BEHAVIORAL AND BRAIN SCIENCES (1997) 20, 1–55

Printed in the United States of America

What memory is for

Arthur M. Glenberg
Department of Psychology, University of Wisconsin, Madison, WI 53706
Electronic mail: glenberg6facstaff.wisc.edu

Abstract: Let’s start from scratch in thinking about what memory is for, and consequently, how it works. Suppose that memory and
conceptualization work in the service of perception and action. In this case, conceptualization is the encoding of patterns of possible
physical interaction with a three-dimensional world. These patterns are constrained by the structure of the environment, the structure of
our bodies, and memory. Thus, how we perceive and conceive of the environment is determined by the types of bodies we have. Such a
memory would not have associations. Instead, how concepts become related (and what it means to be related) is determined by how
separate patterns of actions can be combined given the constraints of our bodies. I call this combination “mesh.” To avoid hallucination,
conceptualization would normally be driven by the environment, and patterns of action from memory would play a supporting, but
automatic, role. A significant human skill is learning to suppress the overriding contribution of the environment to conceptualization,
thereby allowing memory to guide conceptualization. The effort used in suppressing input from the environment pays off by allowing
prediction, recollective memory, and language comprehension. I review theoretical work in cognitive science and empirical work in
memory and language comprehension that suggest that it may be possible to investigate connections between topics as disparate as
infantile amnesia and mental-model theory.
Keywords: amnesia; automatic memory; coherence; embodied meaning; inferences; language comprehension; memory; mental models;
recollective memory; symbol grounding

1. Introduction memory? Because, by ignoring them, we have been making

a big mistake. Most theories of memory treat internal
Most memory theories presuppose that memory is for representations as meaningless symbols such as a string of
memorizing. What would memory theory be like if this zeros and ones that “encode” features (e.g., Hintzman
presupposition were discarded? Here, I approach memory 1986; McClelland & Rumelhart 1986; Metcalfe 1993), as
theory guided by the question, “What is memory for?” The pointlike objects with no structure (Gillund & Shiffrin
answer that I develop is influenced by three sources. The 1984), or as propositions relating intrinsically meaningless
first is Lakoff and Johnson’s (Johnson 1987; Lakoff 1987; symbols (Kintsch 1988). Two problems arise from this
Lakoff & Johnson 1980) cognitive linguistic analysis of treatment. The first is the symbol grounding problem
language, conceptualization, and meaning. They propose (Harnad 1990): How do those meaningless symbols come
that cognitive structures are embodied; they arise from to take on meaning? The answer is not as simple as referring
bodily interactions with the world (cf. Harnad 1990; 1993). the symbol to a lexicon, because words in the lexicon must
After a brief review of the Lakoff and Johnson program, I also be grounded. Also, not all of those meaningless sym-
examine the literature on memory (the second source) for bols are meant to represent words or wordlike concepts;
evidence that cognitive structures are, indeed, embodied, some are meant to represent complex nonverbal displays
and why that is so. I will propose that memory evolved in (Posner & Keele 1968; Schacter et al. 1990). The second
service of perception and action in a three-dimensional problem is that we have not availed ourselves of a golden
environment, and that memory is embodied to facilitate opportunity. By treating internal representation as mean-
interaction with the environment. The third set of ideas ingless symbols, we have not thought about the possibility
comes from research on mental-model theory of language of taking advantage of other forms of representation. In-
comprehension. I relate mental-model theory to the notion stead of meaningless symbols, suppose that representations
of embodied memory by proposing that because language have a structure that is lawfully related to the objects being
acts as a surrogate for more direct interaction with the represented. The structure of the representations might
environment, language comprehension must also result then play an important role in determining, for example,
in embodied representations, which are in fact mental what concepts are easily associated, because their struc-
models. In exploring these ideas, I develop an approach to tures literally fit together. For example, it seems easy to
memory and language comprehension that suggests ways of associate “horse” and “spotted” because horses have sur-
dealing with old problems (e.g., why recollection and com- faces that can be spotted, whereas it is more difficult to
prehension are effortful), as well as new concepts to replace associate “idea” and “spotted.” Note that this sort of think-
old ideas (e.g., an association). ing trades on the analogical nature of the representations
rather than on propositional listings of content (see Palmer
1.1. Why embodiment should matter 1978). That is, we could just as easily assert “the idea was
to cognitive psychologists spotted” as “the horse was spotted.” Nonetheless, one
Why should psychologists interested in language, learning, seems to make sense and the other does not.
and memory care about issues such as embodiment of In the next few sections I develop the case that internal

Q 1997 Cambridge University Press 0140-525X/97 $9.001.10 1

Glenberg: What memory is for

representations are analogically structured (embodied), perceptual system. That is, what a symbol means is what it
that this structure helps to explain memory phenomena, refers to in the “outside” world.
and that in conjunction these ideas suggest that the stan- Lakoff (1987) presents (at least) three arguments against
dard memory paradigms are ill-conceived and that standard the plausibility of generating meaning by this sort of symbol
memory phenomena may be revealing little that is impor- grounding. First, this theory requires that categories be
tant about memory. These sections are followed by a Aristotelian, that they have sharp boundaries. Aristotelian
discussion of the possibilities for analogical representation categories are needed so that we can successfully map
underlying language comprehension. between the arbitrary symbols in the propositions and the
elements in the world. Thus, if a proposition is supposed to
be about a horse, to give the proposition its proper meaning
1.2. Embodiment and the Lakoff and Johnson program
we must be able to map the symbol for horse (X19, perhaps)
A central concern of the Lakoff and Johnson program is the onto horses, and exclude zebras and antelopes and perhaps
concept of meaning. According to Lakoff (1987), the stan- even ponies. In contrast to this requirement, there is a
dard theory of meaning in cognitive science is based on the tremendous amount of empirical work in the psychology of
notion of truth values of propositions, and, as it turns out, human categorization implying that categories in the head
this theory will not work as a theory of human meaning. are not Aristotelian. Instead many significant categories
Explication of why this is so requires a bit of patience, in have fuzzy boundaries (Oden 1984; 1987), graded member-
part because the way psychologists use the term “proposi- ship (Kalish 1995), complex structures (Lakoff 1987), or are
tion” is different from the way philosophers and logicians based on prototypes (Rosch 1973). Furthermore, the exten-
use it. For the psychologist, propositions are relations sions of even basic biological categories are less than
among symbols, that is, an assertion that a relation exists. It certain, and categories based on human culture are even
is these assertions that are supposed to be meaningful. more fuzzy. Thus, categories such as democracy, justice,
Importantly, although the propositions are supposed to and mother (Lakoff discusses biological mothers, birth
capture meaning, the symbols used in the propositions are mothers, adoptive mothers, stepmothers, etc.) seem to have
taken to be, by themselves, meaningless or arbitrary: there structures quite different from the classical Aristotelian
is no intrinsic relation between a particular symbol and its category.
meaning. Thus when illustrating propositions, a psycholo- A second argument against the standard theory as a
gist may use a word to stand for an element in the proposi- theory of human meaning is based on an analysis of Putnam
tion, but that is just a convenience. Indeed, the meanings of (1981). This analysis, however, is directed toward the phi-
the words need to be specified, presumably by other losopher’s meaning of proposition, and so it requires a bit of
propositions. Thus we should replace any words in a psy- new terminology. To the psychologist, a proposition (like
chologist’s proposition with things such as “symbol X19.” aRb) is supposed to have meaning. To the philosopher,
This state of affairs is quite useful because it allows for aRb is a sentence in a formal language. The meaning of
reasoning (the derivation of new propositions) to be based the sentence (its propositional content) corresponds to
on the manipulation of propositions by syntactic rules. the function that determines, for any possible situation,
These rules are thought to operate independent of the whether that sentence is true or false. In plain language,
referents of the elements (nodes and symbols) in the which is not an exact equivalent but close enough, the
propositions. meaning of a sentence such as “the horse is spotted” is
For example, suppose that proposition 1 (P1) asserts that whatever allows one to determine if it applies to particular
element a is in relation R to element b. In shorthand, P1: situations. Furthermore, two sentences have the same
aRb. Furthermore, suppose P2: bRc. Now, if R is a transitive meaning if they have the same truth values for all possible
relation (such as “larger than”), and both P1 and P2 are true, situations.
then by the syntactic rules of transitive inference, P3: aRc is Putnam discovered a serious problem with this truth-
also true. Thus, for the psychologist, we have created new value notion of meaning: it is not difficult to construct pairs
knowledge, namely, that P3 is true. Note that these proposi- of formal sentences whose symbols are mapped to radically
tions have truth values, but they fail a commonsense test of different things, but that have the same truth values in all
what it means to have meaning. Namely, in order for a situations. In other words, even though the sentences are
statement to be meaningful (to us), we must know what the about radically different things on the truth-value account
statement is about. In contrast, although we know that P3 is of meaning, because the sentences have the same truth
true, we have little idea what it is about, because we have no values they are supposed to have the same meaning.
idea what a and b stand for. Clearly, it does not make much sense to assert that sen-
The problem of what a and b stand for is the symbol tences about different things mean the same thing. As it
grounding problem (Harnad 1990): How do we give mean- turns out, the problem is with the arbitrary nature of the
ing to the arbitrary symbols? To know what these proposi- symbols. They only mean when they are mapped onto the
tions are about requires a mapping between the elements in world, and Putnam demonstrated that it is impossible to
the propositions (a, b, c, and R) and the world (or a possible find the one and only correct mapping.
world, or a model of the world). Without this mapping, the Lakoff and Johnson’s third argument against the standard
symbols can only refer to other symbols, which in turn refer theory is based on their analysis of language use and what it
to yet other symbols. Just like trying to learn the meaning of implies about cognition. In brief, people frequently use
a word in a foreign language by using a dictionary written metaphorical language (“He’s trapped in his marriage,”
solely in that language, such a system of symbols will never “Your theory is airtight,” “I’m really high today”). Further-
generate meaning (Searle 1980). Most psychologists don’t more, Lakoff and Johnson propose that metaphorical lan-
see a problem here, because they are happy to point to guage is not just the way people talk, but accurately reflects
perception: the arbitrary symbols are grounded by the the way people think. Given that theories cannot literally be

2 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Glenberg: What memory is for

airtight and that people’s emotional states cannot literally the body. Because the same object may be useful for
be high, it is hard to imagine how cognition could be based accomplishing a variety of goals, categorization can be
on the mappings of arbitrary symbols and produce such flexible and context dependent (Barsalou 1993).
(easy to understand) language. Consider three objections to these claims. The first is
Several other cogent arguments against the use of arbi- that because we have different bodies, we will understand
trary symbols in a theory of meaning can be found in the world in different ways. In fact, that is a valid prediction.
Barsalou (1993; Barsalou et al. 1993) and Shanon (1988). For example, what makes an object a chair for a particular
Barsalou and Shanon note that people have a hard time individual will depend on whether or not that individual is
defining many familiar words, and that the definitions can able to get his or her body into a sitting position using the
vary greatly with context. This finding is difficult to explain object. Thus, depending on the height of the object, the
if one believes that meaning of words is a simple list of well- width of the flat surface, the object’s strength, and so on,
formed propositions. They also note that there is no good the object will be a chair for some people (e.g., a child) but
account of how propositions composed of meaningless and not for others (e.g., an aging grandfather).1 Nonetheless,
arbitrary symbols might have evolved or how a child could bodies are substantially the same around the world and
have discovered them. As Shanon concludes, across cultures. Thus, although there will be variability
Specifically, it appears that the underlying substrate of mental around the edges, our common human endowments and
activity is not a repertory of well-defined, well-structured ab- our common environment ensures a great degree of com-
stract symbols, and that the workings of mind cannot be gener- mon center to cognitive structure.
ally characterized as the computational manipulations of such A second objection to the claim that cognitive represen-
symbols. Rather, the substrate in which mental activity takes
place should be one that meets the following requirement: It
tation is embodied is that the mapping problem has not
should not be fixed by any coding system that is defined a priori, been solved; there is still the problem of mapping (arbi-
it should afford maximal sensitivity to unspecified dimensions trary) words to embodied representations so that we can
and distinctions, it should be context-sensitive, and it should be talk about what we are perceiving and thinking. This is a
embedded in the framework of the organism’s action in the deep problem (e.g., Harnad 1990; Plunkett et al. 1992), but
world. (p. 80) it is not one that I intend to address here. The point of the
That is a call for an embodied approach to meaning. above is that embodied representations allow us to under-
stand how, except for the seriously deranged, we all know
the difference between say, horses and ideas, and contrary
1.3. Embodiment and meaning
to what Putnam’s analysis shows of the standard theory, we
If we dismiss the standard theory, what is left? Lakoff and don’t ever confuse them.
Johnson offer a theory of meaning based on the concept of A third objection is that some things are meaningful (e.g.,
embodied knowledge. Because I will be approaching the a beautiful sunset) even when there is no apparent possi-
problem from the question “What is memory for?” I will bility for bodily interaction.2 The embodied account of
develop an idea of embodied meaning that is distinct from meaning is situated, so that action-oriented meaning can
the Lakoff and Johnson proposal. Nonetheless, the pro- vary greatly with context. Thus, depending on the context, a
posals are clearly related. In outline, my proposal is that Coke bottle can be used to quench thirst, or as a weapon, a
perceptual systems have evolved to facilitate our interac- doorstop, or a vase. That is, its meaning depends on the
tions with a real, three-dimensional world. To do this, the context. Similarly, a beautiful sunset is a context that com-
world is conceptualized (in part) as patterns of possible bines with objects and memories to suggest actions consis-
bodily interactions, that is, how we can move our hands and tent with warmth, relaxation, and a good beer.
fingers, our legs and bodies, our eyes and ears, to deal with Later I will discuss how embodied representations can
the world that presents itself? That is, to a particular person, be extended to represent abstract concepts and how they
the meaning of an object, event, or sentence is what that may provide a novel way of dealing with unanalyzed con-
person can do with the object, event, or sentence. cepts such as association. For now, however, I turn to
How does this approach answer the objections raised to developing a particular sketch of embodied representations
the standard theory of meaning? Importantly, embodied that arises from a consideration of what memory is for. This
representations do not need to be mapped onto the world to sketch is not a fully testable theory. The idea is to show how
become meaningful because they arise from the world. In a type of theory that is not subject to the criticisms leveled at
other words, embodied representations are directly meaningless symbol theory can handle problems of mem-
grounded by virtue of being lawfully and analogically re- ory and comprehension.
lated to properties of the world and how those properties
are transduced by perceptual-action systems (Harnad 1990; 2. What memory is for
1993). Thus, the meaningful, action-oriented component of
conceptualization is not abstract and amodal. It reflects Except for the recent blossoming of interest in indirect
how bodies of our sort can interact with objects. memory (see sect. 5.1), the contemporary psychology of
Given that embodied representations do not need to be memory has been dominated by the study of memoriza-
mapped onto the world to be grounded, there is neither tion.3 In part, this seems to have arisen from a failure of
need for representations to be Aristotelian nor for the many twentieth century memory theorists to consider what
categories in the world to be Aristotelian. Furthermore, memory is for. By the end of section 5, I will have concluded
because embodied representations are not discrete, mean- the following: memory is embodied by encoding meshed
ingless symbols, they can reflect subtle, fuzzy variations in (i.e., integrated by virtue of their analogical shapes) sets of
the world. How then do categories arise? Objects fall into patterns of action. How the patterns combine is constrained
the same (basic) category because they can be used to by how our bodies work. A meshed set of patterns corre-
accomplish the same interactive goal, such as supporting sponds to a conceptualization. Updating memory occurs

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 3

Glenberg: What memory is for

whenever the meshed patterns change (a change in concep- with patterns of interaction based on memory. The two
tualization of the environment) and the updating is in terms patterns can combine because they are both embodied, that
of a change, or movement, or trajectory toward a new set of is, both are constrained by how one’s body can move itself
meshed patterns. Thus, memory records how conceptual- and manipulate objects. The resulting pattern of possible
izations blend into one another. This memory works in two actions is a conceptualization: the possible actions for that
broad modes. First, patterns of action based on the environ- person in that situation. For example, “turn left to get
ment (projectable properties of the environment) are auto- home.”
matically, that is, without intention, meshed with patterns Thus meaning of an object or a situation is a pattern of
based on previous experience. This automatic use of mem- possible action. It is determined by the projectable features
ory corresponds closely to implicit or indirect memory. of the object molded by bodily constraints and modified by
Second, patterns from the environment can be suppressed memory of previous actions. These memories provide the
so that conceptualization is guided by previous experience nonprojectable features. As another example, consider the
encoded as trajectories. This is a conscious and effortful use meaning of the cup on my desk. The embodied meaning is
of memory. The ability to suppress environmental patterns in terms of how far it is from me (what I have to do to reach
contributes to prediction, the experience of remembering, it), the orientation of the handle and its shape (what I have
and language comprehension. to do to get my fingers into it), characteristics of its size and
material (the force I must exert to lift it), and so forth.
2.1. The function of memory in a dangerous Furthermore, the meaning of the cup is fleshed out by
environment memories of my previous interactions with it: pouring in
coffee and drinking from it. Those memories make the cup
We live in a dangerous, three-dimensional world. Given the
mine.
size, density, and physical capabilities of our bodies, the
Note three characteristics of this sort of meaning. First,
natural environment is hostile. We are open to predation,
because bodily actions take place in space, embodied
and our interactions with the world can lead to injury from
meaning captures spatial (or topological) and functional
freezing, burning, drowning, and falling. Clearly, survival
properties. Thus a synonym for this type of embodied
requires the capability to navigate this environment and,
meaning is spatial-functional meaning. Second, because we
just as clearly, our perceptual system has evolved to do just
interact with objects via parts, conceptualization in terms of
that. For example, we have developed impressive abilities
bodily interaction forms the basis for partonomies (Tversky
to use information (e.g., optical flow fields) to guide action
& Hemenway 1984) and basic-level categorization. Third,
so that obstacles are avoided. These abilities may not
conceptualization in terms of patterns of bodily interaction
require any sort of representation of the environment and
is very close to Gibson’s (1979) notion of affordance.
they may not require memory; responding constrained by
Thus, what is memory for? Its primary function is to
characteristics of the environment and our bodies guaran-
mesh the embodied conceptualization of projectable prop-
tees successful action (for a review, see Bruce & Green
erties of the environment (e.g., a path or a cup) with
1985).
embodied experiences that provide nonprojectable proper-
On the other hand, it is frequently the case that we need
ties. Thus the path becomes the path home and the cup
to differentiate. In addition to avoiding obstacles in our
becomes my cup. This meshed conceptualization, the
path, we need to pick out and follow a particular path, avoid
meaning, is in the service of control of action in a three-
a particular location, or approach a particular person. This
dimensional environment.
sort of differentiation requires a memory system. What
makes one person a particular person (to you) or one path
the path to your house, is its relevance to you, that is, how 2.3. Evidence for embodied conceptualization
you have interacted with it in the past. An optical-flow field How far can this account of embodied meaning be pushed?
cannot contain this information; it is the province of mem- At the least, there are intriguing results that fit this account
ory. This distinction is discussed by Epstein (1993), who nicely and that do not seem to have a natural explanation in
uses the term “projectable” to refer to properties of the cognitive accounts based on meaningless symbols. I will
environment that can be specified by information available review some of this literature from the domains of affect,
in the light and “nonprojectable” to refer to properties that memory, and imagery.
must be signaled by other sources. Thus spatial layout is a
2.3.1. Embodiment and affect. Van den Bergh et al. (1990)
projectable property, whereas ownership is a nonproject-
presented typists and nontypists with sets of letter pairs
able property that must come from experience.
(e.g., WX and ZD). The subjects were asked to choose the
one pair (from each set) that was liked the best. Typists
2.2. Embodied conceptualization, memory, showed a clear preference for pairs typed with different
and meaning fingers over pairs typed with the same finger, whereas the
To support action, the perception of projectable properties nontypists showed little preference. (The typing finger was
is in terms of patterns of possible action: how we can determined using AZERTY keyboards in Belgium and
examine, grasp, shove, leap over, or move around an object. QWERTY keyboards in the United States.) Van den Bergh
This coding depends on the capabilities of our bodies, both et al. argued that, for typists, part of the encoding of letters
as a species and as individuals. Because the world is per- is that of a motor program or movement. The incompatible
ceived in terms of its potential for interaction with an movements generated by letters typed with the same finger
individual’s body, it is proper to call the perception “em- resulted in a negative evaluation. It is unlikely that this
bodied.” effect arose from associations to specific letter combina-
Patterns of action derived from the projectable proper- tions because the effect was most robust for pairs of letters
ties of the environment are combined (or meshed, sect. 3.1) with low frequency in the language.

4 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Glenberg: What memory is for

Berkowitz and Trocolli (1990) and Berkowitz et al. (1993) may well be “efferent leakage” that can be measured using
illustrate the influence of the body on affect judgments. In psychophysiological techniques. In support of these ideas,
one experiment, subjects were asked to judge the person- Cuthbert et al. note that psychophysiological responsivity is
ality of a fictitious person described in neutral terms. Half particular to the image being evoked. Thus imagining a
the subjects listened to the description while holding a pen fearful situation evokes sweating, imagining positive situa-
between their teeth without using their lips. This activity tions results in measurable activity in muscles associated
forces the face into a pattern similar to that produced by with smiling, and imagining negative situations results in
smiling. The other subjects listened to the description while activity in the muscles associated with furrowing of the
biting down hard on a towel. This activity forces the face brow. There are analogous effects for imagery related to
into a pattern similar to that produced by frowning. The other perceptual/action systems. Thus, in imaging pendu-
subjects who were smiling rated the person described more lar motion, discharges in the eye muscles follow the appro-
positively than did the subjects who were frowning. It is priate frequency, in imaging bicep curls there are dis-
unlikely that this effect arose due to demand characteristics charges in the biceps, and in imagining the taste of a
of the experiment for the following reason: the effect was favorite food there is an increase in saliva flow. These results
obtained only when subjects were distracted from their are compatible with the notion of embodied, spatial-
activities. When they were asked to focus on the activities, functional encoding. In addition, the idea of embodied
the subjects seemed to compensate for the forced smile encoding has an advantage over Lang’s response proposi-
(frown) and rate the description more negatively (pos- tions. According to Cuthbert et al. (1991), the function of
itively). What can account for this finding? Experienced imagery is to allow new behaviors to be tried out “off-line.”
emotion is embodied. When the body is manipulated into a It is not clear however, how response propositions can be
state that is highly correlated with an emotion, the body integrated (other than by concatenation) to effect this
constrains other cognitive (that is embodied) processing. rehearsal. In contrast, the integration of responses is basic
to the notion of mesh (see sect. 3.1) of embodied encodings.
2.3.2. Embodiment and imagery. Montello and Presson That is, given that the information is encoded in terms of
(1993) asked subjects to memorize the locations of objects bodily interaction, effecting one action (or imagining it)
in a room. The subjects were then blindfolded and asked to necessarily constrains the operation of simultaneous and
point to the objects. Pointing was fast and accurate. Half of successive actions.
the subjects were then asked to imagine rotating 908 and to
point to the objects again. That is, if an object was originally 2.3.3. Embodiment and memory. Effects of embodiment
directly in front of the subject and the subject imagined are revealed by research on memory for subject-performed
rotating 908 clockwise, the correct response would be to tasks (Cohen 1981; Engelkamp & Krumnacker 1980; Saltz
point to a location toward the subject’s left. In this condi- & Donnenwerth-Nolan 1981; see also a special issue of
tion, the subjects were slow and inaccurate. The other Psychological Research, 1989, including Englekamp &
subjects, while blindfolded, were asked to actually rotate Zimmer). The basic finding is that memory for actions
908 and to point to the objects. These subjects were just (performing a command such as “open the book”) is better
about as fast and accurate as when pointing originally. Thus, than memory for the verbal description of the commands.
mentally keeping track of the locations of objects, a task that One interpretation of this finding is that memory special-
many cognitive psychologists would suspect as being cogni- izes in embodied information.
tive and divorced from the body, is in fact strongly affected The nature of our bodies also controls ease of remember-
by literal body movements. ing. Consider a series of studies by Tversky and her col-
Rieser et al. (1994) reported a similar finding for children leagues (e.g., Bryant et al. 1992). In these experiments,
and adults. The participants were tested for the ability to subjects read about and memorized spatial layouts corre-
imagine (while at home) their classrooms and to point to sponding to scenes viewed from particular perspectives
objects from various perspectives. When the perspective (e.g., in the hotel scene, “To your left . . . you see a
change was accomplished by actually changing position (at shimmering indoor fountain”). Objects were located above,
home), the 5-year-olds were correct on 100% of the trials, below, in front, in back, to the left, and to the right of the
the 9-year-olds were correct on 98%, and the adults on observer in the imagined scene. After the scene was mem-
100%. When the perspective change was accomplished orized, the time taken to retrieve a particular object was
solely by imagination, the 5-year-olds were correct on 2% of measured. For equally well memorized locations, one
the trials, the 9-year-olds were correct on 27%, and the might expect the retrieval times to be independent of
adults were correct on 100%. Even the adults showed great location. Another hypothesis is that the times would be
difficulty in terms of the time needed to accomplish the correlated with the degree of mental rotation needed to
imagination-only version of the task. When actually chang- mentally face the object. The results, however, were con-
ing position, 100% of the adult responses required less than trary to both of these hypotheses. Fastest responding was to
two seconds, whereas when imagination was used, only objects located on the head/feet axis, followed by the
29% of the responses required less than two seconds. front/back axis, followed by the left/right axis. Tversky
Findings on the psychophysiology of imagery also point argues that these results follow from using a “spatial frame-
to a close connection between body and cognition. These work” that is sensitive to environmental asymmetries (such
findings are summarized by Cuthbert et al. (1991). Their as gravity) and perceptual asymmetries (we generally look
starting point is Lang’s (1979) bio-informational theory, and attend to the front). In other words, retrieval processes
which asserts that encoding of events includes response appear to be sensitive to how we use our bodies.
“propositions,” and that imagery (visual and otherwise) is Klatzky et al. (1989) demonstrated contributions of the
the activation of those propositions. Furthermore, although body to symbolic or semantic judgments. They trained
overt responding is inhibited during an imagery task, there subjects to make hand shapes corresponding to descrip-

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 5

Glenberg: What memory is for

tions such as “pinch” or “clench.” The verbal descriptions stepping across the rocks), the rocks, soil, and twigs become
were then used as primes for judging the sensibility of (for you) a path.
phrases such as “aim a dart” (sensible) or “close a nail” (not I envision mesh of embodied encodings as being analo-
sensible). The appropriate prime, that is a prime corre- gous to coarticulation in speech production. When pro-
sponding to the hand shape used in the to-be-judged action, nouncing the initial /d/ in “dog,” the articulators are shaped
speeded the sensibility judgment compared to a neutral in part by the requirement to enunciate the following
prime. Thus, the hand shape for “pinch” speeded the vowel, and when pronouncing the vowel, the articulators
sensibility judgment for “aim a dart.” It is unlikely that this are shaped not only by the vowel, but by the preceding and
priming effect derives from any sort of verbal mediation: following consonants. Furthermore, the constraints on ar-
the priming effect was found for subjects trained to make ticulation are not consciously imposed, but are constraints
the hand shapes when signaled by nonverbal primes. Also, that follow from real movements of physical devices: the
when subjects were trained to make verbal responses (but tongue can only be in one place at one time, and how it is
not hand shapes) to the nonverbal primes (e.g., saying the going to move to the next place will depend on where it is
word “pinch” when shown the nonverbal signal for pinch), now. Thus pronunciation of the word requires a mesh of
the priming effect was eliminated. Klatzky et al. suggest real physical actions.
that the sensibility judgment requires a type of mental An example of cognitive meshing is borrowed from
simulation using an embodied, motoric medium. Generat- Barsalou et al. (1993). Imagine a ball; now imagine that it
ing the appropriate hand shape “facilitates constructing the has yellow and white stripes; now imagine that the ball is
representation and/or simulating the action/object pairing” deflated (it is a beach ball). Adding each new descriptor is
(p. 75). not a matter of adding a simple association or adding a
proposition to a list. Instead, each previously constructed
representation constrains how the new descriptor is uti-
3. How embodied memories are used
lized. Thus, the yellow and white stripes surround the ball.
Consider this scenario. You have been wandering in the Then, not just the ball, but the stripes too become de-
woods, and suddenly you are unsure of the way home. You formed when the ball is deflated. The stripes and the ball
see what appears to be a path, but you are not certain if it deflate together because they are encoded as patterns of
really is a path, yet alone the path home. You take a few action subject to the same spatial-functional constraints.
steps and hunt for evidence. As you continue your explora- This meshing occurs not just in imagination, but in memory,
tion, you become convinced that this is the right path: the comprehension, and perception.
patterns of rocks, twigs, and soil align themselves to form a It is the mutual modification of meshed patterns of action
connected pattern that could be a path. Also, as you move that produces emergent and creative features of thought.
along, you are able to conform your own body to the The deflated beach ball is not simply a deflated ball associ-
putative path. For example, the overhanging branches are ated with an unchanging stripes feature. Instead, the fact
not so low that you have to stoop or crawl; when you reach a that the stripes are deflated arises from the operation of
stream, the distance between the rocks forms a series of meshing. Related concepts will mesh easily, because that is
stepping stones that can be used by an animal of your size what it means to be related (sect. 7.3), and, with some
and agility. effort, we can mesh arbitrary concepts. Thus a “tiger
bicycle” is one designed for hunting tigers, and it consists of
a mesh between the actions required to hunt and those
3.1. Mesh of patterns as functional required to ride a bicycle, whereas “colorless green ideas”
constraint satisfaction are uninspired ways of dealing with environmental crises.
Recognition of the path as a path arises from an exploration In short, mesh underlies our ability to understand novel
of the environment and a fit between the environment and conceptual combinations. Note that the type of mesh I am
embodied knowledge. This fit can be conceptualized as a proposing depends on the analog nature of embodied
type of constraint satisfaction, but here the constraints are actions, not just their propositional content.
spatial and functional, not associationistic or probabilistic
(cf. Rumelhart et al. 1986). Thus projectable properties of
3.2. Clamping projectable properties
the environment (arrangement of rocks, twigs, and soil) are
encoded in terms of how you (with your particular body) Meshing patterns of action based on projectable properties
can interact with that environment (e.g., whether the dis- of the environment with those from memory changes the
tances between the rocks in the creek can be broached). way we conceptualize the environment. Thus, the soil,
Other patterns of interaction come from memory, for twigs, and rocks are conceptualized not just as a path, but as
example, patterns representing goals such as “get home the path home. There is a danger, however, in allowing
without getting wet.” In conceptualizing the environment patterns from memory to modify conceptualization: mesh-
as a path, the spatial-functional patterns based on project- ing of patterns can distort the perception of the environ-
able properties from the environment are combined or ment. Clearly, survival requires seeing the environment for
meshed with the patterns from memory. The meshed what it is (soil, twigs, rocks), not just for what it means (the
pattern dictates how (or if ) the body can be moved in a way path home). To keep the system reality-oriented, it is
that simultaneously satisfies both sets of patterns of action necessary to ensure that patterns based on projectable
(e.g., “Can I, with my body, get from rock to rock without properties of the environment are primary. That is, the
getting wet?”). This sort of mesh is a possibility because all meshed conceptualization that is achieved cannot be at the
of the patterns are embodied, that is, they are all encoded in cost of distorting the environmental input. I will refer to this
terms of how your body constrains actions. When the as “clamping” projectable properties of the environment.
patterns can be meshed into a plan for coherent action (e.g., Clamping projectable properties ensures that experi-

6 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Glenberg: What memory is for

ences are individuated or situated. We do not experience and trajectories approaches the ideal of enabling memory
categories, but individual, particular events (cf. Barsalou et to code the full variety of human experience.
al. 1993). We cannot direct our perceptual system to ignore Because updating of trajectories occurs only when there
differences between two paths, just because they are both is a change in conceptualization, memory is sensitive to
paths, or between two chairs just because we can fit our frequency and to novelty. To illustrate this, consider once
bodies into both. Because the projectable properties are again walking the path home. Three phenomena are associ-
clamped, the two chairs, although members of the same ated with repeated actions: (1) memory for the repeated
category, remain separate chairs. action (walking the path) will be an increasing, but nega-
tively accelerated, function of frequency (e.g., Logan 1988);
(2) memory for a particular typical repetition of the action
3.3. Updating memory
will be poor (Glenberg et al. 1977; Naveh-Benjamin &
I have proposed that embodied memory acts as a source of Jonides 1984); (3) memory for a particular unusual repeti-
nonprojectable patterns of action that mesh with patterns tion of the action will be good (Hunt 1995). The frequent
derived from projectable properties of the environment; interactions with the path will result in frequent updating
the mesh is possible because both sets of patterns are (reinforcement of a particular trajectory) and, conse-
constrained by how the body works. If memory is to be quently, a shift toward a stable conceptualization (e.g., a
useful, however, it must be updated. That is, new experi- shift from possibly the path home to definitely the path
ences must affect the system so that we come to learn the home). However, once the conceptualization is stable, little
path home. Because experience is continuous (or at least further updating occurs. Thus each encounter with the path
the environment appears continuous to beings of our size will have less and less of an impact (phenomenon 1).
and abilities), we must deal with how it can be captured by a Because typical encounters result in little new conceptual-
system using a finite brain. ization and little updating, we have little memory for the
Consider this possibility. Projectable properties are individual steps down the path (phenomenon 2). However,
clamped and then embodied memories mesh to produce a if reconceptualization is required (e.g., when a log appears
particular conceptualization (e.g., the path home). At this across the path, so that now, in terms of bodily constraints
point, either an action is taken (e.g., a step along the path) or on action, the path is a blocked path) memory is again
projectable properties of the environment change (e.g., a updated, leading to memory for novel events (phenomenon
barrier appears). In either case, the system is forced to 3).
settle into a new conceptualization. Here is the proposal for
updating memory: memory is updated automatically (that
3.4. Prediction and suppression
is, without intention) whenever there is a change in concep-
of the clamped environment
tualization (mesh). The degree to which updating takes
place is exactly correlated with the degree to which the The meshed conceptualization of the current environment
conceptualization changes. dictates what actions are possible in that environment.
Updating is not encoding a new memory trace. Instead, Prediction, however, requires simulating how an action will
the shift from one pattern of possible actions (one concep- produce a new conceptualization, which in turn can be used
tualization) to the next is reinforced. That is, what is to simulate the next action, and so forth. Two difficulties
updated is how one situation flows into another. I will refer arise. The first is that simulated action does not change the
to this flow as a “trajectory,” using the term to imply that the environment. Thus, changes in projectable properties that
change is not random. Instead, actions humanly possible would have resulted from a real action cannot be clamped
under the current conceptualization are biased by what was to automatically guide further action. A second difficulty is
possible in the previous conceptualization, just as pronun- that currently clamped stimulation provides the wrong
ciation of a vowel is biased by the pronunciation of the constraints, because those constraints are only relevant
preceding consonant. before the simulated initial action. I believe that this is a
The idea of trajectories solves several problems in the major problem, and that it requires a radical (and dan-
psychology of memory. It provides a way of conceptualizing gerous) mechanism: suppression. In particular, I propose
dynamic information in memory that is sensitive to biolog- that in the service of prediction, we have developed the
ical and spatial-functional constraints (Shiffrar et al. 1993). ability to, if not ignore, at least to suppress the overriding
Trajectories can reflect minimal changes in conceptualiza- contribution of the current environment to conceptualiza-
tion, such as from one step along a path to another, or gross tion. This is a risky operation because it loosens the tie
changes such as from a step to a fall. The idea offers the between reality (the current environment) and conceptual-
beginnings of a solution to the problem of features. Most ization. Perhaps because suppression is so dangerous, it is
theories of memory are based on the idea that memories are an effortful process. As we will see, however, suppression
multidimensional, consisting of a vector of features, such as results in several serendipitous abilities, including con-
“animate,” “red,” and “smaller than a bread box.” None of scious autobiographical memory and language comprehen-
these theories, however, is committed to a listing of what sion.4
those basic features might be. In fact, because experience is Once clamping of projectable properties is suppressed,
so varied, it is hard to imagine a complete list. Also, given a multistep prediction arises from following trajectories
feature-based system, it is difficult to understand how guided by bodily constraints on action. For example, by
people can ever learn anything truly new: we must always following trajectories we can envision what will happen
conceptualize using the same basic features. In contrast, when we proceed down the path. We also have the ability to
because embodied patterns of action can be infinitely envision arbitrary events (such as what actions are possible
varied and infinitely meshed with goals (also specified as if the path is washed out by a storm or blocked by strange
patterns of action), a system based on embodied concepts creatures), not just events we have previously experienced.

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 7

Glenberg: What memory is for

Prediction for these arbitrary scenarios is based on seeking Thus, a connectionist system would be equally happy to
a mesh among patterns of action. Some of the patterns are learn that a Coke bottle can be used as a chair or as a
based on well-learned trajectories. Other patterns (e.g., weapon. In an embodied system, constraints arise because
interactions with strange creatures) come from a consider- of analog coding of projectable features and their implica-
ation of how our bodies work. These patterns can mesh to tions for human action. In Palmer’s terminology, these
give a coherent conceptualization because they are all constraints are intrinsic to the operation of the system. For
based on bodily interaction. Keep in mind, however, that in example, how we think about a Coke bottle is constrained
prediction the mesh of these patterns may not be guided by not just by particular experiences with Coke bottles, but by
stable and projectable features of the environment. To the the actual shape and heft of the bottle, too. Thus, an
extent that environmental constraints are suppressed, and embodied system would have little difficulty understanding
to the extent that trajectories are not well-learned, the how a Coke bottle could be a weapon, but it would balk at
predictions will tend to be variable and inaccurate. Thus, it learning that it could be used as a chair.
is easy to predict the outcome of the next step on a well- These differences are not unique to my proposal. Lakoff
traveled path: the simulated mesh is strongly constrained by (1988) argues that connectionist systems need to be
projectable features of the current environment and well- grounded in the body to give meaning to connections and
learned trajectories. It is more difficult to predict what will constraints. As an illustration, he notes that phonology is not
happen many steps down a new path when the projectable arbitrary; instead, it is constrained by the muscles, shapes,
features must be suppressed and trajectories uncertain. and control of articulation. Shepard (1988) makes a
related point regarding the abilities of connectionist
systems to self-organize and generalize: “[N]ontrivial self-
3.5. Mesh and connectionism
programming can take place only if some knowledge about
Many of the ideas and much of the terminology introduced the world in which the system is to learn is already built in.
in section 3 are borrowed from connectionist approaches to Any system that is without structure has no basis for
cognition. Some examples are constraint satisfaction, tra- generalization to new situations” (p. 52). How the body can
jectories as paths through a set of states, and clamping of interact with the world provides just such a basis for
projectable features. There are two other, perhaps deeper, generalization.
similarities. As I will discuss in section 6.2, an embodied These comments should not be taken to mean that an
conceptualization functions as a preparatory state. Given a embodied system cannot be simulated using connection-
particular conceptualization, an organism is better pre- ism. In fact, it may well be that connectionism will be the
pared to act when changes in the situation easily mesh with surest route to formalizing these ideas. Nonetheless, it will
the conceptualization than when changes do not easily have to be a connectionism that differs from the sorts
mesh (i.e., we are surprised). This notion of preparedness currently in use.
underlies priming phenomena, and it is close to connec-
tionist interpretations of semantic priming developed by
4. Memory in the long term and in the short term
Masson (1995) and Sharkey and Sharkey (1992).
The second deeper similarity relates to ideas of con- The system described so far seems to be useful for negotiat-
text and situated representation. For example, Smolensky ing the environment, and it seems to correspond to what
(1988) discusses how a distributed representation of “cof- some have called semantic (Tulving 1983) or generic
fee” will depend on whether the coffee is in a cup, in a can, (Hintzman 1986) memory. Where is episodic memory, that
or in a person. Similarly, as I discuss in sections 3.1 and 7.3, is, our memory for particular, personal experiences? The
action patterns based on projectable features of an object answer: the same place. I propose that episodic recollection
(e.g., a Coke bottle) can mesh with action patterns underly- is a type of pattern completion via meshed bodily con-
ing goals in particular contexts (e.g., drinking or fighting), so straints on action. Furthermore, the episodic character, the
that the resulting meshed conceptualization is context- feeling that a memory is personally relevant, arises from
dependent. suppressing clamped projectable properties of the environ-
Nonetheless, there are important differences between ment. In this case, conceptualization is driven by trajecto-
my use of terminology and connectionist systems. For ries rather than by changes in the environment.
example, connectionist accounts of semantic or meaningful To some cognitive psychologists, this idea will seem
information are based on conceiving of meaning as activa- wrong on the face of it: it denies the difference between
tion of a limited number of features, at least at the input episodic and semantic memory; it denies the idea that
layer. Unfortunately, most theorists fail to specify what the episodic memory is temporally organized; it provides no
features are, and they fail to specify how those features distinction between short-term and long-term memory.
might be learned or changed as a consequence of develop- Before describing how the idea seems right, I will briefly
ment. In the system that I am proposing, initial coding is not address why these problems are more apparent than real.
featural, but analog, in terms of patterns of possible action.
Furthermore, as one learns more about the interactive
capabilities of one’s body, objects and actions can be im- 4.1. Episodic and semantic memory
bued with new meaning: what I can do with that object now. I am explicitly equating episodic and semantic memory in
A second important difference concerns the nature of the sense that there are no separate episodic and semantic
constraints. In standard connectionist accounts, constraints memory systems, hierarchically arranged (Tulving 1984) or
are, in Palmer’s (1978) terminology, extrinsic (but see otherwise. Of course, phenomenal memories differ in con-
Regier, 1995, for an exception). That is, a particular con- tent, accessibility, and so on. But those differences do not
straint represents statistical, or joint-occurrence, informa- imply separate systems. Whereas this equation of memory
tion, not a necessary feature of the operation of the system. systems may have been controversial 10 years ago, data and

8 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Glenberg: What memory is for

mainstream memory theorizing are now moving in this lieved to code information along acoustic/articulatory di-
direction. In short, there is little data to support a distinc- mensions, whereas long-term store coded “semantic” infor-
tion between a memory system devoted to personal experi- mation. However, demonstrations of meaning-like coding
ence and one devoted to general knowledge (McKoon et al. in short-term situations (Shulman 1972), as well as long-
1986). What appeared to be strong evidence for a memory term memory for articulatory and orthographic information
organized by “semantic” dimensions (Collins & Quillian (e.g., Hunt & Elliot 1980) deny this simple distinction. Also,
1969), is now known to reflect frequency of experience the quick forgetting demonstrated using the Brown-
(Conrad 1972). Evidence that was taken to indicate the Peterson distractor technique, is now known to reflect a
storage of prototypes (Posner & Keele 1968) in semantic combination of poor initial coding (Muter 1980) and inter-
memory is now taken to reveal the operation of retrieval ference from previously studied material (Keppel & Under-
processes that can average experiences (Hintzman 1986; wood1962; Watkins & Watkins 1975).
McClelland & Rumelhart 1986). Priming effects that were What are we to make of the impressive body of informa-
thought to reflect the spread of activation along permanent tion on apparently separate short-term modules (e.g., Bad-
semantic links can be easily demonstrated for newly deley’s [1990] articulatory loop, phonological store, and
learned (hence episodic) information (McKoon & Ratcliff visual/spatial sketchpad)? An alternative theoretical posi-
1986a). Thus the distinction between episodic and seman- tion is to consider the evidence as indicative of skills and
tic memory probably reflects a difference in the frequency strategies effective in particular domains (cf. Kolers &
with which the memories are used, the methods of assess- Roediger 1984), rather than of separate modules. This skill-
ment, and the content of the information, rather than any based alternative can easily accommodate findings that
intrinsic differences in memory systems. might otherwise be interpreted as evidence for new work-
ing memory modules. As one example, Reisberg et al.
(1984) demonstrated an increase in “working memory
4.2. Temporal organization of episodic memory
capacity” by instructing subjects how to use their fingers to
If the framework that I have described is the only memory code numbers in a memory span task. This evidence might
system, then it explicitly denies a tenet of theorizing about be interpreted as evidence for a new “finger-control” mod-
episodic memory: memory is a record of events that main- ule, but it seems more sensible to view it as a newly learned
tains some semblance of temporal order (see, e.g., Mur- skill. As another example, Carpenter et al. (1994) speculate
dock’s [1974] conveyor belt model or Glenberg & Swan- that there may be separate working memory capacities for
son’s [1986] temporal distinctiveness theory). Almost language production and language comprehension. Again,
assuredly, the tenet that episodic memory maintains order the alternative that different skills are involved in compre-
derives from the fact that temporally distant information is hension and production would seem to more easily accom-
harder to remember than recent information. This fact modate the data.
does not demand a theoretical explanation that maintains Nonetheless, one must come to grips with our intuitions
time as a dimension of memory, however. In fact, Friedman of immediate access to some information and difficulty
(1993) presents a convincing case that episodic memory in recovering other information. Consider this proposal.
is not organized temporally. First, there is little priming Memory and the perceptual/action system are designed to
between temporally contiguous but otherwise unrelated produce a meshed conceptualization (possible actions) for
experiences. Second, memory for time of occurrence of current stimulation. It is this constantly changing conceptu-
events is not only inaccurate; it shows nonmonotonic scale alization (changing because the stimulation changes in
effects. That is, memory for when an event occurred may be response to action) that gives the illusion of a short-term
accurate for the day, inaccurate for the month, accurate for memory. Because the current conceptualization updates
the season of the year, inaccurate for the year, and accurate memory and provides the starting point for future concep-
for the decade. Third, as Friedman discusses, for most of tualization, it will have a strong influence on performance
human history, memory based on a linear dimension of time over the next few moments (as does a short-term memory).
would serve little useful purpose. Instead, a memory orga- Distraction (a changing environment) does cause a disrup-
nized by functional significance or by recurrent events tion in short-term behavior because it produces a forced
(seasons, migrations, life cycles) would seem to have much change in the current conceptualization. Limits on the
greater adaptive significance. “capacity” of a short-term store are simply the limits on
coherent conceptualization.
This framework also rationalizes some aspects of re-
4.3. Short-term memory
hearsal and control of thought. In particular, it seems that
The idea of a single memory system seems wrong in that some sort of cyclical activity is needed to maintain informa-
there is no mention of separate processes for long-term tion in the forefront of consciousness. Baddeley (1990)
memory and for short-term or working memory. Much of discusses this as an articulatory loop that must reactivate the
the evidential basis for a separate short-term store (or decaying contents of a phonological store. But if memory is
working memory, according to Baddeley, 1990) has been like a box that holds items of information, why should
eroded. For example, the “recency effect” is the enhanced cyclical activity be necessary? The answer comes from the
recall of items from the end of a list. Because it was thought nature of trajectories. They are not static memory traces;
to be easily disrupted by a short period of distraction, it was they are reinforced changes from one conceptualization to
taken to be a hallmark of short-term store. We now know, the next. Thus, there is no holding of trajectories in mind.
however, that recency effects can be very long-term (Glen- Instead, to maintain a thought or a conceptualization in the
berg 1984; Greene 1986; 1992). Another supposed hall- absence of clamped projectable properties, it is necessary
mark of a separate store is acoustic/articulatory encoding to reuse a trajectory, or to replay the same scene over and
(e.g., Hintzman 1967). That is, short-term store was be- over.

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 9

Glenberg: What memory is for

5. Memory in two modes: Automatic and effortful contribution of embodied memories to conceptualization
of the current environment. It is a type of perceptual
The major function of memory is to mesh constraints on fluency in that it affects how aspects of projectable proper-
action based on nonprojectable properties with constraints ties are conceptualized. Because embodied memories do
from projectable properties. This is an automatic function not change the clamped environment, the automatic opera-
of memory in the sense that it is not under conscious tion of memory does not help one to literally see more
control, and it corresponds rather directly to recent work on clearly (that is, with greater acuity): it helps one instead
indirect or implicit memory. There is also an effortful mode understand the environment. That is why repetition prim-
of memory. Effortful suppression of projectable properties ing has negligible effects on accuracy (in the signal detec-
allows conceptualization to be guided by trajectories. The tion sense of ability to discriminate) while affecting inter-
resulting conceptualization is what underlies personal, pretation (or bias; Ratcliff & McKoon 1993; Ratcliff et al.
autobiographic, conscious recollection. 1989).
Repetition priming is modality specific because it is often
based on clamped projectable properties. For example,
5.1. Memory’s automatic contribution
consider an experiment in which both pictures and words
to conceptualization
are presented and later subjects must identify the objects in
When we are walking the path home, we do not need to fragmented pictures. To identify the pictured object, sub-
consciously recall which way to turn at each intersection; jects must use their memories to mesh with the projectable
when we recognize our children in a crowd, it is not because fragments. Clearly, features of the letters used in spelling
we have subjected each face to a conscious check; and as we the name of the pictured object are irrelevant to this task, so
read each word in a sentence, there is no need to try to little priming is expected or found between reading words
remember back to when we might have last encountered a in phase 1 and identifying pictures in phase 2 (Weldon &
similar-looking pattern in order to ascertain the meaning of Roediger 1987).
the word. Memory is automatically, that is, without inten- Use of trajectories may underlie conceptual forms of
tion, creating a mesh between the projectable properties repetition priming as well. For example, presentation of
(the path, the faces, the letters) and patterns of interaction “Amazon” in phase 1 will facilitate answering “What is the
controlled by nonprojectable properties. Research on indi- longest river?” in phase 2. Clearly, words are more than just
rect or implicit memory (Roediger 1990; Roediger et al. marks on a page. In reading “Amazon” we think about what
1994; Tulving & Schacter 1990) is tapping this automatic rivers are in terms of swimming, fording, and so forth. This
mode of functioning. cognitive activity reinforces trajectories from the word
Indirect tests of memory do not require conscious deci- “Amazon” to these activities. Later, in comprehending the
sions that something is remembered. Instead, the tests question “What is the longest river?” we may create a
often measure some form of repetition priming, that is, the similar conceptualization of rivers. Given the previously
extent to which previous exposure to a stimulus facilitates reinforced trajectories, the embodied conceptualization of
current processing. For example, a list of words (or pic- Amazon is easily reachable (that is, meshes with) the
tures) can be presented in phase 1 of a repetition priming embodied conceptualization of “longest river.”
experiment. In phase 2, subjects are asked to identify
degraded stimuli, some of which occurred in phase 1.
5.2. Effortful memory
Repetition priming is the phenomenon that identification
of stimuli actually presented in phase 1 is superior to In section 3.4, I discussed the idea that multi-step predic-
identification of stimuli presented for the first time in phase tion requires suppression to loosen control of projectable
2. This finding occurs whether or not the subjects are properties on conceptualization. I suggested that suppres-
attempting to remember anything about phase 1 (e.g., sion is dangerous because projectable properties that
Jacoby & Dallas 1981; Weldon & Roediger 1987). A more should be controlling action (such as walking) are ignored.
conceptual form of indirect memory can be measured by, This analysis leads to several suggestions. First, because
for example, presentation of a word and the later choice of suppression is dangerous, it is effortful. The effort is a
that word as an answer on a test of knowledge (Blaxton warning signal: Take care; you are not attending to your
1989). Among the many interesting findings generated by actions! Also, the effort forces us to use suppression conser-
this research, several may be particularly important. First, vatively. Second, there are behavioral indices of suppres-
repetition priming can be of very long duration. It is not sion. For example, when working on a difficult intellectual
unusual to be able to demonstrate positive effects over problem (which should require suppression of the environ-
weeks and months (Sloman et al. 1988). Second, repetition ment), we reduce the rate at which we are walking to avoid
priming effects are often sensitive to presentation and test injury. Third, autobiographical memory arises from sup-
modality. For example, pictures prime pictures more than pressing the environment: once the environment is sup-
pictures prime words, and vice versa (Weldon & Roediger pressed, conceptualization is controlled by trajectories and
1987). Third, people with dense amnesia often perform bodily constraints on mesh rather than the projectable
equivalent to nonamnesics on indirect, repetition priming features of the environment. Thus recollection is similar to
tests (e.g., Musen & Squire 1991). I address this last finding prediction. Both are effortful, both depend on trajectories,
in section 5.2.4. and both are constrained by the body. On this view, con-
Jacoby (e.g., Jacoby et al. 1993) has made the case that scious recollection is a type of pattern completion that is
much of repetition priming is due to an automatic compo- inherently reconstructive (Bransford 1979).
nent. Jacoby characterizes this component as “familiarity” The effort in suppressing the environment can be used to
that arises from “perceptual fluency.” In the embodiment explain standard and nonstandard facts of episodic memory.
framework, the automatic component of memory is the As an example of the latter, consider the phenomenon of

10 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Glenberg: What memory is for

averting one’s gaze when engaged in a difficult memory tions for several related phenomena. Consider first infantile
task. When recollection is difficult and unrelated to the amnesia. There is now good evidence that the phenomenon
current environment, clamping of the environment must is not as dramatic as initially proposed. In particular, there is
be suppressed to allow internal control over conceptualiza- evidence for good early retention when it is tested nonver-
tion. Closing one’s eyes or looking toward a blank sky are bally. To the extent that a test trades on the automatic
actions that help to suppress the environment by eliminat- operation of memory, it should reveal substantial memory
ing projectable properties that would normally be clamped. for the infants. In addition, both Howe and Courage (1993)
Glenberg et al. (1995) have demonstrated that people avert and Nelson (1993) have suggested that what changes
their gaze when working on moderately difficult recollec- around ages 2–3 is the child’s ability to code and retrieve
tion tasks (but not easy ones) and that this behavior en- information in ways understandable to adults. For Howe
hances accurate remembering. and Courage, this amounts to developing a self-concept
useful in organizing and retrieving memories. For Nelson,
5.2.1. Encoding paradigms. How people are instructed to this amounts to learning how to use narrative structures to
think about (i.e., encode) to-be-remembered stimuli organize and relate the child’s narrative (i.e., self ) experi-
greatly affects success in conscious recollection. Interactive ences. Nelson notes that this learning is typically guided by
imagery (e. g., Bower 1970), levels of processing (Craik & interactions with adults.
Lockhart 1972), and generation paradigms (e.g., Slamecka Consider the following explanation for the correlation
& Graf 1978) all illustrate this phenomena. As an example, between development of self-concepts and the emergence
consider the use of interactive imagery to memorize arbi- of recollective experience. Recollective experience re-
trary pairings such as “lamp – 88.” Success in remembering quires (1) suppression of environmental input, (2) use of
the pairing is greatly enhanced by imagining, say, a neon self-generated information (trajectories) to drive the con-
light shaped to form the digits 88, compared to rote re- ceptual system, and (3) an attribution that the resulting
hearsal of the words. conceptualization is due more to internally-guided than
Standard analyses based on the notion of abstract, amo- externally-guided construction. I suspect that a major factor
dal symbols have difficulty with these effects because in the development of a concept of self is just the ability to
the abstract propositional description of the to-be- suppress environmental information. Until that skill is mas-
remembered stimuli are the same regardless of the encod- tered, conceptualization is controlled by the clamped envi-
ing task. That is, for both rote rehearsal and imagery one ronment; after that skill is mastered, conceptualization can
must remember the same thing, “lamp – 88.” On an be guided by oneself. That is, one can control what one is
embodied account, constructing an image requires mesh- thinking about. Furthermore, development of language (by
ing a conceptualization of a lamp with that of 88. The interacting with adults) may well be an important experi-
changes in conceptualization from the orthographic stim- ence in learning how to control suppression and recollec-
ulus to the meshed image update memory trajectories. tive experience: development of language facility is tanta-
Later, partial information such as “lamp” may be given as a mount to learning to use words to guide conceptualization.
cue for the pair. Reading and conceptualizing “lamp” will be Thus skill in suppressing the environment is facilitated by
along the lines of the reinforced trajectory. Importantly, the language, and this same skill supports recollective experi-
analog shapes of the successive conceptualizations increas- ence and the development of a notion of self.
ingly specify the final conceptualization of the neon 88. If recovery from infantile amnesia requires learning to
(Interestingly, the ability of young children to use this sort suppress the environment’s control of conceptualization,
of strategy depends on those children manipulating the perhaps adult anterograde amnesia results from a traumat-
objects [Varley et al. 1974].) Contrast this with a situation in ically induced reduction in the ability to suppress. Two
which the encoding task is not interactive imagery but findings are consonant with this speculation. First, amne-
simply reading the two words or engaging in rote rehearsal. sics exhibit poor performance on explicit tests of memory
There is little mesh created by reading the words: the words requiring conscious recollection, but not on implicit tests of
are pronounced separately so that there is not a physical memory (e.g., Musen & Squire 1991). According to the
mesh such as that produced by coarticulation. Further- framework developed here, it is the explicit, recollective
more, there is no conceptual mesh in terms of the patterns tests that require suppression of the environment, not the
of interaction between the two objects named by the words. implicit, automatic tests. Second, although there are nu-
No wonder that little can be reconstructed from the cue merous explanations of amnesic abilities and disabilities,
“lamp” alone.5 none provides any explanation for the feel of memory. That
5.2.2. The feel of memory. Why is there a phenomenal feel is, when it can be demonstrated that amnesics are using past
to conscious recollection? Why does the content of memory experience as effectively as normal rememberers (on im-
appear to reflect personal experience? Why doesn’t percep- plicit or automatic tests), why don’t the amnesics have any
tion or automatic uses of memory feel that way? The feel of sense that they are remembering? Of course, the same
memory comes from the effort of suppressing the environ- question can be asked of the normal rememberers: When
ment and the consequent knowledge that conceptualiza- they perform well on an implicit (automatic) memory task,
tion is being driven by previously created trajectories. This why do they lack the experience of remembering? For the
process has the feel of personal memory because of our normal rememberers, the feel of memory comes from an
belief that the achieved conceptualization is free from effortful suppression of environmental input and the attri-
domination by the projectable properties of the environ- bution that conceptualization is controlled by the self.
ment. When conceptualization is controlled predominately by the
environment, as when performing implicit memory tasks, it
5.2.3. Suppression and amnesia. To the extent that skill in does not feel like memory. And, this is the usual state for
suppressing the environment develops, it suggests explana- amnesics.

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 11

Glenberg: What memory is for

5.3. The Kolers-Roediger program lecture hall. Thus, we must suppress projectable properties
Other memory researchers have proposed ideas similar to of the environment to comprehend language. Second, and
the framework outlined here. A particularly good example perhaps more difficult, we must also suppress the project-
is the “procedures of mind” approach (Kolers & Roediger able properties of the language signal itself. That is, to
1984). In fact, the similarities between the approaches are understand the language, we cannot focus on the shapes of
striking. Kolers and Roediger suggest that many distinc- the letters, the patterns of spaces between the words on the
tions popular in memory theorizing reflect different skills page, or the chirps and squeaks of the speech signal.
rather than different memory stores. Importantly, while Several predictions follow from the claim that language
championing a symbolic account, Kolers and Roediger note comprehension requires suppression of projectable proper-
that abstract, meaningless symbols will not do. Instead, they ties. The first is that good language comprehenders should
prefer symbols that retain characteristics of how they were be good at suppressing the environment. Second, good
acquired: “We claim that knowledge of objects is specific to language comprehenders should be good recollectors,
the means of experiencing them” (p. 419). Thus the sym- given that both require suppression. Third, unavoidable or
bols are in some ways analogical, as I have advocated. nonsuppressable properties of the environment should
Kolers and Roediger also object to modeling knowledge disrupt language comprehension. Of course, distracting
using psychologists’ propositions because “descriptions of noise or sights will impair comprehension, but a more
events rarely if ever tell a person what to do about the subtle effect is discussed by Sanford and Moxey (1995).
events described” (p. 439). Of course, conceptualization in They note that many types of regularity seem to disrupt
terms of patterns of interaction with the environment was language comprehension, and hence those regularities are
designed to overcome this problem. Finally, Kolers and classified as instances of poor style. Repeating patterns of
Roediger eschew the idea that memory is purely a con- articulation (McCutchen & Perfetti 1982), phonemes (e.g.,
scious experience. They propose instead that the most “Crude rude Jude chewed stewed foods,” from Baddeley &
important contribution of memory is to the automatic Hitch, 1974), and excessive repetition of particular sen-
execution of skills. tence structures all seem to slow comprehension. Sanford
Given the similarities between the Kolers-Roediger pro- and Moxey propose that these regularities contribute to the
gram and the embodiment framework, are there any differ- computation of coherence, but because the regularities
ences? One is my emphasis on meaning, that the meaning “are irrelevant to the writer’s message,” the processing
of an object or event is a meshed pattern of possible action. rapidly runs to a halt. Here is a different (but related)
A second difference is the idea of mesh itself. The mesh suggestion. The regularities are regularities in the project-
between the projectable features of an object and nonpro- able properties of the environment. The regularities cap-
jectable features from memory can dramatically change the ture attention and contravene the suppression required for
meaning of an object or event (see sect. 7.3). This sort of conceptualization and comprehension of what the language
combination is made possible by considering both the is about. That is, instead of paying attention to the meaning
projectable and the nonprojectable features to be patterns of the language, we start to pay attention to the language
of action that can combine as physical, bodily actions can be itself.
combined. If separate patterns of action can be forced into Fourth, because language comprehension is seen as a
a coherent pattern of bodily movement, then we can general skill, performance in language comprehension
comprehend the combination; in this way rocks, twigs, and tasks should correlate with performance in other compre-
soil combine to form a path for a particular person. It is not hension tasks. Gernsbacher et al. (1990) have demonstrated
clear how the skills described by Kolers and Roediger can just this.
be combined except through concatenation. Finally, I am
attempting to extend the analysis to language comprehen- 6.1. Mental models in language comprehension
sion.
Suppose, as Taylor and Tversky (1992) claim, that at least
one of the functions of language is that “language is a
6. Language comprehension surrogate for experience” (p. 495). If language is to be a
useful surrogate, it must make contact with the sorts of
I have argued that the same memory system underlies embodied representations that we use to characterize the
perception, semantic memory, and episodic memory. The world, and I propose that language does this relatively
meaning of a situation is given by a meshed pattern of directly: we understand language by creating embodied
possible actions, and that is an embodied conceptualization. conceptualizations of situations the language is describing.
The system is updated whenever there is a change in In fact, this is the only reasonable story for how we can
conceptualization. Thus, the environment is compre- manage to learn from language.
hended as a series of transformations of embodied con- This story works when language is being used as a
ceptualizations. I propose a similar characterization of surrogate for events that are completely absent and when
language comprehension. Language comprehension, like language is being used to enhance current experience.
comprehension of the environment, is the successive trans- Consider a situation in which a mother is instructing her
formation of conceptualizations that are patterns of pos- child. Representations derived from the language must
sible action. smoothly integrate (mesh) with representations derived
Like recollective memory, language comprehension re- from other aspects of the environment. Thus, being told
quires suppression of the environment, but in two ways. “That plate is hot” must modify the embodied representa-
First, the content of the language may have nothing to do tion of the plate in order to modify interactions with the
with the physical environment in which the language is plate. Tannenhaus et al. (1995) demonstrated just this sort
expressed. Lectures, for example, have little to do with the of smooth and immediate integration. Their subjects re-

12 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Glenberg: What memory is for

sponded to verbal commands (e.g., “Put the apple on the The expert’s model incorporates tighter constraints on
towel in the box”) to move actual objects arrayed before action based on trajectories derived from experience.
them. Eye movements were monitored during the task. Given the same text, the expert is able to take (appro-
Movement of the eyes to referent objects was very closely priately constrained) actions that leave the nonexpert
time-linked to the verbal command. Additionally, the envi- baffled. This effect of expertise in language comprehension
ronment was used to smoothly disambiguate the language. parallels the expert guide who can spot the trail (based on
For example, when considering the language alone, the trajectories derived from experience) while the novice sees
phrase “on the towel” is temporarily ambiguous. It may only twigs, soil, and rocks.
describe the location of a particular apple (the apple that is
on the towel) or where an apple is to be put. Indeed, when
6.2. Comprehension, prediction, and priming
there was only one apple in the array, the eye movements
indicated uncertainty. When the array contained two apples I have argued that embodiment in terms of action patterns
(one on a towel and one on a napkin), however, then the is just what is needed to facilitate interaction with the
phrase “on the towel” will almost certainly be meant to environment and prediction. Is prediction an important
specify a particular apple, not a location in which to put the component of language comprehension? Clearly, language
apple. In the two-apple case, the eye movements indicated would be of little use if it did not enable better prediction of
no uncertainty. Thus, understanding of the sentence made the environment. But, the question asked in the literature
virtually immediate use of the context, in contrast to notions on comprehension is different: Does a mental model serve
of modularity of syntactic analysis. This sort of integration is as a source of “on-line” predictions about the upcoming
possible if both the environment and the language are text? In response to this question, one might ask “Should
understood as embodied patterns of action. it?” If the point of language is to be a surrogate for
This sort of reasoning is compatible with work on mental- experience, that is, to help us take appropriate action in real
model theory. The basic claim of mental-model theorists is situations, it makes little sense to expect the representation
that language comprehension results in representations of to predict upcoming text: it should predict changes in the
what the language is about, not representations of the situation. In fact, McKoon and Ratcliff (1992) reached the
language itself (e.g., chirps, words, sentences, or proposi- conclusion that there is little evidence that people make
tions). Johnson-Laird (1989, p. 488) writes that a mental predictive inferences while reading.
model is a representation of a situation such that “its Nonetheless, Keefe and McDaniel (1993) presented
structure corresponds to the structure of the situation that convincing evidence for what appeared to be just those
it represents.” With an important emendation, this defini- inferences. Using the standard logic based on psychologists’
tion can apply to the sorts of representations I have been propositions, Keefe and McDaniel reasoned that pronun-
describing. It seems unlikely that the literal, in-the-head ciation of a probe word would be faster if the word were
structure of the representation could actually be iso- part of a recently made inference than if not. For example,
morphic to the structure of the situation (in contrast to subjects read a sentence such as “After standing through
Glenberg et al. 1994). the three-hour debate, the tired speaker walked over to his
chair.” Following the sentence, subjects pronounced the
6.1.1. Mental models from language and perception. probe word “sat.” Supposedly, pronunciation of the probe
Embodied mental models are “models” in the following word would be facilitated by its having been incorporated
sense: a model is useful if it can be used to predict the effect into an inferred proposition such as “The speaker sat
of an action in the real situation being modeled. One way to down.” In the control condition, for which an inference
ensure accurate prediction is to build into the model including the word “sat” is unlikely, subjects read a sentence
spatial-functional constraints analogous to those of the real such as “The tired speaker moved the chair that was in his
situation. For example, a useful model of an airplane will way and walked to the podium to continue his three-hour
have wings that generate lift when it interacts with air debate.” Indeed, pronunciation of the probe word was
currents, much like a real airplane’s wings generate lift. faster in the predictive condition than in the control condi-
Similarly, a mental model built from language incorporates tion. In fact, pronunciation of the probe following the
embodied constraints on action like those derived from predictive sentence was as fast as when the sentence
comprehension of the environment. This sort of mental explicitly continued with “and (he) sat down.” Murray et al.
model is useful to the extent that it incorporates enough (1993) used a similar methodology and produced a similar
constraints on action to derive predictions. effect when the “to-be-inferred event was in focus at the
One difference between embodied models derived from time of test” (p. 464). Why is evidence for predictive
language and those derived from perception is how com- inferences found only shortly after the predicting sentence?
pletely the meshed pattern of possible actions constrains Does this evidence demonstrate that subjects were at-
further action and prediction. The multiple projectable tempting to predict the upcoming text?
properties of the environment, because they are clamped, Consider an interpretation of these findings from the
tightly constrain conceptualization and action. In language point of view that the goal of language comprehension is the
comprehension, the patterns of possible action that con- creation of a conceptualization of meshed patterns of
tribute to a meshed conceptualization are much looser. action. In this case, interpretation of a word, phrase, or
That is, language is ambiguous in not specifying exact sentence consists of meshing the actions consistent with
parameters of spatial layout, force, and so forth (Talmy that bit of language with the patterns of action derived from
1988). Thus, conceptualizations derived from language do previous text. After comprehending Keefe and McDaniel’s
not constrain action as effectively as conceptualization predictive sentence (“ . . . the tired speaker walked over to
derived from the environment. This is one reason for his chair”), only certain actions can be easily meshed with
differences between expert and nonexpert comprehension. the conceptualization. For example, the actions implied by

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 13

Glenberg: What memory is for

“He began to rake the leaves” does not mesh. In contrast, priming reflects activation along permanent links such as
the action of sitting will mesh, and hence interpretation and between “English major” and “poem.” In one global situa-
pronunciation of the probe word “sat” is quick. After tion, the English major was working on a writing assign-
comprehending Keefe and McDaniel’s control sentence ment, and, indeed, reading of “poem” was facilitated rela-
(“ . . . walked to the podium to continue his three-hour tive to a neutral condition. In another global situation,
debate”), the action of sitting meshes about as well as the however, the English major was working on a computer
action of raking, and so pronunciation of the probe word program. In this case, reading the target “poem” was not
“sat” is slow. facilitated. The implication is that priming reflects ease of
This interpretation of the results is radically different integration (mesh) of concepts, not spread of activation
from that used in standard propositional accounts of infer- along permanent links.
ence making. In the standard account, an inference corre-
sponds to encoding a new proposition, something akin to
6.3. Space in language comprehension
“He sat down,” and one would expect some effect of this
proposition well after it was encoded. The embodied ac- If embodied conceptualization is a pattern of possible
count is that no “inference” in the standard sense is made. actions, then it must incorporate information about spatial
Instead, the action of sitting in the chair is temporarily layout, because actions are played out in space. The data
compatible with the embodied conceptualization. When from several research projects investigating spatial coding
the situation changes, some actions are no longer compati- in mental models provide this evidence. First, there have
ble with the embodied conceptualization and the “infer- been investigations of how language can lead to accurate,
ence” is no longer operative. This notion of temporary analogical representations of a described layout. For exam-
compatibility (how well the probe will mesh with the other ple, Denis and Cocude (1989) had subjects read texts
constraints) may well underlie McKoon and Ratcliff ’s describing the layout of objects on a circular island. After
(1986b) data for “partial” encoding of predictive inferences, several readings, they were asked to mentally simulate
and the temporary effect noted by Keefe and McDaniel scanning from one object to another. The main finding was
(1993) and Murray et al. (1993). Of course, this is not to say of a correlation between distance (if the objects had actually
that language comprehenders might not make forward been arrayed) and simulated scanning time. Morrow et al.
inferences if induced to do so (e.g., one might be asked to (1989; see also Rinck & Bower 1995) had their subjects
“guess what happened next”). These sorts of inferences are memorize the layout of the rooms in a building (and objects
just the sort of predictions discussed in section 3.4. How- in the rooms) before reading a passage describing the
ever, given that language, unlike the environment, only movements of a protagonist throughout the building. Mor-
loosely constrains action, it is more reasonable to wait until row et al. measured time to verify that particular objects
what happens next is described. were located in particular rooms as a function of the
The procedures used by Keefe and McDaniel (1993) and protagonist’s described movements. Interestingly, when a
Murray et al. (1993) follow from the more general notion of described movement (e.g., from Room A to Room C)
“semantic priming.” The standard idea is that processing required passage through an unnamed room on the path of
causes activation to spread along permanent links to seman- the movement, verification of objects located in the un-
tically related information, and this spread of activation named room was faster than verification of objects in other
speeds processing of the related information. Thus, reading unnamed rooms off the path. Apparently, subjects were
the prime, “doctor” speeds the decision that the target using the spatial information in the building layout while
“nurse” is a word (Meyer & Schvaneveldt 1971). The comprehending the text.
standard interpretation of semantic priming is embarrassed Glenberg et al. (1987) demonstrated the contribution of
by demonstrations that priming need not be due to perma- spatial information to language comprehension without
nent links (McKoon & Ratcliff 1986a), and that the effec- prememorization. Subjects read texts describing a protago-
tive relation between the prime and the target may have nist and a target (e.g., a jogger and a sweatshirt) that were
little to do with the presumed semantics of categories either spatially dissociated (the jogger took off his sweat-
(Sheldon & Martin 1992). Might semantic priming be shirt before jogging) or spatially associated (the jogger put
another instance of the operation of mesh? Assume that on his sweatshirt before jogging). After a sentence or two in
language comprehension is an attempt to mesh action which the protagonist was kept foregrounded but the target
suggested by the current word or phrase with the pattern of was not mentioned, accessibility of the target (e.g., the
actions already established. Thus, in thinking about what a sweatshirt) was greater in the associated condition than in
“doctor” is (the actions taken by a doctor and how one the dissociated condition. (McKoon and Ratcliff, 1992,
interacts with a doctor), one sets up a conceptualization in have argued that this effect may reflect a type of salience.
which the actions suggested by “nurse” will mesh. Hence, See Glenberg and Mathew, 1992, for a counter to this
processing of “nurse” is facilitated relative to the case when interpretation.) Along similar lines, O’Brien and Albrecht
it is preceded by an unrelated prime word such as “rake.” (1992) demonstrated sensitivity to spatial location of char-
A report by Hess et al. (1995) strongly suggests that acters in a text well after the spatial information was
semantic priming reflects something akin to mesh rather introduced. Thus, several sentences after reading, “As Kim
than spread of activation along permanent links. Their stood outside the health club, she felt a little sluggish,”
subjects read a sentence describing a local context, such as readers would balk at the sentence, “She decided to go
“To complete the assignment, the English major wrote a outside . . . ”
. . . ” and they then read a target word such as “poem.” The One interpretation of these findings is that they reflect a
question of interest was whether the local context (“English representation that is analogical with respect to space, that
major”) would facilitate reading of the target (“poem”) is, that the mental model is constructed in an inherently
regardless of the global situation. This would be expected if spatial medium. This seems unlikely. Langston et al. (in

14 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Glenberg: What memory is for

press) have demonstrated that spatial contiguity, in the one entity (an agonist) acting against another (an antago-
absence of other relations, does not have strong functional nist) and that the entities may have different strengths and
consequences. In these experiments, subjects read (or different tendencies (either toward action or toward inac-
heard) texts describing the spatial layout of four objects. In tion). Importantly, these basic entities and relations can be
outline, the texts read, “B is to the right of A, C is under B, D based on bodily experiences such as pushing and being
is to the left of/right of C.” The last sentence in the “close” pushed, moving objects, and so forth. Talmy suggests that
condition was “D is to the left of C,” so that the spatial our understanding of causal terms (e.g., “because”) reflects
layout of the objects has D under (that is, close to) A. The an agonist’s tendency (toward action or inaction) being
last sentence in the “far” condition was “D is to the right of overcome by an antagonist. Thus, we understand the sen-
C,” so that object D is separated from A. After reading, tence “The ball kept rolling because of the wind blowing on
subjects were tested for availability of the target object, A. If it” as an agonist (the ball) with a tendency toward inaction
space is represented analogically, and if closeness in that being overcome by the stronger antagonist, the wind. Talmy
space has functional consequences, then the target should also demonstrates how this analysis can be extended to
be more available in the “close” condition than in the “far” psychological instances of causation, social references, and
condition. We tested for availability of A using speeded interpretation of modals such as can, may, must, and
recognition of A and time to read a sentence referring to should. Thus, interpretation of “John cannot leave the
object A. Availability of A was never reliably affected by the house” comes about from assigning John the role of an
condition (“close” versus “far”), even though memory for agonist whose actions are blocked by the unmentioned but
the spatial layout was well above chance. stronger antagonist of social or physical constraint. In the
How are we to understand the contrast between Langs- case of “should not” the antagonist is a value or belief, and
ton et al. (in press) and the other research that clearly points so on. The point is that what has traditionally been treated
to an appreciation of spatial relations during comprehen- as prototypically abstract (e.g., cause, force, modality), can
sion? One possibility builds on the distinction between be conceptualized in embodied terms, and in so doing
mental models encoding space in a spatial medium and brings out important similarities in our understanding of
mental models encoding spatial-functional action and these concepts.
thereby representing space incidentally. Consider a rein- Bowerman (1982; 1985) discusses a number of cases of
terpretation of Glenberg et al.’s (1987) jogger on this children’s late speech errors that imply an understanding of
spatial-functional account. When the jogger puts on the the more abstract in terms of the concrete. Bowerman
sweatshirt, there is a mesh between the jogger and the classifies an error as a late speech error when it occurs after
sweatshirt: wherever the jogger goes, the sweatshirt goes a linguistic form has been used correctly and when the error
too. Then, later facilitation in reading “sweatshirt” is not does not mirror adult usage. She argues that, given these
due to spatial closeness of the jogger and the sweatshirt, but constraints, the error arises from an overextension of the
their functional relatedness. On this account, the texts used adult-sanctioned relation between domains. Typically, the
by Langston et al. (1995) resulted in encoding patterns of spatial domain is extended, so that the pattern of errors is
action between the reader (projected into the situation) and asymmetrical across domains. For example, children im-
each object (A, B, C, and D). Given that spatial layout is not port spatial terms into other domains, but rarely vice versa.
encoded directly, there is little reason to suspect that Bowerman reports that children use the spatial verbs “put”
availability of object A will depend on its spatial distance and “take” to describe state changes, such as “put the door
from object D. In other words, spatial distance only matters locked.” Also, it is commonplace to use spatial terms when
when it corresponds to functional distance. describing time, for example, “the week before” or “be-
The proposal that embodied mental models reflect a tween spring break and finals week.” Is this just a conven-
structured space (that is, a space structured by possible tion, or does it reflect a conceptualization in which we
actions) rather than a uniform space, is consistent with understand time by using spatial dimensions? The late
several research programs. McNamara (1986) and McNa- error “behind the dinner” for (“after the dinner”) would
mara et al. (1989) adduce evidence that spatial memory is seem to imply the latter (Bowerman 1982; 1985). Finally,
structured and perhaps hierarchical. Bryant et al. (1992) Bowerman describes the use of spatial terms to speak of
argue that the time needed to answer questions about memo- nonspatial dimensions, such as looseness of teeth (“They’re
rized spatial layouts reflects an embodied encoding. They all the same length of loose”) and temperature of water (“I
find that retrieval of information aligned on the head/feet want it the same size as Christy’s was”).
axis is faster than for the front/ back axis, which in turn is One final example should suffice. Suppose that we con-
faster than for the left/right axis. They interpret these differ- ceptualize abstract trait information (e.g., that Marta is
ences as reflecting asymmetries of the body. energetic) as a meshing between the person and the trait.
That is, the actions that Marta might perform are meshed
with “energetic” so that her actions are constrained to be
6.4. Comprehension of nonconcrete descriptions
energetic. To test this notion, Fernandez and Saiz (1989) had
If language comprehension is in terms of meshed action, subjects read texts describing the association or dissociation
how is it that we come to understand abstract language that of a main character and a trait. In a text about Marta, an
is not about concrete objects or situations? Here I adopt a expert in international business, the critical sentences in the
version of Lakoff ’s (1987) spatialization of form hypothesis. associated condition read (in translation from the Spanish):
Namely, we understand abstract situations by conceptualiz-
(1) She has just been appointed to a government position.
ing them in concrete ways. Almost everybody considers her an especially energetic person.
Talmy’s (1988) analysis of force dynamics is a good
example of how abstract concepts can be given a bodily Whereas the critical sentences in the dissociated condition
interpretation. He notes that we can conceptualize forces as read:

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 15

Glenberg: What memory is for

(2) She has just been appointed to a government position. and understanding the environment. Consider, for exam-
Almost nobody considers her an especially energetic person. ple, a percept of a drawing of an “impossible object.” There
After reading one or two filler sentences in which Marta was may be no three-dimensional object that could project that
kept foregrounded (but her energy never mentioned), ac- two-dimensional outline. Nonetheless, the percept is not
cessibility of “energetic” was evaluated by speeded recogni- incoherent; the percept is of a drawing that has no corre-
tion of the probe “energetic.” On average, responding in the sponding three-dimensional realization. Percepts may be
associated condition was over 100 milliseconds faster than re- unusual or bizarre, but never incoherent because the per-
sponding in the dissociated condition. Thus, readers may well ceptual/action system is designed to transduce patterns of
have been conceptualizing abstract trait information as em- possible interaction. Similarly, a random collection of words
bodied and meshed with an embodied conception of Marta. (or even phonemes or features) will be perceived coher-
ently, perhaps as chirps and whistles, and a random collec-
tion of sentences will be perceived coherently (correctly) as
6.5. Embodiment and coherence
a random collection of sentences.
Some texts make sense; others do not. The ones that make Nonetheless, we do get the sense that some collections of
sense are judged coherent. But, what produces that sense of sentences are not random. Sentences cohere to the extent
coherence? A standard answer is that it arises from the that they produce continuous transformations (trajectories)
connectedness of the psychologists’ propositions underly- of a meshed set of possible actions. Consider (4) again. The
ing the text; when the propositions are connected (or can be second sentence seems incoherent in that it cannot be
made connected through bridging inferences; Haviland & incorporated into the the standard situational interpreta-
Clark 1974) then the text is coherent. When the proposi- tion of flat wallpaper on a vertical wall in a gravitational
tions do not connect, either bridging inferences need to be field. However, if the initial model is changed so that any of
made to connect them or the text will appear incoherent. these presuppositions about the situation are eliminated
This interpretation of coherence is wrong in several (e.g., the wallpaper has niches in it, the wall is not yet
respects (see Sanford & Moxey 1995). Importantly, the vertical because it will be incorporated into a doll’s house,
account is wrong because whether or not propositions etc.), then the sentences are coherent. Another example is
connect and how they connect depends first on interpreting also adapted from Sanford and Moxey:
the propositions against a situation. Consider the following (5) John ate a banana. The banana was brown. Brown is a good
example adapted from Sanford and Moxey: color for hair. The hair of a dog is drunk to counteract a
(3) While measuring the wall, Fred laid the sheet of wallpaper hangover.
on the table. Then he put his mug of coffee on the wallpaper. Sanford and Moxey use this snippet of text to illustrate that
(4) After measuring the wall, Fred pasted the wallpaper on the sentences that incorporate cohesion markers (e.g., ana-
wall. Then he put his mug of coffee on the wallpaper.
phoric reference) can, nonetheless, be judged incoherent.
A propositional analysis does not reveal that (4) is odd, and The problem is that the sentences do not update a mental
thus a propositional analysis cannot indicate “local incoher- model. That is, the patterns of action suggested by each
ence” and cannot trigger bridging inferences to maintain sentence do not admit to smooth transformation of the
coherence (see also O’Brien & Albrecht 1992). Noticing mesh from one sentence to the next. Note, however, that as
that (4) is odd arises from a consideration of the situation, with the previous example, a change in the initial situation
that once the wallpaper is on the wall, under normal can render the sentences (more) coherent. Imagine that
conditions, it cannot not support a mug of coffee. To state it John engages in free association whenever he eats fruits.
differently, coherence is a relationship among ideas, and Then the list of sentences, as descriptions of his free-
texts do not have ideas – only readers do. associations, seem (more) coherent. Similar examples can
An impressive counter to the claim that coherence de- be constructed for film (e.g., the sequence of cuts seem
rives from connecting propositions can be found in Barton incoherent unless one has the appropriate model of the
and Sanford (1993). Their subjects read about an airplane film) and for events in the world (e.g., changes in the
crash that occurred in the Pyrenees between France and weather seem incoherent unless one has the appropriate
Spain. The subjects were asked for advice on where the model of weather systems). In short, coherence is a prop-
survivors should be buried. In fact, the subjects readily erty of models (the ideas that people have), not a property
offered advice; that is, they understood the text, judged it as of snippets of language.
coherent, and were ready to suggest where the survivors A final example was taken from the abstract of a talk
should be buried. Nonetheless, only about 60% of the given in a computer sciences seminar at the University of
readers noticed that “survivors” are not buried. In a second Wisconsin–Madison.
experiment, when readers were asked where to bury the (6) The talk will concentrate on the design of the communica-
“surviving dead,” only 23% noticed a problem. Clearly, the tions subsystem [of the Meiko CS-2 MPP System]. This utilizes
readers were not forming propositions and checking them a “fat tree” network constructed from high performance cross-
for sensibility, because “surviving dead” cannot make a point switches. Processing Elements interface to this network
sensible proposition. via a communications coprocessor which contains intelligence
An alternative account of coherence is twofold. First, to handle virtual addressing and ensures very low message start
coherence is a matter of degree, and in fact no bit of up times.
language is completely incoherent. Second, the degree of This text may well be very coherent for its intended audi-
coherence can only be computed from the mesh of a ence, but it is at the low-end of the dimension for me. The
situational representation of what the language is about. problem is not that the propositions do not connect. The
The claim that no bit of language is completely incoher- propositional relation between “fat tree” networks and
ent rests on the analogy between understanding language crosspoint switches is virtually transparent; similarly, it is

16 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Glenberg: What memory is for

quite clear that a coprocessor intervenes between the priming phenomena including forward inferencing (6.2),
“Processing Elements” and the network. The problem is and coherence (6.3). Suppression of projectable properties
that I do not know what a “fat tree” network is, or what of the environment is seen as critical to multistep prediction
“crosspoint switches” or “Processing Elements” are. I do (3.4), the feeling of memory (4.0), the decrease in physical
not know the literal shapes of these things, nor do I know activity when thinking (5.2), amnesia (5.2.4), correlation of
the actions they can take or how I can interact with them. language comprehension with recollection (6.0), and ef-
Because I lack that knowledge, I cannot build a coherent fects of incidental patterns on comprehension (6.0). Finally,
spatial-functional model. Presumably, crosspoint switches trajectories are applied to frequency effects in memory
can be arrayed or interconnected in some way so that they (3.3), the nature of rehearsal (4.3), automatic uses of
make up a “fat tree” network. But for me, the mesh is memory (5.1), and expertise (6.1.1).
missing.
The ideas (1) that coherence is a function of the mesh in
7.2. Embodied knowledge, emotions, and social
an embodied model, (2) that the embodied models con-
behavior
structed to understand language are the same as those that
underlie comprehension of the natural environment, and Can embodied patterns of action underlie all conceptual-
(3) that the purpose of perception and memory for the ization? Our experiences of music, taste, and emotions all
natural environment is to guide action, all lead to a sugges- seem to have aspects that do not fit well into a spatial-
tion about how to assess comprehension. Most laboratory functional straitjacket, and one suspects that aspects of
comprehension tests require verbatim reproduction of a these experiences are represented in addition to action
text, reproduction of “idea units,” or speeded responding to patterns. Nonetheless, given the ease with which these sorts
words or phrases. A more sensible comprehension test, of experience combine with spatial-functional experience
however, is one that requires action. To what extent can the (consider the contribution of music and mood to the under-
reader take sensible action (or make sensible predictions) standing of the action depicted in a film), it is not inconceiv-
on the basis of the text? (6) is relatively incoherent for me able that they may eventually be covered by the same sort of
because I can make so few predictions. For example, if the analysis.
type of switches were changed, I don’t know if that would Missing from the discussion is a consideration of hedonic
change the network from a “fat tree” network to some other valence and motivation to act. It is not as yet clear how
kind; if the communications coprocessor was not intel- pleasure and pain should be represented in an action-oriented
ligent, I do not know if the message start up times would be system (but see Lang 1979). What is clear, however, is that
slower or faster. On the other hand, (6) is not completely hedonic valence affects action and how experiences be-
incoherent because there are some predictions that I can come meaningful. Our understanding of pleasurable expe-
make. For example, based on knowledge of part-whole riences is in part action-toward those experiences, whereas
relations, I can predict that if the crosspoint switches are our understanding of aversive experiences is in part action-
eliminated, there will be no “fat tree” network. away. Several ideas follow. Given that action-away does not
necessarily specify what the action is directed toward, it
ought to be more diffused and variable than action-toward.
7. Conclusions
Also, on this analysis, approach and withdrawal are not
I began with a consideration of the Lakoff and Johnson poles of a single dimension: withdrawal from one situation
program and the problem of meaning. In applying their does not imply approach toward another. Thus our under-
insights to a theory of memory and mental models, the standing of emotional experience should reflect at least two
concept of embodiment becomes central. The basic claim is dimensions (e.g., Schneirla 1959).
that an individual’s memory serves perception and action. Malter (in press) applies these ideas to consumer re-
Memory meshes nonprojectable features with projectable search, in particular, to impulse buying. He proposes that
features of the environment to suggest actions for that projectable features of a product automatically mesh with
person in that situation. These patterns of action are what affectively charged memories (perhaps imparted by adver-
make the environment meaningful to that person. This tisements) to produce an irresistible approach-dominated
framework provides a way to address meaning, symbol conceptualization. Thus the consumer experiences a strong
grounding, recollective and automatic uses of memory, and desire to approach and manipulate the object, and in most
language comprehension. cases that can only be accomplished after purchase. Fur-
thermore, Malter notes that overcoming this urge to buy
requires effortful suppression of the projectable features in
7.1. Summary of interpretations and predictions order to deliberately evaluate the purchase. In the face of a
The framework provides alternative accounts of standard strong impulse to buy, however, that effort may be viewed
phenomena and it makes new predictions. Here is a brief as unattractive or not considered at all.
review. The concept of embodied knowledge is used to There is also reason to believe that an embodied, action-
address the problem of meaning and symbol grounding oriented analysis has implications for social psychology.
(sect. 1.3), why people interpret the world differently (1.3), Fiske (1992) traces the history of action-oriented theories
effects of bodily activity on emotions (2.3.1), imagery of social cognition from James (“My thinking is first and last
(2.3.2), memory for actions (2.3.3), sensibility judgments and always for the sake of my doing,” as quoted in Fiske) to
(2.3.3), short-term behavior (4.3), and understanding in Heider (1958) to current “pragmatic” research. Fiske de-
abstract domains (6.4). Mesh of patterns of action is applied fines pragmatism as a framework in which “meaning, truth,
to emergent features of thought (3.1), recollective memory and validity are determined by practical consequences
(5.2), interactive imagery (5.2.1), interpretation of semantic [and] concrete goal-relevant actions” (p. 886).

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 17

Glenberg: What memory is for

Fiske’s own analysis of social cognition is compatible with associate a novel taste with illness than to associate a novel
the ideas I have described, and her analysis suggests an sight with illness (Garcia & Koelling 1966). In contrast,
important extension. According to Fiske (and Heider pigeons find it easier to associate a novel sight, rather than a
1958), the key to social cognition is to view others not just as novel taste, with illness (Wilcoxin et al. 1971). If learning
objects that we can affect, but as beings who can effect us in comes about through meshing patterns of bodily action,
turn. Consonant with this premise, Fiske proposes that our then, given species differences in anatomy, physiology, and
ability to infer traits is in the service of interaction with possible actions, the fact that stimuli will mesh differently
others. for different species is a foregone conclusion.

7.3. Mesh 7.4. Standard memory paradigms

Among the interpretations that stem from a consideration If knowledge is embodied, then commonly used laboratory
of embodied representations, the notion of mesh seems paradigms for studying memory may well be missing the
most important. Ideas mesh to the extent that the pattern of mark. Many of these paradigms use random lists of words as
action underlying one idea can be integrated with the the objects of memory. Whereas there are reasons for using
pattern of action underlying another. The patterns mutually lists of words, these are reasons related to history and
modify and constrain one another because the conjoint convenience, not to any analysis of the design of memory. A
actions must be possible given our bodies. This mutual favorite argument to justify the verbal list format is that
modification of patterns of action is what underlies the each word corresponds to a mini-event, and from memory’s
construction of meaning from words whose senses are point of view these mini-events are similar to other events
jointly modified by the contexts in which they occur. in the world. This argument loses much of its force, how-
Meshing patterns of action provides a new way of think- ever, if memory is embodied and designed for negotiating a
ing about componentiality and productivity in language. As three-dimensional environment.
an example, consider the Coke bottle. Its shape, and thus its The discrepancy between the design of memory and the
affordances for human action, allow it to mesh with many design of the tools used to analyze it may account for
physical situations and goals. It can be used for storing undesirable characteristics of memory research. Impor-
liquid, as a cup, a doorstop, a weapon, a vase, and so on. tantly, memory researchers have not made much progress
Thus the meaning of a Coke bottle (how we can interact in understanding the nature of memory. We know about
with it) is not fixed, but infinitely varied, depending on the many phenomena (Greene 1992), but there is little agree-
context of use. Importantly, however, the meaning is in no ment as to the interpretation of those phenomena, how they
way arbitrary or unconstrained: the meaning of the bottle is fit together, or whether a particular phenomenon is of any
constrained by its shape (heft, fragility, etc.) and the impli- importance. Even with something as basic as the effect of
cations of that shape for action. Thus the spatial-functional repetition, the theoretical diversity is astounding: we have
meaning of a coke bottle is componential in that it will mesh theories in which repetitions enhance the strength of a
with many human contexts. Because that mesh can trans- single representation (Gillund & Shiffrin 1984), theories in
form the meaning, however, its use is creatively productive. which repetitions are individually preserved (Hintzman
This type of componentiality helps us to understand what 1986), and theories which treat memory much like a holo-
Barsalou et al. (1993) term “linguistic vagary.” When people gram (Metcalfe 1993). We have multi-store theories and
are asked to describe the features of a category such as single-store theories; single-system theories and multiple-
“Coke bottle,” there is tremendous variability both across system theories. All of these positions receive support from
people and from time to time in a particular person’s some aspects of the literature. I suspect that this diversity of
descriptions. Linguistic vagary should be the norm if the positions arises because in using inappropriate tools we
meaning of a concept is determined by its mesh with the obtain incompatible views of memory much like the views
context. of the blind men touching the elephant.
The idea of mesh may prove to be a concept that can
replace “association.” Although association has played a
central role in theories of cognition, the term carries little ACKNOWLEDGMENTS
An important impetus for this paper was a pleasant conversation I
theoretical weight. What we mean by an association is little had with Larry Barsalou while sitting on the steps of the Chicago
more than a conditional probability; if B is associated with Art Institute. Most of the ideas were developed further while I was
A, then P(BuA) . P(B). There is little or nothing in our a member of the research group “Mental Models in Discourse
theories to help us understand when “laws of association,” Processing” at the Zentrum für Interdisziplinare Forschung, Uni-
such as frequency and recency, hold, and when they do not. verstät Bielefeld. I thank members of the group, and especially
In contrast, the notion of mesh can, like an association, be Tony Sanford, for many lively discussions. Larry Barsalou, William
used to relate concepts, but the nature of the relation is Epstein, Alan Garnham, and Morton Ann Gernsbacher, in partic-
deeper: when patterns mesh, they modify each other be- ular, provided much help in eliminating the most loopy ideas from
cause they must conjoin in a way that respects constraints an earlier draft. Additional valuable help was received from Jesse
on bodily action. Thus, “Coke bottle” is difficult to mesh Prinz, Peter Gärdenfors, Henry Roediger, III, Robert Lockhart,
G. Dorffner, and several anonymous reviewers. Preparation of this
with “chair.” paper was supported in part by AFOSR AASERT Grant F49620–
Mesh provides a rationale for Thorndike’s (1932) concept 92-J-0310 and University of Wisconsin Graduate Research Com-
of belongingness (see also Ohman et al. 1976) as well as mittee Grant 940109 to Arthur Glenberg. Requests for reprints
various ideas put forward by Gestalt psychologists. Further- should be sent to Arthur Glenberg, Department of Psychology,
more, mesh may help to explicate species-specific differ- University of Wisconsin, 1202 West Johnson Street, Madison, WI
ences in associability of stimuli. Rats find it easier to 53706, or by electronic mail to glenberg^facstaff.wisc.edu.

18 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Commentary/Glenberg: What memory is for

NOTES The function of memory is to bring past experience to bear on present

1. The argument is not that grandfather is unable to interpret a action. This function is manifest in our everyday judgments and deci-
platform as a suitable chair for his granddaughter. Instead, the sions of family and work and in our personal mental life. Memory in
argument is that the grandfather knows perfectly well that the everyday life may thus be called functional memory. Functional mem-
platform is not a suitable chair for himself. Thus, when the young ory is barely recognized in the traditional perspective of reproductive
child refers to the narrow platform as a chair, the grandfather may memory. (Anderson 1996, p. 363; see also Anderson 1991b)
have some difficulty in knowing what the child is talking about. It is
only when the grandfather adopts the child’s perspective (i.e., how Glenberg hit the nail on the head with the criticism that “contem-
the child’s body can interact with the world) that he comes to porary psychology of memory has been dominated by the study of
appreciate how the platform can be a chair for the child. (Adopting memorization” (sect. 2, para. 1). Reproductive memory – memor-
another’s perspective requires suppression, as discussed in sect. ization – is only a small part of what memory is for. Memory is
3.4) the main point is that words and events do not have set mainly for goal-directed thought and action.
meanings for all times, places, and people. Words and events have The opening quote of this commentary comes from a functional
meanings to individuals because of their individual bodies and approach to memory in a spirit similar to that of Glenberg, but one
experiences in the world. based on a very different foundation. Whereas Glenberg seeks a
2. Thanks to Alan Baddeley for raising this objection most foundation in spatial-perceptual function, the present approach is
forcefully. founded on the goal-directed approach-avoidance of everyday life.
3. Thanks to Tony Sanford for this observation. This rests on an exceptionally solid base of experimental applica-
4. Furthermore, Gärdenfors (1995) proposes that the ability to tions in almost every area of psychology (e.g., Anderson 1991a;
form “detached” representations, that is, those unconnected to 1991b).
current environmental stimulation, is a significant evolutionary Approach and avoidance, the dynamics of everyday life, depend
step. Bloom (1994) discusses how the ability to manipulate such on values. One main function of memory is for the construction of
representations motivates language acquisition. values. Values are not generally memorized or reproduced, but
5. This account is very similar to that given to explain how constructed, because they depend on specifics of the operative
thinking about “Amazon” can facilitate answering the question situation. Once constructed, of course, they may be stored for use
“What is the longest river?” That example, however, was used in in future construction. Neither of these two functions has much to
the context of the implicit, or automatic, mode of memory. Given do with reproductive memory, however.
that automatic and effortful memory are clearly dissociable, how It is not necessary to “start from scratch in thinking about what
can the same processes apply? Although dissociable, tasks de- memory is for” (abstract), because a respectable start has already
signed to tap these two modes of memory share many sub- been made. The present functional approach to memory is not a
processes (cf. Jacoby et al. 1993). On the embodied account, both promissory note. It has a cutting edge in the form of cognitive
modes make use of trajectories to develop a meshed conceptual- algebra. Experimental studies in numerous areas have revealed
ization: the memory. The purest automatic case is when the the operation of algebraic laws of cognition (see contributors to
trajectories apply to clamped projectable properties, such as Anderson 1991a). These algebraic laws have provided a validated
fleshing out a fragmented picture. This is what Roediger et al. theory for the psychological measurement of values. As one contri-
(1994) call a data-driven task. The purest effortful case is when bution, a dissociation between memory for given verbal stimuli
nothing in the current environment is related to the sought-after and memory for an integrated response constructed from those
information so that all projectable properties must be suppressed. stimuli has been demonstrated (Anderson 1991b, pp. 9–13, 40–46).
This is an instance of Roediger et al.’s conceptually-driven tasks. One can concur with Glenberg that “commonly used laboratory
The “Amazon” and “lamp” examples are intermediate in terms of paradigms for studying memory may well be missing the mark”
suppression. We cannot completely suppress the projectable (sect. 7.4, para. 1). They are indeed missing the mark – because
properties because we need to read the words. Nonetheless, the they have prejudged memory as essentially reproductive. The grip
orthography of the words is of little use; what matters are the of this preconception appears even in critics such as Neisser (1982,
trajectories they invoke. p. xii), who comments that traditional approaches have been
“accumulating the wrong kind of knowledge.” Despite his em-
phasis on everyday memory, however, Neisser has remained
bound by the reproductive conceptualization of memory.
Glenberg begins too narrowly. His attempt to base memory on
Open Peer Commentary visual perception recognizes the goal-directedness of cognition
only in one respect. Even at the sensory level, many senses besides
vision – especially affective senses such as taste and sex – govern
Commentary submitted by the qualified professional readership of this much of everyday life. These and other affective senses do not
journal will be considered for publication in a later issue as Continuing generally provide the accuracy criteria Glenberg requires in the
Commentary on this article. Integrative overviews and syntheses are visual sense.
especially encouraged. The difference between visual perception and affective senses
is basic. The visual sense does best when it apprehends the
external world as it really is. Affective senses, in contrast, are
Functional memory versus reproductive primarily concerned with an internal world that has a value-based
memory concern with the external world. Externalist accuracy of taste is
denied by idiosyncratic taste aversions. Similarly, the affective
Norman H. Anderson value of sex is clear, whereas the idea of its accuracy seems strained
Psychology Department, University of California, San Diego, La Jolla, CA and unreal.
92093-0109; nanderson66 ucsd.edu A related reservation about Glenberg’s approach arises from
results obtained with the functional theory. The external world
Abstract: A functional theory of memory has already been developed as does not suffice to explain the internal world of cognition. In
part of a general functional theory of cognition. The traditional conception particular, algebraic laws of intuitive physics differ from the laws of
of memory as “reproductive” touches on only a minor function. The
primary function of memory is in constructing values for goal-directedness
physics in the external world. Glenberg’s emphasis on veridical
of everyday thought and action. This functional approach to memory rests spatial cognition is too narrow because it leaves little place for
on a solid empirical foundation. nonveridical cognition.

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 19

Commentary/Glenberg: What memory is for

A similar reservation is raised by Glenberg himself, in his At a more detailed level, the finding that prior generation of a
comment that “hedonic valence and motivation” do not fit his word, as opposed to simply reading that word, leads to a lower
“spatial-functional straitjacket” (sect. 7.2, paras. 2 and 1, respec- probability of being able to produce that word when it is presented
tively). A memory theory adequate to account for everyday tachistoscopically ( Jacoby 1983) – or with letters missing (Blaxton
thought and action cannot be expected to emerge from this 1989) – also seems problematic. It seems integral to Glenberg’s
“spatial-functional straitjacket.” analysis that bodily actions (such as speaking or writing) result in
A theory of memory should indeed begin, as Glenberg’s does, stronger memory representations than do nonphysical actions;
with the issue of what memory is for. At the most general level, the guidance of action, after all, is “what memory is for.”
function of memory is to bring past experience to bear on present With respect to human cognitive–motor capabilities that seem
thought and action. To make this specific requires an understand- difficult to reconcile with the theory, musicians are one source of
ing of the nature of thought and action. Few would disagree that examples. Consider a trained musician, proficient on two instru-
thought and action are mainly goal-oriented. Goal-directed ments, who attempts to execute on the clarinet a simple piece
thought and action embody a fundamental approach–avoidance originally learned on the piano. An experienced musician will do
axis. At bottom, approach and avoidance depend on values. With- this with ease, and even a novice will show some positive transfer.
out a theory of value, memory theory has limited significance. An In such cases, however, the overlap in the physical aspects of the
effective approach to the value problem, far from complete but two sets of actions is minimal. A more trivial example is that
demonstrably effective under certain conditions, has been devel- musicians can hum a new piece they have learned on some
oped along with cognitive algebra. These algebraic rules embody instrument, such as a guitar, yet the requisite movements of the
and provide functional measurement of psychological value, pro- vocal apparata do not overlap the physical actions performed on
viding a new way of thinking about memory. the instrument. If all memory is truly embodied, such transfer
Glenberg has rightly stressed the need for a functional approach should be minimal.
to memory. His visual–functional approach certainly seems to Imagine another musician who, after a long delay, is unable to
bring out significant aspects of functional memory. The present play a piece once played well. Here Glenberg’s theory, as we
functional approach is in the same spirit. understand it, makes an odd prediction: the musician should not
even be able to hear the piece mentally. The ability to predict how
the piece would sound derives, in Glenberg’s theory, from the
capacity to suppress projective characteristics of the environment
Problematic aspects of embodied memory and to “mentally” follow the trajectory that the embodied actions
would otherwise dictate. A failure to access those embodied
Aaron S. Benjamin and Robert A. Bjork memories for the purpose of action implies a de facto inability to
Department of Psychology, University of California, Los Angeles, CA access them recollectively. Both abilities, according to Glenberg,
90095-1563. benjamin66 psych.ucla.edu; rabjork6 6 psych.ucla.edu rely on the same embodied trace, which is “designed” to serve
action, not recollection.
Abstract: Glenberg’s theory is rich and provocative, in our view, but we The symbol-grounding problem, which renders current mem-
find fault with the premise that all memory representations are embodied.
We cite instances in which that premise mispredicts empirical results or
ory models inadequate in Glenberg’s view, seems a much less
underestimates human capabilities, and we suggest that the motivation for serious problem to us, for two reasons.
the embodiment idea – to avoid the symbol-grounding problem – should First, current models are not quite as bad as they may seem with
not, ultimately, constrain psychological theorizing. respect to their representational ability. “Meaningless” strings of
0s and 1s represent their environment in a crudely analog manner.
Glenberg issues a powerful call-to-arms to anyone who is serious Thus, two traces have the potential to be analogous to the degree
about understanding basic issues in human memory. We applaud that the codable features are indigenous to the representational
aspects of his analysis, including his emphasis on the form of system present in the organism. This constraint is not, in our view,
mental representations and his effort to account for phenomena arbitrary; it seems, in fact, more tractable than the notion of
as diverse as dead reckoning, language comprehension, and representations serving action patterns in an infinitely variegated
occasion-setting phenomena. In particular, his concept of “mesh,” way. We grant, however, that the nature of the features involved in
which provides an index to gauge the successes and failures of the representational assumptions of current models remains
arbitrary associations, seems an important idea. The phenomena poorly specified.
he cites demonstrate convincingly that association formation is not Second, the use of an atom that is maximally atheoretical in the
– at least in many instances – arbitrary. representational system approximates the ultimate implementa-
Certain implications of Glenberg’s theory, however, seem in- tional ends of such a theory: to describe its working in terms of
consistent with empirical findings, or seem to predict that humans hardware preinstalled in the human brain. Neurons and synapses
should not have certain cognitive–motor abilities they clearly are no less arbitrary in their symbolic values than nodes or
have. These implications, examples of which we cite below, derive connections, and the use of such “meaning-weak” symbols forces
primarily from the theory’s exclusive reliance on “embodiment” as our theories to describe representations in terms of patterns of
the mechanism by which mental representations take on meaning. activation – much as the brain is likely to. In this sense, it is the
The embodiment mechanism is adopted by Glenberg to avoid the engineer, and not the psychologist, who must face the symbol-
symbol-grounding problem (Harnad 1990), which he portrays as grounding problem.
the most damning feature of present models of human memory, Glenberg’s work is laudable in its scope, coherence, and em-
but we see that problem as more apparent than real – an argument phasis on meaningful representations. The singular emphasis on
we embellish at the end of this commentary. grounded meaning, however, seems to underestimate the flex-
One finding that seems problematic for the theory is that mere ibility and functions of human memory and to ignore certain
exposure to a stimulus can alter subsequent performance on process considerations. Whether the innovative and potentially
certain perceptual or cognitive tasks (often without the per- powerful concept of mesh – the process that guides the combina-
former’s awareness). Such stimulus-driven processing, typically tion of symbols and the combination of symbols with immediate
independent of a subject’s tasks or goals at the time, suggests environmental input – is actually tractable, for example, remains
functions and operations of memory that do not fit neatly in the to be seen. Ultimately, Glenberg’s theory does not remedy the
embodiment framework. In fact, Glenberg’s explanation of current deficiencies in computational theorizing as to the nature
implicit-memory priming effects seems to contain buried within it of mental representation, but it does demonstrate that a serious
an implicit assumption that prior exposure can have automatic- theory, at least in its early stages, can confront such problems
activation (task-independent) effects on memory. head-on.

20 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Commentary/Glenberg: What memory is for

Meshing Glenberg with Piaget, Gibson, Glenberg notes that his view of conceptualization in terms of
and the ecological self possible actions is very similar to J. J. Gibson’s (1979) concept of
affordance. Considering some other aspects of Gibson’s ecological
Richard A. Carlson view of perception would help to address the selectivity of percep-
Department of Psychology, Penn State University, University Park, PA tion and cognition. Gibson noted that selective attention need not
16802-3106; cvy6 6 psu.edu; require suppression (or “filtering”) of information. Instead, he
http:/ /popeye.la.psu.edu/psych/psycweb/hpages/RCarlson/home-htm argued, perception is a process of actively picking up information.
There are two related points here: First, we might consider how
Abstract: Glenberg’s rethinking of memory theory seems limited in its current purposive activity serves to select those projectable prop-
ability to handle abstract symbolic thought, the selective character of
erties meshed in a particular conceptualization. This seems to me
cognition, and the self. Glenberg’s framework can be elaborated by linking
it with theoretical efforts concerned with cognitive development (Piaget) a more plausible general approach to selection than effortful
and ecological perception (Gibson). These elaborations point to the role of suppression. Second, when we consider perceptual exploration
memory in specifying the self as an active agent. (e.g., looking) as a patterned activity (E. J. Gibson 1993), the
problem concerning how conceptualization can include objects
Glenberg’s ambitious target article suggests a major rethinking of that don’t support bodily interactions is at least reduced if not
memory theory, and I am very sympathetic to its general direction. eliminated. For example, Glenberg’s “beautiful sunset” (sect. 1.3)
Glenberg’s proposal seems to have several weaknesses that limit its might be more convincingly unpacked in terms of the informa-
generality, however. My goal in this commentary is to offer some tional interactions we have with it –which are no less patterns of
friendly suggestions for linking Glenberg’s efforts to rethink mem- action than walking along a path or drinking from a coffee cup.
ory theory with other theoretical work. Finally, Glenberg’s framework seems well-suited for relating
I see three major limitations to Glenberg’s framework. First, it is the function of memory to an implication of Gibson’s views that
hard to see how his scheme works for abstract symbolic thought, or Neisser (e.g., 1991) has called the “ecological self.” Consider
for the perception of events and objects that do not support direct Gibson’s (1979, p. 234) statement that “What we see now refers to
bodily interactions. How could the memory representations in- the self, not the environment.” That is, the information available to
volved in abstract, symbolic thought be embodied and thus the visual system specifies both objects in the environment and the
grounded? Second, the target article does not adequately address (typically moving) point of view of an active agent. This percep-
the selective character of cognition, perception, and action. A tually specified point of view is the ecological self. Similarly, the
staggering number (however counted) of projectable properties of information available from memory should be understood as
the environment are presently available to my perceptual systems, specifying not just nonprojectable properties of objects in the
but it seems that at a given moment only a small fraction of these environment, but the experienced self who interacts with these
could be meshed with information available from memory. If we objects. Developing this relation could be an important step
must suppress environmental input to allow cognition to serve its toward relating memory theory to questions about the nature of
functions in guiding activity, wouldn’t suppression also be neces- consciousness (Carlson 1992).
sary to select which projectable properties are currently meshed I believe that Glenberg has posed a very important challenge to
or clamped? Third, like representational theories of mind in standard theories of memory, although I am skeptical about some
general, Glenberg’s framework focuses on representation of the of his specific proposals, such as the central role of suppression.
environment, neglecting the informational basis of the self. Where Meshing Glenberg’s ideas with other theoretical proposals points
is the agent whose possible actions provide the basis for mesh? the way toward integrating memory theory with theories of cogni-
Unlike standard theories of memory, Glenberg’s does lend itself tive development, perception, and consciousness.
to elaboration in ways that address these issues. Two of Glenberg’s
central ideas are that (1) mental representation is grounded in
perceptuomotor activity, resulting in embodied meaning, and
(2) information from the environment is encoded in terms of Glenberg’s embodied memory:
possible actions. Both of these ideas contrast dramatically with
views that have been standard, if often implicit, in memory
Less than meets the eye
theories based on laboratory memory paradigms. However, these Robert G. Crowder and Heidi E. Wenk
central ideas echo important themes in the work of Piaget and
Psychology Department, Yale University, New Haven, CT 06520-8205;
Gibson. Drawing on these resources, I briefly suggest possible 6 minerva.cis.yale.edu; heidi.wenk6
rgcrow6 6 yale.edu
answers to the questions raised in the previous paragraph.
A possible approach to reconciling the hypothesis that all Abstract: We are sympathetic to most of what Glenberg says in his target
meaning is embodied with the facts that we (1) do use arbitrary article, but we consider it common wisdom rather than something radi-
symbols and (2) do engage in abstract thought can be found in cally new. Others have argued persuasively against the idea of abstraction
Piaget’s ideas about the development of mental operations (see in cognition, for example. On the other hand, Hebbian connectionism
Chapman, 1988, for a review). Piaget argued that the mental cannot get along without the idea of association, at least at the neural level.
operations that manipulate symbolic representations share formal
structures with sensorimotor activity that is constrained by the Our commentary title is deliberately provocative: the fact is, we
laws of physics. For example, the reversibility of logical operations approve of and second much of what Glenberg has to say about
that serves to guarantee noncontradiction and logical coherence in memory. Our title is fair enough, however, because Glenberg
symbolic thought develops from – and is thus grounded in – the himself has chosen a title with some provocative grandeur to go
empirical reversibility of displacements in space. This point is above ideas that we thought had become quite commonplace, if
compatible with Glenberg’s observation that memory representa- not common wisdom.
tions are, like current perception, structured in terms of possible The essence of his framework, a general argument against
actions. Studying Piaget reminds us, however, that it is not just abstraction, is a case in point (sect. 1). Although probably not
representations, but also mental operations – not just symbols, but popular in some circles at Carnegie-Mellon University, this atti-
symbol systems – that must be grounded in activity in a three- tude is not unanticipated among modern workers (Brooks 1978;
dimensional environment. And as Shepard (1984) points out, it is a Hintzman 1986; Jacoby & Brooks 1984; Medin & Schaffer 1978;
significant question which regularities in cognition are supported Nosofsky 1988). Accepting that memory lacks well-defined ab-
by information currently available to perception and which are stract symbols indeed favors the idea of embodiment as Glenberg
supported by internal representations fixed by individual learning asserts. But this argument (sect. 1.2) sounds much like a restate-
and development or by evolution. ment of Tulving’s (1983) encoding specificity principle (so a word

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 21

Commentary/Glenberg: What memory is for

is not an abstract concept after all, but rather is encoded according overcome the symbol grounding problem (Harnad 1990). How-
to a momentary context, which must be part of the retrieval cue for ever, the notion of embodiment should be qualified to avoid the
good recall). danger of a theoretical deus ex machina. I find the idea convincing
What are we to make of the accusatory and dismissive charge that embodied (projectable) properties from the environment are
that much of modern memory research has been about memoriza- meshed with embodied memory-based (nonprojectable) knowl-
tion (sect. 2)? Some of us proceduralists maintain that memory is edge in our sensorimotor experience. I do not think, however, that
the delayed aftereffect of mental operations (and we appreciate the clamping of projectable properties of the environment is
Glenberg’s approving appeal to the proceduralist manifesto of primarily “to keep the system reality-oriented” (sect. 3.2, para. 1).
Kolers & Roediger 1984). But regarding the field of memory as Projectable properties (the physical environment and the design
memorization seems to shift this field back to the days when we of our bodies) impose ontological constraints on our actions, but
(almost invariably) used intentional rather than incidental learning they are not sufficient to determine specific courses of action.
procedures, as was typical before the levels-of-processing work. Accepting that the environment is usually full of affordances
Moreover, a view of conceptualization in terms of patterns of (chairs to be sat on, windows to be opened, etc.), the competition
interaction with the environment “meshes” closely with that same among them would be so complex that they would not grant any
proceduralist notion of means-specific representation. Embodied particular action. Using Glenberg’s terminology, there would be
representation might more parsimoniously be considered such many ways to clamp projectable properties for a given environ-
means-specific representation with an emphasis on the physical ment. Our situation would thus be similar to that of Buridan’s
body and its ecological setting. proverbial donkey, unable to chose among equally attractive
Glenberg’s main answer to his question of what memory is “for” courses of action.
is that memory is designed to mesh the embodied conceptualiza- On occasion, some projectable properties may be immediately
tion of projectable properties of the environment with embodied clamped to produce reactive actions, such as avoiding an obstacle,
experiences that provide nonprojectable properties (sect. 2). He but these are not sufficient to determine agentive (self-generated,
says that this process is updated by shifting from one pattern of hierarchically organized, intentional) action. Agentive action re-
possible activations to another, thus establishing a flow among quires that projectable properties become subordinate to active
successive situations. The “streaming” of cognition is and has been nonprojectable states such as intentions, goals, emotions, and the
a salient phenomenon since the time of William James, but any like. For example, because I have the goal of cutting a strip of
such continuous process can be (and has been) described as a paper, I look for the scissors in my room (an environment involving
series of very fine-grained symbols. Current exemplar models of many potentially affordable objects) and, after finding the scissors,
memory can simulate mesh with an ongoing retrieval process that I cut that particular piece of paper instead of any other object in
continuously retrieves multiple representations acting in concert. the environment that can be cut. Thus, the projectable properties
Suitable assumptions about the granularity of the encoded of the environment do not by themselves specify particular pat-
“traces” and the time span each covers would deliver predictions terns of action. Instead, nonprojectable properties are the primary
much like those cited in favor of the idea of mesh. ones for human action.
Likewise, we hold Glenberg’s summary of work on long- and Glenberg shares with other authors (Lakoff & Johnson 1980;
short-term memory (sect. 4) as perfectly justified by the evidence Talmy 1988) the intuition that abstract nonprojectable concepts
but not any sort of radical change (Cowan 1995; Crowder 1993). are also embodied because “we understand abstract situations by
We too are attracted by the emergence of connectionism in today’s conceptualizing them in concrete ways” (sect. 6.4, para. 1). This
theoretical landscape (sect. 3.5) and by the wisdom of many of assumption is necessary for the theory because accepting nonem-
Gibson’s (1979) insights about cognition (sect. 3.2). bodied properties again would bring us to the symbol grounding
Others will undoubtedly argue in defense of the concept of the problem (Harnad 1990). However, I found some problems in the
association, whose refutation Glenberg surely intended to be notion that nonprojectable memories are embodied, namely:
among his most stunning rejections of our past. The age of this (a) the problem of discretization, (b) the lack of an explanatory
(should we say “Aristotelian”?) concept is of course not relevant to theory of conceptual metaphor, and (c) the phenomenon of
its value as an axiom now, but we wonder whether Glenberg suspended embodiment.
rejects association equally at all levels of analysis, including its Discretization. An important function of memory is to reduce
Hebbian incarnation at the neural level (Hebb 1949/1961). And the flow of events to manageable and meaningful parameters. One
how can he express sympathy for connectionism without being way to reduce information is to discretize it, which is obviously
repelled by its associative flavor? Perhaps he would suggest that done in language-mediated experience and mental models. Some
associationism is valid for the neural level of analysis but not for the of Glenberg’s observations are germane to the notion of informa-
“cognitive” or “psychological” levels. It would thus be the opposite tion reduction and discretization: He realizes that neither space
of an “emergent” principle (for such principles are evident at nor time is continuously (analogically) encoded in our memory.
higher levels of analysis but not at lower levels). For this, we Space is encoded by language into a few discrete topological
modestly propose the designation “Vanishing Principle.” regions (front, left, etc.), and time is encoded in terms of gross
categories (now, after, before, etc.). This contrasts with the fine-
grained spatiotemporal tuning with objects we need in our sen-
Embodiment in language-based memory: sorimotor experience (e.g., to move among objects or manipulate
them safely).
Some qualifications In comparison with the analogical perceptual experience, dis-
Manuel de Vega cretization has a cognitive cost. Thus, language-based spatial
Departamento de Psicologı́a Cognitiva Facultad de Psicologı́a, Universidad representations are computed in ego-centered or object-centered
de La Laguna, 38200 S/C de Tenerife, Spain; mvega6 6 ull.es frameworks that involve a relational parsing of space absent in
sensorimotor experience (de Vega et al., in press). Thus, the verbal
Abstract: (1) Non-projectable properties as opposed to the clamping of description of an object’s position involves a sentence that relates
projectable properties play a primary role in triggering and guiding human the target object to a framework object (e.g., A is below C),
action. (2) Embodiment in language-mediated memories should be quali- whereas the equivalent deictic gesture of pointing to an object
fied: (a) Language imposes a radical discretization on body constraints
(second-order embodiment). (b) Metaphors rely on second-order em-
does not require relating the pointed object to any particular
bodiment. (c) Language users sometimes suspend embodiment. framework.
In conclusion, there should be a first-order embodiment (con-
Glenberg’s target article gives a fresh approach to human concep- tinuous, analogical) corresponding to sensorimotor experience
tualization by emphasizing its embodied nature as a way to and a second-order embodiment (discretized and relational) cor-

22 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Commentary/Glenberg: What memory is for

responding to language-based memories. To what extent is the para. 1). The action selection paradigm also views mind as operat-
second-order embodiment really an embodied experience? Do all ing on sensations so as to create information for its own use.
of our memories involve implicit motor and visuospatial patterns, Information is seen, not as taken from the environment and
for instance? The answer to these questions may be crucial for processed, but as created by the individual agent. This assumes
Glenberg’s theory and they deserve urgent empirical research. embodiment in the sense of “structural coupling” (Maturana 1975;
The lack of an explanatory theory of metaphor. The fact that Maturana & Varela 1980; Varela et al. 1991). Glenberg’s notion
many abstract concepts are metaphorical supports Glenberg’s that “the world is conceptualized (in part) as patterns of possible
insightful notion of embodied conceptualization. But we need to bodily interactions” (sect. 1.3, para. 1) can be taken as a partial
know more about how metaphorical conceptualization is possible implementation of this structural coupling.
and what its functional consequences are. Again, I would like to Connections with artificial intelligence. “[T]he meaningful,
know what sort of embodiment, if any, remains in metaphor. action-oriented component of conceptualization . . . reflects how
Notice that embodiment should not be a metaphorical notion bodies of our sort can interact with objects” (sect. 1.3, para. 2).
itself in order to retain its theoretical appeal. This means that the Certain artificial systems conceptualize in this way. Edelman’s
metaphors based on embodied experience should involve real Darwin III (1989) categorizes objects by tracing their outlines
bodily constraints. However, metaphors do not seem to rely on with its hand, by feeling their texture, and by seeing their visual
first-order embodied conceptualization (continuous and analogi- patterns. Drescher’s schema mechanism (1988; 1991) categorizes
cal); instead, they are built on the reduced and discrete concepts in terms of the possible results of actions (see also Franklin 1995,
provided by language. Thus, some of the spatial metaphors de- Ch. 12).
scribed by Lakoff and Johnson (1980) are built on gross topological “[T]he spatial-functional patterns based on projectable proper-
dimensions (e.g., top-down), rather than on a Euclidean metric ties from the environment are combined or meshed with the
space, and the psychosocial causality metaphors explored by patterns from memory [patterns representing goals]” (sect. 3.1,
Talmy (1988) rely on the schematic conceptualization of force para. 1). A Maes net (1990) provides a mechanism for just this
dynamics. kind of meshing. Equipped with a sensory-data-to-percept pre-
Suspended embodiment. The most important limitation to the processor, Kanerva’s (1988) sparse distributed memory could
notion of embodied representations is that language users are furnish another such mechanism.
apparently able to suspend embodied constraints. For example, to “We cannot direct our perceptual systems to ignore differences
understand the sentence, “Mary flew from Madrid to New York,” . . . between two chairs” (sect. 3.2, para. 2). True. But note that we
we need activate neither the mental (embodied) pathway nor the need not treat these two chairs as individuals with, say, name,
body actions and sensations typical of flying. Of course, we can history, and so forth. One can be the-chair-I’m-sitting-in and the
make a quite detailed simulation or “mental traveling” from verbal other the-empty-chair-to-my-right. Computationally, these two
description if we are required to do so (e.g., Denis & Cocude approaches are quite different (see Agre & Chapman 1987).
1989). But in most cases mental traveling seems a superfluous use “[I]n the service of prediction, we have developed the ability to
of mental resources, unnecessary for ordinary comprehension . . . suppress the overriding contribution of the current environ-
(conceptualization). What probably happens in those cases of ment to conceptualization” (sect. 3.4, para. 1). Johnson and Scan-
“suspended embodiment” is that we implicitly know that we can lon (1987) found the need to build this same ability to “think” off-
instantiate the sensorimotor trajectory but we do not need to do it line into their Pacrat. In her spreading activation behavior nets,
on that particular occasion ( just as we know that a check means Maes (1990), uses no-action-below-threshold to achieve offline
money but we do not need to cash it immediately). In conclusion, “thinking.”
comprehension involves a potential embodiment rather than an “Memory’s automatic contribution to conceptualization: When
actual embodiment in some cases. we are walking the path home, we do not need to consciously recall
All three issues – discretization, metaphors, and suspended which way to turn at each intersection” (sect. 5.1, para. 1). Is this
embodiment – are manifestations of the same general principle: Brook’s (1991) “the world’s its own best model?” Continual sam-
the compacted nature of language-mediated conceptualization. pling of the environment is taking place. Yet, knowing which
The data reduction processes in language and mental models branch to take is a not a nonprojectable property of the environ-
should not be neglected in a theory of memory. The ways our body ment. It’s not a property of the environment at all. It’s a conse-
constrains our conceptualizations are probably relaxed or modi- quence of the goal of reaching home. But is it the result of some
fied in language because memories are not built on sensorimotor internal model? In the case of the branch being at hand, I think
experience, but on a schematic reduction of such experience. not. Planning a route in advance would require a model. Maybe
the idea of meshing conceptualizations provides the beginnings of
a mechanism for choice at the branch, while running a trajectory
with environmental input suppressed is the beginnings of a mech-
Action patterns, conceptualization, anism for planning.
and artificial intelligence “When the jogger puts on the sweatshirt, there is a mesh
between the jogger and the sweatshirt: wherever the jogger goes,
Stan Franklin the sweatshirt goes too” (sect. 6.3, para. 4). The frame problem
Institute for Intelligent Systems and Department of Mathematical Sciences, (Brown 1987; Franklin 1995, p. 116) that has so plagued Artificial
University of Memphis, Memphis, TN 38152; stan.franklin6 6 memphis.edu; Intelligence doesn’t arise in humans. Meshing seem to provide an
www.msci.memphis.edu/franklin explanation.
Assumptions. “[T]o a particular person, the meaning of an
Abstract: This commentary connects some of Glenberg’s ideas to similar
ideas from artificial intelligence. Second, it briefly discusses hidden
object, event, or sentence is what that person can do with the
assumptions relating to meaning, representations, and projectable proper- object, event, or sentence” (sect. 1.3, para. 1). The very notion of
ties. Finally, questions about mechanisms, mental imagery, and conceptu- meaning seems to refer back to the discredited attempt to map
alization in animals are posed. arbitrary symbols to the world. I suggest that internal “patterns of
possible bodily interactions” have a causal relationship to a corre-
I’m a proponent of the action selection paradigm of mind (Frank- sponding “object, event, or sentence,” rather than one of meaning.
lin 1995, Ch. 16), whose first tenet is that “The overriding task of To a thermostat, the particular shape of its bimetalic strip has a
Mind is to produce the next action.” Or, as William James puts it, causal relationship to its world. An outside observer might ascribe
“My thinking is first and last and always for the sake of my doing.” a meaning, say 708, to that shape, but this ascription has no
Thus, I find myself quite in tune with Glenberg’s view that relevance to the thermostat (thought of as a particularly simple
“memory evolved in service of perception and action” (sect. 1, agent sensing its particularly simple world and acting on it). To the

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 23

Commentary/Glenberg: What memory is for

thermostat, the shape of the strip is simply its perceptual mecha- source of patterns of action that mesh with the perception of
nism in the service of action selection: so, too, “patterns of possible projectable patterns obtained from the environment. The effortful
bodily interactions.” mode involves suppressing the patterns from the environment and
“Embodied representations” (sect. 1.3, para. 2): Thinking of letting conceptualization be guided solely by trajectories formed
“patterns of possible bodily interactions” as “representing” some- from previous experiences.
thing doesn’t buy us anything. Symbols represent things. These In my writing on the subject (Gärdenfors 1995; 1996), I have
“patterns of possible bodily interactions” causally interact with distinguished between two kinds of mental representations: cued
things. The association of the patterns with the things is a repre- and detached. A cued representation stands for something that is
sentation only to an outside observer. present in the current external situation of the representing
“[S]patial layout is a projectable property, whereas ownership is organism. In general, the represented object need not be present
a nonprojectable property that must come from experience” (sect. in the actual situation, but it must have been triggered by some-
2.1, para. 2). Though the distinction signaled by the terms “pro- thing in a recent situation. When a particular object is categorized
jectable” and “nonprojectable” seems a useful one, the terminol- as food, for example, the animal will act differently compared to
ogy chosen seems less than felicitous. “Projectable” would con- when the same object had been categorized as a potential mate.
note the environment actively projecting the property toward the In contrast, detached representations may stand for objects or
sensing agent. Not so. It’s the sensing that’s active, and perception events that are neither present in the current situation nor trig-
that creates the property in the first place (Franklin 1995, pp. 413– gered by some recent situation. A chimpanzee who walks away
14; Oyama 1985). The agent projects both projectable and non- from a termite hill to break a twig in order to peel its leaves off to
projectable properties onto the environment. The distinction is make a stick that can be used to catch termites has a detached
one of degree of constraint; projectable properties are more representation of a stick and its use. In Glenberg’s terminology, the
constrained by the environment (see comments on sect. 5.1). detached representations are all trajectories the animal can ac-
Questions. “To support action, the perception of projectable tively conceptualize.
properties is in terms of patterns of possible actions” (sect. 2.2, I want to argue that the distinction between cued and detached
para. 1). If so, the job of the perceptual apparatus is to turn representations fits very well with Glenberg’s two modes of mem-
incoming sensory data into appropriate patterns of possible action. ory and that it is useful for understanding many aspects of the role
What could the mechanism be? Skarda and Freeman have shown of memory in higher cognitive functions. Here I will just address
that recognition of an odor by a rabbit depends on which basin of planning and pretense (for an analysis of other functions, see
attraction a pattern of activity on the olfactory bulb falls into [See Gärdenfors 1996).
Skarda & Freeman “How Brains Make Chaos in Order to Make As Glenberg notes in section 3.4, prediction requires the (dan-
Sense of the World” BBS 10(2) 1987.] I suspect that all of our gerous) suppression of inputs from the environment. The ability to
recognitions of objects, events, and sentences happen by such a envision various actions and their consequences is a necessary
mechanism. But recognition is the identification of the appropri- requirement for an animal to be capable of planning. Following
ate concept. This leaves me wondering about the mechanism that Gulz (1991, p. 46), I will use the following criterion: An animal is
takes us from sensory patterns to “patterns of possible bodily planning its actions if it has a representation of a goal and its
interactions.” current situation and is capable of generating a representation of
“[S]tripes . . . are encoded as patterns of action” (sect. 3.1, para. partially ordered sets of actions for reaching the goal. The repre-
3). Patterns of actions rather than patterns of vision? Are the sentations of the goal and the actions must be detached, that is, the
actions eye movements tracing out the stripes as in Edelman’s input from the environment must be suppressed.
Darwin III? And, how do we abstract a purely sensory concept, There are several clear cases of planning among primates and
say red, in terms of “patterns of possible bodily interactions?” some less clear cases in other species. However, all evidence for
And what of mental imagery? Is the perceptual apparatus used planning in nonhuman animals concerns planning for present
inversely to transform action patterns into something closer to needs. Apes and other animals plan because they are hungry or
degraded sensations? I can conjure up an image of Michelangelo’s thirsty, tired, or frightened. Humans seem to be the only animal
David with no hint of action present, though my underlying that can plan for future needs. Gulz (1991, p. 55) calls planning for
conceptualization of the David may well be in terms of possible present needs immediate planning; planning for the future is
actions. called anticipatory planning. Humans can predict that they will be
If humans conceptualize in terms of patterns of possible actions, hungry tomorrow and so they save some food; we realize that the
so must our higher primate cousins. African grey parrots have winter will be cold, so we start building a shelter in the summer.
been shown to conceptualize (Pepperberg 1983). How far afield in The crucial distinction is that for an organism to be capable of
the animal kingdom does this mode of conceptualization extend? anticipatory planning it must have a detached representation of its
What other modes, if any, are there? future needs. In contrast, immediate planning only requires a cued
representation of the current need. There is nothing in the
available evidence concerning animal planning, notwithstanding
all its methodological problems, that suggests that any species
The role of memory in planning and pretense other than Homo sapiens has detached representations of its
desires. In Glenberg’s terminology, this requires that humans be
Peter Gärdenfors able to suppress the feelings and desires of the current situation
Cognitive Science, Lund University, Kungshuset, S-222 22 Lund, Sweden; and evoke memories of other desires during anticipatory planning.
peter.gardenfors66 fil.lu.se; lucs.fil.lu.se/staff/Peter.Gardenfors The point here is that evoking memories of desires and emo-
tions presumes a rudimentary “theory of mind” (see, e.g., Gopnik
Abstract: Corresponding to Glenberg’s distinction between the automatic 1993; Leslie 1987; and Premack & Woodruff 1978). An organism
and effortful modes of memory, I propose a distinction between cued and must somehow realize that the memory of an emotion was its own
detached mental representations. A cued representation stands for some-
emotional experience. Such a memory is also a prime example of
thing that is present in the external situation of the representing organism,
while a detached representation stands for objects or events that are not the embodied conceptualization on which Glenberg focuses in his
present in the current situation. This distinction is important for under- theory of memory. This line of argument thus underpins his claim
standing the role of memory in different cognitive functions like planning in section 5.2.3 that “a major factor in the development of a
and pretense. concept of self is just the ability to suppress environmental
information.”
Glenberg distinguishes between two modes of memory: automatic A distinction similar to that between cued and detached repre-
and effortful. In the automatic mode, memory functions as a sentations has been made by Leslie (1987), who argues that

24 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Commentary/Glenberg: What memory is for

“decoupled” representations are necessary for an organism to would be more feasible. In either case, adults would encode many
engage in pretense. Leslie suggests (1987, p. 415) that when a child elaborations about perceptual features and actions. When com-
pretends that a banana is a telephone, the pretend representation prehending the action “the cook tripped the butler” in a mystery
must be “quarantined” from some of the sensory information. The novel, the reader would generate inferences about how the action
perception of the banana must be complemented with informa- was performed, such as the instrument of tripping (foot), the room
tion about telephones evoked from the child’s memory. Leslie of the tripping, the location of the tripping within the room, the
claims (1987, p. 416) that the emergence of pretense “is an early part of the butler’s body that was initially contacted, and the
symptom of the human mind’s ability to characterize and manipu- trajectory and intensity of the fall. The reader would also generate
late its own attitudes to information. In short, pretense is an early inferences that explain why the action was performed, for exam-
manifestation of what has been called a theory of mind.” ple, to get revenge or to prevent the butler from meeting a guest.
The upshot is that planning and pretense, as well as other higher Available evidence indicates that mental microworlds are em-
cognitive functions, presume Glenberg’s effortful mode of mem- bodied in some respects. Zwaan et al. (1995) had college students
ory, that is, the capacity to suppress patterns from the environment read literary short stories in order to assess whether readers
and to let conceptualization be guided solely by previous experi- monitor various dimensions of the microworld, such as characters,
ences encoded as trajectories. As a consequence, this capacity is a temporality, spatiality, causality, and intentionality (i.e., characters’
sine que non for the development of a “theory of mind.” goals and plans). Reading times for sentences increased when a
new character entered the microworld, when there was a large gap
or transformation in the time line (i.e., flash back or flash forward),
when there was a change in spatial setting, when an incoming
Where is the body in the mental model action was not causally related to the prior context, and when a
for a story? character generated a new goal or plan. In contrast, coherence
breaks at the propositional, textbase level had virtually no impact
Arthur C. Graesser on reading times. Zwaan et al.’s results are compatible with
Department of Psychology and the Institute for Intelligent Systems, The Glenberg’s theory of embodied language comprehension.
University of Memphis, Memphis, TN 38152; a-graesser6 6 memphis.edu; Some negative evidence, however, suggests that mental micro-
www.psyc.memphis.edu/faculty/graesser/graesser.htm worlds are not completely embodied. Graesser et al. (1994)
proposed a constructionist theory that identifies the inferences
Abstract: Researchers in the field of discourse processing have investi- that readers encode on-line when they comprehend narrative.
gated how mental models are constructed when adults comprehend
stories. They have explored the process of encoding various classes of
According to this theory and available empirical evidence, readers
inferences “on-line” when these mental microworlds are constructed construct inferences that explain why actions/events occur and
during comprehension. This commentary addresses the extent to which why the writer would bother mentioning something in the text,
these inferences and mental microworlds are “embodied.” but not how the actions/events occur. For example, when reading
“the cook tripped the butler,” the notion of “revenge” would get
The field of discourse processing has investigated how meaning encoded but not the notion of “foot.” Thus, story comprehension is
representations are constructed in the adult understanding of explanation-driven, but not driven by a need to specify frivolous
connected discourse, such as when a person reads a novel or details about how actions/events occur and about static properties
listens to a lunchroom conversation. The early models were quite of characters and objects. Such details are encoded only to the
disembodied, as portrayed in Glenberg’s description of the stan- extent that they help explain the plot and the rationale for the
dard, symbolic, proposition-based theory of meaning. However, writer explicitly mentioning something.
the standard theory has been abandoned by most researchers in The problem of embedded microworlds also suggests that there
discourse processing. This abandonment can be attributed to are limits on the embodiment of narrative actions. If the narrator
theories that insisted on more grounded representations, as in asserts that the cook tripped the butler, the action receives a direct
situational semantics (Barwise & Perry 1982), procedural seman- focus in the discourse and may very well be embodied. However, it
tics (Miller & Johnson-Laird 1976), and studies of deixis (i.e., is doubtful whether the same action would be embodied in an
references to contextual indexes of speech acts, such as locations, embedded microworld: “The wife was embarrassed when she
times, and who is speaking to whom; see Duchan et al. 1995). heard her husband fumble nervously through the story about the
Today’s discourse processing theories offer richer specifications of cook tripping the butler.” In this case, the meaning representation
the referential content of discourse segments, the situation model, of the cook’s tripping the butler would probably be a meatless
and the communicative context of the discourse (Britton & skeleton or a symbolic token rather than an embodied representa-
Graesser 1996; Gernsbacher 1994). Although some of these theo- tion. The question would also arise as to which body is relevant.
ries have embodied meaning representations, the processing Would it be the body of the wife, the husband, the cook, or the
mechanisms do not incorporate Glenberg’s distinctive claims butler? Contemporary theories of narrative comprehension have
about meshing, trajectories, projections, and suppression. emphasized the importance of tracking the existence and points of
This commentary raises the question of whether the meaning view of multiple agents in the story world, including the narrator
representations that humans construct during discourse compre- and the narratee (Duchan et al. 1995). Glenberg’s theory does not
hension are embodied ones. Story comprehension is worth exam- identify (1) the agent that most constrains the embodied meaning
ining because stories have received the most attention by dis- representation and (2) the conditions in which embodiment is
course researchers, and they are microcosms of events and suppressed.
experiences in the real world. That is, both stories and everyday In closing, most researchers in discourse processing would
experiences include people performing actions in pursuit of goals, resonate with Glenberg’s claims about embodied meaning repre-
events that present obstacles to these goals, conflicts between sentations. However, there are conditions in which the representa-
people, emotional reactions, and spatial scenarios. When a novel is tions are either not embodied or are substantially depleted of flesh
comprehended by an adult, the reader constructs a fictitious and bone. Glenberg needs to identify these conditions in order
microworld of such content. To what extent is the content of the make his theory a serious competitor in the field of discourse
microworld an embodied representation? processing.
If the mental microworld were completely embodied, the
reader would encode many details about the characters, objects,
spatial layout, actions, and events – much as in a high-resolution
videotape. However, stories are read rather quickly, at the rate of
approximately 150 to 400 words per minute, so a fuzzy videotape

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 25

Commentary/Glenberg: What memory is for

Embodied representations are part terference effects depend on the specific combination of task-
of a grouping of representations irrelevant environmental stimuli and the particular kind of infor-
mation needed for the task at hand. This basic finding is difficult to
Christopher Habel, Barbara Kaup, and Stephanie Kelter reconcile with the notion of a single integrated representation, nor
Graduate Program in Cognitive Science, University of Hamburg, D-22527 can it be accounted for by Glenberg’s notion of suppression, in which
Hamburg, Germany; habel66 informatik.uni-hamburg.de; the environment is globally set against internal representations.
www.informatik.uni-hamburg.de/wsv/hp/christopher-english.html Indirectly embodied knowledge. Glenberg’s proposal is not
very clear with respect to a serious problem for any embodiment
Abstract: Glenberg argues for embodied representations relevant to
theory, namely, the meaning of things “when there is no apparent
action. In contrast, we propose a grouping of representations, not neces-
sarily all being directly embodied. Without assuming the existence of possibility for bodily interaction” (sect. 1.3, para. 6). We agree with
representations that are not directly embodied, one cannot account for the Glenberg that Talmy’s “force-concept” (sect. 6.4) can be consid-
use of knowledge abstracted from direct experience. ered an example of knowledge abstracted from experience: start-
ing with the bodily experience of gravitation and mechanical force,
Glenberg’s innovative proposal is grounded in the assumption that it seems possible to expand the concept into other areas, such as
memory is designed to serve particular goals; this involves the electromagnetic force. However, instead of assuming that the
computational level, in the sense of Marr (1982). We will address representation of force stays exclusively on the level of embodied
this aspect of Glenberg’s approach before we turn to the more representations, we prefer the view that knowledge can be “indi-
specific, algorithm-level assertions. rectly embodied.” More precisely, humans are able to abstract
The computational level. In psychological theories of memory from the details of an experience to a greater or lesser extent. We
and cognition, mental representations are usually implicitly ex- accordingly propose replacing the dichotomy of “embodied repre-
pected to represent the world as veridically and completely sentations vs. abstract concepts,” with a spectrum in which knowl-
possible. Empirical evidence of discrepancies between what is edge of different types of abstraction coexists. “Indirectly embod-
mentally represented and what is actually the case are referred to as ied concepts” are those that lack the specific aspects of bodily
“distortions,” “illusions,” or “fallacies” if they cannot be attributed interaction for the experiences from which they originate. Such
to “capacity limitations.” When eventually biological requirements “abstracted representations” are essential for commonsense rea-
are taken into consideration, it often turns out that the discrepancy soning (see Habel & Eschenbach 1995). For instance, abstracting
can be seen as resulting from an advantageous processing principle. from certain aspects of direct experience with sound and heat,
Research on visual attention provides an example. Attentional indirectly embodied concepts such as “radiation” or “transmis-
phenomena have traditionally been viewed as arising from a basic sion” may be formed. Such abstracted concepts of “distant (mate-
capacity limitation, but today it is widely proposed that they may be rial) influence” are useful for commonsense reasoning. For exam-
better understood as a consequence of coordinative processes that ple, the layman’s difficulty with thinking about radioactivity is
protect ongoing action choices from interruption by irrelevant caused by the inability to perceive it, but this can be partially
information (e.g., Van der Heijden 1991). overcome by using abstract conceptualizations.
Glenberg takes the opposite approach. His proposal proceeds Another example: since numbers can be seen as abstract con-
from considerations about the purpose of cognitive processing and cepts, we must consider whether, and by what links, knowledge
thereby specifies principles of information selection right from the about numbers and numerical properties is embodied. Even if
beginning: mental representations contain information relevant to connection between numbers and sets of concrete entities can be
action or action prediction. Representations can only be expected seen as the embodied origin of some numbers, namely, the small
to be veridical or complete to the extent necessary or advan- ones, it is useful to use an abstract number concept for problem
tageous for these purposes. This approach is extremely important solving. Numbers, that is, abstract entities, are the basis for the
for research on memory and language where guidelines for mak- solution of related but clearly distinguished problems with respect
ing hypotheses about information selection are lacking. Many to quantities (e.g., “buying three Cokes, each for $1.00” and
studies have attempted to determine what information gets en- “watching three beautiful sunsets while drinking one martini each
coded during text comprehension, yet there is still no principled time”). Furthermore, it is not plausible to see direct embodiment
and comprehensive account of the data: readers/listeners form as the way to ground large-number concepts. Rather, they are only
neither “veridical” representations of the propositional text base indirectly correlated – via small numbers – with our embodied
nor fully specified representations of real-life-like situations. For environment (cf. the “numerosity” concept of Gallistel & Gelman,
example, it cannot really be explained why readers/listeners put 1992). [See also Davis & Pérusse: “Numerical Competence in
special weight on the protagonist’s goals (Graesser et al. 1994; Animals” BBS 11(4) 1988.]
Kaup et al. 1996) or how the experimental task or instruction To summarize: it is important to combine the level of embodied
influences which information is mentally represented (e.g., Al- representations with representations of a more abstract character
brecht et al. 1995). in a heterogeneous system.
To conclude: the presupposition that mental representations
are ideally veridical and complete has to be abandoned. A discus-
sion of the principles of information selection is urgently needed.
One hopes that Glenberg’s target article will stimulate such a Has Glenberg forgotten his nurse?
discussion.
Arthur M. Jacobs and Johannes C. Ziegler
The algorithmic level. Glenberg’s central concept, embodied
Center for Research in Cognitive Neuroscience, Centre National de la
representations, seems to be rich representations containing var-
Recherche Scientifique, 13402 Marseilles Cedex 20, France;
ious kinds of information (motor programs as well as kinestetic, jacobs66 lnf.cnrs-mrs.fr
visual, auditory, haptic, and other perceptual information, emo-
tions, goals, etc.) whereby relevance to action or action prediction Abstract: Glenberg’s conception of “meaning from and for action” is too
constitutes the selection criterion. This idea is problematic, if an narrow. For example, it provides no satisfactory account of the “logic of
embodied representation is meant to be supported by just a single Elfland,” a metaphor used by Chesterton to refer to meaning acquired by
mental subsystem. Instead, a grouping of representations, each being told something.
supported by a different mental subsystem that is responsible for
providing a specific kind of information (cf. Barwise & Etche- All that we call spirit and art and ecstasy only means
mendy’s [1995] “heterogenous reasoning”), is better compatible with that for one awful instant we remember that we
forget.
numerous findings from the neurosciences and psychological
research with the dual-task paradigm. It is well-established that in- G. K. Chesterton (in Gardner 1994, p. 101)

26 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Commentary/Glenberg: What memory is for

Bolzano, Brentano, and Husserl all had a problem with Kant. They The myriad functions and metaphors
would not believe that logic and psychology have something to do of memory
with one another. They would not believe that the laws of logic are
identical with cognitive processes. For example, Husserl taught Asher Koriat and Morris Goldsmith
that logic has nothing to do with “cognitive acts.” Putnam (1981), Department of Psychology, University of Haifa, Haifa, 31905 Israel;
Lakoff (1987), and Glenberg have a problem akin to the aforemen- 6 psy.haifa.ac.il; mgold6
akoriat6 6 psy.haifa.ac.il
tioned authors. They do not believe that thought is abstract
(disembodied), atomistic, and logical (Lakoff 1987, p. xiii). Lakoff Abstract: Glenberg provides a new and exciting view that is especially
useful for capturing some functional aspects of memory. However, mem-
teaches “experientialism,” the doctrine that reason is made pos-
ory and its functions are too multifarious to be handled by any one
sible by the body. Glenberg buys Lakoff ’s doctrine, together with conceptualization. We suggest that Glenberg’s proposal be restricted to its
Rosch’s (1973) prototype theory (ignoring classic objections to own “focus of convenience.” In addition, its value will ultimately depend
prototype theories of categorization; see Johnson-Laird 1987). He on its success in generating detailed and testable theories.
sets out to promote the experientialist doctrine by proposing that
memory is for, and meaning is from, action. Glenberg proposes a new approach to memory that is predicated
We shall not argue with Glenberg about the old idea that on certain assumptions about the function of memory. He states
meaning can come from doing or using things. What we want to that his proposal “is not a fully testable theory” (sect. 1.3, para. 7),
discuss is his forgetting of the logic of Elfland. “I knew the magic but evaluates it in light of empirical findings in a post-hoc manner.
beanstalk before I had tasted beans; I was sure of the man in the Overall, 23 different phenomena of memory, language, and
moon before I was certain of the moon.” By using these meta- thought are addressed (as summarized in sect. 7.1), ranging from
phors, Chesterton (1994), the inventor of “Father Brown” and the symbol grounding problem to the effects of expertise.
critic of Darwinism, admirably expressed what Popper (1935) We believe that Glenberg’s proposal is best viewed as a meta-
wrote a whole book about: the primacy of deduction over induc- theoretical rather than a theoretical contribution. What he is
tion. As far as Glenberg’s ideas about semantics or language offering is more a metaphor (or set of metaphors) for thinking
comprehension are concerned, Chesterton’s lesson is this: there about memory than an articulated and testable theory. As meta-
are things the meaning of which we do not get by doing or using phors, the concepts of “embodiment,” “mesh,” “trajectory,” and so
them. There are laws we learn before we can speak and shall retain forth allow certain aspects of memory to be construed in terms of
when we cannot write. (In Chesterton’s time and context, these the physical analogs of these concepts. They can help guide our
“laws” were mainly taught by nurses via fairy tales.) These laws, thinking about memory and perhaps inspire the development of a
thus imparted, can give us meaning. Examples include the lesson new genre of memory theories. However, as currently formulated,
of “Beauty and the Beast,” which states that “a thing must be loved the concepts and auxiliary assumptions are too underspecified to
before it is lovable”; or the less poetic lesson that “if Jack is the son constitute such a theory in themselves.
of a miller, a miller is the father of Jack.” Metaphors play an essential role in science (Koriat & Goldsmith
Thus, as long as Glenberg’s neo-Gibsonian, anti-Tarskian theory 1996a). They are cognitive tools that help in abstracting the critical
does not provide an account of getting meaning from being told aspects of the phenomena, in defining the questions of interest,
something (cf. Johnson-Laird 1987), it remains incomplete. Glen- and in guiding the research approach. Moreover, they provide a
berg’s reply to the third objection to his ideas, which is closely general conceptual framework that serves that development of
related to our argument, may leave many readers bewildered. He specific theories and models. Unlike the theories that they breed,
counters the point that some things are meaningful even when however, metaphors are neither right nor wrong: They can be
there is no apparent possibility for bodily interaction with the judged only in terms of their usefulness. This applies to Glenberg’s
insight that meaning depends on context! When dealing with proposal as well. Three main conclusions follow from this view-
mental models, he says that being told “That plate is hot” modifies point.
the embodied representation of the plate in order to modify First, a major benefit of Glenberg’s conceptualization is that it
interactions with the plate. Perhaps if he gave the reader some brings to the fore certain neglected aspects of memory. As he
general notion of how and under what conditions such modifica- points out (sect. 7.4), traditional memory research has generally
tions take place, one would have a starting point for tackling the focused on the study of item memorization. Behind this preoc-
delicate problem of the logic of Elfland. cupation lies the storehouse metaphor of memory, which has
Glenberg’s account is also strangely isolated from other pre- shaped much of the history of memory research (Koriat & Gold-
vious and similar ones. Thus, one wonders how much his theory smith 1996a; Roediger 1980). In recent years, however, many
sketch adds to Piaget’s (1968) genetic epistemological theory of important memory phenomena that have attracted experimental
the development of meaning in infants. How do embodied con- attention – such as procedural memory, implicit or indirect
ceptualizations relate to Piaget’s “schème” (the generalizable memory, and priming effects – do not yield readily to conceptual-
aspect of coordinating acts that can be applied to analogous ization in terms of the storage and retrieval of discrete memory
situations)? When Glenberg says that abstract words are under- traces. To accommodate these and other phenomena, several
stood by analogy to concrete ones, does he mean something like alternative conceptualizations have been put forward, including
Piaget’s transition from the sensorimotor stage of thinking to the Bransford et al.’s (1977) “stage-setting” metaphor, the “pro-
preoperational and operational stages? In this regard, we wonder ceduralistic” or “skill” view of memory (e.g., Crowder 1993; Kolers
how Glenberg acquired the meaning of the term “abstractness,” & Roediger 1984), and the “tool versus object” distinction offered
since he picks the notion of “force” as a prototypical abstract by Jacoby and Kelley (1987).
concept. We can see how this concept can be embodied. But Such efforts notwithstanding, we are still lacking convenient
would concepts like “atom,” “schizophrenia,” or “Greek philoso- metaphors to guide our thinking about the effects of past experi-
phy” not be somewhat more exciting challenges to his theoretical ence on perception and action. Glenberg’s view focuses attention
claims? on precisely this aspect of memory and offers a rich vocabulary for
In terms of Chesterton, one may be tempted to ask: Has construing the working of memory within the functional context of
Glenberg forgotten his nurse, or whoever it was who told him fairy an organism’s active interaction with the environment (cf. Bruce
tales? 1985; Neisser 1988). Hence, one could find the view useful even if
it had no predictive power beyond other existing formulations.
Second, however, metaphors fulfill a critical function by serving
as stepping stones toward testable theories. Here we see Glen-
berg’s proposal as mainly a promissory note. Our feeling is that
there is a jump from a metatheoretical, metaphorical level of

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 27

Commentary/Glenberg: What memory is for

analysis to the empirical level (as but one example, consider the What working memory is for
argument regarding the need for cyclical activity in rehearsal, put
forward in sect. 4.3, para. 4). Although a metaphor both guides and Robert H. Logie
constrains the types of theories that can be developed, different Department of Psychology, University of Aberdeen, Aberdeen, Scotland,
theories – and hence predictions – can be derived from the same United Kingdom; psy11866 aberdeen.ac.uk; www.psyc.abdn.ac.uk
metaphor. Ultimately, then, the value of the proposal will depend
on its success in breeding detailed theories from which testable Abstract: Glenberg focuses on conceptualizations that change from
moment to moment, yet he dismisses the concept of working memory
predictions can be derived.
(sect. 4.3), which offers an account of temporary storage and on-line
A third observation also stems from our view of the function of cognition. This commentary questions whether Glenberg’s account ade-
conceptual metaphors. As a cognitive tool, each metaphor has its quately caters for observations of consistent data patterns in temporary
own “focus of convenience” (Kelly 1955), that is, a domain of storage of verbal and visuospatial information in healthy adults and in
phenomena or processes for which it is best suited. The focus of brain-damaged patients with deficits in temporary retention.
convenience of Glenberg’s proposal seems to be the implicit and
procedural aspects of memory that support a major portion of our If I close my eyes and then try to pick up the pen on the desk in
daily interaction with the environment. These phenomena are front of me, it is immediately apparent that we humans have
salient in many of the examples used to introduce the basic core of temporary representations of our immediate environment. These
the proposal (sect. 2). representations survive the offset of visual perception and support
Glenberg, however, makes a great effort to stretch his conceptu- our interactions with the environment that we have recently
alization to other areas that seem to be well beyond its natural perceived. Memory then offers a means to support this interac-
focus of convenience. Although his desire to provide a compre- tion, but are the temporary representations products of how
hensive and integrative conceptual framework is understandable, memory works or do they arise from emergent properties of the
we feel that by going too far afield he does a disservice to his cognitive apparatus for temporary retention (Logie 1995; Rich-
proposal. Some of the extensions of his framework seem to be ardson et al. 1996)?
rather forced and unconvincing. For example, the idea that ex- Consider another observation. There are individuals who, fol-
plicit, episodic remembering is effortful because it requires sup- lowing brain damage, are unable to retain simple verbal se-
pression of “clamped” impinging stimuli (sect. 5.2) adds little to quences, and who fail to show a range of phenomena linked with
existing explanations and fails to address many of the essential temporary retention of words by healthy brains (e.g., Vallar &
aspects of such remembering that seem to be captured better by Baddeley 1984; for a review, see Della Sala & Logie 1993). Yet
other metaphors. these same individuals can hold normal conversations and seem to
In our opinion, one should not attempt to achieve too much with have little difficulty in finding their way around in the world. A
any single metaphor. Glenberg began his proposal by asking “what different kind of brain damage can result in individuals who have
is memory for?” and his answer led him to a particular view of no difficulty retaining verbal sequences or describing a scene
memory with a particular focus of convenience (see also Alterman while viewing it, yet cannot adequately access information from
1996; Karn & Zelinsky 1996). Others, focusing on different func- parts of the scene once it has been removed (Beschin et al., in
tions and aspects of memory, have been led to rather different press; Guariglia et al. 1993).
views. For example, emphasizing the role of memory in providing The current representations of scenes or words in each of the
a faithful account of past events, we proposed a “correspondence” above scenarios might be likened to Glenberg’s notion of meshing
metaphor (Koriat & Goldsmith 1996a) that is useful in such or conceptualization. But in his analysis there is little to account for
domains as autobiographical memory, eyewitness testimony, and those aspects of memory that, in the absence of the external
metamemory (Koriat & Goldsmith 1994; 1996b). Neisser (1996) physical stimulus, might allow the conceptualizations to be main-
stressing the social functions of memory in everyday life (e.g., tained moment to moment, or to be updated and manipulated.
impression management), proposed to view memory as a form of Various theories of cognition have attributed these kinds of cogni-
“doing” (see also Winograd 1996). Anderson (1996), stressing the tive functions to what is often referred to as working memory.
contribution of memory to the formation of value judgments (e.g., There appears to be no place for this breed of theory in Glenberg’s
attitude formation), opted for a “value metaphor,” in which mem- view, yet he argues that on-line conceptualization is the reason that
ory involves the “on-line construction of values and integration we have memory. His arguments fail to consider the reports of
thereof.” patients with specific deficits of temporary storage. However, the
How should one treat such differences of opinion regarding the contrasting data patterns from amnesics and from patients with
essential nature of memory? Clearly, each view entails its own short-term retention deficits offer strong evidence for functional
unique framework for memory research and theorizing. Nev- dissociations between modules of working memory and a cumula-
ertheless, they can all live together peacefully and contribute to tive collation of knowledge and experiences.
the study of memory in their respective domains. As we have The demonstration of long-term as well as short-term recency
argued previously (Koriat & Goldsmith 1996c), memory is not effects (sect. 4.3, para. 1) does little to erode the case for a separate
monolithic, and any attempt to characterize it in terms of a single working memory. The time scales over which these different
conception or function will certainly not do justice to its inherent forms of recency appear are dramatically different, and I have yet
heterogeneity. Thus, in line with our call for “metaphorical plural- to see evidence of suffix effects or effects of delayed recall in a
ism,” we applaud Glenberg’s proposal as a stimulating new addi- study of long-term recency. Moreover, the demonstration of se-
tion to our arsenal of conceptual tools for understanding memory. mantic coding in temporary storage tasks simply indicates that
However, no approach can claim to have a monopoly on the short-term storage is not limited to the traditional view of a short-
myriad facets and functions of memory. term verbal memory. The finding is entirely consistent with
working memory as a bailiwick of specialized cognitive functions
that support temporary storage and on-line manipulation of repre-
sentations. Other counterarguments can be offered for the re-
maining examples given in this section of Glenberg’s target article.
The notion of working memory offers a framework within which
to account for on-line semantic processing (e.g., Just & Carpenter
1992), for temporary storage of visual and spatial properties of the
environment (Logie 1995), and for temporary storage of verbal
material. In particular, the concept of the phonological loop has
been singularly successful in providing a coherent account of a

28 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Commentary/Glenberg: What memory is for

range of phenomena, including neuropsychological data, using sensory projections to guide its interactions: It cannot infer the
relatively few assumptions (Baddeley 1996). likely consequences of interactions except through spatiotemporal
The argument that the modules of working memory simply correlations in its sensory projections and internal variables.
comprise a range of acquired skills (sect. 4.3, para. 2) begs more An organism’s sensory projections provide multidimensional
questions than it answers. We know at least as much (if not more) analog input from sense organs. However, in order to model
about temporary memory as we do about skill acquisition. For potential interactions, its cognitive system must be able to deter-
example, why should there be such consistent data patterns in mine which changes in its sensory projections were self-induced.
temporary storage performance across individuals if performance (This is necessary, for example, to determine whether to attribute
relies on acquired skills rather than some universal aspect of a a movement on the retinal image to the organism or to a perceived
cognitive architecture? Moreover, could a theory based on ac- object.) Thus, memory must also mesh feedback concerning the
quired skills offer a coherent account both of normal temporary organism’s multidimensional motor signals.
storage and of deficits found in patients such as “PV” (Vallar & Glenberg offers no specific proposal to incorporate internal
Baddeley 1984) or “NL” (Beschin et al., in press)? An alternative feedback into his “spatial-functional” notion of mesh and ex-
role for skill acquisition might be in learning to use components of presses ambivalence concerning whether this would be possible
the cognitive apparatus or in learning to use them more effectively. (sect. 7.2). Given the conclusions of his analysis (namely, that
For example, most human beings have the apparatus for generat- memory is for modeling potential interactions), internal feedback
ing speech, and aspects of this apparatus can be used covertly for and motivation must be included, first, because internal feedback
subvocal rehearsal. The model of the phonological loop then offers is necessary to discriminate survival-relevant differences in an
an account of the nature of this apparatus and how it is used to organism’s sensory projections. One way we learn to distinguish
provide temporary verbal storage. A related argument has been sensory projections is by correlating them with their physiological
made for temporary retention of movement sequences and visual and affective consequences. For example, we learn to discriminate
properties of scenes or objects (Logie 1995). berries not only by the interactions they afford (e.g., picking,
The distinction between working and long-term memory has an chewing) but how they taste and how we feel after eating them. We
extended pedigree, going back at least as far as Locke (1690) who are also naturally predisposed to certain affective reactions (e.g.,
referred to the distinction between “contemplation” and the fear of snakes).
“storehouse of ideas.” Of course, historical precedent does not Suppose memory were to mesh sensory projections and motor
necessarily justify the distinction, but unlike the blind men (sect. signals to the exclusion of their internal consequences. Then
7.4, para. 2), we already have a global view of the elephant (i.e., spatial features that were more relevant to discriminating objects
memory). What we are interested in is how its various parts help it by their physiological effects would be obscured by those that
make a path through the jungle. were less relevant but more pronounced. According to Glenberg,
what makes “one path the path to your house, is its relevance to
you, that is, how you have interacted with it in the past” (sect. 2.1,
para. 2). But the path’s “relevance to you” cannot be equated with
Memory must also mesh affect “how you have interacted with it” unless your affective reactions to
it and the affective consequences of your interactions with it have
Karl F. MacDorman been meshed with your prior interactions.
University of Cambridge, Computer Laboratory, Cambridge CB2 3QG, The second reason memory must incorporate internal feedback
England, United Kingdom; karl.macdorman6 6 cl.cam.ac.uk; and motivation is because we cannot settle the symbol grounding
www.cl.cam.ac.uk/users/kfmll problem (Harnad 1990) until we have explained how goals arise.
Glenberg writes that “[o]bjects fall into the same (basic) category
Abstract: To model potential interactions, memory must not only mesh because they can be used to accomplish the same interactive goal”
prior patterns of action, as Glenberg proposes, but also their internal
(sect. 1.3, para. 3). Presumably, an organism (gradually) sensitizes
consequences. This is necessary both to discriminate sensorimotor infor-
mation by its relevance and to explain how goals about the world develop. itself to the category through achieving the goal. This presupposes
In the absence of internal feedback, Glenberg is forced to reintroduce a that it already has a goal that can be expressed in terms of patterns
grounding problem into his otherwise sound model by presupposing of actions based on projectable properties of the environment.
interactive goals. This is no less solipsistic than presupposing that it already has a
symbol that corresponds to the category (e.g., Fodor’s 1975 nativ-
Glenberg’s target article provides an excellent description of how ism). Truly adaptable behavior requires some interactive goals to
cognitive categories might develop through bodily interactions: develop within the lifetime of the organism, because specific
memory meshes patterns of action with previously meshed pat- interactive goals cannot evolve in anticipation of needs that have
terns of action by virtue of their (analog) shape. I believe Glen- yet to arise. Hence, it would seem reasonable to posit internal
berg’s basic thesis, that memory contributes to survival by model- variables (which indicate general health maintenance and repro-
ing potential interactions, has much to offer. While dictionaries ductive needs) that guide empirical adaptation (i.e., memory)
tend to identify memory with conscious recall, Glenberg provides toward the development of interactive goals.
a framework for modeling not only recollections but virtually any One characteristic of radical behaviorism (and simple reinforce-
kind of empirical sensorimotor adaptation. It is well worth examin- ment learning) is that it excludes interactive goals as theoretical
ing how far Glenberg’s framework can take us in simulating constructs. Instead, it is assumed that patterns of response develop
intelligent behavior and where it may need augmentation. on the basis of stimulus and reward without reference to goals
We first need to define what is meant by “patterns of action.” about the world. It would seem that, within Glenberg’s framework,
Glenberg bases them on “projectable properties of the environ- potential goals are solely a function of an organism’s body and
ment” (sect. 2.1). His use of the term properties is unfortunate environment. An alternative is that an organism develops goals
because it implies that we should conceptualize memory in terms with reference to its conceptualization (or perceptual world)
of properties that exist independently of any particular organism under the influence of internal feedback (Cowley & MacDorman
(with its unique body, sense organs, and life history). The same is 1995; MacDorman 1996).
true when Glenberg writes of the need to remain “reality-
oriented” and to see “the environment for what it is” (sect. 3.2,
para. 1). Although one may certainly assume the existence of an
observer-independent reality, it does not follow that it is possible
or useful to develop cognitive constructs in terms of it. I find no
basis to claim that an organism has anything outside itself and its

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 29

Commentary/Glenberg: What memory is for

Is memory caught in the mesh? Meshing integrates a stimulus/episode/event with an action. Ini-
tially, I was tempted to view meshing as a convolution of stimulus
Colin M. MacLeod and action, like the memory storage scheme in todam2 (Murdock
Division of Life Sciences, University of Toronto at Scarborough, 1993). But then what would it mean to say that a mesh is not an
Scarborough, Ontario M1C 1A4, Canada; association, as Glenberg repeatedly emphasizes? If what is meant
macleod6 6 banks.scar.utoronto.ca; is that associations link two episodes or stimuli, whereas meshes
www.scar.utoronto.ca/tweb/acad/ integrate an episode with an action, then I think this relies on too
life-sciences/psychology/macleod.html
narrow a definition of association. Even Aristotle wrote that “acts
of recollection, as they occur in experience, are due to the fact that
Abstract: Can memory be cast as a system that meshes events to actions? one movement [that is, thought] has by nature another that
This commentary considers the concepts of mesh versus association,
arguing that thus far the distinction is inadequate. However, the goal of
succeeds it in regular order” (in Herrnstein & Boring, 1966,
shifting to an action-based view of memory has merit, most notably in p. 328). How could we ever tell a thought-movement mesh from
emphasizing memory as a skill and in focusing on processes as opposed to an analogous association? Perhaps Glenberg’s intention is to avoid
structures. the excess baggage associated with the concept of association; if so,
it would be valuable to specify more clearly how a mesh is not an
By attempting to tie memory to action, Glenberg is allying himself association, given that this is his core idea.
with his colleagues who study attention (e.g., Allport 1989) and Having thus far been critical, I hasten to add that there is much
who increasingly emphasized this connection. Although I would to commend in Glenberg’s account. I agree that memory is best
not accuse him of false modesty in saying that “this sketch is not a viewed as a set of skills serving perception and action, which
fully testable theory” (sect. 1.3, para. 7), I do applaud the effort and explains why there are no all-purpose, effortless mnemonics. Skills
agree that it is necessary for a full understanding of cognition. In – whether in memory or in tennis – require practice. I also concur
what follows, I consider the success of this particular enterprise that priming due to past experience(s) does not enhance present
with regard to memory only, leaving its relevance to language for perception; rather, it increases the fluency of interpretation and
others to evaluate. the coming to mind of relevant prior episodes (see Masson &
No “sketch” is well defined and absolutely free of speculation; MacLeod 1996). Similarly, conscious recollection produces its
indeed, its purpose is to stimulate the development of a more full- characteristic “feel” via an attribution applied to fluent reprocess-
fledged theory. Nonetheless, I found the frequent reliance on ing (cf. Jacoby & Brooks 1984). I even agree that comprehension
undefended intuition troubling. What is a “commonsense test of may be a general skill (Palmer et al. 1985).
what it means to have meaning” (sect. 1.2, para. 2)? And why is it If Glenberg’s view of memory based on event–action meshing is
“hard to imagine how cognition could be based on the mappings of to prosper, many questions must be addressed at a number of
arbitrary symbols” (sect. 1.2, para. 7)? Likewise, I found the levels. I will end by raising a few of these (without providing any
casting aside of existing accounts to be rather cavalier, as when at hint of an answer). How can we distinguish events from actions, or
the outset it is simply asserted that most theories reveal “little that is this even a reasonable question? How is a mesh not an associa-
is important about memory” (sect. 1.1, para. 2). Which theories are tion? What role do instances play in remembering? Why, if we do
meant to be included in this sorry collection? Or take the case not experience categories, are object pictures categorized faster
where propositional theories are discarded because “a simple list than they are named (Smith & Magee 1980)? What must be done
of well-formed propositions” (sect. 1.2, para. 8) could not possibly to connectionist modeling to make it work for an “embodied
do the necessary job. Would anyone expect this to be a “simple” system”? If suppressing the environment is crucial to successful
task? And then there is the case of feature-based systems, where remembering, might field dependence (Witkin & Goodenough
“it is difficult to understand how people can ever learn anything 1981) be a good predictor of individual differences in memory?
truly new” (sect. 3.3, para. 4). Why would this criticism not also be Should connecting of events to actions, as opposed to other events,
true for a system based on actions (or their components)? lead to better memory for those events (see Cohen, 1989, for
More in the way of concrete comparisons to and criticisms of supporting evidence)? And above all, is mesh strong enough to
existing views would have been welcome. Even explanations that support memory?
would appear to serve as springboards for Glenberg’s views – such
as instance theories (e.g., Logan 1988) or transfer-appropriate skill
views (e.g., Kolers & Roediger 1984) – are only mentioned and not Semantic memory
integrated into the present sketch. I would take this a step farther:
when the behaviorists tried to account for all of cognition in terms Timothy P. McNamara
of behavior, appeals were made to such ideas as subvocal speech to Department of Psychology, Vanderbilt University, Nashville, TN 37240;
explain language (Skinner 1957). How big a step is Glenberg’s mcnamara66 ctrvax.vanderbilt.edu
account from this position, and in what direction?
The central concept of Glenberg’s approach is that of “mesh.” Abstract: Glenberg tries to explain how and why memories have semantic
content. The theory succeeds in specifying the relations between two
major classes of memory phenomena – explicit and implicit memory – but
it may fail in its assignment of relative importance to these phenomena and
in its account of meaning. The theory is syntactic and extensional, instead
of semantic and intensional.

Disguised as a monk from the Monastery of Memory, Glenberg

infiltrates the Citadel of Semantics, bastion of philosophers and
linguists. Glenberg wants memories to be meaningful, to be about
things in the world, and he wants to understand how memories
refer. But prevailing theories of memory do not provide solutions
to these problems. Glenberg’s answer is that meaning is bodily
movement, that memory is a vassal to perception and action, and
that memory is for meshing the “embodied conceptualization of
projectible properties of the environment . . . with embodied
experiences that provide nonprojectible properties (sect. 2.2, para.
5).” Glenberg is after a theory of meaningful memories, a genuine
Figure 1 (McNamara). Four-point path. theory of semantic memory.

30 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Commentary/Glenberg: What memory is for

I will confess that even after several readings of the target 1980). What is missing in all theories of which I am aware,
article, I don’t understand exactly what Glenberg has in mind. It including Glenberg’s, is intension: The stuff that allows us to
occurred to me, for example, that with a bit of care I could communicate about tables, cephid variables, and unicorns; the
substitute proposition-speak (e.g., proposition, argument overlap, stuff that explains why it is possible for someone to believe that O.
elaboration, etc.) for Glenberg-speak (e.g., conceptualization, J. Simpson lives in California and not believe that the murderer of
mesh, embodiment, etc.) and the paper would be transformed into Nicole Brown Simpson lives in California, even though the two
an endorsement of propositional theories. This leads me to con- may be one and the same person. As I understand them, embodi-
clude that I’m missing something big. Hence, my comments will ment, mesh, and symbol grounding are not intensional concepts.
be brief and probably idiosyncratic. Glenberg’s proposal is a species of naturalistic psychology. The
Let me start by identifying what I think are two intriguing ideas, concepts of embodiment, mesh, and so on are not theoretically
one I think is right and one about which I am skeptical. The right grounded until we have some idea about what bodies get embod-
idea is that there is an extremely important but largely unexplored ied in and what it is that representations get meshed with.
relation between perception, action, and memory (e.g., Rieser Glenberg’s psychology of memory would seem to require a theory
1989). Glenberg describes some of the relevant research, but an of the structure of the world; after all, projectible properties will
example is more compelling. Consider the four-point path in be described in the vocabulary of physics, not psychology. There is
Figure 1. Study this diagram for a few seconds, and then answer nothing wrong in principle with naturalistic theories, but they may
the following questions without looking at the figure: have severe practical limitations, such as being the last of all
1. Without rotating your body, point to 1 as if standing at 3 scientific theories to be completed (e.g., Fodor 1980). I appreciate
facing 4. Glenberg’s desiderata of a theory of memory, but I’m inclined to
2. After rotating your body 1808, point to 4 as if standing at 2 think that memory researchers may be better off abandoning the
facing 1. citadel, returning to the monastery, and getting back to work.
The first judgment is hard; the second one is easy. Why? The
answer is clear – bodily movement accomplishes mental computa- ACKNOWLEDGMENT
tion – but the mechanisms are poorly understood (but see Pel- Preparation of this commentary was supported in part by National Science
lizzer & Georgopoulos 1993). Foundation Grant SBR-9222002.
The other idea is that the role of implicit memory is to integrate
representations provided by perceptual systems with representa-
tions of previous experience, and that our experience of explicit
memory arises from the suppression of the environment and the Is memory like understanding?
attendant rise of the influence of representations of past experi-
ence. To my knowledge, this conceptualization of the relation Gail Musen
between implicit and explicit memory is new. My enthusiasm is Department of Psychology, Barnard College, Columbia University, New
York, NY 10027-6598; gm306 6 columbia.edu
dampened, however, by the claim that the most important func-
tion of memory is the implicit, automatic function of integrating
Abstract: There are three major weaknesses with Glenberg’s theory. The
immediate and past perceptions. Here’s my problem: People for
first is that his theory makes assumptions about internal representations
whom this system is intact but the other system is scrambled that cannot be adequately tested. The second is that he tries to accommo-
(anterograde amnesics) are totally helpless. They can’t learn any- date data from three disparate domains: mental models, linguistics, and
thing new (with the exception of Tower of Hanoi, mirror tracing, memory. The third is that he makes light of advances in cognitive
and other equally exciting skills) and must be under 24-hour care. neuroscience.
Amnesics cannot find the path home, let alone follow it. Contrast
the plight of an amnesic patient to that of patient M.S., who has Glenberg attempts to integrate memory, linguistic, and mental-
severely impaired visual implicit memory but intact explicit mem- model research into a coherent model of the practical uses of
ory: He owns a computer software company (Gabrieli et al. 1995). memory. He claims that cognitive properties are “embodied” and
I acknowledge that this comparison is unfair because M.S. has cannot be treated as meaningless symbols that can be manipu-
apparently lost only one of what may be a large number of implicit lated. Glenberg develops the idea that representations are analogi-
memory systems. Even so, the comparison forces me to wonder cally related to properties in the world, stating that the body and its
about the relative importance of each memory system. I don’t interaction with the environment play a crucial role in cognition.
think there is any doubt about who – H. M. or M. S. – we would Glenberg’s theory has three shortcomings. The first is that it
rather be. The automatic, implicit memory system is surely impor- makes assumptions about the nature of internal representations. In
tant for something, but it is fundamentally aplysia memory: it is the the past, debates on this issue have been somewhat fruitless
system that’s been around for hundreds of millions of years, not the because it is impossible to determine their format definitively. The
system that is responsible for the rich complexity of human second is that his theory largely ignores neuropsychological re-
cognition. search and recent advances in cognitive neuroscience. The third is
Although Glenberg pines for a theory of semantic memory, he that there may be important differences between language
formulates his proposal in the language of syntax: The structure of comprehension and mental-models research on the one hand, and
a mental representation is lawfully related to the object repre- memory research on the other, that make an abstract approach to its
sented; a concept is created by the mesh between perception and study less useful than an approach that focuses more directly on the
memory; and ideas can be associated because the structures of brain and behavior. Each of these points will be addressed in turn.
their mental representations fit together. Why are these ideas Glenberg begins his paper by renouncing the idea that cognitive
expressed in the language of form when the goal is a language of representations can be reduced to meaningless symbols. He
content? My hypothesis is that people have no other way of argues instead that internal representations are analogically struc-
conceiving of, or at least talking about, meaning. Indeed, theories tured (or embodied). Although he develops his idea well, and his
of meaning seem to be of two types, syntactic or extensional. model is less abstract than those that treat representations as
Either they specify the form of and the formal operations on meaningless symbols, he still relies on assumptions about how we
semantic presentations (e.g., Katz 1972) or they focus on the mentally represent memories. This type of argument is reminis-
causal relations between agents and objects (e.g., Gibson 1979). cent of the imagery debate in the 1970s that lasted about a decade
Both of these problems are important ones to solve (e.g., Harnad and was never adequately resolved. Although important informa-
1990), but even together they are not sufficient as a theory of tion was acquired about other topics of interest in imagery (e.g.,
meaning. The first type of theory isn’t semantic at all, and the whether or not it was epiphenomenal and whether or not it was
second probably isn’t the right kind of semantic theory (e.g., Fodor functionally equivalent to perception), no answer was established

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 31

Commentary/Glenberg: What memory is for

about the nature of its representation. Anderson (1979) raised the Functional memory: A developmental
important point that one cannot decide the format of images perspective
because for every theory that proposes one type of representation
there can be another theory postulating another representation Katherine Nelson
that makes the same predictions. This is because one can always Developmental Psychology, City University of New York Graduate Center,
alter the process operating on the representation to fit the results. New York, NY 10036; knelson6 6 email.gc.cuny.edu
Anderson’s point is also relevant to Glenberg’s claim because
Glenberg also argues about the format of internal representations. Abstract: The functional theory of memory set out in Glenberg’s target
article accords with recent proposals in the developmental literature with
The nature of internal representations is less commonly de-
respect to event memory, conceptualization, and language acquisition
bated among memory theorists than among those who study from an embodied, experiential view. The theory, however, needs to be
reasoning and linguistics. Neuroimaging techniques and data from supplemented with a recognition of the sociocultural contribution to these
neuropsychological patients have helped advance our knowledge cognitive processes and emerging structures.
about the brain systems underlying memory, and these data have
been the basis of many memory debates. However, this sort of In answering the question “What is memory for?” Glenberg states
integrative approach combining efforts in neuroscience, biology, that “embodiment in terms of action patterns is just what is needed
and psychology is not a focus of research in the fields of reasoning to facilitate interaction with the environment and prediction”
and comprehension – perhaps because both of these cognitive (sect. 6.2). This fits very well with a developmental perspective.
skills initially require the cooperation of lower forms of cognition, For example, Nelson (1993) similarly suggested that the basic
such as memory, perception, and attention. Our knowledge is still function of memory is to support action in the present and to
incomplete about the underlying brain areas responsible for predict future states. Specifically: “Any system of learning and
carrying out these lower cognitive processes, and this information memory conserves information about environmental conditions.
is necessary before we can begin to study reasoning and compre- It enables the organism to undertake action to meet goals under
hension at the brain systems level. It is accordingly more useful to specific variable conditions . . . . The basic . . . functional memory
discuss these two processes at a more abstract level, like the one system directs action in the present and predicts future outcomes”
proposed by Glenberg. However, because we have some informa- (p. 372). This functional view of memory allows for individual
tion about the brain regions responsible for memory, but not for differences – and specifically for differences in perspective and
reasoning and comprehension, a single model may not account for memory between adult and child – while preserving a general
all three cognitive processes. process account (as does Glenberg’s theory).
The discussion of “what memory is for” focuses more heavily on Other parts of Glenberg’s functional theory of memory, concep-
the commonalities between memory and the two higher cognitive tualization and language comprehension, also fit well in a develop-
processes than on the brain systems model of memory. Although mental perspective. For example, the “functional core concept”
this approach has many strengths, any discussion of memory needs outlined by Nelson (1974) could easily be restated in Glenberg’s
to incorporate these findings. Glenberg dismisses much of this terms: “to a particular person, [child] the meaning of an object,
groundwork. For example, he claims that it is unnecessary to posit event, or sentence is what that person can do with the object,
a distinction between short- and long-term memory because event, or sentence” (sect. 1.3, para. 1). Furthermore, the state-
properties of both types of “systems” affect each other (e.g., ment “objects fall into the same (basic) category because they can
recency effects can be long term; errors in short-term memory can be used to accomplish the same interactive goal” (sect. 1.3, para. 3)
be semantic in nature). However, Baddeley (1990) pointed out maps neatly onto the idea of “slot-filler categories” as the origin of
that these results do not necessarily present a problem for the view the abstraction of functionally based superordinate categories
that there is more than one memory store, and it is possible for such as food, furniture, and animals (Nelson 1988; Lucariello et al.
tasks to reflect the operation of more than one store. 1992). These ideas are all consistent with Lakoff ’s construction of
Many would agree that the most crucial data in support of “embodied cognition” as well as with an experiential developmen-
multiple memory systems come from neuropsychological patients. tal psychology and an “experiential semantics” (Nelson 1996).
Not only are there patients who have impaired long-term memory Glenberg’s theory of memory might well have evolved from the
with intact short-term memory (and vice versa), but there is also a script theories of action, event memory, conceptualization, and
growing literature using brain imaging techniques indicating that language comprehension (Schank & Abelson 1977), although he
specific areas of the brain are designed to process specific types of does not refer to this source. The important functions of updating
information. While Glenberg’s theory does not deny this, it cer- and predicting were articulated in terms of scripts almost 20 years
tainly neither speaks to this issue nor treats it as having any ago, and subsequent research based in this framework also pro-
importance. Glenberg states (sect. 4.1, para. 1) that he means duced evidence of poor memory for repeated typical episodes and
explicitly to equate episodic and semantic memory. He states that good memory for unusual or atypical episodes of a recognizable
theorists in recent years have acknowledged that although memo- scripted event, findings Glenberg refers to from other more recent
ries differ in the underlying processes used to retrieve and store paradigms. There is much else in the script literature that is
memories, this does not necessarily imply that separate systems compatible with Glenberg’s theory and that may provide addi-
exist to support such memories. While this statement is in princi- tional support, as well as extend it in new directions and raise new
ple true, even major proponents of the processing view of memory issues. For example, any encounter by an individual is thought to
acknowledge its difficulty with accounting for neuropsychological instantiate a script appropriate to that situation; how does the
data (Roediger 1990). In light of the recent findings using both individual’s memory representation automatically recognize or
neuropsychological patients and neuroimaging techniques, the match an available script to a newly encountered situation? To my
brain system view should not be dismissed. knowledge, this problem remains unsolved and appears to be as
In sum, although it is important to consider the underlying problematic for Glenberg as for Schank and Abelson (1977).
format of memory representations and to compare memory to Although there are many more specific points worthy of com-
other cognitive processes, such endeavors should be undertaken ment, given space constraints I can consider only two of them. One
cautiously. Glenberg’s model invites active discussion among theo- is the case of infantile amnesia. I have addressed this in terms of
rists in these three subfields, but important differences in the level prior event knowledge – a kind of automatic memory – supple-
of complexity and current knowledge between subfields should be mented by a personal, social function of conscious recollection in
acknowledged. And finally, recent advances in memory research narrative form. Such recollection is fostered by exchange of
that have revealed an important interface between cognition and memory talk between adults and children, as much developmental
neuroscience should be considered and incorporated into in any research now attests, leading to the establishment of an auto-
theory of memory. biographical memory system (personal memory) beginning be-

32 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Commentary/Glenberg: What memory is for

tween three and four years (Nelson 1993). I agree that the a representation that is not the direct result of what is “held in
establishment of this memory function is related to the establish- view” (Olson 1993; Russell 1996).
ment of a distinctive sense of self and other. Glenberg’s relating of “Holding in mind” is very similar to clamping. Both allow the
recollection and prediction in terms of the effortful suppression of child to construct a representation of the world that exists inde-
projectible properties is most interesting in light of recent work pendently of the child’s current experience of the world (Russell
suggesting that the establishment of a continuing self with a 1996). Clamping occludes or immobilizes environmental input
history (personal memory) and a personal future takes place at (suppression is the flip side of the same coin). Although “holding in
about the same age (about 4 years), and that these developments mind” does dampen the overriding input of the environment,
are related to the kinds of talk that children and adults engage in unlike clamping, it allows for a dynamic interplay between concept
on these topics. It may be further noted, however, that auto- and reality. The child is able to think about one thing while
biographical memory involves two kinds of temporal order: order experiencing something different (Russell 1996). The develop-
within an episode or narrative and order between the events ment of the ability to hold in mind changes the child’s relationship
represented. Achieving the latter order no doubt requires consid- not only to the world but also to representations of the world.
erable management and effort. But the order within events – how Glenberg glosses over these important points in his treatment of
things go from first to last within a script, for example – appears to childhood amnesia.
be automatic in an embodied memory system for infants and Childhood amnesia may be described as the inability to recol-
young children as well as for adults (Bauer 1996; Nelson 1986). lect any autobiographical memories of the first three years. Glen-
Finally, the claim that the onset of autobiographical memory berg conflates this definition of childhood amnesia with the
(and the offset of infantile amnesia) is related to the discussion of general inability to form memories with episodic “traces,” auto-
past episodes with other people reflects a more general assump- biographical or not. He argues that apart from the ability to
tion that understanding human cognition, in its development as suppress environmental inputs, childhood amnesia is resolved
well as in its developed forms, requires seeing it as part of a world because children develop the ability to control what they are
of sociocultural activity as well as in its embodied state. Language thinking about, which Glenberg links to the acquisition of lan-
becomes a surrogate for direct experience only through the guage. Simply having the ability to control what you are thinking,
intersubjectivity and interactivity of the child and her caregivers. however, does not explain how the child develops a concept of self
The issue that Glenberg deliberately sidestepped – that of the that allows for autobiographical memory. Children must also be
matching of language forms to the individual conceptualizations able to manage representations about sources; that is, information
derived from embodied representations – can only be addressed about the world as well as their current attitude toward that
within a social communicative framework. As such, it is the key to information – it is “I who sees this” (Perner 1991).
many of the issues that he addresses, and specifically to those This fundamental representation of the self as the source of
involved in language comprehension. The symbol grounding perceptions and knowledge about the environment, which is
problem (Harnad 1990) cannot be solved entirely through embod- separate from the perception and formation of knowledge struc-
ied cognition but requires the recognition that words are socially tures about the environment, provides a basic, self-referential
shared symbols that somehow overcome the difficulties posed by duality. It is this duality – “that which is represented” versus “that
the existence of individual experiential worlds (see Nelson, 1996, which is doing the representing” – that forms the basis for a
for further discussion). concept of self. Once this is achieved, the child is able to concep-
tualize knowledge as a representational state of the mind, some-
thing that is “owned” by the self. (Perner 1991). In Glenberg’s
words, the child is beginning to think in “adult ways.”
Against suppression and clamping: It is unclear how Glenberg’s model could provide for such a
A commentary on Glenberg development through the constructs of clamping and suppression.
The general ability to compare two representations of the same
Jason T. Ramsay and Bruce Homer thing, one “held in mind,” the other “in view,” is a prerequisite for
Human Development and Applied Psychology, OISE/University of Toronto, higher-order operations such as formulating a proposition or
6 oise.utoronto.ca;
Toronto, Ontario, M5S-IV6 Canada; jramsay6 reflecting on one’s own thoughts. This does not mean that there is a
bhomer6 6 oise.utoronto.ca sudden point below which children are unable to recall events
from the past. Rather, as this basic skill of holding in mind develops
Abstract: The ability of Glenberg’s model to explain the development of and takes into account more extensive and subtle distinctions
complex symbolic abilities is questioned. Specifically, it is proposed that
the concepts of clamping and suppression fall short of providing an
between the two representations being held, richer representa-
explanation for higher symbolic processes such as autobiographical mem- tions can be formulated, some of which permit episodic memory.
ory and language comprehension. A related concept, “holding in mind” The development of “holding in mind” revolutionizes both what is
(Olson 1993), is proposed as an alternative. stored and the way children access information about the environ-
ment, thus changing their relationship to their memories. The
This commentary will examine two components of Glenberg’s child’s mode of representing events changes in accord with
theory; clamping (sect. 3.3) and suppression (sect. 3.4). Glenberg changes in the ability to relate what is “held in mind” to what is
ties clamping and suppression to the emergence of recollective “held in view.” Memories made before this change in representa-
experience in children (sects. 5.2.1–6.1). Clamping and suppres- tional ability do not have the same sort of information as memories
sion fall short of providing the basis for the symbolic abilities that made at a later stage of representational ability.
are crucial to the waning of early childhood forms of memory and Clamping and suppression also fall short of explaining language
the comprehension of language. Further development and mod- comprehension (sect. 5.2.3). Glenberg assumes that language
ification will be necessary for Glenberg’s model to explain the role comprehension involves decoding the spoken or written message
of a dynamic memory process in linguistic creatures. in a literal fashion; the meaning of an utterance is realized as
Recollective experience in children may be better interpreted potential action in the environment (sect. 6) and abstract language
from the perspective of the general ability to “hold in mind” two is understood in concrete terms (6.4). Good comprehenders are
differing representations (Olson 1993; Russell 1996). It is neither considered to be good at suppressing the environment. One
the ability to cancel out or suppress the contribution of the wonders how these “good comprehenders” would do at compre-
environment to conceptualization, nor the ability to clamp or hending figurative language. Glenberg’s proposal gives short shrift
suppress the conceptualization based on the environment that to language phenomena and avoids the crucial distinction between
bootstraps autobiographical memories and a sense of self, as “what-is-said” and “what-is-meant.” Clearly, “holding in mind”
Glenberg maintains. Rather, it is the basic ability to “hold in mind” allows for the comparison of representations of the utterer’s speech,

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 33

Commentary/Glenberg: What memory is for

gestures, or other metalinguistic information. “Holding in mind” berg’s notion of temporarily suppressing interaction with the
allows a dynamic interplay among many representations of diverse outside world so that simulation can play a dominant role.
information. Clamping may account for the occlusion of infor- Another possibly relevant piece of research is my own
mation but it seems less adequate to account for the carrying for- connectionist work in modeling the acquisition of spatial seman-
ward of information that may be usefully related to current percep- tics (Regier 1996). This model learns to map simple two-
tual experiences. This is where “holding in mind” might be relevant. dimensional percepts of objects moving relative to one another
It is unclear how clamping and suppression could deal with onto spatial terms in a number of languages. The aspect of the
metalinguistic concerns and allow for different levels of meaning. work that is germane to Glenberg’s proposal is its incorporation of
In conclusion, Glenberg succeeds in providing a theory that perceptual structure that constrains its operation, such that not
views memory in relation to the richness of human cognitive everything is learnable. This then is an example of constraint that
processes such as navigation. His thesis fails to explain highly Glenberg would probably view as “bodily” in nature, embedded in
complex symbolic processes such as the development of self and a connectionist model.
the comprehension of language. Thus, these two pieces of connectionist research highlight two
very general implemented principles of the Glenbergian project:
mental simulation and perceptual constraints. But there is nothing
What is modeling for? particularly connectionist about either of these two principles.
Their implementation in connectionist fashion may be reassuring
Terry Regier in that it indicates that such implementation is well within the state
Department of Psychology, University of Chicago, Chicago, IL 60637; of the art, but both mental simulation and perceptual constraints
6 uchicago.edu
regier6 could easily be accommodated within a symbolic system. In fact,
the idea of a perceptual symbol system (Barsalou & Prinz 1996)
Abstract: What would Glenberg’s attractive ideas look like when compu- exemplifies a fundamentally symbolic approach to cognition that
tationally fleshed out? I suggest that the most helpful next step in
has perceptually constrained mental simulation at its heart. De-
formalizing them is neither a connectionist nor a symbolic implementation
(either is possible), but rather an implementation-general analysis of the spite the inspiration Glenberg draws from connectionism, and
task in terms of the informational content required. despite the existence of connectionist systems that embody some
of the principles he espouses, it is those principles themselves,
In an engaging and thought-provoking target article, Glenberg rather than their implementation in this or that system, that
starts by asking what memory is for; in answering this question, he ultimately form the core of the effort.
also suggests what memory might be. The proposed purpose of What, then, is modeling for? If after all the trouble of imple-
memory is clear: It supports perception and action in the world. mentation we are left with the conclusion that it is the general
The general idea behind the posited mechanism is also clear, but principles we began with that count, why bother? The answer is
the details are less so. This is a perfectly reasonable state of affairs that the process of modeling can lead to a clarification of these
for the time being, as Glenberg’s intention was presumably to general principles. Prior to jumping to implementation, a careful
convey the outlines of a simulation-based approach to conceptual- examination at the computational and algorithmic levels (Marr
ization and memory and to indicate a number of radical rein- 1982) would determine the content of the data structures required
terpretations that this approach suggests. He has succeeded in and the nature of the operations running over them. And this,
this, in my estimation. The central concept, of internal simulation rather than the implementation itself, could help in making
aligning with occasionally suppressible perception, is an appealing Glenberg’s attractive but still largely intuitively grasped theoreti-
one. Still, once the basic idea has been absorbed, a question cal notions more concrete. That would definitely be worthwhile.
concerning the concretization of this general notion presents
itself: What sorts of representations might underlie such simulated
embodied conceptualizations, and what operations might manipu-
late them? In a word, what do these ideas look like when fleshed What memory is for action: The gap
out computationally? between percepts and concepts
Glenberg points out that some of the ideas he has adopted can
be traced to connectionism; a natural question then is whether the Yves Rossetti and Emmanuel Procyk
approach as a whole is appropriately modeled through connec- Vision et Motricite, INSERM Unité 94, 69500 Bron, France;
tionism. I would first like to discuss briefly two pieces of connec- 6 lyon151.inserm.fr
rossetti6
tionist research that may be of relevance for the formalization of
these ideas, and would then like to suggest that the aspects of these Abstract: The originality of Glenberg’s theoretical account lies in the
systems that make them appropriate are not intimately bound up claim that memory works in the service of physical interaction with the
three-dimensional world. Little consideration is given, however, to the role
with their connectionist character. A symbolic system could easily of memory in action. We present and discuss data on spatial memory for
exhibit the same properties. So to peek at our punch-line in action. These empirical data constitute the first step of reasoning about the
advance: There are certainly tools available that can aid in the link between memory and action, and allow several aspects of Glenberg’s
formalization of mesh, bodily constraints, and trajectories, but theory to be tested.
those tools can reside on either side of the great connection-
ist/symbolic divide. I believe that the relevant issue for these The theoretical first section of Glenberg’s target article proposes
purposes is not so much the implementational substrate of the that there is a close link between conceptualization and action, yet
mechanism, but rather the informational content of the data relatively little experimental data from the literature about move-
structures involved. ment science are considered. Considering Glenberg’s paper as a
Bartlett Mel’s work on robot control (Mel 1988; 1989) exhibits a whole, one feels that between the idea that concepts are embodied
feature that is quite similar to a central aspect of Glenberg’s in action and their role in language comprehension a step is
program. Mel’s work concerns a connectionist system that learns missing. As pointed out by Glenberg (sect. 7.4), common “labora-
to guide a robot arm to reach objects. The critical point is that once tory paradigms for studying memory may well be missing the
the model has been trained, it can activate a mental image of the mark. Many of these paradigms use random lists of words as the
arm in its environment and can use this mental simulation to object of memory.” This is particularly true if “memory is embod-
evaluate contemplated arm movements without actually moving ied and designed for negotiating a three-dimensional environ-
the arm in the external world. Once the mental simulation is done ment” (sect. 7.4, para. 1) and if “the world is conceptualized (in
and a course of action has been decided upon on, the model can part) as patterns of possible bodily interactions, that is, how we
act. This internal simulation corresponds fairly closely to Glen- can move our hands and fingers, our legs and bodies, our eyes and

34 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Commentary/Glenberg: What memory is for

ears, to deal with the world that presents itself ” (sect. 1.3, para. 1).
The first step for testing Glenberg’s theory should thus be to
investigate what memory is for action. The aim of this commentary
is to make use of examples of interaction between perception and
memory in the guidance of simple actions that may be useful in
bridging theoretical considerations about action and discussions
about semantic memory or language. We will thus show how
different levels of object representation may interact during the
production of simple motor responses to spatial stimuli.
A problem with Glenberg’s theory of meshed patterns of action
is that it suggests that there may be a single way in which memory
and perception interact to guide a given action in a given context.
Once the unified mesh has been built from the projectable and
nonprojectable properties of the spatial layout, this “unified”
representation is used to guide action. In addition, this view
benefits from an automatic contribution of memory. A lot of data,
however, indicate that motor and symbolic representations of a
given stimulus can be dissociated (Goodale & Milner 1992; 1995;
Jeannerod & Rossetti 1993). Patients with visual agnosia or blind-
sight provide an interesting example, being able to reach and grasp
a simple visual object without identifying it (Goodale et al. 1991;
Perenin & Rossetti 1996); they are counterevidence to the idea
that the “meaning of an object or a situation is a particular pattern
of possible action” (sect. 2.2, para. 3). In addition, such patients are
unable to produce a similar action when the target object has to be
memorized (see Milner & Goodale 1995; Rossetti 1997). They
seem able to integrate the metric properties of an object with
action on it, but they remain unable to memorize this percept to
delay the action or to combine the percept with memorized
representations that could allow conscious recognition of the
object. In contrast, optic ataxia patients cannot act on objects they
can describe and recognize normally (Perenin & Vighetto 1988).
These cases do not agree with the statement that “the meaningful,
action-oriented component of conceptualization is not abstract
and amodal” (sect. 1.3, para. 2). In the same way, prefrontal lesions
in the monkey produce a dissociation between visually guided and
memory guided saccade or reaching to the same target (Funahashi
et al. 1993; Procyk et al. 1996). In addition, the neuroanatomy of
the visual areas and their projections suggests that there is an early
segregation between a dorsal stream specialized for vision for
action and a ventral stream specialized for visual object recogni-
tion (see Goodale & Milner 1992; Jeannerod & Rossetti 1993;
Milner & Goodale 1993). These data may be regarded as elemen-
tary examples of segregation between percepts and concepts.
Action implies space perception and representation, but does it
always require conceptualization? The above examples show that Figure 1 (Rossetti & Procyk). Confidence ellipses (95%) of
action can be accurate even when no conceptualization is available pointings toward the same absolute target position in space under
to the patient. There must be a “morphological” level of sensory three different conditions. Target positions were encoded by a
integration that lies between the physical and the symbolic levels brief (300 msec) passive positioning of the left index finger on a
(see Petitot 1990). This level would code metric but not symbolic sagittal panel, and pointing was done with the right hand either 0
properties of spatial layouts. sec (dotted ellipses) or 8 sec (bold ellipses) after target presenta-
In criticizing the distinction between short-term and long-term tion, on the other side of the panel. In the single target condition,
memory, Glenberg argues that different “skills,” rather than differ- Ss had to point to a target that was 30 cm from the starting point
ent “modules,” are involved in producing separate sets of data (arm resting on the table) and 118 anterior to the vertical axis
(sect. 4.3). The motor control literature shows that this is not crossing the starting position. In the two other conditions, Ss had
always the case. According to Glenberg, there would be no a priori to point toward six possible target locations, including the location
reason to believe that a given action would be based on different used in the single target session. In the arc condition, targets were
representations, depending upon the length of a short delay after positioned on an arc centered on the starting position (ø 30 cm), as
stimulus presentation. Several such examples are, however, avail- shown by the broken line. In the line condition, targets were
able in the literature (Bridgeman 1991; Gentilucci et al. 1996; positioned on a line crossing the starting point (118), as shown by
Goodale & Stevens 1992; Goodale et al. 1994; Rossetti & Regier the broken line. The ellipse orientation observed after the 8 sec
1995; Rossetti et al. 1996; White et al. 1994; see review in Rossetti, delay in the arc and line conditions differed significantly. In both
in press). One of the most striking examples made use of the cases, ellipse orientation was aligned with the target array, that is,
Müller-Lyer illusion of length (Gentilucci et al. 1996). Although with the spatial layout used in each particular condition. These
the perceptual effect of this illusion is very vivid, it affects point- specific orientations of the pointing scatter were observed neither
ings from one end of the line to the other only very minimally (see when the delay was 0 sec nor when the spatial layout was reduced
also Aglioti et al. 1995). The effect of the illusion on movement to a single target. Arrows indicate movement direction (individual
length increases, however, when the layout has to be memorized results from a representative subject).
during a longer delay. Another example is illustrated here in
Figure 1. Pointing movements produced either zero seconds or a

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 35

Commentary/Glenberg: What memory is for

few seconds after target presentation should lead to similar perfor-

mance, although longer delay should cause greater variability.
Qualitative differences obtained in such a condition provide a
strong argument for two possible modes of sensory processing:
The spatial layout morphology is integrated in the action only for
the longer memory delay (Rossetti & Régnier 1995, Rossetti et al.
1996). This confirms that “action-oriented meaning can vary
greatly with context” (sect. 1.3, para. 6) and, more specifically, that Figure 1 (Schwartz et al.). Two dimensions of attentional func-
“Patterns of action derived from the projectable properties of the tioning.
environment are combined . . . with patterns of interaction based
on memory” (sect. 2.2, para. 2). If our results confirm that one’s present context with the remembrance of previous embod-
“Patterns of action based on the environment (projectable proper- ied experiences provides the meaningful conceptualizations
ties of the environment) are automatically, that is, without inten- needed to guide effective perception and action. While this
tion, meshed with patterns based on previous experience” (sect. 2, functional approach impressively draws together a wide range of
para. 1), they nevertheless indicate that this may be so only after a empirical findings, one wonders how an embodied mind in its
given delay. natural habitat happens to clamp certain projectable properties
If “embodied mental models reflect a structured space (that is, and not others. How does it select from memory those experiences
. . . structured by possible actions)” (sect. 6.3, para. 5), then it is that bear most usefully upon the current situation? Without the
difficult to explain why the mental model of a target position to be capacity to attend selectively to some external phenomena or
reached should vary according to whether or not the position has memories and to ignore others, a person’s thought and behavior
to be memorized. A possible way to reconcile Glenberg’s theory are confused, disorganized, and largely ineffective, as the experi-
with empirical data about the role of memory in action is to ence of frontal lobe patients, schizophrenics, and others attests.
consider that the “meshing” of percepts and concepts takes time. Glenberg addresses one aspect of the problem of selection
In our experimental conditions (Fig. 1), it is not true that “spatial when he notes that the capacity for prediction sets up a conflict
layout is a projectable property” (sect. 2.1, para. 2), because all between the internal and external contributions to conceptualiza-
targets are never available simultaneously. It may be that our tion; he suggests that cognitive suppression developed as a solu-
shorter delay did not provide enough time for “embodied memo- tion to this problem. We agree that suppression serves this
ries” to interact with “the current environment” (sect. 5.1). A function, but we think it does much more besides. In what follows
possible way to stress the elaboration of a conceptualized repre- we show how considering the role of memory and suppression
sentation of a spatial layout is to require the subject to produce a within a broader attentional framework enables us to incorporate
verbal estimate of the current position to be pointed to. Under this psychological phenomena that do not mesh easily into Glenberg’s
condition, the effect of the spatial layout already clearly appears conceptualization.
for the 0 sec delay (methods similar to those in Figure 1). Thus, the Following Kaplan (Kaplan 1995; Kaplan & Kaplan 1989) and
delay and the forced verbalization produced similar qualitative Cimprich (1992), we derive our attentional framework from the
effects on the pointing, showing that the very same action can taxonomy William James proposed in his Principles of Psychology
depend or fail to depend on contextualized representation. This (1890/1981). James noted that attention can be directed either
result also suggests that “automatic” and “effortful” (sect. 5) “outward” toward the phenomenal world (“sensorial” in his termi-
components of memory contributing to conceptualization may be nology) or directed “inward” toward memories, abstract ideas, and
responsible for similar effects on action. so forth. We refer to this as the internal–external dimension of
Delayed-response tasks do not involve “changing conceptual- attentional functioning. James also noted that attention can be
ization (changing because the stimulation changes in response to summoned (and held) involuntarily or directed willfully. We refer
action)” (sect. 4.3, para. 3). They disagree with Glenberg’s theory, to this as the easy–effortful dimension of functioning, because
however, in that they show changes in the representation underly- willfully directing one’s attention seems always to involve some
ing the action. To be compatible with these experimental data, degree of effort. Treating these two dimensions as orthogonal
Glenberg must integrate the idea that the meshing of percept with yields the matrix shown in Figure 1.
concept requires some time to be accomplished, that is, that there The top left cell encompasses those instances when the mind
may be a time gap between percept and concept. attends effortlessly to external phenomena that intrinsically hold
one’s interest and dominate awareness, such as mountain vistas,
wild animals, dramatic (or dramatized) human social interaction,
and so on. Of the four cells in the matrix, the easy–external form of
Suppression, attention, and effort: attention seems most closely related to Glenberg’s notion of
A proposed enhancement for clamping projectable properties of the environment, inasmuch as
a promising theory the focus is directed outward and the conceptualization does not
involve effort. We should not, however, equate absence of effort
David A. Schwartz,a J. Eric Ivancich,b with absence of suppression, for some forms of suppression can
and Stephen Kaplana,b feel quite effortless. A reader absorbed in a mystery novel, for
aDepartment of Psychology, The University of Michigan, Ann Arbor, MI example, or a hunter stalking his prey, or a person engaged in
48109; david.a.schwartz66 umich.edu; bDepartment of Electrical courting or mating behavior all exhibit reduced sensitivity and/or
Engineering and Computer Science, The University of Michigan, Ann Arbor, responsiveness to stimuli that might otherwise attract their notice.
MI 48109; ivancich66 eecs.umich.edu; stephen.kaplan6 6 umich.edu In such instances, we suggest, the psychological content and/or
process (mystery, hunting, mating) resonates so strongly with
Abstract: Although Glenberg’s theory benefits from the incorporation of a evolved human inclinations that it engages mechanisms of sup-
suppression concept, a more differentiated view of suppression would be
even more effective. We propose such a concept (based on the attention
pression (and thus captures attention) without, or even in spite of,
framework first developed by William James in the late nineteenth one’s conscious intention.
century), showing how it accounts for phenomena that Glenberg describes The bottom right cell, by contrast, corresponds most closely to
and also for phenomena that he ignores. Glenberg’s notion of effortful reflection in the service of predic-
tion or recollection. Here suppression of the external contribution
Memory, Glenberg writes, is for imbuing the “obvious” properties to conceptualization frees thought from the here and now to
of an environment with personal significance, such that an object- facilitate goal-directed mental transformations such as solving
ive world acquires subjective import. The mental meshing of problems. Glenberg speculates that suppression of the external

36 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Commentary/Glenberg: What memory is for

contribution to awareness is experienced as effortful and aversive There’s a lot to like about Glenberg’s proposal. Some concepts do
because ignoring the environment is dangerous. We agree that seem to have “functional-spatial” content: functional fixedness is a
purposeful reflection often does feel effortful and that ignoring robust phenomenon; home is where you hang your hat; and, at
the environment can indeed be dangerous. However, we find least for three-year-olds, a hole is to dig (Krauss 1952). Who
Glenberg’s causal linkage not altogether compelling because, as knows? Maybe there are organisms whose Umwelt is, indeed,
we point out below, reflection is not always effortful and there are entirely furnished by things-of-action. Moreover, it seems that
effortful forms of suppression that actually serve to enhance rather thinking sometimes recruits mechanisms that subserve perception
than to diminish the external contribution to conceptualization. and motor activity, for example, imaging improves motor perfor-
The top right cell of the matrix, for example, encompasses those mance (Finke 1986) and loss of meaning associated with word
instances, such as daydreaming and reverie, when the internal repetition can be staved off by engaging in semantically appropri-
contributions to conceptualization so dominate awareness that a ate gestures (Werner & Kaplan 1963). All in all, it seems plausible
person can seem almost oblivious to external circumstances. (We that (1) bodily interactions with the physical world constitute the
assume, however, that even in such circumstances a person con- content of some concepts, and (2) cognition sometimes recruits
tinues to monitor the external state of affairs preattentively.) perceptual/motoric vehicles.
Clearly at such times the projectable properties of the environ- What then, is not to like? Well, for one thing, Glenberg’s
ment are in some sense suppressed, but this suppression involves proposal that “embodied representations” have the content they
no feeling of effort. Does the absence of effort suggest that do “by virtue of being lawfully and analogically related to proper-
neglecting the environment while daydreaming is somehow less ties of the world . . . [as] transduced by perceptual-action systems”
dangerous than ignoring it when engaged in purposeful thought? (sect. 1.3, para. 2) arrives with not so much as a hint that theories of
Moreover, just as one might at times find it necessary to content that lean on nomological liaisons or similarity relations
suppress awareness of the environment in order to liberate have their own nontrivial problems. When push comes to shove,
thought from the here and now, so might one effortfully suppress Glenberg turns, unannounced, to a different theory. Better, proba-
reflection in order better to attend to what’s going on in the bly, to leave heavy duty semantics to people who do it full time,
environment. When “what’s going on” happens to be a particularly most of whom would flinch at hearing standard philosophical
boring lecture or journal article, or an uneventful vigil for a sentry doctrine on sentence meaning presented as “[it is] whatever allows
keeping watch, the amount of effort involved may reach truly one to determine if [the sentence] applies to particular situations”
heroic proportions. The feeling of effort in such instances, how- (sect. 1.2, para. 5); verificationism is as unpopular as classical
ever, cannot signal the danger of ignoring the environment, description theory these days.
because the effort is associated with attending to the environment. It is also the case that even good evidence that some concepts
Why, then, does suppression, whether of the internal or external have functional-spatial content (or that some cognitive activities
contribution to conceptualization, often feel effortful? We suggest are supported or accompanied by motor activation) does not make
that suppression is effortful because, like swimming upstream, it it equally plausible that this is true of all conceptualization.
involves willfully opposing the inertial flow of undirected thought, Glenberg admits that “experiences of music, taste, and emotions
and overcoming (neurophysiological) inertia, of whatever kind, . . . have aspects that do not fit well into a spatial-functional
requires (physiological) work. For example, consider trying to straitjacket,” but thinks that “the same sort of analysis” may turn
inhibit a response that you have already initiated, as in the stop- out to be applicable (sect. 7.2, para. 1). His idea of what constitutes
signal paradigm (Logan & Cowan 1984). This type of motor success is worth a closer look. How, for example, could an individ-
suppression generally feels effortful, though it seems unrelated ual conceptualize a sunset? Glenberg’s answer is that “a beautiful
either to prediction or to recollection. We suggest, rather, that all sunset is a context that combines with objects and memories to
effortful mental activity consists in deliberately intervening in or suggest actions consistent with warmth, relaxation, and a good
suppressing one’s ongoing thought or behavior for the sake of beer” (sect. 1.3, para. 6). That is, a sunset can signal an individual,
some internally represented goal. Consequently, in the absence of S, that various activities – making a fire, going to the “fridge,” etc. –
any represented goal or problem to solve, thought is largely will (under the circumstances) have as their consequences
exogenous (driven by external cues), habitual (following default warmth, relaxation, and a good beer. Put a bit more formally,
associative structure), and effortless. Glenberg’s solution is that a sunset can be conceptualized because
These differences notwithstanding, we applaud Glenberg for it can serve as a discriminative stimulus. How so? Well, as any
having proposed an ambitious and synthetic theory of cognitive operant theorist knows, discriminative stimuli acquire reinforcing
functioning. By placing the mind in the body and the body in the properties along with their ability to “control” some bit of behavior.
world, he offers an engaging perspective on the hows and whys of Were Glenberg’s story to be that S’s conceptualization of X is a
human memory. While we question a number of his specific matter of S having response tendencies with respect to X, then, if X
assertions, we enthusiastically endorse the spirit of his proposal. To comes to serve as a discriminative stimulus for S, X has ipso facto
those of us for whom symbolic cognitive models consisting of been conceptualized by S. This seems to be what he has in mind:
disembodied formalisms and propositional logic seem as dry as the “Our understanding [conceptualization] of pleasurable experi-
desert sands, Glenberg’s functional approach stands as a refresh- ences is in part action-toward those experiences, whereas our
ing oasis in the conceptual landscape. understanding [conceptualization] of aversive experiences is in
part action-away” (sect. 7.2, para. 2). This alternative characteriza-
tion is unsatisfactory on at least two counts. First, identifying
meanings with responses (or dispositions to respond) notoriously
Conceptualizing a sunset ± using a sunset fails to yield appropriate contents (Alston 1964). The first snowfall
as a discriminative stimulus of the season may also “suggest actions consistent with warmth,
relaxation, and a good beer,” but one’s conceptualization of the
Carol Slater first snowfall is not the same as one’s conceptualization of a
Department of Psychology, Alma College, Alma, MI 48801; beautiful sunset. Second, this account still does not allow concep-
6 alma.edu
cslater6 tualization of items that are neither available for direct physical
interaction nor serviceable as discriminative stimuli.
Abstract: Glenberg offers two different accounts of embodied conceptu-
alization. The first fails in cases where no direct bodily interaction is Informal examples are not the only hard cases. An important
possible. The second fails in cases where the object in question cannot account of object recognition construes it as a parsing of visual
serve as a discriminative stimulus; moreover, it yields inappropriate displays into standard 3-D forms (geons) related to each other in a
content even in cases where it can be applied. Glenberg’s disregard for the limited number of ways (Biederman 1987; 1988). How objects can
conceptual agenda set by the social world is also disquieting. be or have been handled plays no explanatory role. Even in

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 37

Commentary/Glenberg: What memory is for

developmental psychology, the homeland of sensorimotor intel- believe that Glenberg’s idea of “mesh” has two different technical
ligence, it has become increasingly clear that infants’ and young senses that address these two philosophical concerns.
children’s equivalence classes both antedate and crosscut sim- The first philosophical problem concerns how an embodied
ilarities based on overt manipulation (Mandler 1990). Shared theory can represent productivity. Humans produce an indefi-
impalpability, for example, does not incline 4-year-olds to lump nitely large number of expressions or thoughts from a finite
together smoke, shadows, and beeps, on the one hand, with number of components. A theory of representation must be able to
thought contents, on the other (Wellman 1990). explain how primitive representations combine to form larger
Finally, the world in which Glenberg’s embodied cognizers are representational structures. Psychologists and philosophers have
situated is curiously impoverished. One would never suspect that embraced propositional symbols because they can be combined
everyday mind reading, coordination of action, detection of cheat- easily by the syntactical rules of logical languages, such as predi-
ing, identification of kinship, or negotiation of discourse (to name cate calculus. It is more difficult, however, to imagine how analogi-
only a few) set pressing conceptual agendas (Tooby & Cosmides cal or embodied symbols could be combined to produce new
1992). “Social cognition” is characterized with startling brevity as representational structures. What would be the syntax of these
requiring nothing more distinctive than the ability to see people embodied representations?
“not just as objects that we can affect, but as beings who can affect Barsalou (in preparation) and Barsalou and Prinz (in press)
us in turn” (sect. 7.2, para. 5). Nor would one suspect from illustrate how a perceptual theory of knowledge could produce
Glenberg’s account that representation of properties such as being productivity. Glenberg’s construct of “mesh” adds to our under-
a smile, a face, a speech sound, or the sound [ba] are no more standing of this important process (see Prinz & Barsalou, in press,
indebted to individual experience than are properties “specified for a similar proposal). Following Glenberg’s “deflated ball” exam-
by information available in the light” (sect. 2.1, para. 2): Even ple (sect. 3.1.3), combining embodied representations involves
explicitly Gibsonian approaches to social cognition are ignored meshing patterns of action for deflated and ball in a manner that
(McArthur & Baron 1983; Zebrowitz 1990). satisfies constraints on how their referents occur in the world.
Glenberg recommends a pragmatic approach to meaning and These constraints are spatial and functional rather than associative
memory. A pragmatism worthy of the name, however, must be or probabilistic. Thus, the syntax for how new thoughts can be
scrupulous in characterizing the nature of the task (Marr 1982). produced by combining existing ones is dictated by the constraints
on how objects can be manipulated or acted upon in the world.
This is an important claim about productivity because it pro-
vides a means of grounding symbols in meaningful content (Har-
Productivity and propositional construal as nad 1990). Although propositional symbols can be easily com-
the meshing of embodied representations bined through rules of syntax, these combinations can result in odd
or meaningless combinations because the symbols that are being
Karen O. Solomon and Lawrence W. Barsalou manipulated are abstract. For example, under a propositional
Department of Psychology, University of Chicago, Chicago, IL 60637;
syntax, finding a meaningful interpretation for tilted ball should be
6 ccp.uchicago.edu; l-barsalou6
klo16 6 uchicago.edu as easy as finding a meaningful interpretation for tilted candle,
because the same syntactic rule can be used for both combina-
Abstract: Contrary to prevailing views, productivity and propositional tions. But clearly tilted candle is easier to interpret than tilted ball.
construal are not problematic for perceptual views of representation. Because a ball cannot be tilted in the real world, an embodied
Glenberg’s embodied representations contribute to our understanding of theory predicts that tilted ball is more difficult to interpret than
how these two important processes might be implemented perceptually. tilted candle. As this example shows, Glenberg’s embodied theory
of meshing representations suggests testable constraints for how
For at least the last three decades, the dominant view in cognitive concepts are combined productively. According to his theory, the
psychology has been that human knowledge is represented by constraints on the syntax for combining concepts are provided by
systems of amodal propositional symbols. In his target article, the constraints on how objects can be manipulated in the world.
Glenberg argues that human knowledge is represented instead In our laboratory, we have found evidence that people use the
by analogical, perceptually based, “embodied” representations. constraints for how objects behave in world as a syntax for
Glenberg argues that psychologists’ theories of memory have combining concepts. In studies of conceptual combination, Wu
focused on propositional symbols because these symbols reflect (1995) compared feature production for noun phrases in which
the stimuli used in classic memory paradigms, namely, lists of the same modifier (e.g., half ) either revealed the insides of an
words. Glenberg argues that this focus on propositional symbols object (e.g., half watermelon) or left the insides of the object
has caused theorists to ignore the more important and more occluded (e.g., half smile). Wu found that subjects produced more
prevalent “embodied” representations that he discusses in his internal features (e.g., red, seeds) for noun phrases that revealed
paper. While we agree with Glenberg that researchers’ tools and the insides of the objects than for noun phrases that left the insides
paradigms influence their theories (Gigerenzer 1991), we believe occluded. These results suggest that subjects mentally simulate
that other factors, such as philosophical arguments, have also how the two concepts combine in the world and use this simula-
contributed to the widespread acceptance of propositional repre- tion to produce features. These results are in line with Glenberg’s
sentations in psychology, together with the rejection of embodied argument that constraints in the world dictate the productive
theories. In this commentary, we outline and discuss two tradi- combinations that underlie mesh.
tional philosophical arguments against embodied theories, and we A second philosophical problem for embodied theories has
discuss how Glenberg’s embodied theory could handle these kinds been that perceived situations can be construed in many different
of arguments. ways, but an image of a situation only records the situation rather
Embodied theories of representation are not new. Many early than construing it in one particular manner. For example, imagine
philosophers, such as the British empiricists, argued that people’s that you are rearranging your dining room furniture and have the
thoughts and memories are represented by images derived from thought that “the lamp is above the table.” This is clearly different
perceptual experience. These theories and others like them, from the thought “the table is below the lamp.” In the first case the
however, have been dismissed in modern philosophical circles lamp is the focus of thought, whereas in the second case the table is
because of strong objections made by twentieth century philoso- the focus. An analogical representation, such as a picture of the
phers and psychologists. Two important problems for traditional situation, however, cannot distinguish these two possible con-
embodied views are their inability to represent productivity and struals. A theory of embodied representations must be able to
propositional construal (Barsalou, in preparation; Barsalou & represent a particular construal to be a legitimate theory of how
Prinz, in press). Although it is not explicitly stated in his paper, we memories and thoughts are represented (Pylyshyn 1973).

38 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Commentary/Glenberg: What memory is for

We believe that a second sense of “mesh” can handle this There is nothing wrong with the fact that Glenberg is in such
philosophical problem. This sense of mesh corresponds to Glen- good company. The problem is with empirical testing. When he is
berg’s “path home” example (sect. 3.1.1). In this sense, patterns of specific, I am not sure that existing results support his views. Here
action from memory are brought to bear on “clamped properties are several points. First, if the embodied representations (like
of the environment” to conceptualize a situation. Thus a path, Piaget’s earlier sensorimotor schemes) play such a crucial role in
which is a pattern of the environment, can become a path home memory and in conceptualization, why does the cognitive devel-
when it is meshed with memories of following that path home on opment of children with severe (inborn) abnormalities of motor
previous occasions. The path home is a conceptualization and also functions not demonstrate any striking deviation from its normal
a propositional construal of the perceived setting. course (Gouin-Decarie 1969)? Second, if implicit memory is
To further specify how this meshing takes place, we believe that primarily in service of the integration of projectable properties of
Glenberg needs to posit embodied representations that are types, objects and bodily actions, it is surprising that perceptual priming
which can be filled with tokens from the environment. These types (as opposed to explicit recognition) does not depend on the prima
are perceptually based, frame-like structures whose function is to facie “projectable” property of objects, that is, their size (Cooper
construct perceptual simulations of entities and events in their et al. 1992). Third, the presumed suppression of outer activities in
absence (Barsalou, in preparation; Barsalou & Prinz, in press). In the conceptually driven processing mode contradicts reports that
our example, a lamp above a table, the embodied representation such a shift of attention to internal information processing can lead
for above (x,y) is a schematic image having a top and a bottom to more overt behavior, for example, to a disinhibition of opto-
region in which attention is focused on the top region (Langacker kinetic nystagmus (cf. Gippenreiter & Romanov 1972). Fourth, in
1986; Talmy 1983). Objects from the environment (e.g., lamp, view of recent evidence on the dissociation of visual processing
table) can then productively instantiate these particular slots to pathways connected with motor actions and leading to phenome-
form a simulation of what the situation might look like in the nal perception (Bridgeman et al. 1991; Milner & Goodale 1995), I
world. The embodied representation for below (x,y), on the other would expect a similar dissociation in memory performance where
hand, also has a top and a bottom region, but attention is focused Glenberg, in contrast, sees integration.
on the bottom region. Thus, particular propositional construals While some ad hoc explanations may be possible, an obvious
can be created by meshing types from memory with tokens from problem arises with the use of the terms “projectable,” “stimulus-
the perceived environment. To return to Glenberg’s example, “the driven,” and so on. The cognitive approach was born in Koehler’s
path home” is a schematic image of a path that focuses on an criticism of perceptual theories based on projectable properties
endpoint, home. The clamped properties of the environment (e.g., and Chomsky’s analysis of words like “stimulus” and “response.”
a particular path) are meshed with this schematic image to Should we indeed start from scratch?
construct one particular construal of the scene. The views recommended by Glenberg are too undifferentiated,
Contrary to many prevailing views, productivity and proposi- being based on the common dichotomy between data-driven and
tional construal are not intractable for perceptual views of repre- conceptually driven mechanisms. There are more than two levels
sentation. We believe that Glenberg’s two senses of “meshing” in the control of human activity (see Velichkovsky 1990). Thus,
embodied representations contribute to our understanding of how several different levels can be at work in the domain of sensorimo-
these two important processes might be implemented percep- tor mechanisms (Bernstein 1996). Not all these mechanisms are
tually. On the one hand, mesh addresses productivity by assuming data-driven in the straightforward sense, and suppression of some
that combining embodied representations involves meshing two of them could well lead to activation of others.
or more patterns of action from memory so that they satisfy the Hardly more elaborated are Glenberg’s ideas of processing
constraints on how objects are manipulated in the world. On the “above” sensorimotor coordinations. For him this is a relatively
other hand, mesh addresses propositional construal by assuming homogeneous domain where memory takes over the control of
that patterns of action from memory can be brought to bear on conceptualizations despite the disturbing influence of the envi-
perception of the environment to construe situations. ronment. Though this may often occur, we are able to reexamine
our past and we are also able to change our learned attitudes. In
other words, it is an oversimplification to think that the function of
memory is either to “mesh” with environmental contingencies or
The “mesh” approach to human memory: to override their influence. In human cognitive activity, memory
has to be permanently articulated and “overridden,” in service of
How much of cognitive psychology both creative imagination and the reality principle. This selective
has to be thrown away? “suppression of memory” is perhaps one of the main functions of
prefrontal cortex (Deacon 1996). Examples of this “suppression”
Boris M. Velichkovsky include self-referential cognition, linguistic pragmatics, the un-
Unit of Applied Cognitive Research, Dresden University of Technology, derstanding of poetry and intellectual strategies such as reductio
D-01062 Dresden, Germany; velich6 6 psy1.psych.tu-dresden.de;
ad absurdum, which are all connected with some form of “as-if ”
physik-phy-tu-dresden/psycho/bmv-e.html
mental experimentation. The distinction between these groups of
higher-order mechanisms is well supported by the neuropsycho-
Abstract: While sharing the author’s interest in the development of an
action-based framework for memory research, I think the present version
logical dissociation of self-referential experience and semantic
is neither new nor particularly productive. More differentiation is needed memory (one of the cornerstones of memory systems theory – see
to describe memory functioning in a variety of domains and on the many Schacter & Tulving 1994) and by other recent theories of func-
levels of activity regulation. Above all, Glenberg’s proposals seem to tional organization in cognition (Challis et al. 1996).
contradict empirical data.
ACKNOWLEDGMENT
The target article evokes a strong déjà-vu impression. Dissatisfac- The preparation of this commentary was facilitated by a grant from the
tion with the Ebbinghaus tradition is as old as the tradition itself. Japan Society for the Promotion of Science and DAAD (Bonn, Bad
More than 100 years ago Wundt opposed this type of experimenta- Godesberg). I wish to thank Bradford Challis and Nobuo Ohta for a vivid
tion and theoretical synthesis. Later on, Janet, Bartlett, Lewin, the discussion of the issues.
cultural-historical school of Vygotsky, Piaget, and the Gibsonians
elaborated on the theme. In cognitive psychology, a constructive
critical contribution was provided by the levels-of-processing
approach of Craik and Lockhart (1972) and by such (reformed)
classicists as Neisser (1976) and Norman (1993).

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 39

Commentary/Glenberg: What memory is for

Towards a dynamic connectionist whose iteration the image might be regenerated. By extension,
model of memory memory for nonprojectable, action-oriented properties might be
economically coded in a similar way. In both cases, what is stored
Douglas Vickersa and Michael D. Leeb would be the parameters controlling some dynamic process.
aPsychology Department, University of Adelaide, Adelaide, South Australia With respect to the failures of connectionism in modeling
5005; psyc-vic66 arts.adelaide.edu.au; conceptual structures, we agree that Glenberg’s strictures hold for
chomsky.arts.adelaide.edu.au/personal/dvickers/home.htm models that represent stimuli at an input layer, using a set of
bDepartment of Psychology, University of Adelaide; Cooperative Research preabstracted psychological features, for which the act of abstrac-
Centre for Sensor Signal and Information Processing, Adelaide, South tion by the modeler bears the bulk of the model’s explanatory
Australia 5005; psymdlee66 arts.adelaide.edu.au; burden (Brooks 1991; Komatsu 1992). However, there are at least
chomsky.arts.adelaide.edu.au/personal/mlee/home.htm two types of connectionist model that are able to learn and modify
their internal representations. First, there are supervised models,
Abstract: Glenberg’s account falls short in several respects. Besides
sometimes referred to as “connectionist semantic networks,”
requiring clearer explication of basic concepts, his account fails to recog-
nize the autonomous nature of perception. His account of what is remem- which employ architectural “bottlenecks” to reveal underlying
bered, and its description, is too static. His strictures against connectionist representational structures in a set of input/output pairings (e.g.,
modeling might be overcome by combining the notions of psychological Rumelhart & Todd 1993). Second, under the title of “semantic
space and principled learning in an embodied and situated network. map,” the established self-organizing map has been applied to
learning internal conceptual representations (Kohonen 1990).
Glenberg’s view of memory combines notions arising out of The fact that both approaches rely on some uncomfortably arbi-
connectionist modeling with some basic insights from cognitive trary interpretation of the inputs (and outputs, in the supervised
linguistics. It can be evaluated, not only by its cogency, con- case) reinforces Glenberg’s call for the embodiment and situation
sistency, and completeness as an account of memory, but also by its of cognitive models. However, this does not detract from the
coherence with current thinking in these related fields. Some such extent to which the specification of the learning rules constitutes a
integrative conceptualization is necessary if we are to understand complete and formal description of the process by which the
how the various “systems” (e.g., visual, auditory, linguistic, and internal representations in such models are acquired and altered.
motor) can “talk” to each other, and cooperate smoothly and At the same time, such internal representations do not them-
instantaneously. We agree that there is a need to develop cognitive selves constitute principled models of human conceptual struc-
models in situated and embodied agents and to accommodate tures. For example, although the structure of the learned internal
flexible and emergent cognitive structures. We also agree that, representations is highly dependent on the architecture of a
despite its general failure so far to develop embodied and situated connectionist semantic network, the precise forms of the various
models, connectionism represents “the surest route to formalizing architectures used in practice do not appear to be sufficiently
these ideas” (sect. 3.5, para. 6). However, we think that Glenberg’s constrained by an analysis of desired representational outcomes.
attempted integration falls short of a successful achievement in a Arguments for the appropriateness of a learned internal represen-
number of important respects, enumerated below. In addition, we tation are typically based either on intuition (e.g., a semantic map
wish to qualify Glenberg’s criticism that, in the connectionist is observed to use neighboring units to represent words with
modeling of conceptual structures, “most theorists fail to specify subjectively similar meanings) or on some form of post-hoc anal-
what the features are, and . . . how those features might be learned ysis (e.g., a hierarchical clustering analysis of a bottleneck layer in a
or changed as the consequence of development” (sect. 3.5, para. semantic network reveals an appropriate taxonomy of the natural
3). By reviewing some recent connectionist models, we hope to kinds being represented). Therefore, we consider that connec-
extend the discussion of connectionism and to suggest an ap- tionist semantic maps and networks overcome Glenberg’s criti-
proach towards the development of models, which, by virtue of cism regarding the specification of learning principles, but fail
being embodied and situated in an environment, might learn with respect to the specification of representational principles.
psychologically principled internal representations. In contrast, there are connectionist models that use psycho-
Glenberg’s assertion that the perception of meaningful struc- logically principled internal representation but fail to specify the
ture in “projectable” properties depends upon a contextually way such representations are learned and modified. We refer
suggested combination with remembered action patterns seems particularly to models such as alcove (Kruschke 1992) and
inadequate as an approach to perception and is no clearer than others (e.g., Shanks & Gluck 1994), which assume that stimuli are
association as an account of conceptualization and potential repre- represented in a “psychological space” (Shepard 1987a). Whether
sentation in memory. Many phenomena suggest that the visual or not the psychological space construct proves to be an entirely
system can detect the symmetries produced by multiple transfor- adequate model of human conceptual structure, the important
mations (Garner 1974; Leyton 1992; Palmer 1991). Meanwhile, point is that these connectionist models explicitly and consistently
Barnsley and others (e.g., Barnsley & Anson 1993) have shown adopt a set of representational principles that have a strong basis in
that visual images can be encoded by the parameters of a collage of psychological theory.
probabilistically iterated transformations. It is at least possible that A connectionist model able to learn and modify principled
what we perceive as structure or organization is the “resonance” of conceptual representations would seem to require a combination
neural units, which automatically carry out such transformations, of both types of model we have just described. One envisages
thereby generating an output that matches the current sensory constructing something like a connectionist semantic network so
input. Although such transformations may have had their evolu- as to ensure that its internal representations constitute a psycho-
tionary origin in the image changes consequent upon particular logical space. The information required to form these representa-
physical movements, the visual system has evolved to respond to tions (akin to the indices of psychological similarity that drive
such a rich variety of structure in an image that it seems implausi- multidimensional scaling algorithms) would need to be implicit in
ble to suppose that the perception of every organization must still the input/output pairings presented to the model. In line with the
be directly grounded in some actual or potential pattern of action. thrust of the target article, the way to generate these implicit
The author is not explicit enough about what it is that is representational constraints is to consider the model as an embod-
remembered. According to Glenberg, “memory is embodied by ied and situated agent, existing in an environment from which it
encoding meshed . . . sets of patterns of action” (sect. 2, para. 1). receives sensory information (the inputs), and which is influenced
This characterization seems too static and literal – too like a by its actions (the outputs). This conclusion resonates with argu-
videotape – for his purposes. By comparison, in the view of ments by Shepard (1987a; 1987b), who has emphasized the
perception sketched above, the memory of an image may corre- adaptive function of the psychological space approach to human
spond to the parameters of the collage of transformations by conceptual structure and has argued that such representational

40 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Response/Glenberg: What memory is for

structures must evolve subject to constraints implicit in an organ- mechanisms, that is, mechanisms sensitive to the literal
ism’s interaction with its environment. shapes of objects and the capabilities of the body. Thus
In our view, the argument of the target article is subject to cognition is embodied. These ideas contrast to nonanalogi-
shortcomings rather than fundamental flaws. To overcome them cal accounts of cognition that use arbitrary amodal symbols
will inevitably involve, among other things, achieving a detailed to represent objects, actions, and events.
account of the interface between perception and cognition, as well
as that between language (including sign language), gesture, and
We understand language (that is, it takes on meaning) in
action. With respect to connectionist modeling, the representa- the same way we understand a physical situation – in terms
tional dictates of psychological spaces would seem to provide a of possibilities for action. An important function of lan-
promising starting point in the pursuit of a connectionist “mesh” guage is to change the conceptualization of a situation, that
between mental representations, the agent, and the physical is, to change the actions possible. Thus, when your glass is
environment. Despite its limitations, therefore, we feel that this empty, your companion may say, “Here, take mine,”
account is a challenging first step toward a more useful and thereby eliminating the normal social constraint on drink-
realistic view of memory as a dynamic process. ing from someone else’s glass, or if he is particularly
desirous of his own wine, when noting your glance at his
glass he may say, “Don’t touch it,” thereby strengthening
the social constraint. When we read narrative text, we
understand the text in terms of actions available to the
protagonist as the situation described by the text develops.
Author’s Response When we can’t mesh the actions described with our under-
standing of the situation, the text appears incoherent.
Similarly, if your glance was not toward your partner’s wine
glass but beyond it to the salt shaker (i.e., if you were
What memory is for: Creating meaning conceptualizing the situation in terms of the socially per-
in the service of action missible action of reaching for the salt shaker), your part-
Arthur M. Glenberg
ner’s “Don’t touch it” would appear bizarre, or incoherent
Department of Psychology, University of Wisconsin-Madison, Madison, WI
with the situation, prompting you to say, “Huh?”
53706. glenberg66 facstaff.wisc.edu Is this equivalent to “proposition-speak” (McNamara)?
The psychologist’s propositions are built out of abstract,
Abstract: I address the commentators’ calls for clarification of
inherently meaningless, amodal elements that are manipu-
theoretical terms, discussion of similarities to other proposals, and lated by rules allowing operations such as concatenation
extension of the ideas. In doing so, I keep the focus on the purpose and substitution. The elements discussed in the target
of memory: enabling the organism to make sense of its environ- article are not abstract, not inherently meaningless, and not
ment so that it can take action appropriate to constraints resulting amodal; they are not manipulated by rules (although see the
from the physical, personal, social, and cultural situations. discussion of numbers in sect. R3 on Extensions). Instead,
elements fit together on the basis of their analogical
R1. Clarifications shapes.
On the embodied account, meaning is not reducible to
R1.1. Embodiment and mesh. I will discuss only the target bodily movement (McNamara), nor is it equivalent to
article, rather than the general idea of embodiment. The “response tendencies” and “discriminative stimul[i]”
basic assumption is that the cognitive system’s main func- (Slater). As described above, conceptualization of a situa-
tion is to control bodily action; to ensure that actions fit the tion is in terms of the responses available, and availability
physical, biological, personal, social, and cultural con- results from a mesh of many constraints. As Slater notes, a
straints of specific situations. How is this done? These types particular bodily movement can have different meanings in
of constraints are in terms of bodily action. Combining different situations (reaching for the glass, reaching for the
constraints is tantamount to creating a consistent, coherent soap, etc.). But contrary to what Slater presumes, meaning
(that is, doable) action. For example, to quench your thirst is not reducible to bodily movements. Meaning derives
at the dinner table, you reach for a glass in a manner that from actions available in particular situations, and those
reflects the physical shape of the glass, the biological need, actions (as well as the situations) are defined by action-
and your personal history with the glass (that is, you reach based goals and how those combine with the affordances of
for your glass, not your companion’s); moreover, you do so the situation. (I am using the Gibsonian term “affordance”
in a manner that does not breach social and cultural to refer to possible actions as determined jointly by charac-
conventions. The meaning of the glass to you, at that teristics of the physical situation and abilities of the body. I
particular moment, is in terms of the actions available. The extend the meaning of the term in the discussion of affect in
meaning of the glass changes when different constraints on the section Extensions.)
action are combined. For example, in a noisy room, the Crowder & Wenk suggest that mesh is a matter of
glass may become a mechanism for capturing attention (by having fine-grained (but arbitrary) symbols, and that “cur-
tapping it with a spoon), rather than a mechanism for rent exemplar models of memory can simulate mesh with
quenching thirst. The coherent combination of actions is an ongoing retrieval process that continuously retrieves
what I refer to as “mesh.” Note that mesh is nothing like multiple representations acting in concert.” Indeed, many
taking the average of many actions. Nor is mesh simply of the exemplar models are impressive, but mesh has little
concatenating actions. Instead, mesh requires the integra- to do with grain and even less to do with the sorts of retrieval
tion (e.g., coarticulation) of actions to fit the constraints of processes that many of the exemplar theories propose (see
the specific situation. In this way, cognition is situated and discussion of exemplar theories in the sect. R2 on Connec-
individuated. On this account, cognition requires analogical tions). The issue is not grain, but how the symbols (embod-

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 41

Response/Glenberg: What memory is for

ied or arbitrary) come to mean, and consequently the sentences that are easily processed are those that describe
operations permissible. A fine-grained encoding of a situa- ideas that mesh with the current conceptualization. Sen-
tion by arbitrary symbols remains, at best, a description of tences that describe actions that do not mesh with the
the situation that needs to be interpreted to effect action. developing conceptualization are perceived as incoherent,
The requirement for interpretation introduces what appear unless the author takes pains to indicate that the sentences
to be insurmountable problems, as reviewed in the target are not meant to mesh, but are meant to set up an alterna-
article. Perhaps more important, the operations on those tive conceptualization (or mental space, to use terminology
arbitrary symbols appear to be of the wrong sort. For from Fauconnier 1985). These alternatives are signaled by
example, some exemplar models (e.g., Hintzman 1986) locutions such as “he thought about,” “meanwhile,” “in
simulate retrieval from memory by creating a weighted 1886,” or by using the indefinite this (Gernsbacher &
average on each dimension encoded in the retrieval cue. Shroyer 1989) as in “there was this guy.”
The weight is determined by similarity of memory traces to This sort of approach seems to be exactly what is needed
the cue. How would such a model combine actions consis- for Jacobs & Ziegler’s “logic of Elfland,” that is, how we
tent with quenching thirst, personal history, and social can learn by being told, rather than through direct experi-
constraints? The resulting retrieved memory would be an ence. Understanding what we are told is not forming an
uninterpretable mishmash. Also, models such as Hintz- abstract description of the sentences (i.e., the psychologist’s
man’s can only work by specifying, for once and for all, the propositions). Instead, like being told that one can drink
dimensions of encoding. Thus, to make a glass an appropri- from one’s companion’s glass of wine, language changes the
ate response in different situations, all memories of glasses way we understand the world in terms of actions available.
must specify not only that it may contain water but that it Consider Jacobs & Ziegler’s example of the lesson from
rings when tapped, it holds flowers, it breaks and becomes a “Beauty and the Beast” that “a thing must be loved before it
jagged weapon, it can hold down a table cloth, it makes a is lovable.” The lesson seems so divorced from ordinary
fine present, and it amplifies sounds when held to the wall action, so abstract, how can it be understood in an embod-
(at least in old movies). ied system? I suspect that it is not accidental that lessons
For similar reasons, mesh cannot be reduced to convolu- such as this are conveyed through fairy tales. The fairy tale
tion (see Murdock 1993, todam2) as MacLeod suggests. provides a concrete illustration of narrative actions that
todam2 creates mishmashes of meaning. The convolu- allow us to appreciate the meaning of words such as “love.”
tion mechanism combines representations (vectors) using In understanding the fairy tale, that is, in forming an action-
mathematical operations similar to taking a dot product. based conceptualization, the child learns which interactions
The mathematics have no regard for what the elements in a correspond to terms such as “loving” and “being loved,” and
particular vector may represent. As such, todam2 will which actions make one lovable. Telling a child (or adult)
equally well associate (convolve) the representations of the lesson without the illustration does no good: the words
“glass” and “water” and the representations of “glass” and alone are empty, without meaning, and soon forgotten. The
“light-year,” whereas humans would find it much easier to way the child learns from the fairy tale is the way that we
find a connection between the former two than between learn in any abstract domain (e.g., mathematics). Initial
the latter two. learning depends on examples, figures, illustrations, and
Graesser suggests that a completely embodied repre- analogies. In all of these cases, we need to conceptualize the
sentation would be “much like a high resolution videotape” lesson in terms of the concrete and actionable.
(see also Vickers & Lee’s point 2), and that that is not
consistent with observations of people’s understanding of R1.2. Short-term memory. Whereas Crowder & Wenk
texts. To get the representation at an appropriate “resolu- judge my rejection of a separation between long-term and
tion,” de Vega suggests a type of suspended embodiment, short-term memory “as perfectly justified by the evidence
and Habel et al. propose that some representations are but not any sort of radical change,” Logie, Musen, and
indirectly embodied. I will address these proposals in the Rossetti & Procyk object. The latter three commentaries
section R2 but for now I will clarify why an embodied cite neuropsychological evidence consistent with the sep-
representation is unlike a videotape. Our actions in the aration. As the commentators note, I have made little use of
world are based on affordances fleshed out by mesh with neuropsychological evidence in developing my arguments.
nonprojectable properties, not an examination of all prop- This reflects my uncertainty as to how to interpret two of
erties of the environment. Furthermore, memory is up- the primary types of neuropsychological evidence, localiza-
dated by tracking the changes in the possibilities for action; tion of function and dissociations. I will briefly describe
which actions are now possible and which not. Details of some of the reasons for my uncertainty. Sarter et al. (1996)
the physical environment, except as affecting the particular note many problems associated with the interpretation of
experiencer’s actions in a particular situation, are irrelevant. brain imaging. Some of these problems are: (a) complex
The same is true for understanding a situation described cognitive functions may not be isomorphic with neural
in a narrative. As we read, we develop an action-based systems; (b) increasing local activity of excitatory or inhibi-
understanding of the situation described by the text. Unlike tory interneurons may yield similar signals but have di-
a videotape, this is a conceptualization in which certain vergent functions; (c) often a problematic form of the
actions are possible and others not. The sentences that subtractive method is used in data analysis; (d) neural
describe possible actions (that mesh with the conceptual- circuits may be diffuse and have different functions de-
ization), are perceived as coherent. In traditional terminol- pending on the level or pattern of activation, (e) the logical
ogy, coherent sentences seem to be primed, that is, preacti- relation between data and conclusion is not ideal – we wish
vated, preprocessed, preinferred, and so on. But, as to assert that a cognitive function arises from a physiological
discussed in the target article (sect. 6.2), there is no need to process, but we manipulate cognition and observe changes
postulate any sort of automatic preprocessing. Instead, in physiology, rather than vice versa. As Sarter et al. note,

42 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Response/Glenberg: What memory is for

some purchase on the problem can be had by combining Before extending the analogy, consider the following
the imagining studies (in which cognition is manipulated possibility: suppose that memory skills and strategies are
and physiology observed) with studies of patients with developed to deal with common situations such as planning,
lesions (in which physiology is disrupted and changes in taking action, and constructing a conceptualization from
cognition observed). Nonetheless, interpretation of these language. As these situations are common to the human
studies is also far from simple. For example, Farah (1994) condition, so are many of our memory skills. Psychologists,
begins with three well-documented dissociations (e.g., se- however, assume that these skills and strategies reflect the
lective impairment in knowledge of living things). For each operation of separate modules and design procedures to
dissociation, she discusses difficulties in attributing the break down the everyday skills and strategies into compo-
dissociation to impairment in a specific, localized module. nents using procedures such as memory span (recall in
At the same time, she demonstrates how an alternative set order an arbitrary list of words). Note that a memory span
of assumptions about the operation of the brain (assump- task only makes sense if there is something like a limited
tions congenial to network models) provide a different, but capacity short-term store specialized for keeping track of
no less compelling, interpretation. arbitrary lists of information. If there is no working mem-
Farah’s conclusion that dissociations do not necessarily ory, and instead there are skills designed for real language in
imply functionally separate systems is strongly echoed in real contexts, the memory span task may reveal little. At this
the memory literature. One example is the current debate point, an advocate of the theory of working memory might
regarding dissociations between explicit and implicit mem- well ask, “If there is no working memory system, how could
ory tasks. It is now accepted that dissociations can be a memory span task ever result in consistent data and
interpreted as revealing separate systems (e.g., Schacter et consistent dissociations?”
al. 1990) or not (e.g., Roediger et al. 1994). Similarly, we are Consider a laboratory investigation of tennis playing in
recognizing the difficulty in figuring out which dissociations which we extract the player from the complex situation (a
count as critical data. One illustration of these difficulties is tennis game with a real opponent) and try to probe what we
provided by Logie et al. (1996). They note that Baddeley’s believe to be modules underlying the skill. We might begin
model of working memory is supported by a pattern of data by putting the tennis player in a special room with a table-
including (a) a word-length effect (shorter memory span for tennis paddle whose size makes it much more convenient to
long words compared to short words) following both audi- study than a full-sized racquet. The player’s task is to avoid
tory and visual presentation of the to-be-remembered being hit by balls ranging in size from squash balls to
words, (b) a phonological similarity effect (shorter spans for medicine balls. Almost surely performance in this task
similar sounding words) following both auditory and visual would correlate moderately with performance on the tennis
presentation, and (c) in patients with presumed impair- court (as reading span correlates with comprehension skill).
ments to the working memory system, the absence of all but Furthermore, we could easily produce dissociations. For
the auditory phonological similarity effect. In stark contrast example, in a no-impediment condition, the tennis player
to the seemingly clear picture of dissociation of effects and can dodge slow-moving medicine balls, but he is less
impairments, Logie et al. present the following data: (a) in a successful in using the table-tennis paddle to defend against
sample of 251 nonimpaired subjects, 108 failed to show one fast-moving squash balls. In the impediment condition we
or more of the standard effects; (b) when subjects were force the tennis player to stand in one small area. Such an
retested, the appearance or absence of an effect was unreli- impediment is analogous to a neurological insult or the
able; (c) the appearance of an effect was strongly related to imposition of an external constraint in a memory span
the type of strategy reported by the subject; and (d) strategy experiment such as forcing the subject to vocalize “the the
use was variable: about 25% of the subjects reported the . . .” throughout presentation of the stimuli (a proce-
changing strategy within a test session. dure known as articulatory suppression). Confinement to
Logie et al. (1996) choose to interpret these data within one small area results in selective deterioration of perfor-
the working memory framework. For example, in this mance: the tennis player can no longer dodge the medicine
framework the occurrence of a word-length effect may ball, but his defense against squash balls is unaffected. We
require the use of a subvocal rehearsal strategy that engages would also find that people with physical disabilities (e.g., a
the relevant working-memory structures. However, once broken leg) would behave much as the tennis player in the
we recognize that different effects can reflect strategic impediment condition. The point is that this bizarre investi-
choices, to what extent is it necessary to postulate a func- gation would produce stable and complex data and dissocia-
tionally separate working memory? Might not the strategy, tions without revealing much about tennis skill. If our
acting within a general memory system, produce the ef- assumption that memory is designed to deal with arbitrary
fects? Might not dissociations with neurological impair- lists of stimuli is as wrong as the assumption that tennis skill
ment reflect the inability to execute a particular strategy depends on being able to dodge arbitrary missiles, then
rather than the impairment of a particular information- what we learn from the consistent data produced by the
processing module? Consider the following analogy be- memory span procedure may have the worth of what we
tween skilled use of memory and a physical skill, such as learn from table-tennis dodgeball.
playing tennis. The tennis player may use different strokes Logie notes that we do not know very much about the
(strategies) depending on the situation. The different development of skills (tennis or otherwise). Nonetheless,
strokes may be differentially effective, may show different taking a skill-based view of short-term performance sug-
developmental patterns, and may be differentially influ- gests several avenues for testing and reinterpreting existing
enced by various injuries (e.g., to the elbow, the wrist, or the data. First, the causal arrow can be reversed. For example,
knee). Nonetheless, few would claim that there is a back- the correlation between reading span (a type of memory
hand system that is functionally separate from the forehand span) and success at handling ambiguous sentences (Car-
system, and that both are different from the serving system. penter et al. 1994) is taken as support for the claim that

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 43

Response/Glenberg: What memory is for

people vary in working memory capacity and that that Crowder & Wenk suggest that perhaps associations are
capacity affects language processing. In contrast, consider, applicable at the neural level of analysis, if not at the
as do St. John and Gernsbacher (1995), that it is the skill in cognitive level. No doubt there are conditional probabilities
language tasks (c.f. tennis) that determines skill on the at the neural level, but should we take them as explanatory
measurement task (c.f., table-tennis dodgeball). Second, principles? As mechanisms become microscopic, a case can
the skill-based view makes the prediction that training in a be made that analogical shape becomes even more impor-
language skill (e.g., practice in center-embedded clauses or tant. As an example, consider the analysis of genetic expres-
relative clauses) should be reflected in enhanced perfor- sion. At the Mendelian level, one does talk in terms of
mance in a reading span experiment using those types of probabilities. At the chromosomal level, one talks in terms
materials. of information units, and genes, but even here analogical
Rossetti & Procyk also discuss differences between properties, such as the physical distribution of genes on
short-term and long-term performance, but from a differ- chromosomes, become important. When one examines the
ent perspective. In their task, subjects learned to point to a operation of the genes, their analogical shapes take on
single target, one of six targets in a linear array, or one of six the utmost importance. For example, it is because of the
targets arrayed in an arc. In all conditions, on any given trial particular three dimensional folding of the proteins encod-
a single target was presented (by passive movement of the ing genes, and how those foldings constrain interaction with
hand opposite to that which makes the response). With an the constituents of the chemical environment, that the
eight-second delay between target presentation and re- genes express themselves at the time and place that they do
sponding, the distribution of errors mimicked the layout of (Stein et al. 1996). That is, the proteins must literally mesh
the six targets, either linear or arc-shaped. The differences with components of their environment to get the job done.
in the distribution of errors was not evident when there was
no delay between target presentation and responding. Does R1.4. Implicit memory phenomena. Benjamin & Bjork
the difference in error distributions across delay imply a suggest that a problem for embodiment is “that mere
special short-term process? Rossetti & Procyk offer an exposure to a stimulus can alter subsequent performance
alternative answer: there is a measurable delay in the time it on certain perceptual or cognitive tasks . . . [and S]uch
takes to mesh memory trajectories (based on linear or arc- stimulus-driven processing, typically independent of a sub-
shaped training) with projectable features defining the ject’s tasks or goals at the time, suggests functions and
location of the single target presented. A related interpreta- operations of memory that do not fit neatly in the embodi-
tion is that shortly after target presentation, responding is ment framework.” In contrast to this statement, Benjamin
strongly constrained by clamped projectable properties; & Bjork go on to discuss how implicit memory effects are
when the projectable properties are absent (in the delayed dependent on the subject’s tasks and goals. For example,
condition), trajectories play a more significant role in con- asking a subject to read a word out of context (e.g., “peach”)
trolling movement. will enhance the subjects’ ability to identify the word from a
fragmented version of the letters. However, asking a subject
R1.3. Associations. Both Crowder & Wenk and Mac- to (orally) generate the word from a hint (“a fruit with a
Leod ask how completely I eschew associations as an fuzzy skin”) will not much affect (visual) fragment comple-
explanatory construct. My reply: completely. Certainly tion but will enhance performance when more “concep-
events appear to follow one another, actions follow one tual” information is required, such as rapidly naming a
another, and memories follow one another. If that is all that dozen fruits. In the target article, I suggest that these sorts
we mean by an association, then I have no argument, of goals affect whether verbal stimuli are, as in the read
because “association” is used as a descriptive term syn- condition, conceptualized in terms of the projectable prop-
onymous with “conditional probability greater than simple erties, or, as in the generate condition, conceptualized in
probability.” Psychologists and neuroscientists have come terms of the (action-based) meaning of the referents. The
to use the term as a theoretical construct, however. An different conceptualizations (and resulting trajectories)
association has become the mechanism for producing the will only affect performance on later tasks that can mesh
memories that follow from one another. Association is used with those trajectories. As Crowder & Wenk note, this
an explanatory term because we have conceptualized proposal is not different from Tulving’s encoding specificity
knowledge as consisting of amodal, meaningless symbols principle.
that have few ways of relating to one another. That is, the McNamara argues that implicit memory does not de-
only way to connect one such symbol to another is through serve the prominence it is given in the target article,
an association (or a propositional equivalent). If we con- because implicit memory “is fundamentally aplysia mem-
sider analogical representations, however, then other ory: it is the system that’s been around for hundreds of
mechanisms of relating become possible. For example, millions of years, not the system that is responsible for the
mesh relates ideas by virtue of the coordination of patterns rich complexity of human cognition.” This argument fails
of action that underlie meaning. Whether or not there is any for three reasons. First, it implies that a system is to be
merit to action-based conceptualization, it is clear that denigrated because it has worked so well. Thus, we might
associations are not a powerful explanatory mechanism. In also question the importance of the immune system, the
using associative theories, any two ideas (representations) circulatory system, and the visual system, all of which have
ought to be equally associable, or constraints on associa- been around a long time. Second, it is by no means clear
bility are external to the representations. But ideas are not that explicit, conscious retrieval processes contribute much
equally associable, and it seems that no list of external to human cognition. When solving everyday tasks such as
constraints can be complete. One solution is to do away making breakfast or finding our way to the office, it seems
with associations by letting the symbols be analogical and that we use explicit memory only when things go wrong:
intrinsically meaningful. “Where did I put the coffee mug?” Otherwise, most tasks

44 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Response/Glenberg: What memory is for

(putting the key in the ignition, pressing the accelerator, standing of “love,” we can even mesh actions consistent
shifting into reverse) are done relatively automatically. with loving and affordances of tables. This flexibility in our
Similarly in following language we do not consciously construal of “table” is what obviates the need to decide on a
attempt retrieval: the words flow and we understand. It is fixed set of features, dimensions, default values, frames,
only when a word or phrase does not effortlessly mesh with scripts, tracks, MOPs, etc.
the developing conceptualization that we might ask our- It is the flexibility of mesh that helps us to understand
selves, “What does that word mean? Where have I heard it flexibility in language use. Goldberg’s (1995) example “She
before?” the sorts of questions asked on an explicit memory sneezed the foam off the beer” illustrates part of the
task. Finally, Schunn and Dunbar (1996) report that im- phenomena. Given that “to sneeze” is an intransitive verb,
plicit memory priming contributes to complex scientific this sentence is not just ungrammatical, it ought to be next
reasoning. to impossible to understand. Nonetheless, most of us will
have little difficulty with the sentence because we can mesh
R1.5. Flexibility. An emphasis on grounded meaning the actions of sneezing with foam flying off the top of a glass
“seems to underestimate the flexibility and functions of of beer. Similarly we can comprehend Shanon’s (1987) “The
human memory,” suggest Benjamin & Bjork. MacLeod newsboy porched the newspaper” by meshing an action-
questions how people could learn anything truly new in a based conceptualization of newsboys with affordances of
system based on action. De Vega notes that the target newspapers (at least those folded appropriately) and con-
article does not present anything close to a worked-out straints on action appropriate to our culture (at least that of
account of figurative language, and this point is also made the 1950s). In the 2050s, when newsboys, porches, and
by Ramsay & Homer (“Glenberg assumes that language physical newspapers may be things of the past, it will
comprehension involves decoding the spoken or written probably be near impossible to understand such a sentence
message in a literal fashion”) and Vickers & Lee (“This outside of a carefully constructed context.
characterization seems too static and literal”). These criti- Keysar and Bly (1995) have demonstrated that people
cisms reflect the most egregious failure on my part to make are quite flexible in their ability to interpret metaphor. Thus
my ideas clear, because I had intended the embodied some students may be told that “The goose hangs high”
framework to be much more flexible than standard models means that someone is to experience very good luck, and
of memory and comprehension. other students may be told that “The goose hangs high”
Let’s start with MacLeod’s question: How can we learn means that someone is doomed. Both sets of students will
anything new? Consider an infant learning the concept find the interpretation given them first convincing and the
“table.” At first a table may be simply a location where her alternative interpretation strange. One account of this find-
feeding takes place. Then, it may be a stable object that can ing is that much metaphorical language is arbitrary and not,
be used to pull herself up to a standing position. A table may as Lakoff suggests, motivated. There is, however, an em-
become a source for making noise by banging on it, and bodied account equally consistent with the data (and similar
finally a table can become a surface that can support her to the account in Keysar & Bly 1995). We can mesh with the
food while eating and her paper while drawing. The point is projectable properties of high-hanging goose nonproject-
that the table offers different affordances as the child’s able properties, such as the animal’s history. If we imagine
actions become more differentiated: what the child can do the high-hanging goose to be the result of a successful hunt,
with the table, what it means to her, changes as she then it is emblematic of success and good luck. However, if
develops. We don’t have to decide for the child on a few we imagine the family goose strung up by vigilantes during
canonical features or dimensions that represent table for the night, it becomes almost as frightful as a burning cross.
now and evermore. Instead, the meaning of the concept Once a coherent interpretation (meshed conceptualiza-
changes with experience, and will continue to change over tion) has been reached, memory is updated, and that
the course of the child’s life, and as the culture (e.g., how update (trajectory) will bias future interpretation. Our
artists use and portray tables) changes the interactions we conceptualization of the family goose hanging with a bro-
have with tables. What is true for children is true of the ken neck will now be hard to mesh with “good luck.”
adults: objects and events come to mean different things to
us as our skills in dealing with those objects change. The R1.6. Projectable properties. The “use of the term proper-
meaning of computers has changed dramatically with the ties is unfortunate because it implies that we should con-
advent of electronic communication; the meaning of “love” ceptualize memory in terms of properties that exist indepen-
changes with our experiences of being loved. In none dently of any particular organism. . ,” writes MacDorman.
of these cases can meaning be distilled, formalized, and In a similar vein, Franklin writes that it is “perception that
reified; it is always changing. Of course there are well- creates the property in the first place. . . The agent projects
defined and well-behaved concepts in formal, artificial both projectable and nonprojectable properties onto the
systems, but we do not live in such a system. environment. The distinction is one of degree of constraint;
An embodied system that incorporates the idea of mesh projectable properties are more constrained by the envi-
is flexible in the way that memory is used in dealing with the ronment.” I agree with both commentators. I lifted the
world. If meaning were encoded by a list of features (the 0s distinction between projectable and nonprojectable from
and 1s that Benjamin & Bjork champion), the table, Epstein (1993), and in taking it out of his context, did it a
computer, or love experiences would be doomed to the disservice. Perhaps a better term for the sort of theorizing
same inflexible characterization. In an embodied system, in the target article is Gibson’s affordance. I chose not to use
we can mesh the affordances of a table with memories of that term, however, because I did not want to bring along
eating. We can also mesh our embodied understanding of with it Gibson’s aversion to nonprojectable features being
tables with the embodied goal of getting our body elevated added to flesh out a conceptualization. The two terms,
so that we can swat a bug. Given some embodied under- projectable property and affordance, are related. Afford-

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 45

Response/Glenberg: What memory is for

ances, the actions available in a particular situation, depend based on assumptions, explicit or not. Thus, our assump-
jointly on the projectable (physical) characteristics of the tions about associations support the use of paired-associate
environment and the human body. As MacDorman writes, learning, and our assumptions that symbols are abstract
“one may certainly assume the existence of an observer- independent elements support work using lists of arbitrary
independent reality,” but what we make of that reality verbal stimuli to study learning (e.g., Musen & Squire
depends on the particular organism “with its unique body, 1991). It can only help to make these assumptions explicit
sense organs, and life history.” If I had been clearer on this and to examine alternatives.
point, I suspect that McNamara would not have mis- Notwithstanding the benefits of a broad reach, Koriat &
takenly suggested that “Glenberg’s psychology of memory Goldsmith, Musen, Regier, and others point to an impor-
would seem to require a theory of the structure of the tant deficiency. Mine is at best a framework. Any real worth
world; after all, projectable properties will be described remains to be demonstrated through the development of
in the vocabulary of physics, not psychology,” Velich- testable hypotheses.
kovsky would not have confused projectable properties
with retinal image size, and Anderson would not have
R2. Connections
suggested that I attempted “to base memory on visual
perception.” The commentators describe (sometimes in approving
A clearer exposition of projectable properties may well tones, sometimes disparaging) connections with other liter-
have also answered Velichkovsky’s question as to why atures and approaches. Rather than listing all of them, I will
children with “severe abnormalities of motor functions” do single out a few that seem to need some comment.
not demonstrate “striking deviations” in cognitive abilities.
As argued in the target article, joint evolution of the R2.1. Exemplar models. As noted before, Crowder &
cognitive/perceptual system and the body has ensured that Wenk compare my approach to that of exemplar models of
the cognitive/perceptual system is sensitive to projectable memory in which each experience (even separate experi-
properties of the world that are relevant to bodies of our ences of the same nominal event) is registered as a separate
type. Thus, we can expect some ability to detect affordances “trace.” The comparison was based on the ability of exem-
relevant to interaction, whether or not an individual is plar models to retrieve blended information from memory
physically capable of the interaction. specific to the current context, resulting in a streaming of
cognition. There are several ways in which the comparison
R1.7. Role of theory and assumption. By writing the target is not apt. First, exemplar models (as currently instantiated)
article I hoped to accomplish several goals. One is to point model memory traces as a string of arbitrary symbols, in no
out how assumptions about the nature of representation way (or at least in most ways) is not analogous to the event
and meaning have defined the problems and meth- being represented. Second, because the string of symbols is
odologies of cognitive psychology and blinded us to other arbitrary, the blending of information during retrieval does
ways of doing business. Second, I wanted to develop an not depend on the nature of the objects represented, only
approach to memory and language comprehension that the syntax of the representation. Third, trajectories in the
would smoothly integrate with investigations of other com- target article do not correspond to memories of exemplars.
ponents of human experience. Third, as Habel et al. note, A trajectory is the change in conceptualization (the pattern
my framework provides guidelines for generation of hy- of possible actions) produced by taking an action, an inde-
potheses and a rationalization for the findings at hand. For pendent change in the environment, or mesh with other
example, proposing that conceptualization of text, like trajectories. Trajectories are a bias or pathway of change.
conceptualization of the environment, is action-based helps Crucially, the trajectories are not arbitrary because they are
us to understand why readers appear to track protagonist’s constrained by our ideas of possible action. That is, not all
goals and why narrative is easier to follow than exposition. actions are immediately possible from a given starting
Narratives are descriptions of actions, interactions, and position. This characterization of trajectories approximates
changes in the environment (or the protagonist’s interpreta- Vickers & Lee’s call for memories that “correspond to the
tion of the environment) of the sort the cognitive system is parameters of the collage of transformations by whose
designed to deal with. iteration the image might be regenerated.”
Nonetheless, Koriat & Goldsmith and Musen question Vickers & Lee go on to note that current connectionist
the reach of the framework. Koriat & Goldsmith suggest models incorporate components similar to the ones I call
that there are many valid metaphors for the workings of for. Namely, some connectionist models abstract interest-
memory, and each metaphor has its own “focus of conve- ing regularities and relations from the input. As they note,
nience.” In contrast, I believe that a powerful theoretical however, the input is often composed of arbitrary symbols.
description of memory will have implications for language, Other models use representations that are “psychologically
development, planning, and so on. Also, it is almost certain principled,” and Vickers & Lee refer to Kruschke’s (1992)
that by combining constraints from other domains, we will alcove, a connectionist version of an exemplar model of
converge on a correct description more quickly than by categorization. Because alcove uses arbitrary coding of
attempting to model behavior in domains separated solely dimensions (and the relevant dimensions must be specified
for convenience. for the model) it differs from my proposal in the ways noted
Musen agrees with Koriat & Goldsmith’s criticism above for the memory models. Nonetheless, alcove has a
regarding the reach of the framework, and she also finds number of attractive features that may correspond to as-
that assumptions about internal representation are a “major pects of my approach. For example, an important compo-
weakness” of the target article. Indeed, most of those nent of alcove is learning to attend to dimensions that are
assumptions are not currently testable, but that is not an important for a particular task. Attentional learning in
argument against making them explicit. Research is always alcove is close to Solomon & Barsalou’s notion of

46 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Response/Glenberg: What memory is for

“propositional construal” in an embodied approach to cog- supplement the conceptualization. Similarly, the young
nition. child’s thought is characterized as egocentric: reflecting the
relation of objects to the child without concern for other
R2.2 Script Theory. The target article, Nelson suggests,
perspectives. Again, this description corresponds closely to
“might well have evolved from the script theories of action.”
the notion of thought clamped by projectable properties.
Whereas there is some similarity of motivation (e.g., to
There are several differences between Piagetian theory
account for human memory and comprehension of real
and my own. First, I wish to explore the application of an
world events), the mechanisms are disparate. Script theory
action-based, analogical account of cognition to adults and
(e.g., Schank & Abelson 1977) is based on psychologist’s
to the problems of understanding language. Second, I am
propositions, and hence relies on arbitrary symbols and
pushing a single principle underlying cognitive develop-
extrinsic constraints on relations. Initial versions of script
ment: the interactions between bodily growth, experiences
theory were exceptionally brittle; situations had to match
with the environment (including the social environment),
the script (or one of its associated “tracks”) very closely in
and development in the ability to suppress the clamped
order for the script to apply. Connectionist versions of
environment.
schema theory (Rumelhart et al. 1986) overcome some of
the brittleness, but not the arbitrary coding. Thus, they
must rely on extrinsic constraints or frequency of co- R3. Extensions
occurrence as the major features controlling thought.
R3.1. Perceptual symbols and numbers. A basic assertion
In the context of script theory, Nelson raises the ques-
in the target article is that we comprehend situations and
tion, “how does the individual’s memory representation
language in terms of patterns of action: what something
automatically recognize or match an available script to a
means to us is what we can do with it. For the most part, I
newly encountered situation?” I do not postulate that
left unexamined the possibility of other aspects of represen-
thought depends on instantiating an appropriate script.
tation, and whether particular forms of representation (e.g.,
Nonetheless, the general version of the question is legiti-
a visual image) are consistent with meaning as patterns of
mate: How does one select past experience to apply to the
action. Many of the commentaries suggest the need for
current situation? We rarely (if ever) are thrown into a
abstract symbols (Benjamin & Bjork, Carlson, Habel et
situation unrelated to what we were just doing and thinking.
al.), goals (Carlson, de Vega, Franklin, Gärdenfors,
Situations flow and transform sensibly by following real
MacDorman), partially-embodied representations (de
physical and cultural constraints (which is what makes
Vega, Graesser, Habel et al.), and perceptual symbols, or
trajectories useful). For example, although we may de-
analogical representations of characteristics of perceived
scribe a day’s activities as going to work and then going to
information (Franklin, Solomon & Barsalou, Vickers &
lunch, the events are not discrete. The work activities
Lee). As Vickers & Lee put it for visual perception, “the
gradually stop (as my colleague enters my office) and
visual system has evolved to respond to such a rich variety of
transform themselves (as we walk down the hall to the
structure in an image that it seems implausible to suppose
elevator) into lunch activities. Thus, in applying past experi-
that the perception of every organization must still be
ences to the current situation, it is extremely rare that one
directly grounded in some actual or potential pattern of
must select from among all possibilities, and most of the
action.” How can we accommodate an action-based ac-
time there is little selection at all: conceptualization of
count of meaning with the convincing arguments that we
the current situation blends into the next by virtue of
know much about the environment (e.g., colors) in addition
analogical fit.
to patterns of action?
R2.3. Piaget. The connection to Piaget, in particular the An attractive possibility is suggested by Solomon &
similarity between action-based conceptualization and the Barsalou (see also Barsalou 1993). They propose that
Piagetian notion of a sensory-motor schema have been perceptual experience creates perceptual symbols. These
noted by Carlson, Jacobs & Ziegler, Velichkovsky, and symbols do not simply describe the environment, but like
others. In fact, the connection is even closer. The Piagetian Piagetian symbols, they are analogically related to the
symbol is not the abstract amodal symbol of many cognitive environment. Importantly, perceptual symbols are not ex-
theories. Instead, it is arises from action and imitation, and act copies of physical stimulation. Instead, the symbols are
is analogical in the sense that, in some ways, the symbol sensitive to what we are attending and to how we construe
resembles what it stands for. The visual image is a prototypi- the situation (how the projectable properties mesh with
cal Piagetian symbol. For Piaget, action is integral to nonprojectable properties). Thus, perceptual symbols cap-
learning, even in learning the concept of number. In that ture the projectable properties that are relevant for the
case, the physical actions that check one-to-one correspon- actions we are contemplating, and in that sense perceptual
dence are thought to underlie the abstract concept of symbols capture relevant affordances.
number. In addition, a number of Piagetian claims and Perceptual symbols can be made to do a lot of work
interpretations are consonant with the interaction between (Barsalou 1993). Here I will use them in an unusual way to
clamping and suppression I postulated. For example, the account for one domain of human performance that many
preoperational child’s thought can be characterized as cen- commentators assume require abstract, amodal symbols:
tering on the states of the environment, rather than trans- operations on numbers. As Habel et al. note, it is relatively
formations, whereas the child in the stage of concrete easy to imagine that our understanding of small numbers is
operations seems to appreciate the transformations as well based on direct action, such as one-to-one correspondence,
as the states. Such a situation might arise if the preopera- but what are we to do with numbers like “943?” Most of the
tional child is strongly clamped to the current environment. time, we treat such numbers simply as “a lot.” Nonetheless,
As the child learns to suppress the clamping, trajectories we can make fine discriminations, such as between 943 and
based on previous experiences and imagined changes can 944, and this ability does not seem consistent with treating

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 47

Response/Glenberg: What memory is for

all of these numbers as simply “a lot.” Part of our ability is the detailed unpacking of “trip” as actions of extending a
surely the use of one-to-one correspondence to note small foot and so on. If understanding is embodied as action, what
differences and ordinal relations, such as between 943 and has happened to all of the actions?
944, but we are still left with how we deal with larger As a prelude to answering this question, keep in mind
differences. My proposal is that through laborious, time- that we can unpack understanding to finer and finer levels
consuming repetition we have learned how to make arbi- of action. If someone asks me what it means to “fly to New
trary transformations, of the following sort: The three York” I can start describing airplanes, airports, and so forth.
perceptual symbols “9” (i.e., a representation of the shape My ability to unpack will peter out much sooner than an
of the numeral 9, not the concept of nine), “1,” and “4” may airline pilot’s description, and that is why he knows more
be replaced with the perceptual symbol “13.” about flying than do I. Nonetheless, the ability to unpack
Several points follow. First, computational skill is the “fly to New York” does not imply that all of that knowledge
arbitrary manipulation (substitution) of perceptual sym- is precompiled: much of it can be assembled on the fly by
bols. I am characterizing the substitutions as arbitrary meshing trajectories. There is no prespecified schema that
because the operations have nothing to do with natural is brought up upon hearing “fly” or “trip.” Instead, when
constraints based on projectable properties; the numerals comprehending language, the interpretation of “fly” or
“9” and “4” cannot be literally meshed to produce the “trip” is forced to mesh with the interpretation of the
numeral “13.” Learning these substitutions is brute force situation described so far. In the absence of devices that
memorization, that is, brute force creation of trajectories, signal change of topic (e.g., Gernsbacher & Shroyer 1989),
and that is why so many people find learning and doing it is our embodied interpretation of the situation that forces
mathematics to be boring, laborious, and close to meaning- (or at least makes probable) an interpretation of the subse-
less. In fact, if meaning derives from patterns of possible quent language at the appropriate level: the level that
action, then the perceptual symbols of numerals are close to meshes with the current conceptualization. Thus, if the text
meaningless: operations on them do not respect project- describes how the cook gauged the speed of the butler’s
able, action-based properties. Nonetheless, the perceptual walk, raised her leg, and braced for an impact, then mention
symbols used in number manipulation need not be special of “foot” would easily mesh with the understanding of “trip”
amodal cognitive symbols that differ in kind from those whereas as mention of “revenge” would not. However, if the
used in other domains. text describes the cook as considering all of the butler’s past
Second, on this analysis, computational skill (and other insults, then reading “revenge” will be facilitated relative to
formal systems that depend on learning brute force trajec- reading “foot.” In neither case is there priming in the sense
tories) cannot be taken as the paradigm case of human of pre-activation or explicit expectation (see target article,
thinking. Most thinking, most meaning-making, most plan- sect. 6.2.).
ning, and most action arises from following embodied What then is automatically inferred? What parts of the
constraints on mesh, not rule-bound, arbitrary symbol text are always tracked? Which actions always included?
manipulation. Those of us who teach undergraduate math- None. Understanding current language depends on the
ematics courses know that proficiency in mathematical situation set up by the previous language; we understand to
manipulation cannot be a valid index of general intel- the extent that current language generates ideas that mesh.
ligence. There are too many intelligent and competent Thus, if the butler were in Madrid and the cook in New
people who blanch at the sight of numbers. York, asserting that the cook tripped the butler would be
Third, some people have tremendous ability in mathe- hard to comprehend, because it does not respect embodied
matics, and that ability is certainly more than skill in brute understanding of these situations. Running across such a
force symbol substitution. How does the mathematician sentence would force us to consciously unpack our concep-
think? The mathematician has learned to go beyond num- tualization, give up, assume a change of topic, or question
bers as perceptual symbols (i.e., visual images of the nu- the speaker, “What do you mean?” On this account, much of
merals). Instead, the mathematician has learned to form language comprehension skill depends on speaking/writing
embodied representations of quantities and their transfor- skill. That is, an effective speaker uses words that create
mations, much like a child learning about division by conceptualizations that mesh for the listener. The speaker
imaging a pie being cut up. What the mathematician does can do this because of the commonality of human experi-
with numbers is similar to what speakers of a language do ence; he knows the sorts of action-based conceptualizations
with words: we go beyond the immediate (numeral or the listener is creating (because they are like his), and he
word) to an embodied representation whose transforma- uses words that describe embodied transformations of
tions follow natural constraints. This hypothesis explains those conceptualizations. If the speaker is not skillful in
why the teaching of mathematics (and other abstract ideas) considering how his conceptualization may differ from the
is so greatly facilitated by the use of concrete examples: the listener’s (e.g., if the speaker is a child), then communica-
examples give us an easily embodied interpretation of the tion fails.
numbers.
R3.3. Suppression, effort, attention, and propositional
R3.2. Suspended embodiment? The need for some sort of construal. My undifferentiated notion of suppression is
representation that is not fully embodied is suggested by commented on by Carlson, Ramsay & Homer, Schwartz
De Vega, Graesser, and Habel et al. De Vega notes that et al., and Velichkovsky. By considering suppression (of
understanding “Mary flew from Madrid to New York” does the clamping of projectable features) to be a continuous
not seem to require a detailed unpacking of all the actions variable rather than all-or-none, many of the issues raised
(e.g., checking her bags, walking through the metal detec- by the commentaries can be resolved. Thus, in concep-
tor, etc.). Similarly, Graesser suggests that when reading tualizing a situation in terms of which actions are possible,
“the cook tripped the butler” we think about revenge, not our thinking can range from being predominantly con-

48 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 Response/Glenberg: What memory is for

trolled by the projectable properties of the environment (as make a pleasing arrangement, we attempt to center the
seems to be the case with young children and perhaps, as table below the lamp. But, when trying to install the lamp,
Schwartz et al. suggest, frontal lobe patients) to predomi- we attempt to center the lamp above the table. That is, how
nantly controlled by the mesh with nonprojectable proper- we construe the situation depends on the actions we are
ties (i.e., a very thirsty person who refrains from drinking planning. Similarly, the roadway can become the path
from his companion’s glass because of strong social con- home, the path to the grocery store, or a place to throw a
straints). This range of possibilities highlights the need for frisbee. The actions in which we are engaged or contem-
coordination between projectable and nonprojectable plating, when meshed with the environment, determine
properties, rather than suppression of projectable proper- how we construe that environment.
ties. Indeed, it seems that coordination must be the case for Propositional construal is just what McNamara finds
language comprehension to work: to varying degrees we missing from the target article: intension. That is, concep-
must attend to both the projectable properties of the tualizing a situation in terms of actions available, partic-
language signal and the meaning of the signal. ularly when those actions are constrained by mesh with
I argued that suppression of projectable properties of the nonprojectable properties, is exactly what is required to
environment was effortful. Schwartz et al. note that sup- make that situation about something. The situation means
pression is unlikely to be the only determinant of cognitive to us and is about the actions available to accomplish some
effort. When aspects of the environment are suppressed in goal such as centering the table or throwing a frisbee.
the service of an engaging task such as reading a mystery, McNamara also notes that mine is a “language of form [of
“the psychological content and/or process (mystery, hunt- objects] when the goal is a language of content.” But surely
ing, mating) resonates so strongly with evolved human the forms of objects contribute to the actions supported by
inclinations that it engages mechanisms of suppression (and the objects, and hence the meaning of those objects to us.
thus captures attention) without, or even in spite of,
one’s conscious intention” that is, little effort is involved. R3.4. Affect. But what determines the particular proposi-
Schwartz et al. also “suggest that suppression is effortful tional construal, the particular goal that we are acting on?
because, like swimming upstream, it involves willfully op- As noted by Anderson, Carlson, de Vega, Franklin,
posing the inertial flow of undirected thought, and over- Gärdenfors, and MacDorman, some notion of affect or
coming (neurophysiological) inertia, of whatever kind, re- value needs to be included to guide and organize action.
quires (physiological) work.” There are three questions to consider: How is value deter-
These ideas can be accommodated by my framework. mined? How does value affect conceptualization and direct
Translate “resonates . . . with evolved human inclinations” action? And, related, how is behavior guided by long-term
into transformations that follow embodied constraints; and goals? Here is a new proposal: changes in the affec-
translate “willfully opposing the inertial flow” as brute force tive/physiological state of the body changes affordances.
manipulation of perceptual symbols, as in mathematical Consider a terrifying situation. Low-level mechanisms
computation. Thus, daydreaming seems much less effortful detect that the situation is terrifying by noting large loom-
than problem solving because daydreaming follows embod- ing objects, exceptionally loud noises, threatening postures,
ied constraints on transformation, whereas brute force and so on. After detection, our bodily state rapidly changes
problem solving does not. Nonetheless, in both cases we through a wash of chemical and neuronal signals. The heart
need to exert some effort to suppress the clamping of the races, the skin sweats: in a word, we are in a state of high
environment, or physically break the clamping by closing arousal. Given this arousal, some actions are unavailable.
our eyes or looking at a blank wall. We cannot engage in cool reflection, we cannot walk slowly
Coordination between suppression and clamping seems and calmly, we cannot stand still, we may be unable to
close to capturing the distinction Ramsay & Homer make speak. That is, the major paring of options has been made
between “holding in mind” and holding “in view.” Hold- by the body: fight or flight.
ing in view is conceptualization based predominantly on Consider a more prosaic example. You are working hard
clamped projectable properties (compare to Gärdenfors at your computer, well past the lunch hour. As your bodily
“direct” representation); holding in mind is conceptualiza- state of hunger increases, you start to feel more and more
tion based on (relative) suppression of those properties and uncomfortable. As your hunger grows, the possible actions
coordination with nonprojectable properties (compare to afforded by the environment change because your body is
Gärdenfors’s “detached” representation). As I noted, the changing. Soon, the keyboard no longer affords typing and
effort of suppression (or coordination) is a signal that the the chair no longer affords sitting: to a body such as yours,
conceptualization is in part imaginative, and hence, to use the only actions possible are related to food-seeking.
Ramsay & Homer’s word, “owned” by the agent. Alternat- Given these sorts of changes in bodily state and how they
ing between relative suppression (holding in mind) and can change conceptualization, it is easy to understand how
relative clamping (holding in view), provides a way of someone can be blinded by passion (no actions appear
considering the differences in meaning of the conceptual- possible except those directed toward the loved one) or
izations. I have not, however, considered how such an incapacitated by grief (no actions appear). Perhaps in-
alternation would take place. stances of clinical depression submit to a similar analysis.
The coordination between suppression and clamping is The depressed body registers few affordances so that few
also consistent with Solomon & Barsalou’s discussion of actions seem possible. Thus the depressed person spends
propositional construal. That is, the meaning of objects and an inordinate amount of time sleeping, and when awake, he
events depends on how we construe them, but what does it or she feels that nothing can be done. Changing the bodily
mean to construe? As Solomon & Barsalou note, meshing state through drug treatment relieves not just the feeling of
projectable properties of the object with action-based goals depression but the inactivity also. That is, to a different
provides one solution. When trying to move the furniture to body, actions now appear possible. How then can a talking

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 49

References/Glenberg: What memory is for

therapy have any effect? Cognitive therapies induce a mesh “everyday mind reading, coordination of action, detection
of projectable features and actions suggested by the thera- of cheating, identification of kinship or negotiation of
pist. That is, like the dinner companion who allows me to discourse.” Nelson asserts that “The symbol grounding
drink from his glass, the therapist’s words induce the patient problem cannot be solved entirely through embodied cog-
to conceptualize the current state as one in which actions nition but requires the recognition that words are socially
are possible. As another analogy, consider a tired novice shared symbols that somehow overcome the difficul-
(the patient) hiking in the woods with a guide (the thera- ties posed by the existence of individual experiential
pist). When the novice is faced with fording a river, he sees worlds.” I agree fully. Navigating the social environment is
no possibility for action, and he stops. The guide points out as critical as following paths and fording streams, and
that the arrangement of stones in the river afford dry- reading faces is as important as reading texts. Furthermore,
stepping and encourages action. All is well, unless the arguments in the target article for basing cognition on
novice is very tired (very depressed). In this case, the action in the physical environment (e.g., importance of
novice’s body tells him that even the stones pointed out by action to survival, joint evolution of brain and body, mesh of
the guide do not afford stepping for his body. Only a change projectable and nonprojectable properties) pertain with
in the body’s state will change the situation. equal force to taking action in the social world. I apologize
So far, the analysis has suggested how changes in affec- for the neglect, and I hope for success in extending the
tive state can control conceptualization and immediate framework in these directions.
behavior. Remaining questions concern long-term goals,
long-term planning, and taking current action designed to
accomplish those long-term goals, what Gärdenfors calls
“anticipatory planning.” If there is plenty of food around, References
and if action is controlled by conceptualization of the
current situation, why do we bother to plant (or shop) for Letters “a” and “r” appearing before authors’ inintials refer to target article
tomorrow? My proposal uses the same sort of mechanism and response respectively.
outlined above, but with a feedback loop (see Damasio
Aglioti, S., DeSouza, J. F. X. & Goodale, M. A. (1995) Size-contrast illusions
1994 for documentation and discussion of the neural mech- deceive the eye but not the hand. Current Biology 5:679–685. [YR]
anisms underlying such a loop). The argument has three Agre, P. E. & Chapman, D. (1987) Pengi: An implementation of a theory
parts. First, the current state of plenty (or deprivation) can, of activity. Proceedings of the AAAI Symposium on Parallel Models.
with the effort of suppression, be meshed with trajectories AAAI Press. [SF]
Albrecht, J. E., O’Brien, E. J., Mason, R. A. & Myers, J. L. (1995) The role
based on memory. That is, we can daydream about changes of perspective in the accessibility of goals during reading. Journal of
in the current situation. Second, this new conceptualization Experimental Psychology: Learning, Memory, and Cognition 21:364–
will produce characteristic changes in the body (see sects. 72. [CH]
2.3.1. and 2.3.2. in the target article; Ch. 6–10 in Damasio Allport, A. (1989) Visual attention. In: Foundations of cognitive science, ed. M. I.
1994; or simply consider how your own daydreaming can Posner. MIT Press. [CMM]
Alston, W. P. (1964) Philosophy of language. Prentice-Hall. [CS]
change your level of arousal). Third, the changed bodily Alterman, R. (1996) Everyday memory and activity. Behavioral and Brain
state changes conceptualization of patterns of action. Just as Sciences 19:189–90. [AK]
real hunger leads to real actions directed toward food- Anderson, J. R. (1979) Further arguments concerning representations for mental
seeking (because it appears that there is nothing else to do), imagery: A response to Hayes-Roth and Pylyshyn. Psychological Review
86:395–406. [GM]
imagined hunger, through its effects on the current body, Anderson, N. H. (1991a) Contributions to information integration theory: Vol. 1.
leads to real actions directed toward food-seeking. These Cognition, ed. N. H. Anderson. Erlbaum. [NHA]
actions can be characterized as planning for tomorrow. (1991b) Functional memory in person cognition. In: Contributions to
There are several ways in which this account is different information integration theory: Vol. 1. Cognition, ed. N. H. Anderson.
from the simple assertion that thinking about the future Erlbaum. [NHA]
(1996) A functional theory of cognition. Erlbaum. [NHA]
motivates action (although this, in fact, is what I am trying to (1996) Functional memory requires a quite different value metaphor.
explain), and different from accounts such as Anderson’s Behavioral and Brain Sciences 19:190–91. [AK]
cognitive algebra for computing value. First, the account Aristotle (1966) De memoria et reminiscentia. Transl. J. I. Beare. In: Source
provides a theoretical mechanism to translate thought into book in the history of psychology, ed. R. J. Herrnstein & E. G. Boring.
Harvard University Press. [CMM]
motivation: thought literally changes bodily state. Second, Baddeley, A. D. (1990) Human memory: Theory and practice. Allyn &
the account provides a mechanism for how motivation can Bacon. [aAMG, GM]
direct behavior: changes in bodily state change affordances. (1996) Human memory: Theory and practice, rev. ed. Erlbaum. [RHL]
Third, the account does not require the precomputation of Baddeley, A. D. & Hitch, G. J. (1974) Working memory. In: The psychology of
value (or as Gärdenfors would name it: a detached repre- learning and motivation, vol. 8, ed. G. Bower. Academic Press. [aAMG]
Barnsley, M. & Anson, L. (1993) The fractal transform. Jones & Bartlett. [DV]
sentation of a desire) for future outcomes. Fourth, although Barsalou, L. W. (1993) Flexibility, structure, and linguistic vagary in concepts:
the values of different pieces of information are taken into Manifestations of a compositional system of perceptual symbols. In: Theories
account and in some sense averaged, there is no manipula- of memories, ed. A. C. Collins, S. E. Gathercole & M. A. Conway.
tion of values or symbols by rules equivalent to those of Erlbaum. [arAMG]
(in preparation) Perceptual symbol systems. [KOS]
algebra. Instead of computing values like a digital computer Barsalou, L. W. & Prinz, J. J. (in press) Mundane creativity in perceptual symbol
following rules, the body computes value more akin to an systems. In: Conceptual structures and processes: Emergence, discovery, and
analog computer, using the integration, or mesh, of possi- change, ed. T. B. Ward, S. M. Smith & J. Vaid. American Psychological
bilities. Association. [TR, KOS]
Barsalou, L. W., Yeh, W., Luka, B. J., Olseth, K. L., Mix, K. S. & Wu, L-L.
(1993) Concepts and meaning. In: Chicago Linguistics Society 29: Papers
R3.5. Social interaction. Both Nelson and Slater object to from the parasession on conceptual representations, ed. K Beals, G. Cooke,
my relative neglect of the social world. As Slater describes D., Kathman, K. E. McCullough, S. Kita, & D. Testen. Chicago Linguistics
it, the human conceptual agenda is strongly influenced by Society, University of Chicago. [aAMG]

50 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 References/Glenberg: What memory is for

Barton, S. B. & Sanford, A. J. (1993) A case study of anomaly detection: Shallow Cimprich, B. (1992) A theoretical perspective on attention and patient education.
semantic processing and cohesion establishment. Memory and Cognition Advances in Nursing Science 14(3):39–51. [DAS]
21:477–87. [aAMG] Cohen, R. L. (1981) On the generality of some memory laws. Scandinavian
Barwise, J. & Etchemendy, J. (1995) Heterogenous logic. In: Diagrammatic Journal of Psychology 22:267–81. [aAMG]
reasoning: Cognitive and computational perspectives, ed. J. Glasgow, (1989) Memory for action events: The power of enactment. Educational
H. Narayanan & B. Chandrasekaran. MIT Press. 211–234. [CH] Psychology Review 1:57–80. [CMM]
Barwise, J. & Perry, J. (1982) Situations and attitudes. Bradford Books. [ACG] Collins, A. M. & Quillian, M. R. (1969) Retrieval time from semantic memory.
Bauer, P. J. (1996) What do infants recall of their lives? Memory for Journal of Verbal Learning and Verbal Behavior 8:240–47. [aAMG]
specifications by one- to two-year-olds. American Psychologist Conrad, C. (1972) Cognitive economy in semantic memory. Journal of
51:29–41. [KN] Experimental Psychology 92:149–154. [aAMG]
Berkowitz, L., Jo, E. & Troccoli, B. (1993) Feeling and knowing: Evidence Cooper, L. A., Schacter, D. L., Ballesteros, S. & Moore, C. (1992) Priming and
for different psychological processes in emotional reactions. recognition of transformed three-dimensional objects: Effects of size and
Manuscript. [aAMG] reflection. Journal of Experimental Psychology: Learning, Memory, and
Berkowitz, L. & Troccoli, B. T. (1990) Feelings, direction of attention, Cognition 18:43–57. [BMV]
and expressed evaluations of others. Cognition and Emotion Cowan, N. (1995) Attention and memory: An integrated framework. Oxford
4:305–25. [aAMG] University Press. [RGC]
Bernstein, N. A. (1996) Dexterity and its development. Erlbaum. [BMV] Cowley, S. J. & MacDorman, K. F. (1995) Simulating conversations: The
Beschin, N., Cocchini, G., Della Sala, S. & Logie, R. H. (in press) What the eyes communion game. AI and Society 9(3):116–37. [KFM]
perceive, the brain ignores: A case of pure unilateral representational Craik, F. I. M. & Lockhart, R. S. (1972) Levels of processing: A framework
neglect. Cortex. [RHL] for memory research. Journal of Verbal Learning and Verbal Behavior
Biederman, I. (1987) Recognition-by-components: A theory of human image 11:671–84. [aAMG, BMV]
understanding. Psychological Review 94:115–47. [CS] Crowder, R. G. (1993a) Short-term memory: Where do we stand? Memory and
(1988) Aspects and extensions of a theory of human image understanding. In: Cognition 21:142–45. [RGC]
Computational processes in human vision: An interdisciplinary perspective, (1993b) Systems and principles in memory theory: Another critique of pure
ed. S. Pylyshyn. Ablex. [CS] memory. In: Theories of memory, ed. A. F. Collins, S. E. Gathercole,
Blaxton, T. A. (1989) Investigating dissociations among memory measures: M. A. Conway & P. E. Morris. Erlbaum. [AK]
Support for a transfer-appropriate framework. Journal of Experimental Cuthbert, B. N., Vrana, S. R. & Bradley, M. M. (1991) Imagery: Function and
Psychology: Learning, Memory, and Cognition 15:657–68. [aAMG, ASB] physiology. In: Advances in psychophysiology, vol. 4, ed. P. K. Ackles,
Bloom, L. (1994) Meaning and expression. In: The nature and ontogenesis of J. R. Jennings & M. G. A. Coles. JAI Press. [aAMG]
meaning, ed. W. F. Overton & D. S. Palermo. Erlbaum. [aAMG] Damasio, A. R. (1994) Descartes’ error: Emotion, reason, and the human brain.
Bower, G. H. (1970) Analysis of a mnemonic device. American Scientist Avon Books. [rAMG]
58:496–510. [aAMG] de Vega, M., Rodrigo, M. J. & Zimmer, H. (in press) Pointing and labeling
Bowerman, M. (1982) Starting to talk worse: Clues to language acquisition from directions in egocentric frameworks. Journal of Memory and
children’s late speech errors. In: U-shaped behavioral growth, ed. S. Strauss. Language. [MdV]
Academic Press. [aAMG] Deacon, T. W. (1996) Prefrontal cortex and symbol learning: Why a brain
(1985) The spatial metaphor: A universal principle of semantic structuring? capable of language evolved only once. In: Communicating meaning:
Clues from children’s speech errors. Oxford University Psychology The evolution and development of language, ed. B. M. Velichkovsky &
Department Colloquium. [aAMG] D. M. Rumbaugh. Erlbaum. [BMV]
Bransford, J. D. (1979) Human cognition: Learning, understanding and Della Sala, S. & Logie, R. (1993) When working memory does not work:
remembering. Wadsworth. [aAMG] The role of working memory in neuropsychology. In: Handbook of
Bransford, J. D., McCarrell, N. S., Franks, J. J. & Nitsch, K. E. (1977) Toward neuropsychology, vol. 8, ed. F. Boller & H. Spinnler. Elsevier. [RHL]
unexplaining memory. In: Perceiving, acting, and knowing, ed. R. Shaw & Denis, M. & Cocude, M. (1989) Scanning visual images generated from
J. D. Bransford. Erlbaum. [AK] verbal descriptions. European Journal of Cognitive Psychology
Bridgeman, B. (1991) Complementary cognitive and motor image processing. 1:293–307. [aAMG, MdV]
In: Presbyopsia research: From molecular biology to visual adaptation, Drescher, G. L. (1988) Learning from experience without prior knowledge in a
ed. G. Obrecht & L. W. Stark. Plenum. [YR] complicated world. Proceedings of the AAAI Symposium on Parallel
Bridgeman, B., Kirch, M. & Sperling, K. (1981) Segregation of cognitive and Models. AAAI Press. [SF]
motor aspects of visual function using induced motion. Perception and (1991) Made-up minds. MIT Press. [SF]
Psychophysics 29:336–42. [BMV] Duchan, J. F., Bruder, G. A. & Hewitt, L. E., ed. (1995) Deixis in narrative:
Britton, B. K. & Graesser, A. C., ed. (1996) Models of understanding test. A cognitive science perspective. Erlbaum. [ACG]
Erlbaum. [ACG] Edelman, G. M. (1989) The remembered present: A biological theory of
Brooks, L. R. (1978) Nonanalytic concept formation and memory for instances. consciousness. Basic Books. [SF]
In: Cognition and categorization, ed. E. Rosch & B. Lloyd. Engelkamp, J. & Krumnacker, H. (1980) Imaginale und motorische Prozesse
Erlbaum. [RGC] beim Behalten verbalen Materials. Zeitschrift fur Experimentelle und
Brooks, R. A. (1991) Intelligence without representation. Artificial Intelligence Angewandte Psychologie 27:511–33. [aAMG]
47:139–59. [SF, DV] Englekamp, J. & Zimmer, H. D. (1989) Memory for action events: A new field
Brown, F. M., ed. (1987) The frame problem in artificial intelligence. Morgan of research. Psychological Research 51:153–57. [aAMG]
Kaufman. [SF] Epstein, W. (1993) The representational framework in perceptual theory.
Bruce, D. (1985) The how and why of ecological memory. Journal of Perception and Psychophysics 53:704–9. [aAMG]
Experimental Psychology: General 114:78–90. [AK] Farah, M. J. (1994) Neuropsychological inference with an interactive brain:
Bruce, V. & Green, P. R. (1985) Visual perception : Physiology, psychology and A critique of the “locality” assumption. Behavioral and Brain Sciences
ecology. Erlbaum. [aAMG] 17:43–61. [rAMG]
Bryant, D. J., Tversky, B. & Franklin, N. (1992) Internal and external spatial Fauconnier, G. (1985) Mental spaces: Aspects of meaning construction in natural
frameworks for representing described scenes. Journal of Memory and language. MIT Press. [rAMG]
Language 31:74–98. [aAMG] Fernandez, A. & Saiz, C. (1989) Foregrounding of trait information during text
Carlson, R. A. (1992) Starting with consciousness. American Journal of comprehension. Paper presented at The First European Congress of
Psychology 105:598–604. [RAC] Psychology, Amsterdam. [aAMG]
Carpenter, P. A., Miyake, A. & Just, M. A. (1994) Working memory constraints in Finke, R. A. (1986) Mental imagery and the visual system. In: The perceptual
comprehension: Evidence from individual differences, aphasia, and aging. world, ed. I. Rock. Freeman. [CS]
In: Handbook of psycholinguistics, ed. M. A. Gernsbacher. Academic Fiske, S. T. (1992) Thinking is for doing: Portraits of social cognition from
Press. [arAMG] daguerrotype to laserphoto. Journal of Personality and Social Psychology
Challis, B. H., Velichkovsky, B. M. & Craik, F. I. M. (1996) The levels-of- 63:877–89. [aAMG]
processing effects on a variety of memory tasks: New findings and Fodor, J. A. (1975) The language of thought. Crowell. [KFM]
theoretical implications. Consciousness and Cognition 5:142–64. [BMV] (1980) Methodological solipsism considered as a research strategy in cognitive
Chapman, M. (1988) Constructive evolution: Origins and development of Piaget’s psychology. Behavioral and Brain Sciences 3:63–109. [TPM]
thought. Cambridge University Press. [RAC] Franklin, S. (1995) Artificial minds. MIT Press. [SF]
Chesterton, G. K. (1994) The logic of Elfland. In: Great essays in science, ed. Friedman, W. J. (1993) Memory for the time of post events. Psychological
M. Gardner. Prometheus Books. [AMJ] Bulletin 113:44–66. [aAMG]

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 51

References/Glenberg: What memory is for
Funahashi, S., Bruce, C. J. & Goldmam-Rakic, P. S. (1993) Dorsolateral the thalidomide child. In: Determinants of infant behavior, vol. 4, ed. B. M.
prefrontal lesions and oculomotor delayed-response performance: Evidence Foss. Methuen. [BMV]
for mnemonic “scotomas.” The Journal of Neuroscience 13:1479–97. [YR] Graesser, A. C., Singer, M. & Trabasso, T. (1994) Constructing inferences
Gabrieli, J. D. E., Fleischman, D. A., Keane, M. M., Reminger, S. L. & Morrell, during narrative text comprehension. Psychological Review 101:371–95.
F. (1995) Double dissociation between memory systems underlying explicit [ACG, CH]
and implicit memory in the human brain. Psychological Science Greene, R. L. (1986) Sources of recency effects in free recall. Psychological
6:76–82. [TPM] Bulletin 99:221–28. [aAMG]
Gallistel, C. R. & Gelman, R. (1992) Preverbal and verbal counting and (1992) Human memory. Erlbaum. [aAMG]
computation. Cognition 44:43–74. [CH] Gulz, A. (1991) The planning of action as a cognitive and biological
Garcia, J. & Koelling, R. A. (1966) Relation of cue to consequence in avoidance phenomenon. Lund University Cognitive Studies 2. [PG]
learning. Psychonomic Science 4:123–24. [aAMG] Habel, C. & Eschenbach, C. (1995) Abstrakte Räumlichkeit in der Kognition.
Gärdenfors, P. (1995a) Cued and detached representations in animal cognition. Kognitionswissenschaft 4:171–76. [CH]
Manuscript submitted for publication. [aAMG] Harnad, S. (1990) The symbol grounding problem. Physica D
(1995b) Speaking about the inner environment. In: Of thoughts and words, 42:335–46. [aAMG, ASB, MdV, KFM, TPM, KN, KOS]
ed. S. Allen. Imperial College Press. [PG] (1993) Grounding symbols in the analog world with neural nets. Think
(1996) Cued and detached representations in animal cognition. Behavioral 2:12–78. [aAMG]
Processes 36:263–73. [PG] Haviland, S. E. & Clark, H. H. (1974) What’s new: Acquiring new information as
Garner, W. R. (1974) The processing of information and structure. a process in comprehension. Journal of Verbal Learning and Verbal
Erlbaum. [DV] Behavior 13:512–21. [aAMG]
Gentilucci, M., Chieffi, S., Daprati, E., Saetti, M. C. & Toni, I. (1996) Visual Hebb, D. O. (1949/1961) The organization of behavior: A neuropsychological
illusion and action. Neuropsychologia 34:369–76. [YR] theory. Science Editions. [RGC]
Gernsbacher, M. A., ed. (1994) Handbook of psycholinguistics. Academic Heider, F. (1958) Perceiving the other person. In: Person perception and
Press. [ACG] interpersonal behavior, ed. R. Tagiuri & L. Petrullo. Stanford University
Gernsbacher, M. A. & Shroyer, S. (1989) The cataphoric use of the indefinite Press. [aAMG]
this in spoken narratives. Memory and Cognition 17:536–40. [rAMG] Hess, D. J., Foss, D. J. & Carroll, P. (1995) Effects of global and local context
Gernsbacher, M. A., Varner, K. R. & Faust, M. E. (1990) Investigating general on lexical processing during language comprehension. Journal of
comprehension skill. Journal of Experimental Psychology: Learning, Experimental Psychology: General 124:62–82. [aAMG]
Memory, and Cognition 16:430–45. [aAMG] Hintzman, D. L. (1967) Articulatory coding in short-term memory. Journal of
Gibson, E. J. (1993) Ontogenesis of the perceived self. In: The perceived self: Verbal Learning and Verbal Behavior 6:312–16. [aAMG]
Ecological and interpersonal sources of self-knowledge, ed. U. Neisser. (1986) “Schema abstraction” in a multiple-trace memory model. Psychological
Cambridge University Press. [RAC] Review 93:411–28. [arAMG, RGC]
Gibson, J. J. (1979) The ecological approach to visual perception. Houghton- Howe, M. L. & Courage, M. L. (1993) On resolving the enigma of infantile
Mifflin. [aAMG, RAC, RGC, TPM] amnesia. Psychological Bulletin 113:305–26. [aAMG]
Gigerenzer, G. (1991) From tools to theories: A heuristic of discovery in Hunt, R. R. (1995) The subtlety of distinctiveness: What von Reistorff really did.
cognitive psychology. Psychological Review 98:254–67. [KOS] Psychonomic Bulletin and Review 2:105–12. [aAMG]
Gillund, G. & Shiffrin, R. M. (1984) A retrieval model for both recognition and Hunt, R. R. & Elliott, J. M. (1980) The role of nonsemantic information in
recall. Psychological Review 91:1–67. [aAMG] memory: Orthographic distinctiveness effects upon retention. Journal of
Gippenreiter, Y. B. & Romanov, V. Y. (1972) A method of investigation of the Experimental Psychology: General 109:49–74. [aAMG]
internal form of visual activity. In: Perception: Essays in honor of James J. Jacoby, L. L. (1983) Remembering the data: Analyzing interactive processes in
Gibson, ed. R. MacLeod & H. L. Pick, Jr. Cornell University reading. Journal of Verbal Learning and Behavior 22:485–508. [ASB]
Press. [BMV] Jacoby, L. L. & Brooks, L. R. (1984) Nonanalytic cognition: Memory, perception,
Glenberg, A. M. (1984) A retrieval account of the long-term modality effect. and concept learning. In: The psychology of learning and motivation: Vol.
Journal of Experimental Psychology: Learning, Memory, and Cognition 18. Advances in research and theory, ed. G. H. Bower. Academic
10:16–31. [aAMG] Press. [RGC, CMM]
Glenberg, A. M., Kruley, P. & Langston, W. E. (1994) Analogical processes in Jacoby, L. L. & Dallas, M. (1981) On the relationship between autobiographical
comprehension: Simulation of a mental model. In: Handbook of memory and perceptual learning. Journal of Experimental Psychology:
psycholinguistics, ed. M. A. Gernsbacher. Academic Press. [aAMG] General 110:306–40. [aAMG]
Glenberg, A. M. & Mathew, S. (1992) When minimalism is not enough: Jacoby, L. L. & Kelly, C. M. (1987) Unconscious influences of memory for a
Mental models in reading comprehension. Psycoloquy.92.3.64.reading- prior event. Personality and Social Psychology Bulletin 13:314–36. [AK]
inference-2.1. [aAMG] Jacoby, L. L., Toth, J. P. & Yonelinas, A. P. (1993) Separating conscious and
Glenberg, A. M., Meyer, M. & Lindem, K. (1987) Mental models contribute to unconscious influences of memory: Measuring recollection. Journal of
foregrounding during text comprehension. Journal of Memory and Learning Experimental Psychology: General 122:139–54. [aAMG]
26:69–83. [aAMG] James, W. (1890/1981) The principles of psychology. Harvard University
Glenberg, A. M., Schroeder, J. & Robertson, D. (1995) Remembering behavior. Press. [DAS]
Manuscript in preparation. [aAMG] Jeannerod, M. & Rossetti, Y. (1993) Visuomotor coordination as a dissociable
Glenberg, A. M., Smith, S. M. & Green, C. (1977) Type I rehearsal: function: Experimental and clinical evidence. In: Visual perceptual defects,
Maintenance and more. Journal of Verbal Learning and Verbal Behavior ed. C. Kennard. Balliere Tindall. [YR]
16:339–52. [aAMG] Johnson, M. (1987) The body in the mind: The bodily basis of meaning,
Glenberg, A. M. & Swanson, N. C. (1986) A temporal distinctiveness theory of imagination, and reason. University of Chicago Press. [aAMG]
recency and modality effects. Journal of Experimental Psychology: Learning, Johnson, M. & Scanlon, R. (1987) Experience with a feeling-thinking machine.
Memory, and Cognition 12:3–15. [aAMG] Proceedings of the IEEE First International Conference on Neural
Goldberg, A. E. (1995) Constructions: A construction grammar approach to Networks. San Diego. [SF]
argument structure. University of Chicago Press. [rAMG] Johnson-Laird, P. N. (1987) The mental representation of the meaning of words.
Goodale, M. A., Jakobson, L. S. & Keillor, J. M. (1994) Differences in the visual Cognition 25:189–211. [AMJ]
control of pantomimed and natural grasping movements. Neuropsychologia (1989) Mental models. In: Foundations of cognitive science, ed. M. I. Posner.
32:1159–78. [YR] MIT Press. [aAMG]
Goodale, M. A. & Milner, A. D. (1992) Separate visual pathways for perception Just, M. & Carpenter, P. (1992) A capacity theory of comprehension: Individual
and action. Trends in Neurosciences 15:20–25. [YR] differences in working memory. Psychological Review 99:122–49. [RHL]
Goodale, M. A., Milner, A. D., Jacobson, L. S. & Carey, D. P. (1991) A Kalish, C. W. (1995) Essentialism and graded membership in animal and artifact
neurological dissociation between perceiving objects and grasping them. categories. Memory and Cognition 23:335–53. [aAMG]
Nature 154–56. [YR] Kanerva, P. (1988) Sparse distributed memory. MIT Press. [SF]
Goodale, M. A. & Servos, P. (1992) Now you see it, now you don’t: How Kaplan, R. & Kaplan, S. (1989) The experience of nature: A psychological
delaying an action system can tronsform a theory. Behavioral and Brain perspective. Cambridge University Press. [DAS]
Sciences 15:335–36. [YR] Kaplan, S. (1995) The restorative benefits of nature: Toward an integrative
Gopnik, A. (1993) How we know our minds: The illusion of first-person framework. Journal of Environmental Psychology 15(3):169–82. [DAS]
knowledge of intentionality. Behavioral and Brain Sciences 16:1–14. Karn, K. S. & Zelinsky, G. J. (1996) Driving and dish washing: Failure of the
[PG] correspondence metaphor for memory. Behavioral and Brain Sciences
Gouin-Decarie, T. (1969) A study of the mental and emotional development of 19:198. [AK]

52 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 References/Glenberg: What memory is for

Katz, J. J. (1972) Semantic theory. Harper & Row. [TPM] Mandler, J. (1990) A new perspective on cognitive development in infancy.
Kaup, B., Kelter, S. & Habel, H. (1996) The resolution of plural pronouns American Scientist 78:236–43. [CS]
during discourse comprehension. Graduiertenkolleg Kognitionswissenschaft Marr, D. (1982) Vision: A computational investigation into the human
Hamburg 55. [CH] representation and processing of visual information. Freeman.
Keefe, D. E. & McDaniel, M. A. (1993) The time course and durability [CH, CS, TR]
of predictive inferences. Journal of Memory and Language Masson, M. E. J. (1995) A distributed memory model of semantic priming.
32:446–63. [aAMG] Journal of Experimental Psychology: Learning, Memory, and Cognition
Kelly, G. A. (1955) The psychology of personal constructs, vol. 1 & 2. 21:3–23. [aAMG]
Norton. [AK] Masson, M. E. J. & MacLeod, C. M. (1996) Contributions of processing fluency
Keppel, G. & Underwood, B. J. (1962) Proactive inhibition in short-term to repetition effects in masked word identification. Canadian Journal of
retention of single items. Journal of Verbal Learning and Verbal Behavior Experimental Psychology 50:9–21. [CMM]
1:153–61. [aAMG] Maturana, H. R. (1975) The organization of the living: A theory of the
Keysar, B. & Bly, B. (1995) Intuitions of the transparency of idioms: Can one living organization. International Journal of Man-Machine Studies
keep a secret by spilling the beans? Journal of Memory and Language 7:313–32. [SF]
34:89–109. [rAMG] Maturana, H. R. & Varela, F. (1980) Autopoiesis and cognition: The realization
Kintsch, W. (1988) The role of knowledge in discourse comprehension: A of the living. Reidel. [SF]
construction-integration model. Psychological Review 95:163–82. [aAMG] McArthur, L. S. & Baron, R. M. (1990) Toward an ecological theory of social
Klatzky, R. L., Pellegrino, J. W., McCloskey, B. P. & Doherty, S. (1989) Can you perception. Psychological Review 90:215–247. [CS]
squeeze a tomato? The role of motor representations in semantic sensibility McClelland, J. L. & Rumelhart, D. E. (1986) A distributed model of human
judgments. Journal of Memory and Language 28:56–77. [aAMG] learning and memory. In: Parallel distributed processing: Explorations in the
Kohonen, T. (1990) The self-organizing map. Proceedings of the IEEE microstructure of cognition: Vol. 2. Psychological and biological models, ed.
78(9):1464–80. [DV] J. L. McClelland, D. E. Rumelhart & the PDP Research Group. MIT
Kolers, P. A. & Roediger, H. L., III (1984) Procedures of mind. Journal of Press. [aAMG]
Verbal Learning and Verbal Behavior 23:425–49. [aAMG, RGC, AK, McCutchen, D. & Perfetti, C. A. (1982) The visual tongue-twister: Phonological
CMM] activation in silent reading. Journal of Verbal Learning and Verbal Behavior
Komatsu, L. K. (1992) Recent views of conceptual structure. Psychological 21:627–87. [aAMG]
Bulletin 112(3):500–26. [DV] McKoon, G. & Ratcliff, R. (1986a) Automatic activation of episodic information
Koriat, A. & Goldsmith, M. (1994) Memory in naturalistic and laboratory in a semantic memory task. Journal of Experimental Psychology: Learning,
contexts: Distinguishing the accuracy-oriented and quantity-oriented Memory, and Cognition 12:108–15. [aAMG]
approaches to memory assessment. Journal of Experimental Psychology: (1986b) Inferences about predictable events. Journal of Experimental
General 123:297–316. [AK] Psychology: Learning, Memory, and Cognition 12:82–91. [aAMG]
(1996a) Memory metaphors and the real-life/laboratory controversy: (1992) Inference during reading. Psychological Review 99:440–66. [aAMG]
Correspondence versus storehouse conceptions of memory. Behavioral and McKoon, G., Ratcliff, R. & Dell, G. S. (1986) A critical evaluation of the
Brain Sciences 19:167–88. [AK] semantic-episodic distinction. Journal of Experimental Psychology: Learning,
(1996b) Monitoring and control processes in the strategic regulation of Memory, and Cognition 12:1173–90. [aAMG]
memory accuracy. Psychological Review 103:490–517. [AK] McNamara, T. P. (1986) Mental representation of spatial relations. Cognitive
(1996c) The correspondence metaphor of memory: Right, wrong or useful? Psychology 18:87–121. [aAMG]
Behavioral and Brain Sciences 19:211–28. [AK] McNamara, T. P., Hardy, J. K. & Hirtle, S. C. (1989) Subjective hierarchies in
Krauss, R. (1952) A hole is to dig. Harper & Row. [CS] spatial memory. Journal of Experimental Psychology: Learning, Memory,
Kruschke, J. K. (1992) alcove: An exemplar-based connectionist model of and Cognition 15:211–27. [aAMG]
category learning. Psychological Review 99(1):22–44. [DV, rAMG] Medin, D. L. & Schaffer, M. M. (1978) Context theory and classification
Lakoff, G. (1987) Women, fire, and dangerous things: What categories reveal learning. Psychological Review 85:207–38. [RGC]
about the mind. University of Chicago Press. [aAMG, AMJ] Mel, B. (1988) MURPHY: A robot that learns by doing. Neural Information
(1988) Smolensky, semantics, and the sensorimotor system. Behavioral and Processing Systems 544–53. [TR]
Brain Sciences 11:39–40. [aAMG] (1989) Further explorations in visually-guided reaching: Making murphy
Lakoff, G. & Johnson, M. (1980) Metaphors we live by. University of Chicago smarter. Advances in Neural Information Processing Systems
Press. [aAMG, MdV] 1:348–55. [TR]
Lang, P. J. (1979) A bio-informational theory of emotional imagery. Metcalfe, J. (1993) Novelty monitoring, metacognition, and control in a
Psychophysiology 16:495–512. [aAMG] composite holographic associative recall model: Implications for Korsakoff
Langacker, R. W. (1986) An introduction to cognitive grammar. Cognitive Science amnesia. Psychological Review 100:3–22. [aAMG]
10:1–40. [KOS] Meyer, D. E. & Schvaneveldt, R. W. (1971) Facilitation in recognizing pairs of
Langston, W., Kramer, D. C. & Glenberg, A. M. (in press) The representation words: Evidence of a dependence between retrieval operations. Journal of
of space in mental models derived from text. Memory & Cognition. Experimental Psychology 90:227–34. [aAMG]
[aAMG] Miller, G. A. & Johnson-Laird, P. N. (1976) Language and perception. Harvard
Leslie, A. M. (1987) Pretense and representation: The origins of “theory of University Press. [ACG]
mind.” Psychological Review 94:412–26. [PG] Milner, A. D. & Goodale, M. A. (1994) Visual pathways to perception and action.
Leyton, M. (1992) Symmetry, causality, mind. MIT Press. [DV] In: Progress in brain research, ed. T. P. Hicks, S. Molotchnifoff & T.
Locke, J. (1960/1975) An essay concerning humane understanding, ed. P. H. Ono. [YR]
Nidditch. Clarendon [RHL] (1995) The visual brain in action. Oxford University Press. [YR, BMV]
Logan, G. D. & Cowan, W. B. (1984) On the ability to inhibit thought and Montello, D. R. & Presson, C. C. (1993) Movement and orientation in
action: A theory of an act of control. Psychological Review 91(3)295– surrounding and imaginal spaces. Manuscript. [aAMG]
327. [DAS] Morrow, D. G., Bower, G. H. & Greenspan, S. L. (1989) Updating situation
Logan, G. D. (1988) Toward an instance theory of automatization. Psychological models during narrative comprehension. Journal of Verbal Learning and
Review 95:492–527. [aAMG, CMM] Verbal Behavior 28:292–312. [aAMG]
Logie, R. H. (1995) Visuo-spatial working memory. Erlbaum. [RHL] Murdock, B. B., Jr. (1974) Human memory: Theory and data. Wiley. [aAMG]
Logie, R. H., Della Sala, S., Laiacona, M., Chalmers, P. & Wynn, V. (1996) (1993) todam2: A model for the storage and retrieval of item, associative,
Group aggregates and individual reliability: The case of verbal short-term and serial-order information. Psychological Review 100:183–203.
memory. Memory and Cognition 24:305–21. [rAMG] [CMM, rAMG]
Lucariello, J., Kyratzis, A. & Nelson, K. (1992) Taxonomic knowledge: What kind Murray, J. D., Klin, C. M. & Meyers, J. L. (1993) Forward inferences in
and when. Child Development 63:978–98. [KN] narrative text. Journal of Memory and Language 32:464–73. [aAMG]
MacDorman, K. F. (1996) How to ground symbols adaptively. In: Readings in Musen, G. & Squire, L. R. (1991) Normal acquisition of novel verbal
computation, content and consciousness, ed. S. O’Nuallain, P. McKevitt & information in amnesia. Journal of Experimental Psychology: Learning,
E. MacAogain. Benjamins. [KFM] Memory, and Cognition 17:1095–1104. [arAMG]
Maes, P. (1990) How to do the right thing. Connection Science 1:3. [SF] Muter, D. (1980) Very rapid forgetting. Memory and Cognition
Malter, A. J. (in press) An introduction to embodied cognition: Implications 8:174–79. [aAMG]
for consumer research. In: Advances in consumer research, vol. 23, Naveh-Benjamin, M. & Jonides, J. (1984) Maintenance rehearsal: A two-
ed. K. P. Corfman & J. Lynch. Association for Consumer Research. component analysis. Journal of Experimental Psychology: Learning, Memory,
[aAMG] and Cognition 10:369–85. [aAMG]

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 53

References/Glenberg: What memory is for
Neisser, U. (1976) Cognition and reality. Freeman. [BMV] following muscimol injections in macaque prefrontal cortex. Unpublished
(1982) Memory observed: Remembering in natural contexts, ed. U. Neisser. manuscript. [YR]
Freeman. [NHA] Putnam, H. (1981) Reason, truth, and history. Cambridge University
(1988) The ecological approach to perception and memory. New Trends in Press. [aAMG, AMJ]
Experimental and Clinical Psychiatry 4:153–66. [AK] Pylyshyn, Z. W. (1973) What the mind’s eye tells the mind’s brain: A critique of
(1991) Two perceptually given aspects of the self and their development. mental imagery. Psychological Bulletin 80:1–24. [KOS]
Developmental Review 11:197–209. [RAC] Ratcliff, R. & McKoon, G. (1993). Bias in implicit memory tasks. Paper
(1996) Remembering as doing. Behavioral and Brain Sciences presented at the meeting of the Psychonomic Society, Washington,
19:203–4. [AK] D.C. [aAMG]
Nelson, K. (1974) Concept, word and sentence: Interrelations in acquisition and Ratcliff, R., McKoon, G. & Verwoerd, M. (1989) A bias interpretation of
development. Psychological Review 81:267–85. [KN] facilitation in perceptual identification. Journal of Experimental Psychology:
(1986) Event knowledge: Structure and function in development. Learning, Memory, and Cognition 15:378–87. [aAMG]
Erlbaum. [KN] Regier, T. (1995) A model of the human capacity for categorizing spatial
(1988) Where do taxonomic categories come from? Human Development relations. Cognitive Linguistics 6:63–88. [aAMG]
31:3–10. [KN] (1996) The human semantic potential: Spatial language and constrained
(1993a) The psychological and social origins of autobiographical memory. connectionism. MIT Press. [TR]
Psychological Science 4:7–14. [aAMG] Reisberg, D., Rappaport, I. & O’Shaughnessy, M. (1984) Limits of working
(1993b) Explaining the emergence of autobiographical memory in early memory: The digit digit-span. Journal of Experimental Psychology:
childhood. In: Theories of memory, ed. A. Collins, M. Conway, S. Learning, Memory and Cognition 10:203–21. [aAMG]
Gathercole & P. Morris. Erlbaum. [KN] Richardson, J. T. E., Engle, R. W., Hasher, L., Logie, R. H., Stoltzfus, E. R. &
(1996) Language in cognitive development: The emergence of the mediated Zucks, R. T. (1996) Working memory and human cognition. Oxford
mind. Cambridge University Press. [KN] University Press. [RHL]
Norman, D. A. (1993) Things that makes us smart. Addison-Wesley. [BMV] Rieser, J. J. (1989) Access to knowledge of spatial structure from novel points of
Nosofsky, R. M. (1988) Exemplar-based accounts of relations between observation. Journal of Experimental Psychology: Learning, Memory, and
classification, recognition and typicality. Journal of Experimental Psychology: Cognition 15:1157–65. [TPM]
Learning, Memory, and Cognition 14:700–8. [RGC] Rieser, J. J., Garing, A. E. & Young, M. F. (1994) Imagery, action and young
O’Brien, E. J. & Albrecht, J. E. (1992) Comprehension strategies in the children’s spatial orientation: It’s not being there that counts, it’s what one
development of a mental model. Journal of Experimental Psychology: has in mind. Child Development 45:1043–56. [aAMG]
Learning, Memory, and Cognition 18:777–84. [aAMG] Rinck, M. & Bower, G. H. (1995) Anaphora resolution and the focus of attention
Oden, G. C. (1984) Integration of fuzzy linguistic information in language in situation models. Journal of Memory and Language 34:110–31. [aAMG]
comprehension. Fuzzy Sets and Systems 14:29–41. [aAMG] Roediger, H. L. (1980) Memory metaphors in cognitive psychology. Memory and
Oden, G. C. (1987) Concept, knowledge, and thought. Annual Review of Cognition 8:231–46. [AK]
Psychology 38:203–27. [aAMG] (1990) Implicit memory: Retention without remembering. American
Ohman, A., Fredrikson, M., Hugdahl, K. & Rimmo, P-A. (1976) The premise of Psychologist 45:1043–56. [aAMG, GM]
equipotentiality in human classical conditioning: Conditioned electrodermal Roediger, H. L., III., Guynn, M. J. & Jones, T. C. (1994) Implicit memory:
responses to potentially phobic stimuli. Journal of Experimental Psychology: A tutorial review. In: International perspectives on psychological science:
General 105:313–37. [aAMG] The state of the art, vol. 2, ed. G. d’Ydewelle, P. Eelen & P. Bertelson.
Olson, D. R. (1993) The development of representations: The origin of mental Erlbaum. [arAMG]
life. Canadian Psychology 34:1–14. [JTR] Rosch, E. (1973) Natural categories. Cognitive Psychology 4:328–50. [aAMG,
Oyama, S. (1985) The ontogeny of information. Cambridge University AMJ]
Press. [SF] Rossetti, Y. (in press) Implicit perception in action: Short lived motor
Palmer, J., MacLeod, C. M., Hunt, E. & Davidson, J. E. (1985) Information representations, In: Consciousness and brain circuitry, ed. P. Grossenbacher.
processing correlates of reading. Journal of Memory and Language Benjamins. [YR]
24:59–88. [CMM] Rossetti, Y., Gaunet, F. & Thinus-Blanc, C. (1996) Early visual experience affects
Palmer, S. E. (1991) Goodness, gestalt, groups and Garner: Local symmetry memorization and spatial representation of proprioceptive targets.
subgroups as a theory of figural goodness. In: The perception of structure, Neuroreport 7(6):1219–23. [YR]
ed. G. R. Lockhead & J. R. Pomerantz. American Psychological Rossetti, Y. & Regnier, C. (1995) Representations in action: Pointing to a target
Association. [DV] with various representations. In: Studies in perception and action 3, ed.
Palmer, S. E. (1978). Fundamental aspects of cognitive representation. In: B. G. Bardy, R. J. Bootsma & Y. Guiard. Erlbaum Associates. [YR]
Cognition and categorization, ed. E. Rosch & B. Lloyd. Erlbaum. [aAMG] Rumelhart, D. E., Smolensky, P., McClelland, J. L. & Hinton, G. E (1986)
Pellizzer, G. & Georgopoulos, A. P. (1993) Mental rotation of the intended Schemata and sequential thought processes in PDP models. In: Parallel
direction of movement. Current Directions in Psychological Science distributed processing: Explorations in the microstructure of cognition: vol.
2:12–17. [TPM] 2. Psychological and biological models, ed. J. L. McClelland, D. E.
Pepperberg, I. (1983) Cognition in the African grey parrot: Preliminary evidence Rumelhart & the PDP Research Group. MIT Press. [arAMG]
for auditory/vocal comprehension of the class concept. Animal Learning and Rumelhart, D. E. & Todd, P. M. (1993) Learning and connectionist
Behavior 11:179–85. [SF] representations. In: Attention and performance 14: Synergies in experimental
Perenin, M.-T. & Rossetti, Y. (1996) Grasping in an hemianopic field. Another psychology, artificial intelligence, and cognitive neuroscience, ed. D. E.
instance of dissociation between perception and action. Neuroreport Meyer & S. Kornblum. MIT Press. [DV]
7(3):793–97. [YR] Russell, J. (1996) Development and evolution of the symbolic function: The role
Perenin, M.-T. & Vighetto, A. (1988) Optic ataxia: A specific disruption in of working memory. In: Modeling the early human mind, ed. P. A. Mellars
visuomotor mechanisms: 1. Different aspects of the deficit in reaching for & K. R. Gibson. Cambridge: MacDonald Institute for Archaeological
objects. Brain 111:643–74. [YR] Research. [JTR]
Perner (1991) Understanding the representational mind. MIT Press. [JTR] Saltz, E. & Donnenwerth-Nolan, S. (1981) Does motoric imagery facilitate
Petitot, J. (1990) Le physique, le morphologique, le symbolique. Remarques sur memory for sentences? A selective interference test. Journal of Verbal
la vision. Revue de Synthese 4:139–83. [YR] Learning and Verbal Behavior 20:322–32. [aAMG]
Piaget, J. (1968) Genetische Erkenntnistheorie. Suhrkamp. [AMJ] Sanford, A. J. & Moxey, L. M. (1995) Aspects of coherence in written language:
Plunkett, K., Sinha, C., Möller, M. F. & Strandsby, O.(1992) Symbol grounding A psychological perspective. In: Coherence in spontaneous text, ed. M. A.
or the emergence of symbols? Vocabulary growth in children and a Gernsbacher & T. Givón. Benjamins. [aAMG]
connectionist net. Connection Science 4:293–312. [aAMG] Sarter, M., Bernston, G. G. & Cacioppo, J. T. (1996) Brain imaging and cognitive
Popper, K. R. (1935) Logik der Forschung. Springer-Verlag. [AMJ] neuroscience: Toward strong inference in attributing function to structure.
Posner, M. I. & Keele, S. W. (1968) On the genesis of abstract ideas. Journal of American Psychologist 51:13–21. [rAMG]
Experimental Psychology 77:353–63. [aAMG] Schacter, D. L., Cooper, L. A. & Delaney, S. M. (1990) Implicit memory for
Premack, D. & Woodruff, G. (1978) Does the chimpanzee have a theory of unfamiliar objects depends on access to structural descriptions. Journal
mind? Behavioral and Brain Sciences 1:515–26. [PG] of Experimental Psychology: Human Learning and Memory
Prinz, J. J. & Barsalou, L. W. (in press) Acquisition and productivity in 16:5–24. [arAMG]
perceptual symbol systems: An account of mundane creativity. In: Creativity Schacter, D. L. & Tulving, E. (1994) What are memory systems of 1994?
and computation, ed. D. M. Peterson & T. Dartnall. MIT Press. [KOS] In: Memory systems, ed. D. L. Schacter & E. Tulving. MIT Press.
Procyk, E., Tanaka, Y. & Joseph, J.-P. (1996) Spatial problem solving impairments [BMV]

54 BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1

 References/Glenberg: What memory is for

Schank, R. C. & Abelson, R. P. (1977) Scripts, plans, goals, and understanding. Tannenhaus, M. K., Spivey-Knowlton, M. J., Eberhard, K. M. & Sedivy, J. C.
Erlbaum. [KN, rAMG] (1995) Integration of visual and linguistic information in spoken language
Schneirla, T. C. (1959) An evolutionary and developmental theory of biphasic comprehension. Science 268:1632–34. [aAMG]
processes underlying approach and withdrawal. Nebraska Symposium on Taylor, H. A. & Tversky, B. (1992). Descriptions and depictions of environments.
Motivation 7:1–42. [aAMG] Memory and Cognition 20:483–96. [aAMG]
Schunn, C. D. & Dunbar, K. (1996) Priming, analogy, and awareness in complex Thorndike, E. L. (1932) Fundamentals of learning. Teacher’s College
reasoning. Memory and Cognition 24:271–84. [rAMG] Press. [aAMG]
Searle, J. R. (1980) Minds, brains and programs. Behavioral and Brain Sciences Tooby, J. & Cosmides, L. The psychological foundations of culture. In: The
3:417–24. [aAMG] adapted mind: Evolutionary psychology and the generation of culture,
Shanks, D. R. & Gluck, M. A. (1994) Tests of an adaptive network model for the ed. J. H. Barkow, L. Cosmides & T. Tooby. Oxford University Press. [CS]
identification and categorization of continuous-dimension stimuli. Tulving, E. (1983) Elements of episodic memory. Oxford University
Connection Science 6(1):59–89. [DV] Press. [aAMG, RGC]
Shanon, B. (1988) Semantic representation of meaning: A critique. Psychological (1984) Precis of Elements of episodic memory. Behavioral and Brain Sciences
Bulletin 104:70–83. [arAMG] 7:223–68. [aAMG]
Sharkey, A. J. C. & Sharkey, N. E. (1992) Weak contextual constraints in text and Tulving, E. & Schacter, D. L. (1990) Priming and human memory systems.
word priming. Journal of Memory and Language 31:543–72. [aAMG] Science 247:301–6. [aAMG]
Shelton, J. R. & Martin, R. C. (1992) How semantic is automatic semantic Tversky, B. & Hemenway, K. (1984). Objects, parts, and categories. Journal of
priming? Journal of Experimental Psychology: Learning, Memory and Experimental Psychology: General 113:169–97. [aAMG]
Cognition 18:1191–1210. [aAMG] Vallar, G. & Baddeley, A. D. (1984) Fractionation of working memory:
Shepard, R. N. (1984) Ecological constraints on internal representation: Neuropsychological evidence for a phonological short-term store. Journal of
Resonant kinematics of perceiving, imagining, thinking and dreaming. Verbal Learning and Verbal Behavior 23:151–61. [RHL]
Psychological Review 91:417–47. [RAC] Van den Bergh, O., Vrana, S. & Eelen, P. (1990). Letters from the heart:
(1987a) Toward a universal law of generalization for psychological science. Affective categorization of letter combinations in typists and nontypists.
Science 237:1317–23. [DV] Journal of Experimental Psychology: Learning, Memory, and Cognition
(1987b) Evolution of a mesh between principles of the mind and regularities 16:1153–61. [aAMG]
of the world. In: The latest on the best: Essays on evolution and optimality, Van der Heijden, A. H. C. (1991) Selective attention in vision. Routledge. [CH]
ed. J. Dupre. MIT Press. [DV] Varela, F. I., Thompson, E. & Rosch, E. (1991) The embodied mind. MIT
Shepard, R. N. (1988) How fully should connectionism be activated? Two Press. [SF]
sources of excitation and one of inhibition. Behavioral and Brain Sciences Varley, W. H., Levin, J. R., Severson, R. A. & Wolff, P. (1974) Training imaging
11:52. [aAMG] production in young children through motor involvement. Journal of
Shiffrar, M., Heptulla, S., O’Shaughnessy, M. & Freyd, J. (1993) What does it Educational Psychology 66:262–66. [aAMG]
mean to be sensitive to biological motion? Paper presented at the meeting Velichkovsky, B. M. (1990) The vertical dimension of mental functioning.
of the Psychonomic Society, Washington, D.C., November. [aAMG] Psychological Research 52:282–89. [BMV]
Shulman, H. G. (1972) Semantic confusion errors in short-term memory. Journal Watkins, O. C. & Watkins, M. J. (1975) Build-up of proactive inhibition as a cue-
of Verbal Learning and Verbal Behavior 11:221–27. [aAMG] overload effect. Journal of Experimental Psychology: Human Learning and
Skarda, C. A. & Freeman, W. J. (1987) How brains make chaos in order to make Memory 1:442–52. [aAMG]
sense of the world. Behavioral and Brain Sciences 10(2):161–95. [SF] Weldon, M. S. & Roediger, H. L. (1987) Altering retrieval demands reverses the
Skinner, B. F. (1957) Verbal behavior. Appleton-Century-Crofts. [CMM] picture superiority effect. Memory and Cognition 15:269–80. [aAMG]
Slamecka, N. J. & Graf, P. (1978) The generation effect: Delineation of a Wellman, H. (1990) The child’s theory of mind. MIT Press. [CS]
phenomenon. Journal of Experimental Psychology: Learning, Memory and Werner, H. & Kaplan, B. (1963) Symbol formation. Wiley. [CS]
Cognition 4:592–604. [aAMG] White, J. M., Sparks, D. L. & Stanford, T. R. (1994) Saccades to remembered
Sloman, S. A., Hayman, C. A. G., Ohta, N., Law, J. & Tulving, E. (1988) target locations: An analysis of systematic and variable errors. Vision
Forgetting in primed fragment completion. Journal of Experimental Research 34:79–92. [YR]
Psychology: Learning, Memory and Cognition 14:223–38. [aAMG] Wilcoxin, H. C., Dragoin, W. B. & Kral, P. A. (1971) Illness-induced aversions in
Smith, M. C. & Magee, L. E. (1980) Tracing the time course of rat and quail: Relative salience of visual and gustatory cues. Science
picture-word processing. Journal of Experimental Psychology: General 171:826–28. [aAMG]
109:373–92. [CMM] Winograd, E. (1996) Contexts and functions of retrieval. Behavioral and Brain
Smolensky, P. (1988) On the proper treatment of connectionism. Behavioral and Sciences 19:209–10. [AK]
Brain Sciences 11:1–74. [aAMG] Witkin, H. A. & Goodenough, D. R. (1981) Cognitive styles: Essence and origins.
St. John, M. F. & Gernsbacher, M. A. (1995) Syntactic comprehension: Practice International Universities Press. [CMM]
makes perfect and frequency makes fleet. Presented at the meeting of the Wu (1995) Perceptual representation in conceptual combination. Unpublished
Society for Cognitive Neuroscience. [rAMG] doctoral dissertation, University of Chicago. [KOS]
Stein, G. S., Stein, J. L., van Wijnen, A. J. & Lian, J. B. (1996) The maturation Zebrowitz, L. A. (1990) Social perception. Brooks/Cole. [CS]
of a cell. American Scientist 84:28–37. [rAMG] Zwaan, R. A., Magliano, J. P. & Graesser, A. C. (1995) Dimensions of situation
Talmy, L. (1983) How language structures space. In: Spatial orientation: Theory, model construction in narrative comprehension. Journal of Experimental
research, and application, ed. H. Pick & L. Acredolo. [KOS] Psychology: Learning, Memory, and Cognition 21:386–97. [ACG]
(1988) Force dynamics in language and cognition. Cognitive Science 12:49–
100. [aAMG, MdV]

BEHAVIORAL AND BRAIN SCIENCES (1997) 20:1 55