modalities of mental imagery do not reflect a genuine, general imagery trait, but
are to be ascribed to general suggestibility or to social desirability; cf.
Richardson, 1969, p. 46; Sheehan, 1967a; Wagman and Stewart, 1974; White,
Ashton and Law, 1974; White, Sheehan and Ashton, 1977.)
I 26 Mental Imagery and Human Memory
The internal consistency of this test is good, and the test-retest reliability is
adequate (Juhasz, I972; McKelvie and Gingras, I974; White, Sheehan and
Ashton, I977)· However, given that the questions are limited to the visual
modality, it is perhaps surprising that the test appears to have a complex internal
structure. Factor analyses by Ashton and White ( I974) and by White and
Ashton (I 977) identified four separate factors, which they called 'movement'
(Questions 6, 7 and 8), 'misfortune' (Questions 4, 9, I I and I2), 'colour'
(Questions 2, 3 and 10), and. 'stationary' (Questions I, 5 and possibly 4).
However, it should be mentioned that unpublished studies at Brunei University
have failed to replicate this factorial solution.
Gordon's test of visual imagery control has shown a significant correlation
with the QMI in some studies, but not in others (Ernest, I977; Morelli and Lang,
I97I; Morris and Gale, I974; A. Richardson, I977; Spanos, Valois, Ham and
Ham, I973; Starker, I974; White, Sheehan and Ashton, I977). However, they
are loaded on the same factor in factor analytic studies (Di Vesta et al., I97I;
Individual Dij]erences
Forisha, 1975; A. Richardson, 1977). A strong positive correlation has also been
reported between Gordon's test and Marks' VVIQ (McKelvie and Gingras,
1974). Nevertheless, each of these is a pencil and paper test of mental imagery,
and so it is entirely possible that an instrument factor might have been operating
(White and Ashton, 1977; White, Sheehan and Ashton, 1977). Finally, the
available research on effects of social desirability suggests that Gordon's test is
rather less vulnerable to variations in demand characteristics than the QMI (Di
Vesta et al., 1971; Durndell and Wetherick, 1975; A. Richardson, 1977; White,
Sheehan and Ashton, 1977).
Once again, the predictive validity of Gordon's test has been investigated in a
variety of cognitive tasks (Ernest, 1977; Richardson, 1972), but only the very few
studies of memory performance need to be mentioned here. Morelli and Lang
(1971) compared scores in Gordon's test with paired-associate learning, and
found a significant correlation only when the stimulus pairs were supplemented
by relevant pictures; they concluded that the ability to control one's mental
imagery was only employed in learning when imagery was directly aroused by
pictorial presentation. Morris and Gale (1974) reported a positive correlation
between imagery control and the incidental recall of individual words. However,
experiments at Brunei University have failed to find any sign of a relationship
either between imagery control and the recall of paired associates, or between
imagery control and the effect of stimulus imageability. Even when the total
scores in Gordon's test were broken down into the four factors suggested by
White and Ashton ( 1977), the only correlation to reach statistical significance
was that between the 'movement' score and the recall of concrete pairs. It is
really too soon to decide whether Gordon's test will prove useful in predicting
memory performance. Nevertheless, it should be explored, since the questions
contained in the test do not seem to be subject to the conceptual and
methodological criticisms which may be directed at the QMI.
There are certain other procedures which have been suggested for eliciting
introspective reports on the use of mental imagery, such as the Imaginal
Processes Inventory (Singer and Antrobus, 1972), for the analysis of daydream-
ing, the Imagery Survey Schedule (Tondo and Cautela, 1974), for use in
behaviour therapy, and the Individual Differences Questionnaire, which was
devised by Paivio (1971C, p. 495) to investigate cognitive style.. However, these
techniques have not been adequately tested in the context of experiments on
human learning and memory. There is nevertheless one further method which
has been found to have excellent predit:tive capacity, and which is totally
consistent with the methodological attitudes of this book.
In chapters 6 and 7, experiments were described in which the subjects received
post-learning questionnaires which asked them to describe the sorts of strategies
128 Mental Imagery and Human Memory
which they had used to remember different stimulus items. These reports were
analysed in terms of the different sorts of item used and in terms of the subjects'
performance. The main findings were as follows. First, the number of subjects
who report the use of images as mediators is greater with concrete, imageable
material. Second, the number of subjects who report the use of imagery for a
given item correlates with the memory performance on that item. Third, items
for which subjects report images as mediators are recalled better than items for
which subjects report verbal mediators. Finally, subjects will tend to report the
use of images as mediators in the case of concrete, imageable items, even when
instructed to use other learning strategies.
These empirical results seem to have an immediate corollary, that the number
of items for which a subject reports the use of images as mediators should correlate
with that subjeds memory performance. Although open to immediate empirical
verification and although clearly relevant to the study of individual differences in
the use of mental imagery, this idea has hardly ever been seriously considered, let
alone experimentally tested. One reason for this, of course, has been the idea that
such reports are subjective, mentalistic and invalid as a source of scientific data.
Another reason, and one that is rather better founded, is the feeling that the
causal determiners of psychological performance may not be open to conscious
inspection (cf. Pylyshyn, 1973). Nevertheless, while other areas of psychological
research have systematised the collection of introspective reports in the method
of protocol analysis (Newell and Simon, 1972), such reports have not been
collected in research on human memory.
One of the first psychologists to relate memory performance to the subjects'
learning strategies was Bartlett (1932, pp. 59--61, 109-12). He found that his
subjects could be classified roughly as either visualizers, who claimed to rely
mainly upon visual imagery in remembering, or vocalisers, who claimed to rely
mainly upon words rather than mental images. Although the vocalisers tended
to be less confident in their recall, the two groups produced comparable recall
performance. However, Bartlett's experimental procedure was rather peculiar,
involving the learning of arbitrary associations between words and simple
figures, and rather different results might be obtained with more conventional
learning tasks.
One experiment in recent years which compared subjects in terms of the sorts
of mediators which they employed was carried out by Hulicka and Grossman
(1967). They compared a group of old subjects (whose median age was 74.1
years) with a group of young subjects (whose median age was 16.1 years) in a
paired-associate learning task with pairs of common nouns. In a post-learning
questionnaire, it was found that the older group reported more verbal mediators
and fewer imaginal mediators than the younger group. However, it is obvious
that the difference in age between the two groups was entirely adequate to
explain any difference in memory performance, without any appeal to
differences in preferred encoding strategy.
Another experiment which briefly considered the subjects' reported use of
Individual Differences 129
mental imagery was carried out by Wells (1972). She showed that the recall of
verbal stimuli was more disrupted by an interpolated backwards-counting task
than the recall of pictorial stimuli, and suggested that this was because the
pictorial stimuli had been retained as mental images. However, she mentioned
that many subjects also seemed to retain the verbal stimuli in the form of images,
and found that the greater their reported use of mental imagery, the better their
recall (but cf. Hasher, Riebman and Wren, 1976).
Marks (1972) pointed out a possible problem with the use of post-
experimental reports, which is that a subject's responses may be influenced by
whether his recall was actually successful. Demand characteristics might operate
either if the subject cannot remember the mediators he employed, or if he fails to
understand the experimenter's instructions. In either event, 'his perception of
the purpose of the experiment would presumably be based on the naive
(although probably correct) assumption that vivid imagery is expected to
accompany high accuracy', and any correlation between recall performance and
the use of mental imagery would be artefactual. To illustrate this, Marks
repeated the experiment of Sheehan and Neisser ( 1969), and found that, within
subjects, variations in the reported vividness of mental imagery only correlated
with variations in recall when reports were obtained after the recall test.
Nevertheless, in the same paper, Marks reported an experiment by McKellar,
Marks and Barron which showed that the improvement in recall resulting from
the use of the location mnemonic was not related to the subjects' scores on the
VVIQ, but that it was highly correlated (r = +o.67) with the number of
reported visual images. In this instance, Marks accepted the evidential value of
post-learning reports, and concluded: 'This result strongly supports the notion
that visual imagery is the effective mediator of the improved recall.'
Richardson (1978e) carried out an experiment to test the hypothesis
mentioned above in a direct fashion. The subjects learned a list of paired
associates, attempted to recall the response terms when cued with the stimulus
terms, and then completed a post-learning questionnaire of the sort devised by
Paivio, Smythe and Yuille (1968). Various measures of the use of mental
imagery were considered, but the best predictor of a subject's recall performance
(r = +o.8o) was the absolute number of pairs for which he reported the use of
imaginal mediators. A subsequent experiment showed that this correlation
applied in the case of concrete material, but not in the case of abstract material.
This goes some way to rebutting an objection of the sort made by Marks (1972),
that the mediator reports might have been based upon the accuracy of recall; on
this interpretation, it is difficult to see why the frequency of reported imaginal
mediators did not correlate with the recall of abstract material, or why the
frequency of verbal mediators did not show a similar relationship with recall
when the subjects had as much opportunity to report such mediators as they did
the use of mental imagery (Paivio, 197 1c, p. 359; personal communication). In
general, as Paivio (1972) has pointed out, the usefulness of mental imagery as a
mnemonic device will be demonstrated by interactive relationships between
Mental Imagery and Human Memory
One way of trying to study mental imagery in a more 'objective' manner would
be to investigate how subjects carry out tasks which plausibly can only be carried
out by manipulating some pictorial representation (that is, tasks for which more
abstract verbal encoding would be relatively unhelpful). This sort of task was
discussed at length in chapter 4· Although the processes operating can only be
inferred from the subjects' behaviour, success in these tasks can be defined
according to an absolute, public criterion, namely the correctness of the subjects'
responses. Moreover, one can also measure the latency of these responses, and
thus evaluate the speed with which the tasks are carried out. It is therefore not
surprising that psychologists have used such measures as indices of imagery
ability. (In contradistinction to introspective methods, these sorts of task have
elsewhere been referred to as 'tests of objective imagery ability': Richardson,
I978c.)
In his analysis of factors underlying human intelligence, Thurstone ( Igg8)
defined the space factor as the ease with which a subject employs spatial and
visual imagery. This factor is measured by tests of spatial thinking, an example of
which is the Flags test (Thurstone and jeffrey, I 956). This contains 21 questions,
each of which comprises one test figure and six examples, where each figure is the
schematic representation of a flag (perhaps a rectangle with a symbol in one
corner). The subject has to say whether each example depicts the same side of the
flag as the test stimulus (usually rotated) or the opposite side (the mirror-image of
the test stimulus, again usually rotated). The subject's performance is derived
from the number of correct responses given during a five-minute period. This test
is obviously quite similar to the mental rotation task devised by Shepard and
Metzler (I 97 I). Other tests that have been employed in experimental research to
define imagery ability include the Minnesota Paper Form Board (Likert and
Quasha, I94I) and the Space Relations test (Bennett, Seashore and Wesman,
I947). Such tests correlate moderately well with one another (Durndell and
Wetherick, I976a; Ernest, 1977), they load on the same factor or factors in
investigations which have employed factor analysis (Di Vesta et al., I 971;
Forisha, 1975; Paivio, 1971c, p. 496), and scores are often combined from two or
more of a battery of tests. On the other hand, tests of spatial thinking have failed
to show any consistent relationship with ratings produced in introspective
questionnaires on mental imagery (Danaher and Thoresen, 1972; Durndell and
Individual Dijfirences
and the spatial ability of the subjects (and cf. Christiansen and Stone, I968).
However, in the paired-associate learning of noun-adjective pairs, Di Vesta and
Ross (I 97 I) found a clear negative correlation, such that subjects of high spatial
ability were superior to subjects oflow spatial ability only in learning pairs oflow
rated imageability. Again, Stewart's original study had compared the recall and
recognition of pictures and words; Paivio, Rogers and Smythe (I 968) found no
effect of spatial ability upon the free recall of either sort of material, and
Richardson (I978c) found no effects in a test of recognition memory. One
interesting, though as yet unreplicated result was obtained by Di Vesta and
Sunshine (I 974), who investigated serial learning assisted by mnemonic devices
of the 'one-bun' variety (see chapter 6), and who defined imagery ability
according to three spatial tests. This study found that subjects of high imagery
ability produced better performance than subjects of low imagery ability when
operating under imagery mnemonic instructions, but not when operating under
instructions to use verbal mediators. Imagery instructions produced poorer
performance than verbal instructions in the case of subjects of low imagery
ability, but not in the case of subjects of high imagery ability.
This last result is consistent with the idea that tests of spatial thinking predict
the extent to which the subject is able to employ mental imagery as an effective
elaborative code in long-term memory. Nevertheless, it clearly stands in need of
further investigation, and contrasts markedly with the negative findings
obtained by other researchers on the value of spatial tests in memory research. It
should also be pointed out that research on the effects of spatial ability may be
just as vulnerable to experimenter effects as the investigations reviewed earlier on
introspective reports. Unfortunately, as Ernest (I977) has pointed out, there is
simply no experimental evidence on this point. In the circumstances, one must
remain sceptical of the vast majority of the experiments on the effects of spatial
ability on the grounds that the experimental groups were identified before the
administration of the criterion task. When the experimental groups are
identified after they have completed the recall task, variations in spatial ability
seem to have no effect upon performance in long-term memory (Richardson,
I978c).
This conclusion is entirely consistent with the general theoretical suggestions
made in chapters 6 and 7, that the constructive and elaborative uses of mental
imagery are functionally independent of one another. On this analysis, one
would expect a person's performance in long-term memory to vary with his use of
mental imagery as a form of elaborative encoding, and this receives excellent
confirmation from the use of post-learning questionnaires (see above).
Nevertheless, it is by no means clear how the ability to employ mental imagery as
a form of elaborative encoding might be measured objectively, except (trivially)
by performance in another long-term memory task. The next section describes
an alternative approach.
Individual Dij]erences I33
If one distinguishes between these two concepts of coding ability and coding
preference, it is clear that almost all of the research on individual differences in the
use of mental imagery has been concerned with the former.lt has been assumed
that the preference for using mental imagery is correlated with the ability to use
mental imagery (Ernest and Paivio, I97Ih), but there has been little serious
attempt to measure imagery coding preference directly.
An exception is the second experiment reported by Stewart (I 965), which
investigated recognition memory for pictures and words. The test required her
subjects to judge not only whether an item had been presented for learning, but
also whether it had been originally presented as a picture or as a word. If subjects
vary in their preference for mental imagery as a way of encoding information,
this should affect the extent to which they can make the latter sort ofjudgement
with accuracy. Specifically, the judgement which a subject makes should reflect
the sort of coding employed. The results indicated that the subjects of high
spatial ability were more likely to encode a word as a pictorial image than were
those of low spatial ability, whereas subjects of low spatial ability were more
likely to encode a picture verbally than were those of high spatial ability.
However, Di Vesta et al. (I 97 I) mentioned that they had been 'only partially
successful' in replicating Stewart's findings. Richardson ( I978c) also asked his
subjects to identify whether an item had originally been presented as a picture or
as a word. Although the judged presentation modality accurately reflected the
original presentation modality, the degree of accuracy was not influenced in any
way either by introspective reports (as measured by the QMI) or by spatial
ability (as measured by the Flags test). Richardson concluded that coding ability
and coding preference are relatively independent dimensions along which
individuals may vary, at least with regard to the use of mental imagery. If coding
preference is the crucial determiner of performance in memory tasks, than this
would explain why research employing measures of imagery ability has produced
confusing, contradictory and unreplicable findings.
Richardson went on to argue that, in order to understand how a subject
134 j\1entallmagery and Human Memory
carried out a verbal-learning task, one must measure the extent to which he tends
to encode word stimuli as mental images. Stewart's recognition task enables one
to derive just such a measure of coding preference. To be more specific, a
subject's preference for imaginal encoding can be estimated from his accuracy in
judging the original presentation modality of verbally presented items. To the
extent to which he tends to encode such items as pictorial images, his capacity for
making such judgements will be reduced. Richardson measured his subjects'
preference for imaginal coding in just this fashion, and found that subjects who
preferred to encode verbally produced better recall performance than subjects
who preferred to encode pictorially. He suggested that the latter subjects were
employing mental imagery in a separative manner, and, to support this
conclusion, he pointed out certain fine-grain similarities between the results of
his experiment and the effects of separative imagery instructions. It was
concluded that coding preference is an important determiner of performance in
memory tasks, but that a preference for imaginal encoding, as measured
according to the above procedure, leads to discrete representations in human
memory, which actually impairs the subject's ability subsequently to recall the
information thus encoded.
Although this approach appears to be promising, three points should be made
concerning Richardson's procedure. First, it is basically a measure of response
bias, and there is no guarantee that such bias reflects only variations in
mnemonic coding. Second, the superiority of pictorial material (see chapter 5) in
both the retention test and the test of presentation modality meant that
performance was at a ceiling for all subjects. Only the items presented verbally
provided a basis for discriminating among the subjects on the basis of their
coding preference. Third, the subjects were therefore discriminated on the basis
of whether they could remember that particular items had been presented
verbally. However, the fact that a subject could not remember that an item had
been presented verbally does not entail that it was remembered in a pictorial
form, as a mental image, nor in any other specific representation. The results
thus show at most that subjects who could not remember the modality in which
an item was presented in one test remember fewer items in another test; they do
not show that these subjects employ mental imagery, and they certainly do not
show that mental imagery employs a quasi-pictorial representation in human
memory.
verbal memory are also illustrated by findings obtained with two maze tasks (cf.
Ratcliff and Newcombe, I973). In a locomotor maze, the subject walks among
points arrayed on the ground along a specific path shown on a map. The map is
held in a constant orientation relative to the subject, but its orientation relative
to the ground changes as he walks. The subject does not have to 'learn' the maze
in any sense, but he has to carry out mental rotations in immediate memory in
order to relate his perception of the map with his perception of the maze.
Performance on this task is affected by parietal lesions of either hemisphere, but
not by temporal lesions (Semmes, Weinstein, Ghent and Teuber, I955, I963;
Teuber, I963). On the other hand, in a visually-guided stylus maze, the subject
has to learn a specific path among a set of points in a rectangular array, tracing a
path with a stylus over a series of trials. This requires the subject to encode a
complex visual pattern in long-term memory, but he does not have to transform
the visual array in any way. Patients with lesions of the right temporal lobe are
impaired in this task, but those with lesions confined to the parietal lobes or to the
left temporal lobe obtain normal scores (Milner, I 965).
As was mentioned earlier, the neuroanatomical basis oflinguistic functioning
tends to be localised within the left cerebral hemisphere in most subjects. It is
certainly true that verbal learning and memory tend to be disrupted by damage
to the left hemisphere (Newcombe, I969, chap. 6); and, as in the case of non-
verbal retention, the structures of the temporal lobe appear to be of particular
importance (Miller, I972, PP· 56--g; Milner, I97I). As Milner (Ig66) com-
mented: 'It is now well established that a left temporal lobe lesion in the
dominant hemisphere for speech impairs the learning and retention of verbal
material, whether aurally or visually presented, and regardless of whether
retention is measured by recognition, free recall, or rate of associative
learning.'
This functional dissociation of the two cerebral hemispheres with respect to
the processing of verbal and non-verbal information has been taken to support
the dual coding theory of symbolic functioning in a fairly direct manner (Paivio,
I97IC, pp. 522-3; I978b; Sheikh, I977l· However, the assumption of
independence-interconnectedness of the two systems does not appear to be
strictly in accord with the neuropsychological evidence. Specifically, the notion
of interconnectedness, as encapsulated in the coding redundancy hypothesis,
implies that there should not be a radical hemispheric dissociation. This
argument may be spelled out in the form of two premises, each of which is a
central plank of dual coding theory (though neither of which is clearly supported
by empirical findings), together with a conclusion which is completely
contradicted by the available evidence.
I. Dual coding theory interprets many findings in verbal learning and
memory as showing that mental imagery is involved in such tasks. This is a
consequence of the coding redundancy hypothesis, which states that both
imaginal and verbal processing are involved in verbal remembering. Although
many of the findings discussed in chapters 6-8 are consistent with this
Mental Imagery and Human Memory
framework, very little of the evidence serves to distinguish dual coding theory
from common coding theory.
2. Dual coding theory also assumes that mental imagery is the form of
mnemonic representation which is involved in pictorial memory. Although it
might be parsimonious to explain the effects attributable to the operation of
mental imagery in terms of the mnemonic coding aroused by pictorial
presentation, the research discussed in chapter 5 failed to provide any convincing
evidence for this. Insofar as pictorial memory in particular, and non-verbal
cognitive processing in general are vulnerable to lesions of the right cerebral
hemisphere, it would be reasonable to conclude that the neuroanatomical
vehicle of mental imagery is contained in that hemisphere. The only evidence for
this is an experiment on short-term recognition memory reported by Seamon
and Gazzaniga ( 1973). They required their subjects to use either interactive
imagery or subvocal rehearsal, and presented recognition probes to either the left
or the right cerebral hemispheres by means of a tachistoscope. The subjects
produced faster responses for probes to the left hemisphere when they were using
subvocal rehearsal, but they produced faster responses for probes to the right
hemisphere when they were using interactive imagery. This is the only clear
evidence directly associating mental imagery with the functions of the right
cerebral hemisphere, but it is actually contradicted by evidence from studies of
temporal lobectomy which will be discussed presently.
3· Nevertheless, the two premises which have just been outlined imply that
verbal learning and memory should be affected not only by damage to the left
cerebral hemisphere (because of the obvious verbal component), but also by
damage to the right cerebral hemisphere (because of the involvement of mental
imagery). However, the neuropsychological evidence on this matter is quite
unequivocal: damage to the right or non-dominant hemisphere virtually never
gives rise to an impairment in verbal learning tasks (Miller, 1972, p. 57; Milner,
1966; Newcombe, 1969, chap. 6; Walsh, 1978, p. 174). Thus, the two
hemispheres do not seem to contribute jointly to verbal learning and memory.
Conversely, their complete functional dissociation seems to be quite inconsistent
with usual accounts of dual coding theory.
The two cerebral hemispheres do seem to contribute jointly to the recognition
of nameable pictures. Milner (1966) devised a test in which the subjects were
asked to name the objects depicted in each of four pictures, and subsequently
(after an interval occupied by a digit repetition task) to select the original four
pictures from a series of nine pictures. The task was administered to 123 patients
who were undergoing carotid amytal tests prior to neurosurgery. (In this
procedure, the functioning of one of the cerebral hemispheres is selectively
interrupted by the injection of a solution of sodium amytal into the common
carotid artery.) Failures of recognition tended to occur only when the patient
had a pre-existing unilateral temporal lesion, and when the injections were made
into the contralateral hemisphere. This suggests that the relevant information is
normally stored in both cerebral hemispheres and can be utilised provided at
Individual Dif.ferences I39
least one of the temporal lobes is functioning. Thus, there is some neuropsycholo-
gical evidence for a coding redundancy account of the retention of nameable
pictures.
Unfortunately, it is unlikely that such evidence can be more than suggestive
by itself. As Anderson (I978) has pointed out, the demonstration of a functional
dissociation of the two cerebral hemispheres is entirely consistent with the idea
that all information has a propositional form, but that propositions encoding
visual information are stored in the right hemisphere and propositions encoding
verbal information are stored in the left hemisphere; or with the idea that all
information is stored in a propositional form in both hemispheres, but that
procedures for performing verbal tasks are located in the left hemisphere and
procedures for performing spatial tasks are located in the right hemisphere.
Either version of common coding theory would be consistent with the available
evidence. Unless one is to accept Anderson's conclusion that it is in principle
impossible to decide between the two frameworks, future research must devise
more intricate experimental techniques for investigating the neuroanatomical
basis of psychological function.
Although research on the role of mental imagery in verbal memory has
received little support from clinical studies of memory impairment, several
recent studies have approached the problem by applying experimental pro-
cedures to specific clinical populations. Four of these inyestigations will be briefly
summarised.
1. Baddeley and Warrington (I970, I973) studied patients with a severe
global deficit in learning tasks, the amnesic syndrome (Warrington, I 97 I). Their
first series of experiments used tests of verbal learning, and identified this
condition as a selective impairment of long-term memory. The second series
demonstrated that amnesic patients were able to improve their recall perform-
ance when the material was organised on either a phonemic or a semantic basis;
however, they did not benefit at all when unrelated words were linked by
sentences which they were instructed to visualise. This suggested that amnesic
patients were specifically impaired in the use of mental imagery as a form of
coding in long-term memory.
2. Jones (I974) examined the usefulness of an imagery mnemonic in patients
following unilateral and bilateral temporal lobectomy, who were required to
learn heterogeneous lists of concrete and abstract paired associates. As one would
have expected, the patients with bilateral lesions showed no retention through-
out the experiment. The patients with lesions of the left temporal lobe showed a
significant impairment, but were able to improve their performance when
instructed to use mental imagery. Indeed, they showed a normal superiority on
concrete material, and a normal benefit from imagery mnemonic instructions.
This implies that their deficit cannot be attributed to a failure to use mental
imagery, and, conversely, that the neuroanatomical basis of mental imagery is
not contained within the structures of the left temporal lobe. The patients with
lesions ofthe right temporal lobe showed no impairment, a normal superiority on
Mental Imagery and Human Memory
neurological damage. On the other hand, the work ofjones ( 1974) and the more
general background of neuropsychological literature seem to have important
consequences for the status of current theories of imagery and memory.
9· 7 CONCLUSIONS
Conclusions
143
Mental Imagery and Human Memory
verbal storage to explain the capacity for recognising pictorial stimuli which
cannot be readily described, named or labelled. Excellent evidence for such a
representation comes from the investigation of face recognition, which seems to
be almost entirely a visual process. On the other hand, there is fairly good
experimental evidence for a coding redundancy hypothesis, according to which
nameable pictures and picture stories receive both non-verbal, pictorial
representations arid verbal, conceptual representations in long-term memory.
This idea of functionally independent mnemonic representations is supported by
neuropsychological evidence of a 'double dissociation' between pictorial long-
term memory, whose neuroanatomical basis appears to lie in the right temporal
lobe, and verbal long-term memory, whose neuroanatomical basis appears to lie
in the left temporal lobe. However, there is little direct experimental evidence to
suggest that this pictorial representation is the form of mnemonic coding aroused
by the use of mental imagery, and there is neuropsychological evidence which
strongly implies that this identification should not be made. Although it is
possible in principle to specify a version of propositional or common coding
theory which could handle the experimental and neuropsychological findings on
pictoriaLmemory, the various adjustments needed to ensure this consistency with
the empirical data are unlikely to constitute well motivated revisions of current
examples of common coding theory. A dual coding model may therefore be
regarded as the most appropriate theoretical approach for the future investig-
ation of pictorial memory.
In the area of verbal learning, three sorts of experimental procedure have been
employed for investigating the role of mental imagery: the administration of
instructions to the experimental subjects to use mental imagery in their learning;
the comparison of samples of stimulus material in terms of their concreteness or
imageability; and the comparison of individual subjects in terms of their ability
or preference to use mental imagery in learning. Under the appropriate
circumstances, each of these experimental procedures has been found to produce
substantial and reliable effects upon memory performance; moreover, they also
have been found to yield meaningful interactions in terms of their effects upon
recall. Nevertheless, several important considerations need to be appreciated in
evaluating this research. First, the relational organisation of mental imagery
appears to be important in determining its mnemonic efficacy, rather than its
subjective vividness. Second, the empirical evidence does not necessitate any
additional principles oflearning which are specific to the use of mental imagery
as a mnemonic strategy. Rather, all of the findings may be discussed in terms of
such general notions as organisation, integration and semantic representation.
Third, this elaborative use of mental imagery appears to be functionally distinct
(on both experimental and neuropsychological evidence) from the constructive
use of mental imagery in immediate memory, and from the representation of
pictorial information in long-term memory. Indeed, a clear neuropsychological
basis for this function of mental imagery has yet to be identified. Since both
common coding and dual coding positions appear to be consistent with the
Conclusions 145
findings which present grave difficulties for that position. Propositional theories
appear to be a more promising means of developing future investigations of the
faculty of memory.
References
visual working memory. In Rabbitt, P.M. A. and Dornic, S. (eds) Attention and
Performance V. London: Academic Press.
Baddeley, A. D. and Warrington, E. K. (I970). Amnesia and the distinction
between long- and short-term memory. Journal of Verbal Learning and Verbal
Behavior, g, I 76-89.
Baddeley, A. D. and Warrington, E. K. (I973)· Memory coding and amnesia.
Neuropsychologia, 11, I 59--65.
Baggett, P. (I975)· Memory for explicit and implicit information in picture
stories. Journal of Verbal Learning and Verbal Behavior, 14, 538-48.
Bahrick, H. P. and Boucher, B. (I968). Retention ofvisual and verbal codes of
the same stimuli. Journal of Experimental Psychology, 78, 4I 7-22.
Baker, G. P. ( I974). Criteria: A new foundation for semantics. Ratio, 16, I56-89.
Banks, W. P. (I 977). Encoding and processing of symbolic information in
comparative judgments. In Bower, G. H. (ed.) The Psychology of Learning and
Motivation: Advances in Research and Theory, Vol. I I. New York: Academic Press.
Banks, W. P. and Flora, J. (I 977). Semantic and perceptual processes in
symbolic comparisons. Journal of Experimental Psychology: Human Perception and
Performance, 3, 278--90.
Bartlett, F. C. (I932). Remembering: A Study in Experimental and Social Psychology.
London: Cambridge University Press.
Beech, J. R. (I977l· Effect of selective visual interference on visualization.
Perceptual and Motor Skills, 45, 95I-4.
Beech, J. R. (in press). Image latency and recall as functions of array size.
American Journal of Psychology.
Beech, J. R. and Allport, D. A. (I978). Visualization of compound scenes.
Perception, 7, I 29-38.
Begg, I. ( I97I). Recognition memory for sentence meaning and wording.
Journal of Verbal Learning and Verbal Behavior, 10, I 76-8 r.
Begg, I. (I 972). Recall of meaningful phrases. Jounal of Verbal Learning and Verbal
Behavior, 11, 43 I--9.
Begg, I. (I 973). Imagery and integration in the recall of words. Canadian Journal
of Psychology, 27, I 59--67.
Begg, I. (I 978). Imagery and organization in memory: Instructional effects.
Memory and Cognition, 6, I 74-83.
Begg, I. and Paivio, A. ( I969). Concreteness and imagery in sentence learning.
Journal of Verbal Learning and Verbal Behavior, 8, 82 I-7.
Begg, I. and Young, B. J. (I 977). An organizational analysis of the form class
effect. Journal of Experimental Child Psychology, 23, 503-I9.
Bennett, G. K., Seashore, M.G. and Wesman, A. G. ( I947). Differential Aptitude
Tests. New York: Psychological Corporation.
Benton, A. L. ( I969). Constructional apraxia: Some unanswered questions. In
Benton, A. L. (ed.) Contributions to Clinical Neuropsychology. Cl\icago: Aldine
Publishing Co.
Berger, G. H. and Gaunitz, S.C. B. ( I977). Self-r~ted imagery and vividness of
Mental Imagery and Human Memory
Janssen, W. (1976). On the Nature qfthe Mental Image. Soesterberg: Institute for
Perception TNO.
Jarvella, R.J. (1970). Effects of syntax on running memory span for connected
discourse. Psychonomic Science, 19, 235-6.
Jarvella, R. J. and Herman, S. J. (1972). Clause structure of sentences and
speech processing. Perception and Psychophysics, 11, 381-4.
Johnson, D. M. ( 1939). Confidence and speed in the two-category judgment.
Archives of Psychology, 241, 1-52.
Johnson, M. K., Bransford, J. D. Nyberg, S. E. and Cleary, J. J. (1972).
Comprehension factors in interpreting memory for abstract and concrete
sentences. Journal qf Verbal Learning and Verbal Behavior, u, 451-4.
Johnson-Laird, P. N. (I970). The perception and memory of sentences. In
Lyons, J. (ed.) New Horizons in Linguistics. Harmondsworth: Penguin Books.
Johnson-Laird, P. N. (I974)· Experimental psycholinguistics. Annual Review qf
Psychology, 25, I35-6o.
Johnson-Laird, P. N. and Stevenson, R. (I970). Memory for syntax. Nature,
227, 4I2-3·
Jones, G. V. (1978). Tests ofa.structural theory of the memory trace. British
Journal qf Psychology, 6g, 35I-67.
Jones, M. K. ( I974). Imagery as a mnemonic aid after left temporal lobectomy:
Contrast between material-specific and generalized memory disorders.
Neuropsychologia, 12, 2 I -30.
Jorgensen, C. C. and Kintsch, W. ( I973)· The role of imagery in the evaluation
of sentences. Cognitive Psychology, 4, I IO-I6.
Juhasz,]. B. (1972). On the reliability of two measures of imagery. Perceptual and
Motor Skills, 35, 874.
Karchmer, M.A. ( I974). Proportion of unitization as an index of cued recall
level. Journal qf Experimental Psychology, 103, 35 I --7.
Katz,J.J. and Fodor,J. A. ( I963). The structure of a semantic theory. Language,
39, I 70-210
Katz, J. J. and Postal, P. M. ( 1964). An Integrated Theory of Linguistic Descriptions.
Cambridge, Mass.: MIT Press.
Kelly, R. T. and Martin, D. W. ( I974). Memory for random shapes: A dual task
analysis. Journal qf Experimental Psychology, 103, 224-g.
Kenny, A. J. P: (I97I). The Verification Principle and the private language
argument. In Jones, 0. R. (ed.) The Private Language Argument. London:
Macmillan.
Kenny, A. J. P. (I972). To mind via syntax. In Kenny, A. J. P., Longuet-
Higgins, H. C., Lucas, J. R. and Waddington, C. H. The Nature of Mind.
Edinburgh: Edinburgh University Press.
Kenny, A.J. P. (I973). Wittgenstein. London: Allen Lane.
Kenny, A. J. P. ( I975). Will, Freedom and Power. Oxford: Blackwell.
Kerst, S. M. ( I976). Interactive visual imagery and memory search for words
and pictures. Memory and Cognition, 4, 573-80.
References I 57
Paivio, A. (I97IC). Imagery and Verbal Processes. New York: Holt, Rinehart and
Winston.
Paivio, A. (I 972). A theoretical analysis of the role of imagery in learning and
memory. In Sheehan, P. W. (ed.) The Function and Nature ofImagery. New York:
Academic Press.
Paivio, A. (I975a). Coding distinctions and repetition effects in memory. In
Bower, G. H. (ed.) The Psychology ofLearning and Motivation: Advances in Research
and Theory, Vol. g. New York: Academic Press.
Paivio, A. (I975b). Imagery and long-term memory. In Kennedy, A. and
Wilkes, A. (eds) Studies in Long Term Memory. London: Wiley.
Paivio, A. ( I97SC). Imagery and synchronic thinking. Canadian Psychological
Review, 16, I47-63.
Paivio, A. (I975d). Neomentalism. Canadian Journal of Psychology, 119, 263-g1.
Paivio, A. (I975e). Perceptual comparisons through the mind's eye. Memory and
Cognition, 3, 635-4 7.
Paivio, A. ( I976a). Imagery and recall and recognition. In Brown,]. (ed.) Recall
and Recognition. London: Wiley.
Paivio, A. (I976b). Imagery, language, and semantic memory. Research
Bulletin no. 385, Department of Psychology, University ofWestern Ontario.
Paivio, A. ( I976c). Images, propositions, and knowledge. In Nicholas,]. M. (ed.)
Images, Perception and Knowledge. Dordrecht: Reidel.
Paivio, A. (I 976d). On exploring visual knowledge. Research Bulletin No. 398,
Department of Psychology, University of Western Ontario.
Paivio, A. ( I978a). Comparisons of mental clocks. Journal of Experimental
Psychology: Human Perception and Performance, 4, 6I-7I.
Paivio, A. (I978b). Dual coding: Theoretical issues and empirical evidence. In
Scandura,J. M. and Brainerd, C.J. (eds) Structural/Process Models of Complex
Human Behavior. Leiden: Nordhoff.
Paivio, A. (Ig7&). Imagery, language, and semantic memory. International
Journal of Psycholinguistics, 5, 3 I -4 7.
Paivio, A. (I 978d). Mental comparisons involving abstract attributes. Memory
and Cognition, 6, Igg-2o8.
Paivio, A. (I979)· The relationship between verbal and perceptual codes. In
Carterette, E. C. and Friedman, M.P. (eds) Handbook of Perception. Vol. IX:
Perceptual Processing. New York: Academic Press.
Paivio, A. and Begg, I. (I 97 I). Imagery and comprehension latencies as a
function of sentence concreteness and structure. Perception and Psychophysics, ro,
408-I2.
Paivio, A. and Csapo, K. (Ig6g). Concrete-image and verbal memory codes.
Journal of Experimental Psychology, 8o, 279-85.
Paivio, A. and Csapo, K. (I 973). Picture superiority in free recall: Imagery or
dual coding? Cognitive Psychology, 5, I 76-206.
Paivio, A. and Foth, D. (I970). Imaginal and verbal mediators and noun
concreteness in paired-associate learning: The elusive interaction. Journal of
Mental Imagery and Human Memory
Segal, S.J. and Fusella, V. (I970). Influence ofimaged pictures and sounds on
detection of auditory and visual signals. Journal of Experimental Psychology, 83,
458--64.
Segal, S.J. and Fusella, V. (I97I). Effect of images in six sense modalities on
detection of visual signal from noise. Psychonomic Science, 114, 55--6.
Segal, S. J. and Gordon, P. (I969). The Perky effect revisited: Paradoxical
threshold or signal detection error. Perceptual and Motor Skills, a8, 79I-7.
Sekuler, R., Rubin, E. and Armstrong, R. (I97I). Processing numerical
information: A choice time analysis. Journal of Experimental Psychology, 8g, 75-
8o.
Semmes, J., Weinstein, S., Ghent, L. and Teuber, H.-L. (I955). Spatial
orientation in man after cerebral injury: I. Analyses by locus oflesion. Journal
of Psychology, 39, 227-44.
Semmes, J ., Weinstein, S., Ghent, L. and Teuber, H.-L. ( I963). Correlates of
impaired orientation in personal and extrapersonal space. Brain, 86, 747-72.
Senter, R.J. and Hoffman, R. R. (I976). Bizarreness as a nonessential variable
in mnemonic imagery: A confirmation. Bulletin of the Psychonomic Society, 7,
I63-4·
Shallice, T. and Warrington, E. K., ( I970). Independent functioning of verbal
memory stores: A neuropsychological study. Quarterly Journal of Experimental
Psychology, aa, 26I-73.
Sheehan, P. W. (I 966a). Accuracy and vividness of visual images. Perceptual and
Motor Skills, 113, 39I-8.
Sheehan, P.· W. (I966b). Functional similarity of imaging to perceiving:
Individual differences in vividness of imagery. Perceptual and Motor Skills, ~~~
IOI I-33·
Sheehan, P. W. (I 967a). A shortened form of Betts' Questionnaire Upon Mental
Imagery. Journal of Clinical Psychology, 113, 38~.
Sheehan, P. W. ( I967b). Reliability of a short test of imagery. Perceptual and Motor
Skills, as, 744·
Sheehan, P. W. (I967c). Visual imagery and the organizational properties of
perceived stimuli. British Journal of Psychology, 58, 247-52.
Sheehan, P. W. (I 972). A functional analysis of the role of visual imagery in
unexpected recall. In Sheehan, P. W. (ed.) The Function and Nature of Imagery.
New York: Academic Press.
Sheehan, P. W. (1978). Mental imagery. In Foss, B. M. (ed.) Psychology Survey,
No. 1. London: George Allen and Unwin.
Sheehan, P. W. and Neisser, U. (196g). Some variables affecting the vividness of
imagery in recall. British Journal of Psychology, 6o, 71-80.
Sheikh, A. A. ( 1977). Mental images: Ghosts of sensations? Journal of Mental
Imagery, 1, 1-4.
Shepard, R.N. ( 1966). Learning and recall as organization and search. Journal
of Verbal Learning and Verbal Behavior, 5, 201-4.
Shepard, R.N. ( 1967). Recognition memory for words, sentences, and pictures.
References
Imagery, ability 118, 130, 133 Minnesota Paper Form Board 130
control 126-7 Mnemonics 71, 80-2
mnemomc mstructions 3, 9, 67-8, Modality of presentation 66
71-83,94-7,114-5,139-40,144, Modality-specific interference 55, 57
145 Motivation 72, 122
preference 133
Survey Schedule 127 Narrative 111-6
varieties of 2
Neobehaviourism 23
Image superiority hypothesis 62-3
Neomentalism 7, 23
Imaginal Processes Inventory 127
Neuropsychology 39, 118, 134-41
Immediate Memory 3, 43-59, 135-7,
Nominalisations I 02-3
144
Non-identifiability 20-1
Inalienability 26-8
Noun phrases 101-3
lncidentalleaming 72
Incommunicability 27-8
Objective imagery ability 130
Incorrigibility 32
Objective mentalism 7
Independence-interconnectedness 14,
'One-bun' mnemonic 80
137
Operationism 22-4
Indicants 35
Operations 9-13
Individual differences 3, 51-2, 117-42
Ordinary language 24,35
144
Organisation 2, 61,75-7,90-2,99, 144
Individual Differences
Questionnaire 127
Paradigm I, 7-9
Inferred images II, 35 Perception 12-3, 54-5, 57-8,60
Intelligibility I 05-6 Phenomenal experience II, 16, 23,
Intentionally 39-41, 53, 82-3, 89,
26-9,38,42
99-100, 143, 145
Pictorial memory 3, 60-70, 115-6,
Interaction 61, 68, 69, 73-6, 86 135-9, 141, 143-5
Introspection 5, II, 38 Picture metaphor I 7-8
Introspectionism 4-5, 23, 25-7 Picture-word comparisons 49-51,61
Introspectivequestionnaires 118-27, Post-learning questionnaires 127-30
130, 141 Presentation rate 78, 97-8
Prima facie reasons 31
Kinetic imagery 43 Primary memory 2, 71, 74, 85, 140
Privacy 26-8, 31
Labelling 62-3 Propositionaltheory 15-6, 40, 77, 82-3,
Latency of image arousal 85, 87 87, 107, 115-6, 139, 144, 145, 147
Lexical complexity 93-4 Prosopagnosia 64
Literal function of imagery 43 Protocols 35
Location 66, 69 Psychological states 32-4, 36-7
mnemonic 80, 81 Psychophysicaljudgements 47, 54