A Dual Coding Theoretical Model of Decoding

Reading Psychology, 33:465–496, 2012
Copyright
C Taylor & Francis Group, LLC
ISSN: 0270-2711 print / 1521-0685 online

DOI: 10.1080/02702711.2011.557330
A DUAL CODING THEORETICAL MODEL OF DECODING

IN READING: SUBSUMING THE LABERGE AND SAMUELS
MODEL
MARK SADOSKI and ERIN M. McTIGUE

Department of Teaching, Learning, and Culture, Texas A&M University,
College Station, Texas
ALLAN PAIVIO
Department of Psychology, University of Western Ontario, London, Ontario,
Canada
In this article we present a detailed Dual Coding Theory (DCT) model of de-
coding. The DCT model reinterprets and subsumes The LaBerge and Samuels
(1974) model of the reading process which has served well to account for decod-
ing behaviors and the processes that underlie them. However, the LaBerge and
Samuels model has had little to say about comprehension processes. Its subsump-
tion into DCT provides an account of decoding in DCT terms and connects it to
the established DCT explanation of comprehension. Relevant research evidence is
reviewed. We propose that this constitutes an incremental advancement toward
unifying theories in reading.
The purpose of this article is to present a detailed account of de-

coding in reading as explained by Dual Coding Theory (DCT),
and to show that it subsumes the familiar LaBerge and Samuels
(1974) model of the reading process, as currently revised (we will
refer to it hereafter as the LS model). We will show that the DCT
explanation accounts for empirical evidence about learning words
that the LS model alone does not explain. We will propose that
this constitutes an important advancement toward a unified the-
ory of reading.
DCT is a general theory of all human cognition that has
been extended into many domains (Paivio, 2007). It has been
applied in detail to various aspects of literacy including reading,
composing, and spelling (see Paivio 1971, 1986, 2007; Sadoski,
1992; Sadoski, Goetz, Stricker, & Burdenski, 2003; Sadoski,
Address correspondence to Mark Sadoski, Office of Educational Development,
Texas A&M Health Science Center College of Medicine, 8447 State Highway 47, Bryan,
TX 77807. E-mail: [email protected]
465
466 M. Sadoski et al.
Kealy, Goetz, & Paivio, 1997; Sadoski & Paivio, 1994, 2001, 2004;
Sadoski, Willson, Holcomb, & Boulware-Gooden, 2005). DCT
has challenged single-code theories such as schema theory and
construction-integration theory (Kintsch, 1998, 2004) as an
explanation of comprehension and memory in reading (Sadoski
& Paivio, 1994, 2001, 2004; Sadoski, Paivio, & Goetz, 1991). How-
ever, to date, the DCT explanation of reading has not articulated
a fully detailed account of decoding such as that provided by the
LS model. One purpose here is to show that DCT can absorb
the LS model into its established framework without altering any
DCT principles. That is, the resulting model is not a hybrid of
DCT and the LS model, but a reinterpretation and integration
of the LS model into DCT. This subsumption supplies a detailed
DCT account of decoding and its relationship to comprehension.
This degree of theoretical unification has seldom been attempted
in reading theory (Sadoski & Paivio, 2007).
We emphasize that we deal here only with how DCT sub-
sumes the LS model as revised. Other theories of the decoding
process, while sharing much with the LS model, deserve their own
treatment. We therefore will not deal with these other theories
in this article, but we have commented on this class of theories
elsewhere (see Sadoski & Paivio, 2001, pp. 130–133, 2004, 2007).
Two contemporary models that share much with the LS model
are the dual route model of Coltheart, Rastle, Perry, Langdon,
and Zeigler (2001) and the parallel distributed processing model
of Harm and Seidenberg (2004). These models are connection-
ist models designed primarily for computer simulations of word
reading and have little to say about comprehension in context.
Seidenberg (2005, p. 238) summarizes this class of models: “Note
that the term ‘reading’ covers many more phenomena than are
addressed by our models, which focus on comprehending isolated
words.” Our mission here is a detailed explanation of decoding in
DCT terms, and to show how DCT can absorb what is often re-
garded as the most widely accepted and cited account of the de-
coding process (Samuels, 2004) into an established architecture
that addresses comprehension, reader response, and many other
aspects of literacy.
The original LS model of reading was designed as a detailed
account of cognition during decoding that culminated with the
meanings of words and word groups (LaBerge & Samuels, 1974).
Dual Coding Theoretical Model of Decoding in Reading 467
The original model was a bottom-up model that did not include
mechanisms for higher-order processes such as text comprehen-
sion to affect lower-order process such as grapheme-phoneme cor-
respondences. Samuels (1977, 1994, 2004) subsequently revised
the model by adding feedback loops to make it more interactive,
but otherwise the model remained intact with its emphasis on de-
coding. We will detail these revisions later, and for the remainder
of this article we will refer mainly to the current model as presently
revised by Samuels.
The LS model is quite comprehensive in its account of decod-
ing, covering a broad spectrum of decoding behaviors. However,
a continuing limitation of this model has been acknowledged by
Samuels (2006a, pp. 334–335): “Although our model shows how
the information from the page is processed and moved along to
comprehension, the model has almost nothing to say about the
comprehension process.” We will show that subsumption of the
LS model by DCT resolves this problem in a theoretically consis-
tent manner.
Unifying Macrotheories and Microtheories
The absorption of microtheories that are limited to particular

phenomena into macrotheories with more inclusive architecture
is a necessary condition of science. In fact, the progressive uni-
fication of theories is a goal of all branches of science (Holton,
1986; Judson, 1980; Sadoski & Paivio, 2007). For example, both
Kepler’s theory of the motion of the planets and Galileo’s theory
of the motion of earthbound objects were subsumed by Newton’s
larger theory of physics, because all the explanatory power of
the former theories was incorporated in the latter theory. That
is, Newton’s theory accounted for the motion of all massive ob-
jects whether celestial or terrestrial, and it went on to explain even
more events than either Kepler’s theory or Galileo’s theory could
explain alone. Einstein’s theory of general relativity later reinter-
preted the behavior of massive bodies with still more explanatory
power, and so on.
The goal of attaining powerful, unified theories is the same
in the social sciences (Maleske, 1995; Van Dalen, 1979). However,
this progression has moved more slowly in the social sciences due
to their relative youth and to the complexity in defining, measur-

ing, and replicating their theoretical constructs. Scientific theo-
rizing in cognition took shape in the 1950s and 1960s, and ever
since it has adopted a part-to-whole strategy. That is, theory and
research have been primarily directed toward carefully limited
pieces of cognition, under the assumption that first we should
understand all the pieces separately and then merge them into
a unified whole (Newell, 1973).
Newell (1973, 1990) questioned the value of this piecemeal
approach. He felt that rather than advancing toward unification,
the field was becoming increasingly bogged down in small issues
that never seemed to get settled. He proposed that greater theo-
retical progress would be made by using the opposite approach.
Developing unified theories that addressed all aspects of cogni-
tion would allow the pieces to be more productively studied by un-
derstanding how they were constrained to fit into the whole. This
situation seems a particularly apt description of reading, where
for decades different theories of decoding, comprehension, and
response have been advanced with few candidates for a unified
theory to encompass them (cf. Sadoski & Paivio, 2007; Samuels &
LaBerge, 1983).
There has been at least some effort to formulate theories of
reading that offer broad, encompassing frameworks. Over several
decades, Carver (1978, 2000) developed rauding theory, a statis-
tical model derived from Holmes’s (1970) earlier factor analytic
work. Carver’s final path model predicted reading comprehen-
sion efficiency scores with a high degree of accuracy using a set of
quantitative variables, including general verbal ability, decoding
ability, listening comprehension, and reading rate, among others.
Hoover and Gough (1990) proposed the simple view of reading,
a statistical model that predicted reading comprehension scores
from decoding scores and listening comprehension scores. Carver
(1993) was able to show exactly how the simple view of reading was
encompassed by the larger architecture of rauding theory. How-
ever, rauding theory remains a limited explanation of cognition in
reading because it offers no explanation of theoretical constructs
beyond test scores (i.e., it is positivistic).
Sadoski and Paivio (2007, p. 350) suggested that the LS
model of reading “would be readily encompassed by DCT” as an
incremental step toward a unified theory of cognition in read-
ing. That is, DCT could explain the reading process in the same
VERBAL STIMULI NONVERBAL STIMULI
SENSORY SYSTEMS
REPRESENTATIONAL CONNECTIONS
N
O
V N
Logogens Imagens
ASSOCIATIVE STRUCTURE
ASSOCIATIVE STRUCTURE
E V
R E
B R
A REFERENTIAL B
L CONNECTIONS A
L
S
Y S
S Y
T S
E T
M E
M
VERBAL RESPONSES NONVERBAL RESPONSES

FIGURE 1 General Model of Dual Coding Theory. Figure 4-1 from Paivio
(1986). Reprinted by permission of Oxford University Press.
comprehensive detail as the LS model, but it could go well be-

yond the LS model in explaining additional reading behaviors. In
the remainder of this article we will: (1) review DCT as it applies
to reading, (2) review the LS model of decoding, (3) provide a
model that uses DCT principles to reinterpret and subsume the
LS model, and (4) show how the subsumed model expands its ex-
planatory power for decoding while combining it with a consistent
and established model of reading comprehension.
Review of DCT
Figure 1 presents the general model of DCT. It is a model of men-

tal structures and processes, with processing indicated by the ar-
rows between the structural units. The processes account for men-
tal functions such as memory and communication. As applied to
the reading process, the model begins with verbal stimuli in the
form of written language that is detected by our senses (i.e., vision

in the case of print, touch in the case of Braille). As we see writ-
ten shapes, mental representations called logogens are activated.
These are indicated by the representational connections in Figure
1. Logogens are indicated by the rectangles in Figure 1 within the
associative structure of the verbal system in long-term memory.
In DCT, a logogen corresponds to anything learned as a
unit of language in some sense modality. In the visual modal-
ity, this could be familiar letters, familiar subword letter combi-
nations, written words, familiar unified written phrases, or even
larger familiar written language units. In the auditory modality,
these could be phonemes, syllable pronunciations, word pronun-
ciations, familiar unified phrase pronunciations, or even larger fa-
miliar speech units. Logogens derived from speech are theorized
to be both auditory and motor in modality, because speech units
may be learned as much or more by motor production (i.e., articu-
lation) as by listening (Liberman & Mattingly, 1985). That is, audi-
tory and motor logogens used in speech can be closely connected
and rapidly coactivated in both directions. Logogens vary in size,
with words perhaps being the modal length. Therefore, logogens
can be thought of as one variant of the concept of the mental lex-
icon (Paivio, 2010; Paivio & Sadoski, 2011). Although recognized
as wholes up to a point, logogens also can be analyzed into smaller
units or synthesized into larger units. Considerable neuropsycho-
logical evidence and brain imaging research supports the assump-
tion that logogen units assume various sizes (Paivio, 2010).
Visual letter features distinguish one alphabetic letter from
another (as strokes and radicals distinguish one character from
another in logographic languages, etc.). For example, the letter E
is distinguished from the letter F by the additional horizontal fea-
ture at the bottom of the E. This same feature distinguishes the
letter L from the letter I , and so on. The detection of such fea-
tures distinguishes otherwise identical or comparable letter units,
so that features play a role in distinguishing logogens for letters.
Letters can be recognized as wholes, but they may be analyzed into
constituent visual features as well. Neuropsychological and behav-
ioral evidence suggests that we tend to recognize wholes before
parts, but this bias can be rapidly changed so that we can focus
first on parts when the situation compels (Lupker, 1979; Paivio,
2007; Rafal & Robertson, 1995; Sadoski, 2003).
Auditory-motor (i.e., speech) logogens are associated with vi-

sual logogens with varying degrees of probability. In English and
other languages with irregular orthographies, spellings can be as-
sociated with various pronunciations depending on orthographic
conventions and context. Most graphemes represent some variety
of phonemes, with morphology and context frequently determin-
ing which (e.g., c in came, cent, cello; ph in telephone or haphazard;
minute detail vs. minute waltz). Based on our reading experience,
some associations become so high in probability that they become
overlearned and “automatic.” However, the verbal system must be
flexible enough to deal with new orthographic and contextual sit-
uations as they are encountered, so that relationships must remain
probable but not certain. This is the DCT account of automaticity,
the term used in the LS model to indicate the rapid association of
a spelling with a pronunciation with little or no conscious atten-
tion. These associations occur within the verbal associative struc-
ture and are part of long-term memory.
Logogens are associated with other logogens within the ver-
bal system in a wide variety of other ways. Such memory asso-
ciations are readily shown by tests of free association, defining,
sentence completion, and cloze, among many others. Associative
processing within the verbal system is particularly important in
understanding the meaning of abstract words for which there are
no concrete referents. For example, the abstract word true gains
its general meaning from verbal associations such as factual, not
false, accurate, and so on. Concrete words such as tree have access
to this form of meaning as well, but they have the additional ad-
vantage of referential associations with mental images of objects
and events.
The units in the nonverbal system are called imagens. Im-
agens are memory representations of nonverbal objects, events,
and situations. When activated, they take the form of mental
images of environmental scenes, sounds, tactile and kinesthetic
experiences, smells, and tastes. Imagens are modality-specific and
vary in size, and they tend to be activated in nested sets. For ex-
ample, an imagen of a familiar face is a holistic unit that can
be decomposed into eyes, a nose, a mouth, and other features.
Likewise, orchestral or band music can be decomposed into the
sounds of percussion, strings, brass, or other auditory elements.
As with actual experience, images are often embedded in larger
images that either share a common sensory modality (e.g., a vi-

sual image of a meadow of different kinds of wildflowers) or are
composed of images in different sensory modalities (e.g., imagin-
ing the sights, sounds, touches, smells, and tastes of a fine dining
experience). Imagens are indicated by circular shapes in Figure 1.
In reading, imagens are activated via the referential connec-
tions from logogens to imagens (Figure 1). This can be direct, as
in the case of tree, or as part of a processing chain, as when the
abstract word religion is associated with the concrete word church,
which in turn evokes an image of a church. Once activated, ima-
gens can reinforce the activation of the initial logogen or activate
new logogens as indicated by the bidirectional arrows in Figure
1. That is, mental images both reinforce and constrain the set of
probable language associations that might be activated in a given
context.
For example, the phrase lay up can have several common
meanings, including (a) to stock provisions for future use or (b)
to hit a golf ball short so as to avoid a hazard. Depending on con-
text, mental images of these different situations might be activated
and in turn activate contextually appropriate language such as pro-
visions or sand trap. For basketball fans, the compound word layup
activates a still different logogen that is associated with (c) a one-
handed basketball shot, evoking a quite different image and/or
activating logogens such as backboard. In DCT, these constrained,
probabilistic interconnections constitute the basis of meaning.
Hence, the reader comprehends text through a combination of
lexico-grammatical associative activity in the verbal system and any
referential activity in the nonverbal system.
The responses in Figure 1 indicate verbal and nonverbal reac-
tions to cognitive events, such as overtly or covertly pronouncing
language units or forming mental images of them. We will elabo-
rate more on the elements in Figure 1 throughout this article.
Review of the LS Model
Figure 2 and Figure 3 present the LS model (Samuels, 2004). As

noted earlier, the model focuses on decoding, and the description
of comprehension is “sketchy in the extreme” (Samuels, 2004,
p. 1140). This theory contains four main components: visual
memory (represented as VM in the diagrams), phonological
FIGURE 2 The LaBerge and Samuels Model. Figure 2 from Samuels (2004).
Reprinted with permission of the International Reading Association.
memory (PM), episodic memory (EM), and semantic memory

(SM). A contributing process, the reader’s attention (A), is also
represented.
Visual memory is the first step of the process, where atten-
tion is directed to the visual information in the written text. At
the next step, phonological memory, the appropriate phonolog-
ical representation of the visual information is activated. Simul-
taneously, episodic memory may be activated by specific contex-
tual details. The final step is semantic memory, which is a general
store of information such as word meaning alternatives. While the
FIGURE 3 Model of Visual Memory. Figure 3 from Samuels (2004). Reprinted

with permission of the International Reading Association.
directional process so far describes a “bottom-up” model of read-

ing, top-down processes are also included. Top-down information
arrives via feedback loops, which are represented in the model
by the arrows connecting the semantic memory to the phonolog-
ical memory and visual memory (Figure 2). The addition of the
feedback loops accounts for evidence that context can semanti-
cally prime word recognition. Therefore, the feedback loops al-
low higher-order processing, such as comprehension, to influence
lower-order processing, such as visual or phonological processing.
To consider the model’s components in greater detail, we next
describe them individually.
Visual Memory (VM)
The first stage of processing occurs within visual memory and is

more fully represented in Figure 3, a detailed cut-out from Figure
2. In Figure 3, visual memory is hierarchically organized, with fea-
tures at the beginning and words at the end. The reader’s eye de-
tects details of visual information in the print—the features. Fea-
tures are the individual shapes, which comprise letters, such as the
arc of the c or the intersection of the x. Because the same letter
features are present in multiple letters, such as the same arc in c

and d, the relative location between the features is also processed.
Letters are recognized when features are detected, analyzed, and
synthesized into each letter’s unique set of visual properties. The
letter level of the hierarchy is represented by l. Proceeding for-
ward in Figure 3, letters are synthesized into consistent spelling
patterns (sp). Spelling patterns include familiar graphemic com-
binations such as digraphs (e.g., ch, th), rimes (e.g., -at -in), and
morpheme units (e.g., un-, -ing). At the end of the hierarchy, the
spelling patterns combine into words at the level of visual word
codes(v (w)).
However, not all words proceed through all four levels of the
hierarchy, because some words are recognized immediately. In
Figure 3 these shortcuts can be observed from f1 and f2 , in which
the unique configuration of a whole word or a word part triggers
more immediate word recognition. For example, the word pizza
on a lunch menu is frequently (and enthusiastically) recognized
by emergent, kindergarten prereaders due to its general configu-
ration, including the unusual “zz” spelling pattern in the middle
of the word.
In addition, the LS model specifies whether processing re-
quires attention or not. In both Figure 2 and Figure 3, a filled
circle represents a well-learned code that does not require active
attention to process. The information flow between such codes
is shown by solid lines (all routes to v (w1 ) in Figure 3) and
indicates that this processing can proceed without attention. In
contrast, the empty circle routes with dotted lines (several routes
to v (w2 ) in Figure 3) indicate that the reader has not learned
these codes as well and therefore needs to more actively attend to
them.
Phonological Memory (PM)
Phonological memory, the next step in the process, receives input

from multiple sources. The information units within the phono-
logical memory are in auditory and articulatory form and in-
clude articulatory features, phonemes, syllables, and words (these
are not illustrated in Figure 2). Analogous to visual memory,
these phonological units are organized hierarchically. Within the
phonological memory system, features comprise the bottom of
the hierarchy, then phonemes, syllables, and words. Phoneme

units are defined by a combination of sound and motor articu-
lation features. For example the phonemes /s/ and /z/ are both
alveolar fricatives, but the voicing differs between the sounds. In
comparing /s/ and /th/ (as in thin) the voicing and frication are
the same, but the place of articulation differs because the /th/ is
produced by locating the tongue between the teeth. That is, each
phoneme is a unique set of articulatory and auditory features.
Within phonological memory, processing can proceed in ei-
ther direction, bottom-up or top-down. For example, during a
blending task, the student is engaging in a bottom-up process by
producing phonemes individually and synthesizing them into a
larger unit such as a word. During a phonemic analysis task, such
as removing the /d / from /drag/, the student is engaged in top-
down processing as she removes a sound unit from a whole word
unit.
Episodic Memory (EM)
Episodic memory differs from the previously described compo-

nents, because it is not within the direct, bottom-up reading route
from visual recognition to meaning. Instead, episodic memory
engages the contextual information connected to memories
of specific events. For example, if you are reading about crew
boats, you may recall a particularly memorable time when you
rowed a boat, such as the river you were rowing on and anyone
who was with you. In this theory, episodic memory differs from
semantic memory, because semantic memory retains the knowl-
edge but not the referential memories associated with learning
that knowledge. For example, you likely know the ABC song but
may not recall learning it, and therefore it would be stored in
semantic memory, according to this theory. This leads to the final
component in this model, semantic memory.
Semantic Memory (SM)
According to the LS model, semantic memory is where the mean-

ing primarily occurs in the reading process. Individual words pro-
cessed in visual memory and phonological memory activate ap-
propriate meanings in semantic memory. This process may be
automatic if the word is familiar, or it may require the attention

of the reader. Additionally, the meaning of grammatical phrases
and sentences is synthesized in semantic memory. Attention is re-
quired to organize both the individual meanings of words and
their relationships within the grammatical units. Samuels (2004)
explains that when you have a noun phrase (e.g., global warming ),
both global and warming individually produce meanings, but the
phrase global warming is more than simply a sum of the two parts,
it has a meaning that is unique. Attention is therefore required to
(1) determine the possible meanings of the words, (2) determine
the grammatical relationships within a sentence, and (3) combine
this knowledge into contextual meaning. As noted earlier, in com-
parison to the rich description provided for the previous compo-
nents, the LS model’s description of contextualized comprehen-
sion lacks a robust description that is broadly applicable.
The DCT Model of Decoding
Overview of the Model
The diagram in Figure 4 presents the DCT model of decoding

that integrates and subsumes the LS model. The elements of the
DCT model that are included are: sensory stimuli, logogens, im-
agens, the associative structure of the verbal code, the associative
structure of the nonverbal code, representational connections, as-
sociative connections, and referential connections. The elements
of the LS model that are included here are: sensory stimuli, let-
ter features, letters, spelling patterns, words, word groups, and
feedback loops. Elements of the LS model not included in this
diagram are: undefined meaning codes, indicators of attention
and its flow, and the distinction between semantic memory and
episodic memory. The way DCT deals with all the elements of the
LS model that are not included in Figure 4 has been discussed
earlier and will be discussed further.
The integrated model illustrated in Figure 4, although
complex, is by necessity an incomplete and imperfect account of
reading in its full complexity. In the explanations that follow, we
will attempt to address some of the integrated model’s limitations
by discussing some aspects of reading that are not illustrated in it
but that are covered in the general theory.
FIGURE 4 DCT Model of Decoding.
In Figure 4, processing moves from bottom to top. As in the

LS model, letter features are abbreviated by F in the diagram, let-
ters by L, spelling patterns by SP, words by W, and word groups by
WG. As in the DCT model, a logogen is represented by a box and
an imagen is represented by a circular shape. Each logogen box or

imagen circle has a letter in it that abbreviates its sensory modal-
ity. For example, the visual modality is abbreviated by V, auditory
by A, and haptic by H. The auditory/motor nature of phonemes
or larger pronunciation units is abbreviated by A/M. For simplic-
ity, the sensory modalities of smell and taste are not included in
the diagram, although they possibly could be evoked in reading
as mental images of smelly or tasty objects in a multimodal image.
The set of overlapping circles surrounded by a larger circle indi-
cates one example of a multimodal image of an object or event in
context.
For example, each visual logogen of a letter is represented
by a V in a box with an L beneath the box. The number following
each L indicates a separate visual letter. Visual logogens of spelling
patterns are represented by a V in a box with an SP and a number
beneath it, and so on. Imagens are not numbered, so as to indicate
their holistic, embedded nature. In our discussion, we will simply
combine the letter abbreviations for explanation. Hence, in the
following discussion, we will use VL1 to refer to a single visual
letter logogen, A/MWG1 to refer to an auditory/motor logogen
at the word-group level, V to refer to a visual imagen, and so on
(also see the legend in the figure).
Note that separate visual letter features are not placed in lo-
gogen boxes in Figure 4. Isolated visual letter features, such as the
horizontal line at the bottom of the E that distinguishes it from
the F , are not letter-level language units (i.e., logogens) by them-
selves. Rather, they combine in various ways to form letter units.
We will refer to isolated visual letter features as VF1 and VF2 in
text, but this does not imply that letter features themselves are
logogens. However, exactly what constitutes a feature in written
language is not always easy to define. This will be explained more
as we proceed.
Logogens of a given size and modality are at the same hori-
zontal level in the diagram. Visual features are at the lowest level,
visual letters at the next level, visual spelling patterns at the next
level, visual words at the next level, and so on. Note that the audi-
tory/motor pronunciation logogen corresponding to each visual
logogen is offset one level above. This implies that the mental pro-
nunciation of a language unit typically follows its visual processing
by a brief time interval (i.e., 100–300 milliseconds).
The arrows between logogens (boxes) indicate associative

connections within the verbal code, and the arrows between the
imagens (circles) indicate associative connections within the non-
verbal code. Arrows running between logogen boxes and imagen
circles indicate referential connections between the codes. Note
that the arrows are unidirectional, from the bottom up. This is in
keeping with the LS model but does not imply that reading is a
strictly bottom-up process. Context effects are indicated by feed-
back loops from the LS model, to be discussed shortly.
Memory systems are indicated on the left side of the model.
The two memory codes of DCT are indicated: the verbal code and
the nonverbal code. As the labels indicate, both are theorized as
associative networks of the various units in that code. The large
brackets indicate the part of the diagram devoted to each mem-
ory code, so that the bracket for the verbal code covers all the
logogens (boxes), and the bracket for the nonverbal code covers
all the imagens (circles).
Note the feedback arrows immediately outside the verbal and
nonverbal code brackets. They correspond to the interactive feed-
back arrows in the LS model used to indicate that higher-level pro-
cesses can affect lower-level processes in reading (i.e., interactiv-
ity). As modeled here, feedback can occur between levels within a
code and also between codes. To avoid redundancy in the model,
they are used in place of certain bidirectional arrows in the DCT
model that represent interactivity. This will be discussed in the
context of individual examples later.
Processing Operations Illustrated in the DCT Model
In explaining the reading process, both the DCT model and the
LS model begin with external stimuli in the form of written lan-
guage. Readers detect the stimuli and process them at the various
levels already described. Although we start our discussion at this
point, we note that interactive models of reading allow that con-
text effects can influence where the eyes fixate in a line of print,
how much of the available visual stimuli are detected, and how
those stimuli are interpreted (see reviews by Paulson & Goodman,
1999; Rayner, 1997). With this in mind, we begin at the level of vi-
sual letter features.
FEATURE LEVEL PROCESSES

What comprises a “feature” of written language is somewhat a
matter of orthographic familiarity, and any model must take into
account a variety of options. For example, fonts differ widely in
the features used to represent letters (e.g., a, a ), as a visit to a
greeting card shop will amply demonstrate (Sadoski, 2003). Cur-
sive handwriting runs letters together and sometimes uses the
same shape for different letters or letter combinations. That is,
perceptions of letters often depend on the surrounding letters
(Nash-Weber, 1975; Rumelhart, 1977; Sadoski, 2003). However,
features in some form or another are used to discriminate letters,
numbers, punctuation marks, and so on.
VF1 and VF2 represent separate features that combine to
make a letter, VL1, or other written symbol with multiple features.
Examples of such letters or symbols might include the combina-
tion of the features c and - to make e , or - and l to make t. Note
that a feature can appear in more than one letter or symbol, as -
can be the crossbar in either e or t. Analysis at this level would be
mainly for those learning an orthography or for situations where
an individual feature can be critical. For example, the sentences
Charge the battery and Change the battery differ by only one letter
feature but evoke different situations (Sadoski & Paivio, 2001).
LETTER LEVEL PROCESSES

Letters can be activated holistically, without analysis of their
features. In DCT terms, visual letter logogens are activated directly
from print stimuli as represented by the arrow from the visual sen-
sory system to VL2. This would necessarily be the case with single-
feature letters such as c, o, or l. Context can affect recognition at
this level, as in recognizing the difference between the number 1
and a lowercase letter l, both of which might have identical form
in a given font.
Letters activate associated phonological features and
phonemes depending on contexts. For example, letter contexts
determine whether the hard or soft c is activated (e.g., soft c
before i, e, or y, generally). The activation of some phonological
representation is indicated in the diagram by the arrow from
VL1 to A/ML1 and from VL2 to A/ML2. However, a letter might
activate no phoneme, as in the case of silent letters. Also many
nonletter written symbols have no graphophonemic counterpart,
such as &, #, @, and others (i.e., such symbols may be associated

with a word label, but they are not “sounded out”). For simplicity,
silent letters and symbols are not directly represented in Figure 4.
Letters also combine into visual spelling patterns. This is in-
dicated by the arrows from VL1 and VL2 to VSP1. For example,
the letter s and the letter h can combine into the familiar digraph
sh, or the letter t and the letter h can combine into the digraph
th. However, visual unity is not automatic, because these separate
letters can retain separate identities depending on morphological
context, such as the sh in mishap, the th in sweetheart, and so on.
Again, such associations are probabilistic based on orthographic
conventions.
Just as letters may be recognized directly without analysis of
their features, spelling patterns may be recognized directly with-
out analysis of their individual letters. This is represented by the
arrow from the visual sensory system to VSP2.
Accordingly, spelling patterns can activate unified phonolog-
ical associations, as indicated by the arrows from VSP1 to A/MSP1
and VSP2 to A/MSP2. Individual letter phonemes that retain sep-
arate phonological identities in spelling patterns are represented
in the diagram by the arrow from A/ML1 to A/MSP1 and A/ML2
to A/MSP1. For example, the common subword spelling pat-
tern in (i.e., used as a prefix or rime) is pronounced with two
phonemes that can be “sounded out” separately, as opposed to
digraphs like sh or th.
WORD LEVEL PROCESSES

Some word level processes have already been implied, be-
cause letters comprise words and words affect letter recognition
and pronunciation. In Figure 4, a visual word logogen, repre-
sented by VW1, also can be perceived as a whole without analysis
(i.e., sight words). This is represented by the arrow from the visual
sensory system to VW1. A word can also be synthesized from sepa-
rately perceived spelling patterns, like prefixes, roots, and/or suf-
fixes. This is indicated by the arrows from VSP1 and VSP2 to VW3.
For example, the word bicoastal may be recognized as a sight word
for those familiar with it, or it may be analyzed into its syllabic-
morphemic parts (bi, coast, al) and synthesized back into a pro-
nunciation and/or meaning.
Visual words activate their contextually appropriate pronun-

ciations, as indicated by the arrows from VW1 to A/MW1, VW2 to
A/MW2, and VW3 to A/MW3. For example, the word minute is
pronounced differently depending on context (e.g., minute detail
vs. minute waltz). Word pronunciations may also be synthesized or
“sounded out” from spelling pattern pronunciations, as indicated
by the arrows from A/MSP1 and A/MSP2 to A/MW3 (e.g., /th-/
and /-in / = /thin/). This can also be affected by context. For
example, the vowel in the familiar morpheme con takes different
pronunciations depending on whether the word is used as a noun
or a verb (e.g., convert, conflict).
Words are associated with other words in meaningful ways.
This is one aspect of meaning in DCT. For example, an abstract
word such as integrity might be associated with the abstract word
honesty as a meaningful associate (i.e., synonymy). This is indicated
by the arrow from VW1 to VW2. In Figure 4, these abstract words
have no direct connections to imagens. However, concrete words
could be referentially associated with an imagen. This is another
aspect of meaning in DCT. For example, the concrete word phone
could activate a visual image of a telephone. This is indicated by
the arrows from VW3 and A/MW3 to V. This image may in turn
activate images in other modalities, such as the auditory image of
a familiar ringtone, as indicated by the arrow from V to A. While
not separately illustrated in Figure 4, a concrete noun like phone
might also evoke other words such as cell, and so on. As discussed
earlier, the capacity of concrete language to access both verbal
and nonverbal codes is an important part of meaning in DCT.
WORD GROUP LEVEL PROCESSES

Word groups also can be perceived as units (e.g., rock’n’roll).
In Figure 4, this is represented by the arrow from the visual sen-
sory system to VWG1. Word groups could in turn evoke their asso-
ciated complete pronunciations, as indicated by the arrow from
VWG1 to A/MWG1. Word groups might also be recognized as
comprising individual words (e.g., rock and roll). This is indicated
in Figure 4 by the arrows from VW1 to VWG1 and from A/MW1
to A/MWG1.
Like words, word groups may evoke mental images. In Figure
4, either VWG1 or A/MWG1 or both might activate a multimodal
image. The compound word cellphone, for example, might activate
an image of a device with a certain visual appearance (V imagen),

one or more sounds such as a ringtone or the sound of beeps as
the keys are pressed (A imagens), and a haptic sensation such as
holding the device (H imagen). Others are possible. While not di-
rectly illustrated in Figure 4, the multimodal mental image might
activate still other associated language units and images such as
smart phone and any related images.
Word groups can be large phrases. The extent to which larger
word groups are recognized holistically is a matter of familiarity.
For example, to a contract attorney, familiar phrases such as “the
party of the first part” or “the part of the second part” may be
recognized as units without further analysis. Clichéd phrases such
as “at this point in time” are other common examples.
PROCESSING OPERATIONS NOT ILLUSTRATED OR EMPHASIZED IN

THE DIAGRAM
One limitation of the diagram is its inability to show the
full amount of interactivity that actually occurs in reading. The
spreading activation between units and the constraining effects of
context were noted but perhaps not emphasized enough.
For example, we pointed out that context determines
whether th is interpreted as a digraph, as in toothbrush, or sepa-
rately pronounced letters, as in sweetheart. Word level (i.e., mor-
phemic) processing determines the pronunciation of th in these
words. This is represented in Figure 4 by the feedback arrow from
the word level to the spelling pattern level. Likewise, we pointed
out that the pronunciation and meaning of the spelling minute,
like other heteronyms, is determined by context beyond the word
level. This is represented in Figure 4 by the feedback arrow from
the word group level to the word level.
Feedback also can have effects within a level. For example,
the appearance of the word The at the beginning of a word
group affects the probabilities of subsequent words according
to grammar conventions and any prior context. Likewise, the
perception of certain letters can interactively affect each other
within a word. For example, an initial p, k, g, or m followed by
an n simultaneously decreases probable word candidates and
increases the probability that the initial p, k, g, or m will be
silent. Alternatively, the perception of certain initial letters can
be altered by subsequent letters within a word. For example, ch is
pronounced /ch/ when followed by -oice, but /k/ when followed

by -oir , and so on. The general DCT model (Figure 1) uses
bidirectional arrows to emphasize some of this interactivity, but
these were eliminated in Figure 4 because of their redundancy
with the feedback loops from the LS model. The point is that the
full complexity of interactivity in reading is impossible to model
in any diagram, but it is not slighted in either theory.
Because the LS model primarily explains decoding and word
meaning, Figure 4 mainly emphasizes these. However, much more
is implied by the associations within and between the verbal code
and the nonverbal code.
For example, comprehension processes are much more ex-
plicit in DCT than in the LS model. Rather than the relatively
nonspecific use of the term “meaning codes” in the LS model,
meaning is explained more specifically in terms of the verbal and
nonverbal modality-specific codes in DCT and connections be-
tween them. For example, we noted that the abstract word integrity
may gain defining meaning from its association with the word hon-
esty. Integrity could have other meaningful verbal associates includ-
ing principled, dedicated, and so on. All those terms are networked
in a verbal semantic neighborhood that defines them all to
varying degrees.
However, as contexts change so do meanings. When used in
the context of a physical structure like a bridge, the word integrity
takes on a different shade of meaning, that is, physical structural
strength. In this case, defining verbal associations would more
likely be strength, soundness, stability, and so on. Mental images of
bridges may be referentially evoked as well. In the context of the
artistic architecture of buildings, integrity may evoke both shades
of meaning and a whole different set of images. In the reading
of extended texts, all these associations constrain each other
into complex mental models. The main point is that meaning
is specifically modeled here but not in its full complexity. For
further extended discussions of the DCT account of meaning,
mental models, and a comparison to other theories see Sadoski
and Paivio (1994, 2001, 2004).
CONSCIOUS VS. UNCONSCIOUS PROCESSES IN THE MODEL

All the activations and associations modeled may or may
not receive or require conscious attention. They are part of the
unconscious, spreading activation in memory as theorized in both

DCT and the LS model. However, they could be consciously expe-
rienced, as in “sounding out” words, mentally paraphrasing sen-
tences to clarify, intentionally constructing images in order to
comprehend, and so on. All of this depends on the skill of the
reader, the familiarity of the text, the purpose for reading, the
strategies applied, and other factors. This was explained in our
earlier discussions of the concepts of memory and automaticity.
Empirical Support for the DCT Subsumption of the LS Model
If the DCT model can theoretically account for all the decoding
and fluency phenomena that the LS model can, it logically fol-
lows that all such empirical support for the LS model is support-
ive of the DCT reinterpretation. We will not attempt a review of
all this evidence here, but behavioral evidence for the DCT ac-
counts of reading phenomena and research results as detailed
above has been extensively reviewed (Paivio, 1971, 1986, 2007,
2008; Sadoski, 2003, 2005, 2008; Sadoski & Paivio, 1994, 2001,
2004). Here we will briefly deal with selected research issues that
favor the subsumed model but would be difficult for any version
of the LS model to explain without the support of DCT.
Behavioral Evidence in Decoding by Young Children
One issue is the learning of sight words by young readers, a

strength claimed for the LS model (Samuels, 2004). Specifically,
the issue is the consistent research finding that word concreteness
strongly affects sight word learning in young children at the early
stages of reading, even when experimentally controlling for word
length, word frequency, spelling regularity, grammatical class, and
known meaning. Word concreteness, or word imageability, is the
ability of words to evoke mental images in the mind of the reader.
This effect is explained and predicted by DCT but not dealt with
in the LS model.
One of the first studies was conducted with 63 kindergarten
and beginning first-grade children (van der Veur, 1975). One
thousand words of all parts of speech that comprised approxi-
mately 92% of first-graders’ speech vocabularies were first normed
for imagery. A different set of 12 words was generated for each
child, with four of the words rated high in imagery, four medium,
and four low. No children received any words they could al-
ready read. The set of words was shown to the child and pro-
nounced by the experimenter five times in a different random
order each time. After a 10-minute interval, the child was again
shown each word for 5 seconds and asked to pronounce it. Words
were learned in decreasing order from high imagery to low.
Hargis and Gickling (1978) taught nine average-ability
kindergartners a set of 20 concrete sight words and a set of 20
abstract sight words that were matched for word length and fre-
quency. All the words were nouns that were familiar to the chil-
dren in speech but unknown to the children by sight. The chil-
dren: (a) were presented the words in random order 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 after instruction was com-
pleted, more than three times as many concrete words as abstract
words were correctly named by sight. Ten days after training, more
than four times as many concrete words were correctly named by
sight. These results were replicated with another group of kinder-
gartners of average ability and a group of older mentally chal-
lenged children (Gickling, Hargis, & Alexander, 1981).
Kolker and Terwilliger (1981) taught 45 first-grade children
and 53 second-grade children familiar concrete and abstract
words without context. Each child received a different set of 6
words drawn from a list of 147 high-frequency words. The words
were nouns that were unknown by sight but had meanings known
to the child. The examiner (a) presented the word for 1 second
on a flash card, (b) pronounced it, (c) had the child repeat the
word, and (d) provided correction if needed. The first graders
took about 60% more trials to learn the abstract words than the
concrete words. The second graders required only 8% more trials
for abstract words, but the difference was still highly statistically
significant. Terwilliger and Kolker (1982) replicated these results
with 94 beginning first graders while additionally varying word
confusability (i.e., same or different initial consonants).
Several other early studies similarly found concreteness ef-
fects on the sight word learning of young children (Kiraly &
Firlong, 1974; Richmond & McNinch, 1977; Wolpert, 1972; Yore &
Ollila; 1985). However, these studies were criticized for using very
small samples of words and not controlling for familiarity, among

other limitations. Two other studies specifically addressed these
limitations. McFalls, Schwanenflugel, and Stahl (1996) used 44
matched concrete and abstract words taken from the basal read-
ers in use and from outside reading sources with 62 second-grade
students. The concrete and abstract words were matched for fre-
quency, word length, and grammatical class. Controlling for gen-
eral reading ability (i.e., standardized reading test scores), con-
crete words were read with more accuracy than abstract words on
both lexical decision and naming tasks. Laing and Hulme (1999,
Experiment 2) taught 20 beginning reading children of 4–5 years
of age a set of 24 concrete and abstract words matched for famil-
iarity, age of acquisition, number of letters, number of syllables,
number of phonemes, and grammatical class. Words were not ini-
tially known to the children by sight. The set of words was shown
to the child and pronounced by the experimenter twice and the
child once. The words were then presented to the child in random
order. The concrete words were named with significantly more ac-
curacy.
Whereas we have reviewed studies involving very young
children, similar results with word naming accuracy have been
found with middle grade children (e.g., Hargis, Terhaar-Yonkers,
Williams, & Reed, 1988), and with adults in various languages
(e.g., Shibahara, Zorzi, Hill, Wydell, & Butterworth, 2003). The
finding therefore has considerable generalizability.
The important point is that these results are theoretically
consistent with DCT but not with the LS model. Because of the
inclusion of both verbal and nonverbal processes in DCT, con-
crete words can be mentally represented in more ways, making
for memories with more retrieval routes. Learners will more often
have experienced mental images to concrete words than abstract
words already known through speech, and this activation differ-
ence transfers in learning the new written words. These results
have implications for researchers and teachers in sight vocabulary,
but the crucial theoretical point is that no version of the LS model
alone could account for this body of evidence, because referential
mental imagery and word concreteness were not included in the
LS model.
On the other hand, DCT could account for the evidence
claimed as support for the LS model. One key case in point is the
success of various forms of repeated readings on building word

recognition and fluency (Samuels 2006b). In DCT, repeated read-
ing of the same words would facilitate the formation of whole-
word logogens in the verbal system (i.e., sight words) that would
be recognized with reduced effort. The nonverbal system would
play a secondary but important role in facilitating the memory
connections for concrete words, consistent with the sight vocabu-
lary learning studies cited earlier. Word frequency and regularity
are important in establishing word-level logogens in the verbal
code of DCT (Sadoski & Paivio, 2001, 2004).
Reading Comprehension
As noted earlier, the LS model is limited by lack of an account of

the comprehension process (Samuels, 2006a). The DCT account
of the comprehension process was summarized earlier and has
been elaborated and supported by extensive empirical evidence
(reviewed in Paivio, 2007, 2008; Sadoski, 2008; Sadoski & Paivio,
1994, 2001, 2004). We will not attempt to review all this evidence
here, but we will explain in some detail how DCT provides a the-
oretically consistent extension of the LS model’s limited account
of comprehension that other theories cannot.
The original LaBerge and Samuels model (1974, p. 570) di-
rectly implied that mental imagery is a form of reading compre-
hension:
When reading is flowing at its best, for example, in reading a mystery novel
in which the vocabulary is very familiar, we can go along for many minutes
imagining ourselves with the detective walking the streets of London, and
apparently we have not given a bit of attention to any of the decoding
processes that have been transforming marks on the page into the deeper
systems of comprehension.
More recently, Samuels (2006b, p. 24) explicitly stated that im-

agery is a form of reading comprehension:
In order to understand the sentence “The executive ate his steak in the
corporate dining room,” the reader may form a mental image of a scenario
that goes beyond the text information. For example, the reader may form
an image of a well-dressed man who is wearing a business suit and holding
a fork in his left hand and a knife in his right hand while cutting the steak
on a plate that sits atop a white tablecloth.
These descriptions directly implicate the comprehension of an

episode in the form of a mental image that the reader may have
never experienced. DCT directly implicates imagery in the con-
struction of elaborate, novel, imagined situations that go beyond
what may be stored in our personal episodic memories (i.e., cre-
ative imagination). However, whereas all versions of the LS model
have postulated the modality-specific difference between visual
and phonological codes for letters, spelling patterns, words, and
larger language units, no version of the LS model extends the
modality-specific principle to “word meaning codes” in seman-
tic or episodic memory. The form of these codes is not specified
in the LS model. In contrast, DCT assumes that modality-specific
representations are the basis of semantic, episodic, and procedu-
ral memory, or any other memory distinctions. The critical the-
oretical point is that the modality-specificity in the LS model is
extended to a consistent account of comprehension when inte-
grated into DCT but not some other current theories.
A basic distinction between DCT and other current theories
of comprehension and memory is that other theories are single
code theories (e.g., schema theory, context availability theory).
Single or common code theories assume that knowledge in mem-
ory is basically abstract and amodal, existing in a disembodied
state that is associated with no sensory modality. Such theories
either do not deal with concreteness effects in reading compre-
hension (e.g., schema theory) or explain it in different terms
(e.g., context availability theory). In contrast, concreteness is a
key factor in the DCT account of reading comprehension (as ex-
plained earlier, other factors include networks of language associ-
ations such as synonyms, antonyms, etc.). In DCT, the two codes
are theorized to be independent and additive in their effects, pre-
dicting that concrete language should be about twice as compre-
hensible and memorable as abstract language, other factors being
equal. The empirical record over many decades has consistently
demonstrated this (Paivio, 1971, 1986, 2007; Sadoski & Paivio,
1994, 2001, 2004). Here we will focus specifically on how predic-
tions made by DCT have prevailed in experimental tests against
an example of a common code theory.
The case in point involves the competing predictions of DCT

and various versions of context availability theory (e.g., Kieras,
1978; Schwanenflugel & Shoben, 1983). Rather than assuming
separate verbal and nonverbal codes, context availability theory
assumes that all language, whether abstract or concrete, is com-
prehended and remembered by incorporation into a common ab-
stract code. Further, context availability theory assumes that con-
crete language is easier to comprehend and remember because it
can be more readily associated with other items in the network—it
simply has more memory connections. But this advantage can be
neutralized when abstract language is very familiar or when ab-
stract language is presented in a supportive context, because the
connections are then facilitated for the reader. The differing as-
sumptions of DCT and the context availability model present a sit-
uation all too rare in reading research: two rival theories that are
sufficiently well articulated to make testable, contradictory predic-
tions.
A direct test of these competing predictions was performed
by Sadoski, Goetz, and Avila (1995). They used two sets of ex-
tended, factual paragraphs about historical figures that were
matched for number of sentences, words, and syllables; sentence
length; information density; cohesion; and rated comprehensi-
bility. In one set, two paragraphs were rated equal in familiarity
but unequal in concreteness. In this case, DCT predicted that
the more concrete paragraph should be recalled better than the
abstract paragraph due to the advantage provided by imagery,
whereas context availability theory predicted that they should be
recalled equally, because the abstract language paragraph was
equally familiar and presented in supportive 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 as well as the unfamiliar concrete paragraph,
whereas context availability theory predicted that the abstract
paragraph would be recalled better than the concrete paragraph
(i.e., any disadvantage of abstractness would be offset by higher
familiarity).
The results confirmed DCT but not context availability
theory. In the set where the concrete and abstract paragraphs
were equally familiar, the concrete paragraph was recalled nearly
twice as well. In the case where the abstract paragraph was more
familiar, the paragraphs were recalled equally well. Other exper-
imental findings using widely different methods and materials,
including neuroimaging, have confirmed these results (e.g., Gies-
brecht, Camblin, & Swaab, 2004; Holcomb, Kounios, Anderson,
& West, 1999; Kounios & Holcomb, 1994; Nelson & Schreiber,
1992; Paivio, Khan, & Begg, 2000; Paivio, Walsh, & Bons, 1994).
Conclusion: Unifying the Science of Reading
The ultimate goal of science is powerful, unified theories that ex-

plain phenomena in a given domain accurately, comprehensively,
and parsimoniously. Efforts to unify theories of different aspects
of a domain are central to the conduct of science. The science of
cognition in reading has been beset by a lack of unifying theories
to provide an overall explanatory and predictive framework for its
phenomena, including decoding, comprehension, and response
(Sadoski & Paivio, 2007). This article has provided an extended
explanation of one such unifying effort and a step toward that
goal.
The LS model has served as a venerable theory of the decod-
ing aspects of reading, but it has been limited by its account of
comprehension. Its subsumption into the macrotheoretical archi-
tecture of DCT provides an explanation that is consistent with the
LS model and demands no change in DCT. This constitutes an
advancement in theory toward a more powerful, parsimonious,
and unified account of a large variety of reading phenomena un-
der the aegis of a set of general theoretical principles that are
empirically established. Much remains to be learned, but the sub-
sumption of the LS model into DCT serves as an incremental step
toward that goal.
References
Carver, R. P. (1978). Toward a theory of reading comprehension and rauding.

Reading Research Quarterly, 13, 8–63.
Carver, R. P. (1993). Merging the simple view of reading with rauding theory.
Journal of Reading Behavior, 25, 439–455.
Carver, R. P. (2000). The causes of high and low reading achievement. Mahwah, NJ:
Lawrence Erlbaum Associates.
Coltheart, M., Rastle, K., Perry, C., Langdon, R., & Zeigler, J. (2001). DRC: A dual
route cascaded model of word recognition and reading aloud. Psychological
Review, 108, 204–256.
Gickling, E. E., Hargis, C. H., & Alexander, D. R. (1981). The function of im-
agery in sight word recognition among retarded and nonretarded children.
Education and Training of the Mentally Retarded, 16, 259–263.
Giesbrecht, B., Camblin, C. C., & Swaab, T. Y. (2004). Separable effects of seman-
tic priming and imageability on word processing in human cortex. Cerebral
Cortex, 14, 521–529.
Hargis, C. H., & Gickling, E. E. (1978). The function of imagery in word recog-
nition development. Reading Teacher, 31, 870–874.
Hargis, C. H., Terhaar-Yonkers, M., Williams, P. C., & Reed, M. T. (1988). Repeti-
tion requirements for word recognition. Journal of Reading, 31, 320–327.
Harm, M., & Seidenberg, M. S. (2004). Computing the meanings of words in
reading: Division of labor between visual and phonological processes. Psycho-
logical Review, 111, 662–720.
Holcomb, P. J., Kounios, J., Anderson, J. E., & West, W. C. (1999). Dual-coding,
context-availability, and concreteness effects in sentence comprehension: An
electrophysiological investigation. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 25, 721–742.
Holmes, J. A. (1970). The substrata-factor theory of reading: Some experimental
evidence. In H. Singer & R. B. Ruddell (Eds.), Theoretical models and processes
of reading (pp. 187–197). Newark, DE: International Reading Association.
Holton, G. (1986). The advancement of science and its burdens. Cambridge, UK:
Cambridge University Press.
Hoover, W. A., & Gough, P. B. (1990). The simple view of reading. Reading and
Writing: An Interdisciplinary Journal, 2, 127–160.
Judson, H. F. (1980). The search for solutions. New York, NY: Holt, Rinehart and
Winston.
Kieras, D. (1978). Beyond pictures and words: Alternative information-
processing models for imagery effects in verbal memory. Psychological Bulletin,
85, 532–554.
Kintsch, W. (1998). Comprehension: A paradigm for cognition. Cambridge, UK: Cam-
bridge University Press.
Kintsch, W. (2004). The construction-integration model of text comprehension
and its implications for instruction. In R. B. Ruddell and N. J. Unrau (Eds.),
Theoretical models and processes of reading (5th ed., pp. 1270–1328). Newark, DE:
International Reading Association.
Kiraly, J., & Firlong, A. (1974). Teaching words to kindergarten children with pic-
ture, configuration, and initial sound cues in a prompting procedure. Journal
of Educational Research, 67, 295–298.
Kolker, B., & Terwilliger, P. N. (1981). Sight vocabulary learning of first and
second graders. Reading World, 20, 251–258.
Kounios, J., & Holcomb, P. J. (1994). Concreteness effects in semantic process-
ing: ERP evidence supporting dual-coding theory. Journal of Experimental Psy-
chology: Learning, Memory, and Cognition, 20, 804–823.
LaBerge, D., & Samuels, S. J. (1974). Toward a theory of automatic information
processing in reading. Cognitive Psychology, 6, 293–323.
Laing, E., & Hulme, C. (1999). Phonological and semantic processes influence
beginning readers’ ability to learn to read words. Journal of Experimental Child
Psychology, 73, 183–207.
Liberman, A. M., & Mattingly, I. G. (1985). The motor theory of speech percep-
tion revisited. Cognition, 21, 1–36.
Lupker, S. J. (1979). On the nature of perceptual information during letter per-
ception. Perception and Psychophysics, 25, 303–312.
Maleske, R. T. (1995). Foundations for gathering and interpreting behavioral data.
Pacific Grove, CA: Brooks/Cole.
McFalls, E. L., Schwanenflugel, P. J., & Stahl, S. A. (1996). Influence of word
meaning on the acquisition of a reading vocabulary in second grade children.
Reading and Writing, 8, 235–250.
Nelson, D. L., & Schreiber, T. A. (1992). Word concreteness and word structure
as independent determinants of recall. Journal of Memory and Language, 31,
237–260.
Nash-Weber, B. (1975). The role of semantics in automatic speech understand-
ing. In D. G. Bobrow and A. Collins (Eds.), Representation and understanding
(pp. 351–382). New York, NY: Academic Press.
Newell, A. (1973). You can’t play 20 questions with nature and win: Prospec-
tive comments on the papers of this symposium. In W. G. Chase (Ed.),
Visual information processing (pp. 463–526). New York, NY: Academic
Press.
Newell, A. (1990). Unified theories of cognition. Cambridge, MA: Cambridge Uni-
versity Press.
Paivio, A. (1971). Imagery and verbal processes. New York, NY: Holt, Rinehart, and
Winston. Hillsdale, NJ: Lawrence Erlbaum Associates.
Paivio, A. (1986). Mental representations: A dual coding approach. New York, NY:
Oxford University Press.
Paivio, A. (2007). Mind and its evolution: A dual coding theoretical approach. Mahwah,
NJ: Lawrence Erlbaum Associates.
Paivio, A. (2008). Looking at reading comprehension through the lens of neu-
roscience. In C. C. Block & S. R. Parris (Eds.), Comprehension instruction (2nd
ed., pp. 101–113). New York, NY: Guilford Press.
Paivio, A. (2010). Dual coding theory and the mental lexicon. The Mental Lexicon,
5, 205–230.
Paivio, A., Khan, M., & Begg, I. M. (2000). Concreteness and relational effects
on recall of adjective-noun pairs. Canadian Journal of Experimental Psychology,
54, 149–159.
Paivio, A., & Sadoski, M. (2011). Lexicons, contexts, events, and images: Com-
mentary on Elman (2009) from the perspective of dual coding theory. Cogni-
tive Science, 35, 198–209.
Paivio, A., Walsh, M., & Bons, T. (1994). Concreteness and memory: When and
Why? Journal of Experimental Psychology: Learning, Memory, and Cognition, 20,
1196–1204.
Paulson, E. J., & Goodman, K. S. (1999). Influential studies in eye-movement
research. Reading Online [electronic journal of the International Read-
ing Association]. Retrieved from http://www.readingonline.org/research/
eyemove.html
Rafal, R. T., & Robertson, L. (1995). The neurology of visual attention. In M.

Gazzaniga (Ed.), The cognitive neurosciences (pp. 625–648). Cambridge, MA:
MIT Press.
Rayner, K. (1997). Understanding eye movements in reading. Scientific Studies of
Reading, 1, 317–339.
Richmond, M., & McNinch, G. (1977). Word learning: Concrete vs. abstract ac-
quisition. Perceptual and Motor Skills, 45, 292–294.
Rumelhart, D. E. (1977). Toward an interactive model of reading. In S. Dornic
(Ed.), Attention and performance VI (pp. 573–603). New York, NY: Academic
Press.
Sadoski, M. (1992). Imagination, cognition, and persona. Rhetoric Review, 10,
266–278.
Sadoski, M. (2003). The role of orthography in a dual coding theory of reading.
In R. M. Joshi, C. K. Leong, & B. L. J. Kaczmarek (Eds.), Literacy acquisition:
The role of phonology, orthography, and morphology (pp. 67–81). Amsterdam, The
Netherlands: IOS Press.
Sadoski, M. (2005). A dual coding view of vocabulary learning. Reading and Writ-
ing Quarterly, 21, 221–238.
Sadoski, M. (2008). Dual coding theory: Reading comprehension and beyond.
In C. C. Block & S. R. Parris (Eds.), Comprehension instruction (2nd ed., pp.
38–49). New York, NY: Guilford Press.
Sadoski, M., Goetz, E. T., & Avila, E. (1995). Concreteness effects in text recall:
Dual coding or context availability? Reading Research Quarterly, 30, 278–288.
Sadoski, M., Goetz, E. T., Stricker, A. G., & Burdenski, T. K. (2003). New find-
ings for concreteness and imagery effects in written composition. Reading and
Writing: An Interdisciplinary Journal, 16, 443–453.
Sadoski, M., Kealy, W. A., Goetz, E. T., & Paivio, A. (1997). Concreteness and
imagery effects in the written composition of definitions. Journal of Educational
Psychology, 89, 518–526.
Sadoski, M., & Paivio, A. (1994). A dual coding view of imagery and verbal pro-
cesses in reading comprehension. In R. B. Ruddell, M. R. Ruddell, & H.
Singer (Eds.), Theoretical models and processes of reading (4th ed., pp. 582–601).
Newark, DE: International Reading Association.
Sadoski, M., & Paivio, A. (2001). Imagery and text: A dual coding theory of reading
and writing. Mahwah, NJ: Lawrence Erlbaum Associates.
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.
Sadoski, M., & Paivio, A. (2007). Toward a unified theory of reading. Scientific
Studies of Reading, 11, 337–356.
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.
Sadoski, M., Willson, V., Holcomb, A., & Boulware-Gooden, R. (2005). Verbal
and nonverbal predictors of spelling performance. Journal of Literacy Research,
36, 461–478.
Samuels, S. J. (1977). Introduction to theoretical models of reading. In W. Otto,
C. Peters, & N. Peters (Eds.). Reading problems: A multidisciplinary perspective
(pp. 7–41). Reading, MA: Addison-Wesley.
Samuels, S. J. (1994). Toward a theory of automatic information processing

in reading, revisited. In R. B. Ruddell, M. R. Ruddell, & H. Singer (Eds.),
Theoretical models and processes of reading (4th ed., pp. 816–837). Newark, DE:
International Reading Association.
Samuels, S. J. (2004). Toward a theory of automatic information processing in
reading, revisited. In R. B. Ruddell and N. J. Unrau (Eds.), Theoretical models
and processes of reading (5th ed., pp. 1127–1148). Newark, DE: International
Reading Association.
Samuels, S. J. (2006a). Looking backward: Reflections on a career in reading.
Journal of Literacy Research, 38, 327–344.
Samuels, S. J. (2006b). Toward a model of reading fluency. In S. J. Samuels &
A. Farstrup (Eds.). What research has to say about reading fluency (pp. 24–46).
Newark, DE: International Reading Association.
Samuels, S. J., & LaBerge, D. (1983). A critique of A theory of automaticity in
reading: Looking back: A retrospective analysis of the LaBerge-Samuels Read-
ing Model. In L. M. Gentile, M. L. Kamil, & J. S. Blanchard (Eds.), Reading
research revisited (pp. 39–55). Columbus, OH: Charles E. Merrill Publishing
Co.
Schwanenflugel, P. J., & Shoben, E. J. (1983). Differential context effects in the
comprehension of abstract and concrete verbal materials. Journal of Experimen-
tal Psychology: Learning, Memory, and Cognition, 9, 82–102.
Seidenberg, M. S. (2005). Connectionist models of word reading. Current Direc-
tions in Psychological Science, 14, 238–242.
Shibahara, N., Zorzi, M., Hill, M. P., Wydell, T., & Butterworth, B. (2003). Se-
mantic effects in word naming: Evidence from English and Japanese Kanji.
Quarterly Journal of Experimental Psychology, 56A, 263–286.
Terwilliger, P. N., & Kolker, B. S. (1982). The effects of learning confusable
words on subsequent learning of high or low imagery words. Reading World,
21, 286–292.
Van Dalen, D. B. (1979). Understanding educational research: An introduction (3rd
ed.). New York, NY: McGraw-Hill.
van der Veur, B. W. (1975). Imagery rating of 1,000 frequently used words. Jour-
nal of Educational Psychology, 67, 44–56.
Wolpert, E. M. (1972). Length, imagery values, and word recognition. Reading
Teacher, 26, 180–186.
Yore, L., & Ollila, L. (1985). Cognitive development, sex, and abstractness in
grade one word recognition. Journal of Educational Research, 78, 242–247.
Copyright of Reading Psychology is the property of Routledge and its content may not be copied or emailed to
multiple sites or posted to a listserv without the copyright holder's express written permission. However, users
may print, download, or email articles for individual use.

A Dual Coding Theoretical Model of Decoding

Uploaded by

Copyright:

Available Formats

A Dual Coding Theoretical Model of Decoding

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Dual Coding Theoretical Model of Decoding

Uploaded by

Copyright:

Available Formats

Reading Psychology, 33:465–496, 2012

ISSN: 0270-2711 print / 1521-0685 online

A DUAL CODING THEORETICAL MODEL OF DECODING

MARK SADOSKI and ERIN M. McTIGUE

The purpose of this article is to present a detailed account of de-

Unifying Macrotheories and Microtheories

The absorption of microtheories that are limited to particular

to their relative youth and to the complexity in defining, measur-

VERBAL STIMULI NONVERBAL STIMULI

VERBAL RESPONSES NONVERBAL RESPONSES

comprehensive detail as the LS model, but it could go well be-

Figure 1 presents the general model of DCT. It is a model of men-

form of written language that is detected by our senses (i.e., vision

Auditory-motor (i.e., speech) logogens are associated with vi-

images that either share a common sensory modality (e.g., a vi-

Review of the LS Model

Figure 2 and Figure 3 present the LS model (Samuels, 2004). As

memory (PM), episodic memory (EM), and semantic memory

FIGURE 3 Model of Visual Memory. Figure 3 from Samuels (2004). Reprinted

directional process so far describes a “bottom-up” model of read-

Visual Memory (VM)

The first stage of processing occurs within visual memory and is

features are present in multiple letters, such as the same arc in c

Phonological Memory (PM)

Phonological memory, the next step in the process, receives input

the hierarchy, then phonemes, syllables, and words. Phoneme

Episodic Memory (EM)

Episodic memory differs from the previously described compo-

Semantic Memory (SM)

According to the LS model, semantic memory is where the mean-

automatic if the word is familiar, or it may require the attention

The DCT Model of Decoding

Overview of the Model

The diagram in Figure 4 presents the DCT model of decoding

FIGURE 4 DCT Model of Decoding.

In Figure 4, processing moves from bottom to top. As in the

an imagen is represented by a circular shape. Each logogen box or

The arrows between logogens (boxes) indicate associative

Processing Operations Illustrated in the DCT Model

FEATURE LEVEL PROCESSES

LETTER LEVEL PROCESSES

such as &, #, @, and others (i.e., such symbols may be associated

WORD LEVEL PROCESSES

Visual words activate their contextually appropriate pronun-

WORD GROUP LEVEL PROCESSES

an image of a device with a certain visual appearance (V imagen),

PROCESSING OPERATIONS NOT ILLUSTRATED OR EMPHASIZED IN

pronounced /ch/ when followed by -oice, but /k/ when followed

CONSCIOUS VS. UNCONSCIOUS PROCESSES IN THE MODEL

unconscious, spreading activation in memory as theorized in both

Empirical Support for the DCT Subsumption of the LS Model

Behavioral Evidence in Decoding by Young Children

One issue is the learning of sight words by young readers, a

small samples of words and not controlling for familiarity, among

success of various forms of repeated readings on building word

As noted earlier, the LS model is limited by lack of an account of

More recently, Samuels (2006b, p. 24) explicitly stated that im-

These descriptions directly implicate the comprehension of an

The case in point involves the competing predictions of DCT