Professional Documents
Culture Documents
Dual Coding Theory - Mark Sadoski PDF
Dual Coding Theory - Mark Sadoski PDF
Mark Sadoski
Allan Paivio
Reference:
Sadoski, M. & Paivio, A. (2004). A dual coding theoretical model of
reading. In R. B. Ruddell & N. J. Unrau (Eds.), Theoretical models and processes of
reading (5th ed.) (pp. 1329-1362). Newark, DE: International Reading Association.
dual coding theory 2
Dual Coding Theory (DCT) is an established theory of general cognition that has
been directly applied to literacy. This theory was originally developed to account for
verbal and nonverbal influences on memory, but it has been extended to many other areas
of cognition through a systematic program of research over many years (Paivio, 1971,
(Sadoski & Paivio, 1994; Sadoski, Paivio, & Goetz, 1991), as an account of written
composition (Sadoski, 1992), and as a unified theory of reading and writing (Sadoski &
Paivio, 2001). For the fullest understanding of the theory these references and the
specific studies they cite should be consulted. This paper briefly discusses the DCT
response.
The value of explaining reading under the aegis of a theory of general cognition is
compelling. Reading is a cognitive act, but there is nothing about reading that does not
occur in other cognitive acts that do not involve reading. We perceive, recognize,
interpret, comprehend, appreciate, and remember information that is not in text form as
well as information that is in text form. Cognition in reading is a special case of general
cognition that involves written language. Theories specific to reading must eventually
conform with broader theories of general cognition for scientific progress to advance.
comprehension, and response. Theories of reading often focus on one or another of these
aspects of reading but not all. As we shall see, the same basic DCT principles apply to
mental models of text episodes, and even imaginative responses to text. In this chapter
we will briefly explain the theory’s basic assumptions; provide accounts of decoding,
comprehension, and response; compare and contrast DCT with other theories of reading;
Basic Assumptions
A basic premise of DCT is that all mental representations retain some of the
concrete qualities of the external experiences from which they derive. These experiences
separate mental systems, or codes, one specialized for representing and processing
language (the verbal code) and one for processing nonlinguistic objects and events (the
nonverbal code). The latter is frequently referred to as the imagery system or code
because its functions include the generation, analysis, and transformation of mental
images. Each code has its own characteristic units and hierarchical organization.
Together, the two codes account for knowledge of language and knowledge of the world.
The two mental codes and our five senses are orthogonal in DCT. This means
that the two codes each have subsets of mental representations that are qualitatively
different due to the different sensory experiences from which they originated. Because
sensory systems are linked to motor response systems in perception (e.g., eye
movements, listening attitudes, active touch) these subsets have sensorimotor qualities.
We develop visual representations in the verbal code for language units we have seen
such as letters, words, or phrases (e.g., baseball bat). But we also develop visual
representations in the nonverbal code for nonlinguistic forms that we have seen such as
develop auditory representations in the verbal code for speech units we have heard such
as phonemes and their combinations (e.g., the phoneme /b/, the rime /-at/, the word /bat/),
sounds we have heard (e.g., the crack of a wooden bat hitting a ball or the clink of an
in the verbal code for linguistic motor acts (e.g., pronouncing /b/ or writing the letter b or
touching the Braille sign for b), and we develop haptic representations in the nonverbal
code for the active “feel” of objects, textures, and movements (e.g., the heft and swing of
a baseball bat). We do not represent language in the chemical sense modalities (smell
and taste), but we have nonverbal representations for them (e.g., the smell and taste of a
juicy hot dog at a baseball game). Images in these modalities are typically less vivid for
most people. To these should be added affect – emotional feelings and reactions. These
are nonverbal by definition, although we have many names for emotional states. We also
have imagery for such states, and it forms an important component of meaning. We
might imagine the excitement of an enthusiastic fan at a baseball game, for example.
interpretation of reading. The overall system can be imagined as a set of modality- and
code-specific subsystems that are laced with interconnections. These subsystems are
Table 1
Mental Codes
brain. For example, some persons with alexia cannot read the phrase baseball bat but can
recognize the phrase when it is spoken and even write it, providing evidence of
with anomia can recognize a baseball bat but not be able to name it, providing evidence
neuropsychological evidence, see Paivio (1986, 1991) and West, O’Rourke, and
mental codes and sensory modalities has sometimes led to the inaccurate characterization
of DCT as being about the verbal and visual codes. The correct distinctions are between
verbal and nonverbal (imagery) codes, and between the visual modality and the other
sensory modalities.
ability to do two things in the same modality at once. For example, it is difficult to listen
to two conversations at once -- our verbal, auditory capacity is quickly overcome and we
must “shuttle” between the two (i.e., the “cocktail-party effect”). In reading, it is
somewhat difficult to visually process the print and extensively visualize its semantic
content at the same time, particularly for the unskilled reader. Either the reader tends to
slow down or the number of oral reading miscues increases (Denis, 1982; Eddy & Glass,
1981; Hodges, 1994; Sadoski, 1983, 1985). More discussion of this phenomenon and its
implications will appear later. We will next discuss the basic units that comprise each
dual coding theory 7
system, their organization in each system, and the kinds of interconnections and
Basic Units
The basic units in the verbal system are logogens, and the basic units in the nonverbal
system are imagens. These terms are merely jargon for the way the brain represents
different types of information, but DCT assumes that they are concrete as opposed to
being abstract and amodal. The terms are not meant to imply static units. Although
memory representations have some permanence, they are better thought of as evolving
and flexible, such as the way our vocabulary knowledge is constantly enriched by
experiencing words in different contexts or the way images are often of novel scenes
Language units vary in size, although some sizes are more familiar than others (e.g.,
words). Hence, we have visual logogens for letters and written words and phrases;
auditory logogens for phonemes and word and phrase pronunciations; haptic logogens for
may be represented in closely associated auditory-motor form. That is, a phoneme may
Logogens are derived from the perception of language and influence its
perception. A charming example is the child learning to say the alphabet who perceives
perceived as one auditory-motor unit; with more learning it will be perceived as four
separate units. Similarly, words can be learned before their individual letters or
phonemes are learned, and words can be identified as rapidly, or more rapidly, than their
Imagens are modality-specific and vary in size as well, and they tend to be
perceived in nested sets. That is, mental images are often embedded in larger mental
images. Hence we can visually imagine a baseball bat, the bat in the hands of a batter,
and the batter at home plate in a crowded stadium. In the auditory modality, we can
imagine the crack of the bat or the crack of the bat over the noise of the crowd. These
transformed into a sequence of the bat being swung, the crack of the bat and roar of the
crowd, and the batter running to base as the ball speeds off. That is, while the imagens
remain modality-specific, they can be associated into a larger mental structure that
activated by direct sensory input such as seeing printed language, or imagens can be
activated by seeing familiar objects. However, both types of mental representation can
Both internal and external contexts can also prime language or imagery. Seeing the word
bat, glove, game, cap and so on. All these could in turn activate related imagens.
External contexts would serve to limit the activated set to the most relevant members.
Therefore, both bottom-up and top-down inputs can activate mental representations in
dual coding theory 9
interactive ways. When enough input is received from any one source or a combination
of the strength of the inputs or how often or how recently a representation has been
excited. Many of these assumptions are common to most network theories of cognition,
but DCT is unique in its emphasis on the modality-specific, verbal, and nonverbal
Logogens and the verbal system into which they are organized are characterized
by sequential constraints. In all languages, units are combined into certain conventional
sequences at all levels. Hence, the letters b, a, t can be sequenced as bat or tab but not
tba; the words a, baseball, and bat can be sequenced as a baseball bat or bat a baseball
but not baseball a bat, and so on. A hierarchy characterizes the verbal system such that
smaller units can be sequentially synthesized into larger units (e.g., letters to words) or
larger units can be sequentially analyzed into smaller units (e.g., words to letters).
However, units at each level retain a degree of independence such that a spoken word, for
example, can be recognized without necessarily analyzing its phonemic structure. The
of speech or the linear nature of print that we experience in encounters with language. A
common example is that it is easier to spell a long word forward than backward from
principle is reversed in languages that are read and written right to left).
dual coding theory 10
Imagens are represented and organized in a more continuous, integrated way and cannot
The developmental basis for this system is the generally more holistic nature of
senses. An example is the baseball episode given earlier. We can synthesize images of
smaller units (a baseball bat) into embedded or nested units (a bat in the hands of a batter)
to still larger sets (the batter in a stadium with a roaring crowd), or we can analyze the
scene in reverse. In the visual modality, this often takes a cinematic form where we
sequences can also be represented, as the episode of the batter hitting the ball with a loud
Therefore, both the verbal and nonverbal systems have modality-specific units of
various size that are hierarchically organized, but the respective units and their
hierarchies are qualitatively different. Logogens and their verbal hierarchy are heavily
sequentially constrained, whereas imagens and their nonverbal hierarchy are more
Processing Operations
levels metaphor is only partly useful because associative and referential processing can
be seen as spreading activation at the same “level” but involving different codes. In
level is analogous to simply recognizing something as familiar and does not necessarily
stimulus situation and individual differences. In reading text, the stimulus would be the
text characteristics, and individual differences would include reading ability, background
knowledge, instructions, and so on. Therefore, the activation of a visual logogen for a
printed word would involve the legibility of the printed form, the reader’s familiarity with
the word’s visual features and configuration, and any priming effects of context. If the
visually recognized word was also familiar from speech, its associated auditory-motor
phonological logogen usually would be activated rapidly in turn (e.g., baseball). All this
would be carried out in milliseconds and perhaps without conscious attention. If the
visual word was not familiar, visual and phonological logogens at lower levels such as
letter combinations would be activated, requiring more time and attention (e.g., base—
ball). This may implicate higher-order processes in its recognition. On the other hand,
whole familiar phrases can be recognized and named at a glance by the skilled reader
example of associative processing that does not necessarily involve meaning and is
visual word may involve its comprehension in some cases. Heteronyms have one
spelling but different meanings and respective pronunciations, and their phonological
dual coding theory 12
recoding depends on which meaning is implied by context (e.g., bass drum, largemouth
Meaningful associative processing within the verbal code involves the activation
example, the word single has many verbal associations, but only a subset will be
activated in a given context. In a baseball game, the word single would activate verbal
associations such as hit, first base, advance a runner, and so on. In other contexts, the
word single could activate very different verbal associates such as one-dollar bill,
unmarried, hotel room, and so on. Meaning is both constrained and elaborated by the set
logogens in turn activate imagens in the same way they activate other logogens. The
phrase baseball bat can activate mental images of a wooden or aluminum baseball bat;
and single can activate an entire dynamic, nested set of images of a batter hitting a
baseball and running to first base in a stadium. That is, there is not a one-to-one
referentially activate few imagens, while other logogens might activate many. Some
logogens might activate no imagens at all. This is particularly true of language that is
highly abstract; it is difficult to form images of basic idea, for example. Without the
context of a concrete situation, such phrases lack any referential meaning and can only be
dual coding theory 13
verbally defined. This implies that concrete language should be generally better
imagens in the nonverbal system, associative processing may occur within that system
and, in turn, refer back to the verbal system. For example, the set of imagens
nonverbal system to include a batter running to first base in a crowded, cheering stadium.
These imagens might in return referentially activate logogens such as stadium or crowd
or cheers. In this way, spreading activation between and within codes defines and
substance to the meaning; taken literally, this is what “making sense” in reading is all
about.
Figure 1 shows a theoretical model of these units and processes. Verbal and
nonverbal stimuli are perceived by the sensory systems and logogens and imagens are
logogens. These units are modality-specific and of different sizes so that smaller ones
may be representations for graphemes or phonemes, lager ones may be visual words or
by the arrows are of many kinds: grapho-phonemic associations (b-/b/), compound word
associations (e.g., base-ball) common sequences (e.g., first, second, third, home),
antonym associations (e.g., batter-hitter; safe-out), and on and on. The nonverbal system
dual coding theory 14
is illustrated as a series of overlapping and nested sets of imagens (e.g., a baseball bat
being swung by a batter in a crowded, cheering stadium) or other imagens not associated
with a given set. Referential connections are illustrated as arrows running between the
coding systems. Verbal and nonverbal responses are shown as output of the respective
systems.
This figure illustrates some of the most basic assumptions of DCT, but the
illustration is necessarily simple. In actuality, a model for reading even a simple text
would be interlaced with connections and abuzz with activity. Further discussion can be
comprehension, and response in turn. We use a single sentence here, but the reader
should keep in mind that such sentences are more realistically read in much richer,
extended contexts.
dual coding theory 15
SENSORY SYSTEMS
REPRESENTATIONAL C O N N ECTIONS
N
O
V N
Logogens Imagens
E V
ASSOCIATIVE
R E
ASSOCIATIVE
B R
A REFERENTIAL B
L CONNECTIONS A
L
S
Y S
S Y
T S
E T
M E
M
FIGURE 1
General Model of Dual Coding Theory
dual coding theory 16
Consider the skilled reading and in-depth comprehension of the sentence The
batter singled to center in the first. The process begins as the eyes fixate on the printed
forms, probably The batter in the first fixation (eye movement studies indicate an average
span of about nine characters per fixation). Visual logogens for the familiar words The
and batter are activated at the representational level, and immediately associated with
their auditory-motor logogens. “The batter” may be experienced as inner speech, recoded
in phonological form. Perhaps equally quickly, the words are syntactically associated as
a simple noun phrase. Spreading activation to semantic associations also occurs rapidly,
with the different associates of batter activated as options. Common verbal associates of
batter could be baseball player or cake mixture or strike repeatedly, and nonverbal
referential connections could be images of the same. However, in the present context the
word batter has been used in connection with other baseball terms, and context effects
would prime the former option and inhibit the latter ones. Also, The preceding batter
signals a noun usage of batter, syntactically inhibiting the verb option strike repeatedly.
Within perhaps 500 milliseconds (an estimate from electroencephalographic studies), the
words The batter are phonologically recoded and provisionally comprehended in both
A word here about decoding. In reading, this term is theoretically imprecise. The
term recoding is often preferred because it indicates converting the printed form to the
implies (i.e., to decode a message). Conformably, DCT assumes that in reading, the
associations but not necessarily their semantic associates and referents. For highly
familiar words this happens without conscious effort, hence the metaphor “automatic.”
Very familiar phrases such as The batter may be activated as a single unit similar to hot
words may require more grapheme-phoneme level processing, more conscious effort, and
possibly some semantic and syntactic processing. Thus, DCT accommodates multiple-
route models of phonology in word reading (e.g., Coltheart, Curtis, Atkins, & Haller,
1993).
Returning to our example, the next fixation falls on the word singled, perhaps
already noticed in the parafovea of the first fixation. This word appears after the noun
phrase and is marked as a verb by the -ed suffix. Associative processing syntactically
connects The batter with singled and the familiar subject-verb syntactic pattern is
recognized.
conscious, and it may be less complex than is commonly assumed. The verb here may be
comprehended simply as a modifier of the noun phrase. That is, a mental model of the
sentence thus far may be forming in which the batter is imagined in action, hitting the
ball and running to first. This emerging mental model takes the form of a verbal-
nonverbal, syntactic and semantic episode in short term memory. For those less familiar
with baseball terms, the word singled may need to be syntactically paraphrased to hit a
single for clarification, but the result would be the same. In this sense, grammar need not
sequences that evoke a comprehensible image can account for much. This is the DCT
The next fixation includes to center. The various associates and referents of to
center (e.g., between left field and right field; to balance) are part of the spreading
activation, with context inhibiting less appropriate ones. Again, a less familiar reader
may verbally elaborate this elliptical phrase into to center field. The words The batter
singled to center (field) may then be syntactically parsed as the familiar subject-verb-
modifier pattern, recoded phonologically as inner speech, and imagined to now include
the ball speeding to the middle outfield as the batter runs to first base.
The final fixation falls on in the first. Familiar verbal elaborations of in the first
such as in the first inning or in the first place may be associatively activated with context
inhibiting the latter. The phrase will be recognized as a modifier by association with the
syntax established so far and the entire sentence parsed and cumulatively recoded as
inner speech. However, in the first would probably add little to the imaginal mental
model of the episode except possible time cues such as the fresh uniforms and unscuffed
baselines of the early innings of a baseball game. However, it might involve another
As described so far, the processing of the simple sentence The batter singled to
center in the first would take about two seconds at a typical reading rate of 250 wpm
(longer for readers unfamiliar with baseball). For a skilled reader reading for full
comprehension and recall, the result would probably include the sentence recoded as
inner speech and comprehended as a verbal-nonverbal mental model of the episode. Note
that neither is necessarily experienced consciously, or perhaps only barely so, and both
dual coding theory 19
may be rehearsed in memory after the last fixation. These responses differ with readers
A few words here about meaning, comprehension, and mental models. As noted
in the present example, the text would be mentally represented in two codes and in at
Both might be elaborated in various ways. As noted, the word center may be mentally
elaborated to center field, and in the first may be elaborated to in the first inning. Beyond
this, verbal elaboration may take the form of a related set of activated associations in the
verbal system such as baseball, pitcher, swing, hit, fly, grounder, outfield, run, first base,
safe, stadium, crowd, and so on. The imagery representation might be in modalities other
than the visual, depending on the degree of elaboration. A more fully elaborated image
might include the crack of the bat and the roar of the crowd, for example. Associative
connections and referential connections between the verbal associates and the nonverbal
representations and their defining interconnections the more “meaningful” our response.
associations and the set of nonverbal associations correspond and restrict each other
coherence. The term “mental model,” as used here, applies to the total verbal-nonverbal
representations and the associative and referential connections between them. The term
does not imply any theoretical construct beyond what has already been explained. A
mental model in DCT is not an abstraction; the modality-specific units activated and
However, the discussion of our example does not end here. Consider the
inferences that may occur as a mental model is formed for the sentence The batter singled
to center in the first. The sentence does not specify if the hit was a fly ball or a grounder.
It does not specify if the game was a professional baseball game or a little league game.
It does not specify if the stadium was opened or closed or if there was a stadium and
spectators at all. It does not specify if the game was at night or during the day. Yet our
mental models are often specific on such points. Many of these inferences can be
attributed to mental imagery – imagining the general situation described by the language
Readers read with varying degrees of depth and elaboration based on purposes and
might. In other cases is our comprehension is deeper, richer, and more precise.
dual coding theory 21
This leads us back to the subject of response. In many ways, this term implies the
formation of a mental model, a coherent and elaborate rendition of the text. A reader
may fully experience even a simple text such as The batter singled to center in the first
by imagining the event as described. In a still more elaborate response, one might “feel”
oneself as the batter, haptically sensing the heft of the bat and the jolt as it connects. But
response often implies more than sensory imagery or a mental model. We noted earlier
that the phrase in the first adds a time cue to the sentence. This simple time cue may
evoke an evaluative and mildly affective response. A single in the first is not as critical
as a single in the bottom of the ninth with the game tied. That is, there is a different
emotional significance to the two time settings. This introduces a nonverbal, emotional-
evaluative dimension to the response. In full and complete narrative texts such as stories,
such emotional responses are an important aspect of experiencing the text (e.g., Sadoski,
In other contexts, response may take a more logical and rational form. Analyzing
in inner expressions such as “I don’t get this,” or “Now I see,” or “But you haven’t
considered….” Our critical and evaluative powers are also exercised here, and the
experience may also be emotional. We are impressed with a well-argued position with
difficult problem with a one-syllable solution. In its fullest sense, response involves the
reader as a part of the authoring, a partner who stands toe-to-toe with the author and
answers back. This may take the reading beyond what the text language may have
highly abstract language. The answer is that the encoding abstract language is primarily
a matter of verbal associations. Consider the abstract sentence The basic idea remained
vague. As with a more concrete sentence, this sentence can be phonologically recoded,
grammatically parsed, and associated with other language units (e.g., basic idea = main
thought, remained vague = stayed unclear). But beyond such mental parsing and
referent to concretize the abstract, it remains a verbalism with unrealized potential. Such
sentences may be integrated as verbal units and achieve a degree of meaning at the
associative level, but their fuller meaning and response awaits a more concrete context.
Empirical Evidence
The constellation of predictions derived from the DCT model of reading has only
been partially developed and tested, but relevant evidence is available on several research
fronts. We will next review certain empirical evidence in the areas of decoding,
Decoding
generally assumed to occur at the representational level because they usually can be
involve associative processing, and the time required for this processing will presumably
Therefore, spreading activation could theoretically reach and activate still other
readers read concrete, imageable words more accurately than abstract words, with these
effects more prominent for poor readers (Coltheart, Laxon, & Keating, 1988; Juel &
Holmes, 1981; Jorm, 1977). Neurological patients with severe phonological deficits,
whose reading ability is assumed to rely mainly on direct access from orthography to
imageable words than abstract words (e.g., Coltheart, Patterson, & Marshall, 1980;
Funnell & Allport, 1987; Plaut & Shallice, 1993). This evidence is supportive of
multiple-route models of the oral reading of words (cf. Coltheart, Rastle, Perry, Langdon,
If a word’s concreteness can have any influence on the process of naming it, this
Strain, Patterson, and Seidenberg (1995) tested this prediction. They found that,
controlling for word familiarity and linguistic variables (i.e., initial phoneme, word
length, positional bigram frequency), adults were slower and more error prone when
naming abstract irregular words (e.g., scarce) than when naming abstract regular words
(e.g., scribe) or imageable irregular words (e.g., sword). That is, imageable words were
named sooner and more accurately when they were irregular because the spelling-sound
dual coding theory 24
processing time was sufficient to allow for the activation of corresponding imagens.
However, when the words were regular, the associations between spelling and sound
rapidly achieved whether the words were concrete or abstract, so the activation of
imagens for the concrete words did not produce a similar effect in naming. Overall, they
words.
This study has been subject to considerable replication and scrutiny. It was
replicated and extended by Strain and Herdman (1999), who found the same interaction
between frequency, regularity, and imageability in word naming. They also found an
even stronger effect for imageability in the naming of low-frequency regular words,
although this was still less than for low-frequency irregular words. They also found that
the more skilled in decoding participants were, the less strongly imageability influenced
word naming, although the effect was still present even for highly skilled decoders.
Overall, they interpreted their findings to mean that imageability plays a role in naming
words when the connections between orthography and phonology are weak, whether this
Monaghan and Ellis (2002) replicated the original study and again found the
acquisition of the word because when this variable was covaried the interaction was
nonsignificant. They argued that irregular words and low frequency words would be
acquired later and this would largely account for the effect. However, Strain, Patterson,
& Siedenberg (2002) questioned the use of age of acquisition because when they
reanalyzed their original data with age of acquisition as a covariate the interaction of
dual coding theory 25
regularity and imageability persisted. Furthermore, they presented new data confirming
the interaction when age of acquisition was controlled. Ellis and Monaghan (2002)
rejoined on methodological grounds but presented no new data. Overall, these results
suggest a persistent but qualified interaction between imageability and regularity in word
naming.
languages differ in the degree to which they represent spoken language. In Persian,
words are sometimes written with consonant letters only (opaque words) and sometimes
with full vowels (transparent words). The opaque words therefore present more decoding
difficulties. A significant imageability effect was found by Baluch and Besner (2001) in
naming both high and low frequency opaque words in Persian, while the effect was not
present for matched transparent words. These findings have been extended to Turkish, a
perfectly transparent orthography in which each of its 29 letters corresponds to only one
spoken sound invariantly and independent of context. Raman, Baluch, and Besner (1997)
found no significant effects of imageability on word naming of Turkish high and low
frequency words, consistent with the prediction. However, Raman and Baluch (2001)
additionally investigated reading skill and found that skilled readers in Turkish named
similar to what Strain and Herdman (1999) found in English and also consistent with
prediction. Less skilled readers did not, a result different from Strain and Herdman, but
this was attributed to differences in reading Turkish and English. Janyan and Andonova
(2002) still found independent effects of imageability and frequency in naming words in
Bulgarian, an orthography with more graphophonemic consistency than English but less
dual coding theory 26
than Turkish. Furthermore, these researchers found that imageability was more
In sum, these studies indicate that imageability plays a role in naming words when
the connections between orthography and phonology are labored, whether this is due to
The effect of decoding skill also interacts with these factors differently in different
orthographies. These results are theoretically consistent with the interactive DCT model
visual logogens activated by printed concrete words spread their activation to both
auditory-motor naming logogens and imagens for referential meaning. The activated
imagens can significantly contribute to the activation of the naming logogens when the
route between the visual logogens and the naming logogens is slow. The effect is
also complex, predicting at least a triple interaction. As Strain et al. (2002, p. 212)
concluded, “almost all of the factors in the domain of word familiarity and
meaningfulness are correlated with one another, and it is a brave experimenter who
attempts to establish the prominence of one while denying any impact for the others” (cf.
Venezky & Massaro, 1987). Multiple factors influence word decoding of which
compatible with DCT but incompatible with theories that posit modality-independent
word units, concept nodes, or abstract propositions. Moreover, they have implications for
Comprehension
associatively activate mental language, whereas abstract has relatively less access to the
imagery code. Hence, concrete language such as baseball bat should be easier to
understand than abstract language like basic idea – concrete language can be dually
encoded. In DCT, the two codes are assumed to be independent and additive in their
effects, predicting that concrete language should be nearly twice as comprehensible and
memorable as abstract language, other factors being equal. The empirical record over the
last 30 years has shown just this. Concrete words, phrases, sentences, paragraphs, and
longer texts have been consistently shown to be more than twice as comprehensible and
memorable as abstract language units matched for readability, familiarity, and other
variables. The evidence relevant to reading has been extensively reviewed (Sadoski &
Paivio, 2001). Here we will focus on how predictions made by DCT in the area of
comprehension and recall have prevailed in experimental tests against the predictions of
other theories.
availability theory (Kieras, 1978). Rather than assuming separate verbal and nonverbal
codes, context availability theory assumes that all language, whether abstract or concrete,
comprehend and remember because it can be more readily associated with other
propositions in the network – it simply has more connections. But this advantage can be
dual coding theory 28
offset when abstract language deals with highly familiar information or when abstract
language is presented in a supportive context because the connections are then enhanced
for the reader. The differing assumptions of DCT and the context availability model
present a situation all too rare in reading research: two rival theories that are sufficiently
Avila (1995). They used four factual paragraphs about historical figures (Michelangelo
and James Madison) that were matched for number of sentences, words, and syllables,
sentence length, information density, cohesion, and rated comprehensibility. In one set,
two paragraphs were rated equal in familiarity but one paragraph was rated more concrete
than the other. In this case, DCT predicted that the concrete paragraph should be recalled
better than abstract paragraph due to the additional integrating medium provided by
imagery, whereas the context availability model predicted that they should be recalled
equally because the abstract language was equally familiar and presented in context. In
the other set, the paragraphs differed in both familiarity and concreteness, with the
abstract paragraph being the more familiar of the two. In this case, DCT predicted that
the familiar abstract paragraph should be recalled about as well as the unfamiliar concrete
whereas the context availability model predicted that the abstract paragraph would be
recalled better than the concrete paragraph (i. e., the disadvantage of abstractness would
In the case where the concrete and abstract paragraphs were equally familiar,
study participants recalled the concrete paragraph nearly twice as well. In the case where
dual coding theory 29
the abstract paragraph was more familiar, study participants recalled the paragraphs
equally well. These results were consistent with DCT but inconsistent with context
availability theory. Other experimental findings using different methods and materials
have confirmed these results (e.g., Holcomb, Kounios, Anderson, & West, 1999; Kounios
& Holcomb, 1994; Nelson & Schreiber, 1992; Sadoski, Goetz, & Fritz, 1993; Sadoski,
Another case of competing predictions between DCT and another theory involved
integration of abstract text. Abstract sentences such as The basic idea remained vague
can sometimes be encoded as integrated verbal units as well as concrete sentences, even
though concrete sentences enjoy more comprehensible and memorable content. These
questions about the DCT explanation of the integration and recall of abstract text were
first raised by Marschark and Paivio (1977). They studied the cued recall of concrete and
abstract sentences and found that, whereas recall was higher for concrete sentences, the
two types of sentences did not differ in the extent to which recall was integrated as
measured by the retrieval effectiveness of verbal cues that were related to the whole
sentence or to only one content word in the sentence. The higher recall for concrete
sentences was consistent with the DCT assumption that dually encoded information is
more memorable, but the equal integration effect was inconsistent with the assumption
Marschark and Paivio (1977) concluded that dual coding accounted for the higher
recall of concrete sentences, but some other explanation was needed to explain the
equivalent integration effects. Verbal associative processes were a likely DCT candidate
dual coding theory 30
because these processes apply equally to concrete and abstract language. An alternative
theory was posed by Marschark and Hunt (1989). This theory assumed that
distinctiveness and relatedness are different forms of mental processing that acted as
between the language units. Hence, this theory, like context availability theory, predicts
concrete language.
A resolution to the issue was obtained by Paivio, Walsh, and Bons (1994). Their
experimental results showed that strong verbal associations are necessary to produce
concrete word pairs even when verbal associations are not present. The results were not
consistent with the relational-distinctiveness view because integration still occurred for
concrete word pairs where the pair members were not related. These findings provide a
DCT explanation for the integration of abstract language, but they further offer a possible
For example, consider again the abstract sentence The basic idea remained vague
and the concrete sentence The batter singled to center in the first. The former sentence
might be integrated grammatically because the predicate adjective vague links directly
back to the subject noun idea; the modifier is part of the sentence kernel and binds the
subject and the predicate closely together. In the latter sentence, the kernel is The batter
This sentence might be integrated more though its imagery, with singled to center and in
the first adding imaginal and verbal elaboration as discussed previously. Stated
differently, the integrative mental model of the abstract sentence may be more verbal-
associative, and the integrative mental model of the concrete sentence may be more
integrated units but for theoretically different reasons. However, this interpretation is
In sum, DCT can provide a coherent account of the critical role played by
language concreteness in reading comprehension. The evidence cited poses problems for
theories that propose that all language is mentally encoded in abstract, propositional
form. Furthermore, DCT can provide experimental predictions about the comprehension
and recall of concrete and abstract text, and about the integration of both concrete and
Response
Mental imagery, and its correlate, emotional response, are vital to aesthetic
response to text. Imaginative and affective processes are how a text is “realized” or
“lived through” or “brought to life.” As discussed earlier, some aspects of response may
be contemporaneous with the formation of a mental model of the text, so that the
distinction between comprehension and response is somewhat fuzzy. Response can take
more objective forms as well, such as critical evaluation against some standard as in
The empirical evidence for the relationship between imaginative and affective
processes in responding to text was reviewed by Goetz and Sadoski (1996). A program
dual coding theory 32
of research carried out over ten years revealed that imagery and affective response to text
can be measured reliably and validly using conventional methods. Both the strength of
The core of this research program was a set of complementary studies using
literary short stories. In one study (Sadoski & Goetz, 1985), participants read and later
rated each paragraph in an adventure story for either the degree of imagery experienced,
the degree of emotional response experienced, or the relative importance to the story as a
whole. The alpha reliabilities of these ratings, for this story as well as other similar
stories, has been regularly found to exceed .90 (Sadoski et al., 1988; Goetz, Sadoski,
Stowe, Fetsco, & Kemp, 1993). In all these studies, imagery ratings, affect ratings, and
importance ratings were moderately to strongly correlated with each other in an overall
response aggregate. The correlation between imagery ratings and affect ratings persisted
even after controlling for paragraph length and the importance ratings. However, the
relationship between either imagery ratings or affect ratings and importance ratings was
considerably attenuated when the effects of paragraph length and the remaining rating
were partialed. This means that imagery and affective response can generally be seen as
Qualitative reports were the focus of another related study (Sadoski, Goetz,
Olivarez, Lee, & Roberts, 1990). Using the same story as Sadoski and Goetz (1985),
participants read and then produced written recalls and imagery reports. Using the most
extensive coding system for imagery reports yet devised, imagery was coded into
categories such as (a) directly related to a paragraph, (b) a synthesis of information from
dual coding theory 33
two or more paragraphs, (c) distortion of story information, (d) an importation consistent
with the story, or (e) an importation inconsistent with the story. The imagery reports
were also categorized according to modality (e.g., visual, auditory, tactile, affective).
Recall protocols were similarly coded in categories including (a) gist, (b) synthesis from
across text units, (c) distortion, (d) an importation consistent with the story, or (e) an
importation inconsistent with the story. Interrater reliability between independent raters
ranged from .84 to .95 for all codings. A factor analysis of the imagery and recall
affective imagery (i.e., imagining the feelings of the characters), imported imagery
consistent with the story, and distortion recall. Verbal recall categories loaded on these
factors as well, but consistently lower. Hence, story response could be seen as primarily
experienced in the form of mental imagery and affect and somewhat less in the form of
verbal recall.
A subsequent study used the ratings from Sadoski and Goetz (1985) as predictors
of the imagery reports of Sadoski et al. (1990). Goetz, Sadoski, & Olivarez (1991)
and variance due to surface level text factors and found that the paragraph level imagery
ratings of the first group of readers were highly significant predictors of the respective
paragraph level imagery reports of the second, independent group of readers. Another
finding using the same methodology was that importance ratings and surface level text
factors were better predictors of verbal recalls than of imagery ratings. Hence, the
In sum, this program of research has shown that imaginative responses are central
to reading literary stories, and that they can be defined, measured, and interpreted
reliably and validly using conventional research methods. Imagery and emotional
response are moderately to strongly correlated in story response. Further, they are both
As Sadoski et al. (1998, p. 333) noted, “… these three response types often occurred
plot salience, and vicarious, emotionally infused, and fully perceived experiences of story
events.” Together, they form an overall response that involves both the intellect and the
emotions.
also has been extensively compared with schema theory as an explanation of reading
comprehension (Sadoski & Paivio, 1994; 2001; Sadoski, et al., 1991). DCT has also
been extensively compared with the theories of Rumelhart (1977) and Kintsch (1988) in
Sadoski and Paivio (2001). An updated version of Kintsch (1998) has also been
addressed from a DCT perspective (Sadoski, 1999). We will summarize and comment on
These theories of reading have much in common with DCT. A major similarity is
that they are all interactive theories in which reading is served by bottom-up and top-
down processes working in combination. All these theories are similar in assuming a
dual coding theory 35
prominent place for prior knowledge structures in reading. Another similarity is that they
allow, to different degrees, for the representation of knowledge in more than one form.
However, a basic distinction between DCT and these other theories is that they
assume that most knowledge in memory is abstract and amodal, existing in a state that is
has no objective reality and is associated with no sensory modality. How any knowledge
that is not innate becomes divorced from sensory input is an important theoretical and
epistemological question that has not been well explained. These theories propose no
apparent answer to this question; rather, they simply assume the existence of abstract,
amodal knowledge. One apparent reason for this is so that these theories can be modeled
semantic knowledge. The forms that these different knowledge sources take is not
specified, but examples provided suggest that they are ultimately all in an identically
that determines the Baysian probability of testing a set of multilevel hypotheses for the
programs are cited for the source of these formalisms including Kaplan’s (1973) General
Syntactic Parser and the HEARSAY speech recognition program (Lesser, Fennel, Erman,
& Reddy, 1974). That is, whatever the source of the original input (e.g., visual features
of letters), all information is theoretically converted to, and processed in, a common
computational form. As a result, this theory does not discuss the roles of phonological
dual coding theory 36
recoding, inner speech, or mental imagery in any form. This would call for at least some
propositional text base, and the mental model. Verbatim information is surface structure
information including specific words and syntactic arrangements. The propositional code
instantiated with surface structure information and the surface structure is quickly lost.
Individual propositions are then connected into a propositional text base. The mental
that is somehow derived from the text base. Hence, this theory can be seen as a triple-
coding theory that assumes that verbal language and mental imagery may be inputs and
outputs, but the central processing unit is propositional in nature. Hence, it is primarily a
Recent modifications of this theory have made it still more amenable to computer
multidimensional statistical space with each number indicating the strength with which
the proposition is linked to another proposition (i.e., latent semantic analysis). This
brings the theory closer to DCT on the one hand because associational strength between
individual mental representations is critical, but farther away on the other hand because
dual coding theory 37
the entire system is based in computational formalism. As with Rumelhart (1977), this
theory does not discuss the role of phonological recoding or inner speech, and it treats
Computer implemented models that coincide with human data are interesting exercises,
but the name of the game in theory is to know why. Such exercises run the risk of
reification – the fallacy of explaining something and then treating the explanation as real
rather than the thing being explained. In any event, it is useful to remember that
abstractions such as schemata, propositions, and abstract concept nodes are difficult to
operationalize and empirically test with human data. Their interpretive power lies largely
in the assumption that they exist in the first place, mentally analogous to computer
experienced to some degree and for which plausible human assessments can be devised:
relative newcomer to the scene of reading theory. Despite a strong empirical record over
the last several decades in accounting for verbal behavior, much more research is
needed. In this section, we will pose certain issues deserving of further research in
Decoding
dual coding theory 38
written language is a factor in its phonological recoding. This is consistent with the
interactive nature of reading, where top-down semantic and syntactic factors and bottom-
up decoding factors interact at all levels. This also reinforces the assertion that there may
be more to the phenomena of phonological recoding and inner speech than providing us
with a strategy for lexical access – meaning may precede phonology more than we
realize. While much phonological recoding may occur at a deep, cortical level, its more
Despite the pervasiveness of inner speech in reading, too little is known about this
phenomenon. Huey (1908) devoted a chapter to it, regarding inner speech as a ubiquitous
measured by surgically inserting electrodes into the speech musculature, Edfeldt (1960)
found that: (a) all readers appear to engage in inner speech to varying degrees, especially
as reading becomes increasingly difficult, (b) inner speech has no detrimental effect on
reading, and (c) good readers engage in less inner speech than poor readers. Both Huey
and Edfeldt pointed out that inner speech was not inevitable in reading. Later research
reviews by Gibson and Levin (1975) and Rayner and Pollatsek (1989) arrived at the
same general conclusions: Inner speech seems to be a useful but not obligatory vehicle
for recoding text in short-term memory in order to parse sentences, associate words
But these explanations raise unanswered questions. If inner speech is not strictly
necessary, why not simply an “inner semantics” that operates directly on print input, at
least for good readers past the early developmental stages of reading? Speed reading
dual coding theory 39
courses have long advocated “breaking the sound barrier” and reading purely visually to
increase rate and improve comprehension and retention (Frank, 1994). However, Carver
(1982) empirically determined that for skilled readers the optimal rate of comprehending
prose while reading was identical to the optimal rate of comprehending prose while
listening. This rate was about 300 words per minute – about the maximum rate at which
speech can be comfortably produced. What accounts for the pervasiveness of inner
speech, its increase as text difficulty increases, and its convergence with rate of
comprehension, if it is unnecessary?
The DCT approach to these issues might pose inner speech as mental imagery in
the auditory-motor modality being used to retain the surface form of a sentence while
higher-order comprehension processes occur. Recoding printed language from the visual
modality to auditory-motor form is needed because the visual modality can be overloaded
if it has to simultaneously: (a) hold sentence segments already seen in visual memory,
(b) process upcoming print, and (c) construct visual mental images as needed or
interference, the difficulty of trying to perform different tasks in the same sensory
modality. The principle may apply strongly here: There is too much activity in reading
for the visual pathways in the brain to handle alone. Recoding to the auditory-motor
modality allows non-visual rehearsal and speech-like parsing to occur while new text is
Of course, inner speech may serve other functions as well, such as strategically
“sounding out” unfamiliar words or “hearing” vocal phrasings, intonations or other forms
dual coding theory 40
of expressive interpretation. It may also serve as an internal surrogate for the original and
modality without which reading could not optimally occur. Educationally, this implies
that inner speech should be encouraged and taught when in fact many instructional
programs have been introduced to eliminate it. Theoretically, this implies still more
problems for propositional theories. If the surface form of the text is immediately
converted to abstract, amodal propositions and the surface structure lost, what theoretical
purpose could inner speech possibly serve and why should it be so common?
Comprehension
An issue in need of renewed research from the point of view of all cognitive
theories is that of the nature of grammar and its role in comprehending text. Over the
ages, grammars of various kinds have been formulated including traditional grammar,
structural grammar, case grammar, and transformational grammar. The similarities and
grammar and was contrasted with the descriptive patterns of structural grammar, enjoyed
considerable popularity during the second half of the twentieth century. However, its
explanation of sentence comprehension has become heavily strained and has not met with
empirical support.
verbal. Complex sentences are mentally broken down into their deep-structure kernels
and understood in terms of transformational rules that are basically innate and universal.
dual coding theory 41
number and nature of transformations that separate a sentence from its underlying
structure. However, empirical research has found that after controlling for sentence
length and implied changes in meaning, transformations had little effect on processing
time (reviewed in Williams, 1998). That is, neither deep structures nor transformations
consistently showed that word meaning plays a crucial role in comprehension, even
that imagery may play a substantial role in sentence grammar, particularly in the case of
concrete sentences.
described sentences as relationships between cases such as agent, object, instrument, and
the basis of various propositional approaches to cognition and language (e.g., Kintsch,
1998). However, Fillmore (1977, 1984) revised his case grammar theory by putting
sentences refer. Case grammar with these modifications met with more empirical
support than transformational grammar. For example, Black, Turner, and Bower (1979)
had subjects read sentences with a single vantage point such as Terry finished working in
the yard and went into the house and sentences with a changing vantage point such as
Terry finished working in the yard and came into the house. Sentences with changing
vantage points took longer to comprehend, were rated harder to understand, and were
dual coding theory 42
likely to be recalled from a single vantage point. Such evidence provides more support
for the view that both verbal and nonverbal processes are involved in grammar and
comprehension.
More recent developments are still more consistent with the DCT view.
(Langacker, 1987, 1990). Cognitive grammar simplifies the field by rejecting the rule-
governed model of mind and language and replacing it with an associational model. In
this model, networks of association evolve with the experience of the individual
governed by patterns of regularity that develop from childhood. The number of common
English). All sentences are variations on a few patterns that may be represented as
exemplars and varied by analogy. Descriptive, structural grammar is best suited to this
More significantly from the DCT point of view, cognition is not seen as
representations and mental models of reality that may be in the form of imagery.
knowledge of the world to which language refers, rather than a propositional deep
capitalizes on verbal and nonverbal mental representations and syntactic patterns that are
dual coding theory 43
readily experienced and do not require the burdensome and unverified assumptions of
and future research in this area great holds promise for understanding the nature and role
of grammar in reading.
Response
as recently as twenty-five years ago (Kruez & MacNealy, 1996). Indeed, the study of
reader response to literature has become one of the more popular subjects in reading
research. Mental imagery and affect are obviously crucial to literary response because
the sensuous realization of setting, episode, character, and conflict is central to the “lived
through” experience of a literary text. If evidence is needed to support this point, Miall
and Kuiken (1995) factor analyzed a questionnaire of 68 items covering a broad spectrum
of literary responses and found factors for imagery and empathy. They grouped these
into a higher-order factor they called experiencing, the dimension of being absorbed in a
literary work. Conformably, much research from the DCT perspective has shown that
imagery and emotional response are persistently related in responding to literature. This
research was reviewed by Goetz and Sadoski (1996) and was summarized earlier.
An issue that deserves attention is the challenge posed by this research for some
considerable attention in recent years as an explanation for the way readers approach
texts with different stances in mind. Specifically, this theory proposes that the reading
transaction exists on a continuum between efferent reading and aesthetic reading. In the
dual coding theory 44
efferent stance, the reader is mainly concerned with information to be extracted and
retained after the reading event. In the aesthetic stance, the reader is mainly concerned
with the evocation of sensations, images, and scenes as they unfold in the moment. The
reader may vary stances so that, for example, poetry could be read as a source of
sounds, and emotions of the historical events. Moreover, readers can slide along the
continuum from moment to moment within a reading, so that no reading is probably ever
The challenge involved is that considerable evidence exists to show that what is
most imaged and felt in a reading is what is most retained over the long term. For
example, Sadoski and Quast (1990) had their participants read and then rate the
paragraphs in three feature journalism stories for the imagery experienced, the emotions
experienced, or the importance to the story as a whole. Readers were assigned to read the
articles as they would normally read an article about current events and people, not for
testing. Sixteen days later, the readers were given a surprise recall task: They were asked
to write down whatever they most remembered about the stories. What was recalled was
overwhelmingly associated with the emotions and imagery experienced during the
reading as determined by their own statements and earlier ratings. What was rated as
important was recalled poorly. That is, what was retained long after the reading event
If this continuum works as a true continuum, it follows that as the reader moves toward
one pole, he or she must move away from the other. If efferent reading is determined by
dual coding theory 45
what is retained, than the aesthetic aspects of the reading, including images and emotions,
should have been poorly recalled, and information seen as important should have been
better recalled. But these results indicate that what was carried away from the reading
was aesthetic. That is, the transactional theory of reader response does not appear to
sufficiently take into account the nature of memory for what is read whatever the reader’s
stance may be. We do not necessarily remember what we read for information unless we
make efforts to memorize or otherwise record the information. Mental imagery and its
frequent correlate, emotional response, can be evoked even when reading for superficial
information. For example, Sadoski et al. (1990) found that students who were assigned
to read to find typographical errors intentionally inserted in the text of a literary story
reported as much imagery and affect as students assigned to read the story normally for
pleasure. In short, stance may be less consequential than is assumed; a good story is
response. A growing body of empirical support for DCT principles in each of these areas
of teaching reading has been identified and synthesized, and more research is
Decoding
Decodable texts use a high proportion of graphophonemically regular words that are
intended to assist students in learning to decode and gain a sight vocabulary. However,
dual coding theory 46
word imageability has also been found to strongly affect sight word learning. That is,
Hargis and Gickling (1978) taught kindergartners a set of concrete sight words
and a set of abstract sight words that were matched for length and frequency. All the
words were familiar words that were initially unknown to the children by sight. During
training, the children: (a) were shown the words on flash cards, (b) heard each word
pronounced, (c) heard each word used in a sentence, (d) used the word in a sentence of
their own, and (e) repeated the word. Two days later, more than three times as many
concrete words were correctly named. Ten days later, more than four times as many
concrete words were correctly named. These results were later replicated with both
kindergartners of normal ability and older mentally retarded children (Gickling, Hargis,
Kolker & Terwilliger (1981) also taught first-and second-grade children concrete
and abstract sight words. The words were initially unknown by sight but were familiar in
speech. The words were presented for one second on a flash card and pronounced for the
child. The child then repeated the words and corrections were provided as necessary.
This continued until a learning criterion was reached. The first graders took about 60%
more trials to learn to correctly name the abstract words than the concrete words. The
difference for the second graders was less (about 8% more trials for abstract words) but
still statistically significant. Terwilliger and Kolker (1982) replicated their results while
manipulating word confusability (i. e., same or different initial consonants). Similar
results were obtained with kindergartners and beginning first graders by van der Veur
the experiment included manipulations for word decodability and word presentation in
story context or in isolation. Results determined that: (a) concrete words were learned
about 12% faster than abstract words, (b) words presented in story contexts were learned
about 12% faster than words presented in isolation, and (c) decodable words were
learned about 6% faster than nondecodable words. An interaction between these factors
indicated that abstract words presented in isolation took the longest to learn regardless of
context.
The results of these studies are very consistent with DCT and suggest that
and context in learning sight vocabulary. Overall, concrete words are learned faster even
without the benefit of context, especially if decodable. Abstract words that vary in
decodability are learned more slowly and benefit more from being learned in context.
These results are compatible with similar findings with word recognition by adults as
discussed earlier (e.g., Strain et al., 1995, 2002; Strain & Herdman, 1999).
Comprehension
throughout history, often intuitively. Sadoski and Paivio (2001) reviewed the history of
teaching text comprehension from ancient times to modern and found an alternation
between emphasis on the abstract and the verbal (e.g., outlines, epitomes) and the
concrete and the imaginal (e.g., object lessons, imagery training). Clark and Paivio
dual coding theory 48
determined that mental imagery, concreteness, and verbal associative processes play
major roles in the representation and comprehension of knowledge, learning and memory
(1997), in his summary of his research program on influential teachers of reading, found
that such teachers used highly effective strategies including the use of concrete examples,
their extension to more abstract examples, and the analysis of abstract concepts in
concrete terms.
The use of induced mental imagery to enhance student understanding and learning
has gained an increasing record of acceptance (see reviews by Gambrell & Koskinen,
2002; Sadoski & Paivio 2001). The general conclusion of both of these reviews is that
instructions to form mental images significantly enhances the reading comprehension and
memory of both children and adults in various ways. Numerous studies have shown, for
example, that elementary-age students know how to induce mental imagery and that only
brief training and teacher scaffolding is necessary for most children to effectively use
For example, Gambrell (1982) had first and third graders read short stories in
sections. Before each section, the experimental group was told to make pictures in their
heads to help remember, whereas the control group was told to think about what they
read in order to remember. After reading each section, the participants were asked a
prediction question, "What do you think is going to happen next?" Their responses were
scored for factual accuracy and number of accurate predictions. Third graders in the
imagery group reported twice as many facts and made twice as many accurate predictions
dual coding theory 49
as controls. First grades in the imagery group also tended to outperform controls on both
measures, but the differences were not statistically significant. This study indicates that
even with minimal inducement, children can use mental imagery comprehend factual
Mental imagery has also been used as a remediation technique for poor
progressively larger units of language including words, phrases, sentences, and texts.
used including prediction, recall, explicit questions, implicit questions, and a standardized
reading comprehension test. While many of the between-group contrasts did not
answering implicit questions. The reciprocal teaching group outperformed the imagery
training group only on answering explicit questions. The researcher concluded that both
students with problems in this area. Further research in this area would be useful.
Response
Teaching reader response is an area fraught with controversy. Miall (1996) based
on his extensive research into reader response, questioned whether response can or
should be taught at all. He bemoaned the status of teaching literary response in many
classrooms:
dual coding theory 50
loggers in one of our northern forests, there are teachers in too many classes
whose work succeeds only in clear cutting every shoot of literary interest, leaving
hardly a stump behind, mainly for the sake of that giant pulp mill, the testing and
There is no easy solution to this problem. Images and feelings are deeply personal, and
instruction in what to imagine and feel is surely less appropriate than to imagine and feel.
affective fallacy. Finding a way to do this is a pressing problem for researchers and
educators. One educationally valid method from a reading point of view might involve
the avoidance of didactically explaining the effects of a literary work, and instead
Sadoski (2002) discussed how DCT principles might explain the way poets use
commonly taught poems are presented, showing that the mental images evoked by the
poem are sometimes in contrast with the language of the poem. For example, John
Masefield’s poem “Sea Fever” deals with a sailor’s yearning to go back to sea, although
the reasons for not doing so are unclear. Images are evoked of tall sailing ships, the kick
of the wheel, the song of the wind, and the clouds flying by. But the rhythm of the poem
alternates between rollicking, fast-paced lines and slower, languorous lines. The images
are all of freedom, but the lines constrain the reader every time the poem gets going
dual coding theory 51
again. The language therefore is inconsistent with the images evoked, and the contrast is
provocative. Does this mean that the sailor is somehow incapacitated and either
because it is all inferred and ambiguous. Rather, the effects obtained in the readers are
more appropriate for some poems than others, but as a general approach, the scrutiny of
the effects produced by language and imagery in collaboration or in contrast may be one
Fazit
psychological aspects. Few theories offer this scope and have achieved its broad base of
programs of scientific testing are young, little more than thirty years old. They may be
compared to the first comprehensive theory of gravitation that dates back over three
hundred years or to the first comprehensive theory of evolution that dates back nearly one
hundred and fifty years. New theories are subject to change as evidence accumulates,
and they can be expected to make few clear and unambiguous predictions (a statement
that applies to all the theories in this volume). However, DCT specifies to a considerable
extent constructs of mind that have been of interest since ancient times and will
undoubtedly continue to be of interest in the future. Both language and mental imagery
in its various forms are among these constructs. The predictions of DCT have held up
favorably against other theories to date and its future in advancing knowledge of reading
is promising.
dual coding theory 52
References
Baluch, B., & Besner, D. (2001). Basic processes in reading: Semantics affects
Black, J. B., Turner, T. J., & Bower, G. H. (1979). Spatial reference points in
language comprehension. Journal of Verbal Learning and Verbal Behavior, 18, 187-198.
Clark, J. M., & Paivio, A. (1991). Dual coding theory and education.
Coltheart, M., Curtis, B., Atkins, P., & Haller, M. (1993). Models of reading
100, 589-608.
Coltheart, M., Laxton, V. J., & Keating, C. (1988). Effects of word imageability
and age of acquisition on children’s reading. British Journal of Psychology, 79, 1-12.
Coltheart, M., Patterson, E. K., & Marshall, J. C. (Eds.) (1980). Deep dyslexia.
Coltheart, M., Rastle, K., Perry, C., Langdon, R., & Ziegler, J. (2001). DRC: A
dual route cascaded model of visual word recognition and reading aloud. Psychological
Eddy, J. K., & Glass, A. L. (1981). Reading and listening to high and low
imagery sentences. Journal of Verbal Learning and Verbal Behavior, 20, 333-345.
Edfeldt, A. W. (1960). Silent speech and silent reading. Chicago, IL: University
of Chicago Press.
Ellis, A. W., & Monoghan, J. (2002). Reply to Strain, Patterson, and Seidenberg,
(2002). Journal of Experimental Psychology: Learning, Memory, & Cognition, 28, 215-
220.
Fillmore, C. J. (1968). The case for case. In E. Bach & R. T. Harms (Eds.),
Universals in linguistic theory (pp. 1-88). New York: Holt Rinehart & Winston.
Fillmore, C. J. (1977). The case for case reopened. In P. Cole & J. M. Sadock
(Eds.), Syntax and semantics (vol. 8) Grammatical relations (pp. 59-81). New York:
Academic Press.
(Ed.), New directions in linguistics and semiotics (pp. 123-147). Houston, TX: Rice
University Studies.
Frank, S. D. (1994). The Evelyn Wood seven-day speed reading and learning
Funnell, E., & Allport, A. (1987). Non-linguistic cognition and word meanings:
Prinz, & E. Scheerer (Eds.), Language perception and production: Relationships between
listening, speaking, reading, and writing (pp. 367-400). New York: Academic Press.
performance of first and third graders. In J. A. Niles & L. A. Harris (Eds.), New inquiries
in reading research and instruction (pp. 131-135). Thirty-first Yearbook of the National
Goetz, E. T., Sadoski, M., & Olivarez, A., Jr. (1991). Getting a reading on reader
response: Relationships between imagery, affect, and importance ratings, recall and
Goetz, E. T., Sadoski, M., Stowe, M. L., Fetsco, T. G., & Kemp, S. G. (1993).
Imagery and emotional response in reading literary text. Poetics, 22, 35-49.
Gibson, E. J., & Levin, H. (1975). The psychology of reading. Cambridge, MA:
MIT Press.
dual coding theory 55
imagery in sight word recognition among retarded and nonretarded children. Education
53-61.
Holcomb, P. J., Kounios, J., Anderson, J. E., & West, W. C. (1999). Dual-coding,
Portugal.
Juel, C., & Holmes, B. (1991). Oral and silent reading of sentences. Reading
processing models for imagery effects in verbal memory. Psychological Bulletin, 85,
532-554.
Kolker, B., & Terwilliger, P. N. (1981). Sight vocabulary learning of first and
Lesser, V. R., Fennell, R. D., Erman, L. D., & Reddy, D. R. (1974). Organization
abstract sentences. Journal of Verbal Learning and Verbal Behavior, 16, 217-231.
Monaghan, J., & Ellis, A. W. (2002). What exactly interacts with spelling-sound
Nelson, D. L., & Schreiber, T. A. (1992). Word concreteness and word structure
Paivio, A. (1971). Imagery and verbal processes. New York: Holt, Rinehart, and
Paivio, A. (1991). Dual coding theory: Retrospect and current status. Canadian
Paivio, A., Walsh, M., & Bons, T. (1994). Concreteness effects in memory:
Plaut, D. C., & Shallice, T. (1993). Deep dyslexia: A case study of connectionist
Raman, I., & Baluch, B. (2001). Semantic effects as a function of reading skill in
Raman, I., Baluch, B., & Besner, D. (1997). Imageability and frequency effects
reading (4th ed.) (pp. 1057-1092). Newark, DE: International Reading Association.
Hinchman, & D. J. Leu (Eds.), Inquiries in literacy theory and practice (pp. 37-53).
(Ed.), Attention and performance VI (pp. 573-603). New York: Academic Press.
Sacks, O. (2002, October 7). The case of Anna H. The New Yorker, pp. 62-73.
imagery and the comprehension and recall of a story. Reading Research Quarterly, 19,
110-123.
266-278.
Sadoski, M. (2002). Dual coding theory and reading poetic text. Journal of the
Sadoski, M., & Goetz, E. T. (1985). Relationships between affect, imagery, and
importance ratings for segments of a story. In J. A. Niles & R. Lalik (Eds.), Issues in
Sadoski, M., Goetz, E. T., & Avila, E. (1995). Concreteness effects in text recall:
comprehensibility, interest, and memory for text: Implications for dual coding theory and
Sadoski, M., Goetz, E. T., & Kangiser, S. (1988). Imagination in story response:
Sadoski, M., Goetz, E. T., Olivarez, A., Lee, S., & Roberts, N. M. (1990).
Imagination in story reading: The role of imagery, verbal recall, story analysis, and
Sadoski, M., Goetz, E. T., & Rodriguez, M. (2000). Engaging texts: Effects of
Sadoski, M., & Paivio, A. (1994). A dual coding view of imagery and verbal In
reading (4th ed.) (pp. 582-601). Newark, DE: International Reading Association.
Sadoski, M., & Paivio, A. (2001). Imagery and text: A dual coding theory of
Sadoski, M., Paivio, A., & Goetz, E. T. (1991). A critique of schema theory in
reading and a dual coding alternative. Reading Research Quarterly, 26, 463-484.
dual coding theory 61
Sadoski, M., & Quast, Z. (1990). Reader response and long term recall for
journalistic text: The roles of imagery, affect, and importance. Reading Research
359.
Strain, E., Patterson, K., & Seidenberg, M. S. (1995). Semantic effects in single-
21, 1140-1154.
Strain, E., Patterson, K., & Seidenberg, M. S. (2002). Theories of word naming
words on subsequent learning of high or low imagery words. Reading World, 21, 286-
292.
van der Veur, B. W. (1975). Imagery rating of 1,000 frequently used words.
speaking, reading, and writing (pp. 159-179). New York: Academic Press.