Sie sind auf Seite 1von 18

Integrating Memory-Based

and Constructionist Processes in


Accounts of Reading Comprehension
Paul van den Broek, David N. Rapp, and Panayiota Kendeou
University of Minnesota
Memory-based and constructionist processes have both been proposed as essential
components of the activation of concepts (e.g., propositions) and the establishment
of meaningful connections between concepts during reading. In this article, we argue
that a comprehensive theory of reading comprehension should include both sets of
processes. In support of this view, we summarize the results of several studies that
support the contribution of both processes during reading. In addition, we describe
the conceptual framework of the Landscape Model, in which memory-based and
constructionist processes are explicitly interconnected and dynamically interact in an
account of reading comprehension.
In the memoir Baghdad Express, author Joel Turnipseed (2003) described his ex-
periences in the Marine Corps during the Gulf War. His story begins during the
first morning of the air strike on Iraq:
We had gotten up early, packed our gear, and put the finishing touches on our ALICE
packs and our H-harnesses: flashlights and Ka-Bar fighting knives and entrenching
tools and ammo pouches and canteens and first-aid kits buckled and snapped and
duct-taped in their places on nylon straps clipped to a cartridge belt, forming an H
over our bodies. We checked our chemical weapons gear and gas masks. We cleaned
our M16s. We dressed for the first time in our desert camouflage uniforms, which
we called chocolate chips. We were all reservists activated as truck drivers, but the
Corps mission states that every Marine is a combat warrior. (p. 5)
DISCOURSE PROCESSES, 39(2&3), 299316
Copyright 2005, Lawrence Erlbaum Associates, Inc.
Correspondence and requests for reprints should be sent to Paul van den Broek, University of Min-
nesota, Department of Educational Psychology, 211 Burton Hall, Minneapolis, MN 55455. E-mail:
pvdbroek@umn.edu
Turnipseeds equipment only seems to get heavier during his tour of duty:
I carried three seabags, one like a backpack over each shoulder, my rifle in one hand
and a cot in the other. With this load I struggled around and beside and between tents.
I stumbled over ropes and chains, the cement blocks to which they were attached .
(p. 112)
To understand the information provided in a text, a reader must make connections
between text elements as well as between text elements and prior knowledge. The
first excerpt contains several examples of connections between text elements. For
example, various pieces of Turnipseeds gear are individually mentioned, provid-
ing referential connections for items that are contained in ALICE packs. In addi-
tion, the anaphoric term chocolate chips provides a referential connection to the
antecedent uniforms, affording an alternative term for Marine Corps garb. The ex-
cerpt also illustrates possible connections to background knowledge. For example,
the terms ALICE packs, chocolate chips, and the specific pieces of gear described
in the paragraph may activate prior knowledge for other military-based concepts
and perhaps even for other dessert- (and desert-) related terms. Thus, each item
from the list of ALICE components may activate other types of military gear (e.g.,
helmet, boots, grenade, etc.) frommemory. These types of connections, developed
between information in the text as well as between information in the text and
background knowledge, can become readily available during moment-by-moment
reading processes.
The second excerpt provides an illustration of a different kind of frequently
made connection, one that bridges the conceptual gap between two parts of a text.
Most readers of this excerpt will assume that Turnipseed stumbled over the tent
ropes because of the heavy load he was carrying. The information in the text, as
well as background knowledge (e.g., that people usually do not stumble without
cause, that heavy loads make walking difficult), provide the constraints for the in-
ference that establishes this connection.
In addition to these types of connections, readers can develop more elaborate
associations or expectations about events that they believe may occur as the story
unfolds. For example, when reading the first excerpt readers could generate the
prediction that, despite the fact that reservists traditionally serve as truck drivers
during military operations, Turnipseed will become involved in firefight activity
(otherwise, why would he have mentioned all of his gear and his training as a com-
bat warrior?). Another predictive inference, driven by extensive prior knowledge
about the Gulf War, would lead a reader to expect that Turnipseed will actually en-
counter little ground activity (and, perhaps, interpret the description of his gear as a
commentary on the absurdity of military situations). These are just a fewexamples
of elaborate inferences that readers can construct as they read this text. The spe-
cific inferences that readers develop will depend on factors such as the readers
300 VAN DEN BROEK, RAPP, KENDEOU
prior knowledge, willingness to hypothesize about future text events, strategies or
goals for comprehension, and the information provided in the text itself.
As the aforementioned examples illustrate, a central component of comprehen-
sion is the identification of semantic connections between the various pieces of in-
formation in the text, and between this information and readersbackground knowl-
edge. In the discourse processing literature, two sets of processes have been
proposed as providing the foundation for the identification of such meaningful con-
nections: memory-based processes and constructionist processes. According to the
memory-based viewof text processing, as a text is read, information in the text (and
any other information already activated in working memory) will trigger a spread of
activation through the readers knowledge base, activating associated information
(Gerrig & McKoon, 1998; McKoon, Gerrig, & Greene, 1996; Myers & OBrien,
1998; OBrien, 1995). This spread of activation can occur through the episodic
memory representation that the reader has constructed of the text so far as well as
through his or her semantic (background) knowledge. This memory-based activa-
tion process is generally described as occurring for free (McKoon & Ratcliff,
1992), because it is passive and involves little or no influence of strategy. According
tothe constructionist view, readers have explicit andimplicit goals or standards they
actively attempt to satisfy when they read a text. These goals or standards have been
labeled as a search/effort after meaning. Readers can use information from prior
text, their developing memory representation of the text, and/or background knowl-
edge in an effort to achieve these goals or standards (Graesser, Singer, &Trabasso,
1994; Long, Seely, & Oppy, 1996; Singer, Graesser, & Trabasso, 1994; van den
Broek, Risden, & Husebye-Hartmann, 1995). These constructionist processes are
active and strategic.
In this article, we propose that both sets of processes are necessary for a com-
plete account of naturalistic reading and that a comprehensive theory of reading
comprehension should include both. Moreover, we argue that memory-based and
constructionist processes dynamically interact, borrowing from, supporting, and,
possibly, conflicting with each other. We illustrate how this dynamic interaction
can be conceptualized by outlining the Landscape Model, a theoretical framework
in which the two sets of processes are integrated.
MEMORY-BASED AND CONSTRUCTIONIST
PROCESSES IN TEXT COMPREHENSION
The processes associated with memory-based and constructionist views differ in a
number of ways. According to the memory-based view of reading, general pro-
cesses associated with memory influence the availability of information during
reading. Memory-based processes are autonomous and passive (as mentioned,
they occur for free). They are dumb in the sense that activation is the result
INTEGRATING READING PROCESSES 301
only of the strength of the signaling information and its associations to other infor-
mation (either in the developing discourse representation or in prior knowledge),
and not of a desired goal or as a function of comprehension. Constructionist pro-
cesses, on the other hand, are not automatic; they are outcome-oriented and ac-
tively guided by readers propensities for establishing meaning. In general, mem-
ory-based processes are considered passive whereas constructionist processes are
considered strategic (Graesser et al., 1994; McKoon et al., 1996; McKoon &
Ratcliff, 1995; OBrien & Myers, 1999; Singer et al., 1994).
There is considerable evidence that memory-based text processing is a power-
ful factor in determining the availability of information during reading. This evi-
dence has been derived fromgeneral research on the basic properties and processes
associated with memory. Classic research on knowledge representation in memory
demonstrates that similarity among concepts leads to faster activation and compar-
ison latencies (e.g., Collins & Loftus, 1975; Collins & Quillian, 1969; Smith,
Shoben, & Rips, 1974). Similarly, during text experiences, concepts (e.g., words)
mentioned in a text reactivate information that has become associated to those con-
cepts during earlier processing cycles (Albrecht &OBrien, 1993; Cook, Halleran,
&OBrien, 1998; Myers &OBrien, 1998; OBrien &Albrecht, 1991). Moreover,
the more features a current segment of text shares with preceding segments, the
more quickly that earlier information is reactivated. In addition, and in line with
other classic findings on memory (e.g., temporal memory described by Estes,
1972; levels of processing described by Craik & Lockhart, 1972), preceding text
information is more quickly and reliably reactivated with decreases in distance be-
tween earlier and later text information, and with greater elaboration of the earlier
information. Thus, there is considerable evidence for the influence of mem-
ory-based processes.
Likewise, there is strong evidence that readers frequently engage in construc-
tionist processes aimed at creating meaning. For example, readers more frequently
and more strongly reactivate information from prior processing cycles when this
information contributes to the understanding of a text than when it does not (e.g.,
Goldman & Saul, 1990; OBrien, Albrecht, Hakala, & Rizzella, 1995; Suh &
Trabasso, 1993; Sundermeier, van den Broek, & Zwaan, in press; van den Broek,
Rohleder, & Narvaez, 1996). Readers also routinely activate background knowl-
edge frommemory when it is required for comprehension (e.g., Lucas, Tanenhaus,
& Carlson, 1990; van den Broek, Rohleder, et al., 1996). Thus, readers often
(re)activate information when it contributes to, or is necessary for, implementing
strategies to facilitate comprehension. In addition, variations in readers attitudes
or approaches to a text influence the way they process the text. For example, online
processes differ as a function of the type of text a reader expects to read: reading la-
tencies for identical texts vary as a result of expectations about text genre (e.g., fic-
tion or a newspaper article; Zwaan, 1994). Readers goals also influence the online
processing of texts. The same texts evoke different patterns of inferential activity,
302 VAN DEN BROEK, RAPP, KENDEOU
as evidenced by think-aloud protocols and reading times, depending on whether
subjects read a text for entertainment or for study (Linderholm & van den Broek,
2002; Lorch, Lorch, & Klusewitz, 1993; Narvaez, van den Broek, & Ruiz, 1999).
In addition, subtle text cues (e.g., time shifts or event duration) can lead readers to
develop particular expectations or desires for text events (e.g., preferences for nar-
rative situations). These reader responses, in turn, influence expectations for story
outcomes (Gerrig, 1993; Rapp & Gerrig, 2002; Rapp, Gerrig, & Prentice, 2001;
Rapp & Taylor, 2004).
There is also a growing body of evidence that both memory-based and con-
structionist processes simultaneously operate during reading. For example, van
den Broek, OBrien, Halleran, and Kendeou (2004) presented subjects with texts
that varied in two respects: First, target sentences were immediately preceded by a
sentence that provided either a strong or a weak explanation for the target and, sec-
ond, a potential alternative explanation earlier in the text was either elaborated or
not. The first variation was designed to affect constructionist processing, the sec-
ond to affect memory-based processing. The results showed that the availability of
earlier information (as measured by speed of recognition) generally was influ-
enced by the extent of elaborationreflecting memory-based processes
whereas the processing of the target sentences (as measured by reading speed)
generally was influenced by the explanatory power of the immediately preceding
contextreflecting constructionist processes. Thus, both memory-based factors
and constructionist factors influenced readers processing of these texts.
As a result of findings such as these, it is generally accepted that both mem-
ory-based and constructionist processes operate during reading. There is much less
agreement about the circumstances in which each occurs and howthey can be cap-
tured in a single conceptual framework. In the remainder of this article, we contend
that the two sets of processes co-occur and dynamically interact.
INTERACTIONS BETWEEN MEMORY-BASED AND
CONSTRUCTIONIST PROCESSES
In our view, memory-based and constructionist processes both operate during
reading and, moreover, both play an important role in the identification of the se-
mantic connections between the various pieces of information in the text and be-
tween this information and readers background knowledge. To conceptualize how
memory-based and constructionist processes interact, a number of years ago we
introduced the notion of standards of coherence (van den Broek et al., 1995; see
also van den Broek, Lorch, Linderholm, & Gustafson, 2001). Standards of coher-
ence reflect a readers knowledge and beliefs about what constitutes good com-
prehension as well as the readers specific goals for reading the particular text
(van den Broek, Virtue, Everson, Tzeng, & Sung, 2002, p. 137). These standards
INTEGRATING READING PROCESSES 303
influence the extent to which readers will engage in constructionist processes.
They reflect the types of coherence (e.g., referential, causal, spatial) as well as the
strength of coherence that a reader aims to maintain. Standards vary as a function
of reader variables including (but not limited to) reading goals, expectations,
working memory resources, and perspective, as well as contextual variables in-
cluding text genre, presentation rate, and task variables (van den Broek, Fletcher,
& Risden, 1993; van den Broek et al., 2001). In our interactive view, mem-
ory-based activation processes take place as outlined by models such as the Reso-
nance Model (Albrecht & OBrien, 1993; OBrien & Myers, 1999; OBrien,
Rizzella, Albrecht, & Halleran, 1998) or the Landscape Model (van den Broek,
Risden, Fletcher, & Thurlow, 1996; van den Broek, Young, Tzeng, & Linderholm,
1999). The information activated through this autonomous process, constrained
only by the strength of the concepts that trigger it and the strengths of their associa-
tions, is evaluated with respect to a readers standards of coherence. If the activated
information is sufficient to meet those standards, then the connections are easily
and quickly identified. However, if the activated information is insufficient for sat-
isfying those standards, then more effortful, strategic processes will ensue in an at-
tempt to attain the standards. Consequently, if a reader maintains strict standards
(as a function of reader or contextual variables), then constructionist processes will
be more common, with the benefit of deeper understanding but at the cost of
greater effort by the reader. If, however, the reader maintains relatively relaxed
standards of coherence, then constructionist processes will be less common, with
the benefit that reading will be less effortful but at the cost of a potentially less co-
herent text representation (van den Broek et al., 2001). Standards of coherence
vary between readers as well as within readers across reading situations (see van
den Broek et al., 1995), but, as illustrated by the passage at the outset of this article,
standards of referential and causal referential coherence have been found to be the
most commonly maintained standards in narrative reading.
To summarize, memory-based and constructionist processes play important,
yet distinct roles in the comprehension process. To establish connections, readers
must activate the to-be-connected information and determine what, if any, connec-
tions exist between the activated pieces of information. Memory-based processes
are central to the first component, whereas constructionist processes are central to
the second. To put it differently, memory-based processes provide the input to the
constructionist processes, and the product from the constructionist processes de-
termines whether the memory-based input is sufficient for comprehension. The
standards of coherence that a reader has in a particular reading situation provide
constraints, in addition to those provided by textual information and background
knowledge.
There is empirical evidence for the interactive nature of memory-based and
constructionist processes. For example, the previously described recognition and
reading time results obtained by van den Broek et al. (2004) indicated that reading
304 VAN DEN BROEK, RAPP, KENDEOU
involves two steps: (a) passive, autonomous, memory-based activation of informa-
tion, followed by (b) an active, constructionist, coherence-building process. Fur-
thermore, much of the evidence on memory-based processes described earlier is
based on the use of an inconsistency paradigm and, thereby, also supports the no-
tion that memory-based and constructionist processes interact (Albrecht &
OBrien, 1993; OBrien et al., 1998). In the inconsistency paradigm, readers en-
counter information early in the text (e.g., Bill was old and weak) that is incon-
sistent with later information (Bill quickly ran and picked up the boy). By sys-
tematically manipulating variables that are hypothesized to affect activation of the
earlier information and observing whether a slowdown in reading of the later sen-
tence occurs, predictions about discourse and knowledge activation in memory can
be assessed (Myers, OBrien, Albrecht, & Mason, 1994). The logic is that the in-
creased reading time reflects reading difficulty that occurred because readers rec-
ognized the inconsistency between an established characteristic of the protagonist
(e.g., Bill was old and weak) and the critical sentence (e.g., Bill quickly ran and
picked the boy up), and they needed to engage in some sort of inferential process
to attempt to reestablish coherence (Albrecht & OBrien, 1993, pp. 10661067).
Thus, the dependent variable in the inconsistency paradigm actually is one that re-
flects a constructionist process and, therefore, the results in these studies provide
evidence that the simultaneous activation of consistent and inconsistent informa-
tion results in additional processing by the reader (as evidenced by the slowdown
in reading), because inconsistencies violate the standards of coherence that the
readers in these studies maintained.
Standards of coherence also influence activation of information. For example,
spatial information presented early in a text is more availableas indicated by
naming latencyat later points in the text if the early information helps explain a
later event (e.g., the location of an object explains a later event) than if it does not
serve an explanatory function (Sundermeier et al., in press). Similar patterns have
been observed for objects or events.
A final example of evidence for the interaction between memory-based and
constructionist processes comes from finding that the effects of reading goals are
moderated by memory factors: Readers with high working memory capacity show
greater sensitivity to variations in reading goals than do readers with low working
memory capacity (Linderholm & van den Broek, 2002).
It is important to note that both memory-based processes and constructionist
processes are assumed to operate on knowledge structures, either in the formof ep-
isodic memory representations in a discourse model or in the form of pre-existing
semantic (background) knowledge. Conversely, these processes lead to updating
of the ongoing mental representation of a text (e.g., the discourse model) and, pos-
sibly, of the readers background knowledgethe basis for educational practice
(Kintsch, 1998; McNamara & Kintsch, 1996). Both the memory-based and con-
structionist views focus on the factors and processes that determine what is acti-
INTEGRATING READING PROCESSES 305
vated and retrieved during moment-by-moment reading, but may lack an account
of how these processes can lead to changes in the mental representation of
long-term knowledge structures (even though such changes are presumed to influ-
ence subsequent processes). A complete description of memory-based and con-
structionist processing must provide an account not only of how activation and
comprehension occur, but also howthese processes both lead to and depend on de-
veloping memory structures. This becomes particularly important when attempts
are made to capture the memory-based and constructionist views in a single archi-
tecture. We turn to this next.
THE ARCHITECTURE OF THE LANDSCAPE MODEL
OF READING COMPREHENSION
To describe the contributions of memory-based and constructionist processes we
have developed a conceptual model called the Landscape Model (van den Broek,
Risden, et al., 1996; van den Broek et al., 1999). The architecture of the Landscape
Model assumes that as a reader proceeds through a text in reading cycles (with
each cycle roughly corresponding to the reading of a newsentence or proposition),
concepts fluctuate in activation as a function of four sources of information: the
current processing cycle, the preceding cycle, the current episodic text representa-
tion, and readers background knowledge. With the reading of each cycle, particu-
lar concepts are activated and added as nodes to the episodic memory representa-
tion of the text. If a concept is already part of the text representation and is
reactivated, its trace is strengthened. In addition, co-activation of concepts leads to
the establishment (or strengthening) of connections between those concepts. The
resulting network representation influences subsequent activation patterns. These
cyclical and dynamically fluctuating activations lead to the gradual emergence of
an episodic memory representation or discourse model of the text, in which textual
propositions and inferences are connected via semantic relations (such as causal
and referential links). Thus, the model captures the fluctuations of concepts during
reading, as well as the evolving discourse model. The resulting memory represen-
tation is the product of iterative and reciprocal relations between fluctuations of ac-
tivations and the episodic text representation.
Two types of mechanisms guide access to these sources of activation. The first
type is cohort activation. The architecture of the model assumes that when a con-
cept is activated during reading, all other concepts currently activated become as-
sociated with it. Thus, each concept connects with other, related concepts becom-
ing a cohort. In turn, when any of the individual concepts in a cohort become
active, the other concepts are also activated. This mechanism is passive and oper-
ates under a limited pool of activation. Thus, cohort activation is memory-based
306 VAN DEN BROEK, RAPP, KENDEOU
and similar to the activation mechanism described by the Resonance Model
(Myers & OBrien, 1998; OBrien & Myers, 1999; OBrien et al., 1998).
The second type of mechanism is coherence-based retrieval. Unlike mem-
ory-based activations that are based on, for example, featural overlap, coher-
ence-based retrieval is a strategic mechanism by which information is retrieved
with the aim of meeting a readers standards or goals (Linderholm, Virtue, Tzeng,
& van den Broek, 2004). Such retrieval can be from an episodic text representa-
tion, from background knowledge, or from the text itself (e.g., via look-backs in a
text). This mechanism operates under a limited pool of activation that can be dis-
tributed over concepts and, unlike cohort activation, is strategic (and can be
effortful). Thus, coherence-based retrieval is similar to search/effort after mean-
ing mechanisms described by the constructionist view of reading (Graesser et al.,
1994; Singer et al., 1994).
Acentral factor in the model that determines which sources of activation are ac-
cessed consists of the standards of coherence that the reader maintains during
reading (van den Broek et al., 1995). The architecture of the model allows for the
adoption of different types of coherence that a reader may establish (including, but
not limited to, referential, causal, temporal, and spatial connections). Reader stan-
dards can vary as a function of individual differences, text types, reading goals, and
so on (Linderholm & van den Broek, 2002; Narvaez et al., 1999; van den Broek et
al., 2001), but for narratives and many other types of text, referential and causal
standards of coherence are often central. During reading, a readers standards can
at times be met entirely by the information currently activated in the model through
cohort activation (i.e., memory-based processing), whereas in other cases the
reader may need to actively search the episodic text representation and/or back-
ground knowledge to maintain these standards through coherence-based retrieval
(i.e., constructionist processing).
In summary, the architecture of the Landscape Model incorporates a dynamic
interaction of both memory-based and constructionist processes. Thus, these pro-
cesses are not just complementary views but are part of a single theoretical account
of text comprehension.
APPLICATIONS OF THE LANDSCAPE MODEL
In prior research, the conceptual framework of the Landscape Model has been vali-
dated by comparing computational simulations and human data for narrative read-
ing (van den Broek et al., 1999). With regard to online measures, the models pre-
dictions about the patterns of activation of propositions over the course of reading
were strongly related to the activations reported by actual readers. With regard to
off-line measures, in several studies recall of text propositions was found to be
strongly related to actual recall by readers. In addition to frequency of recall, the
INTEGRATING READING PROCESSES 307
model also predicted the actual order in which participants recalled the text propo-
sitions, with the most strongly represented concepts being recalled first and the
strength of semantic relations determining subsequent recall. Thus, the Landscape
Model captures important aspects of the cognitive processes that take place during
the reading of narrative texts and of the resulting representation in memory.
Since its original conceptualization, the Landscape Model has been extended to
various other reading situations and types of texts. We describe some of these now.
Scientific Text
Whereas initial studies using the Landscape Model have focused on narrative read-
ing, similar findings have been obtained on reading of scientific texts. For exam-
ple, we have simulated the comprehension of scientific texts and compared the
models final text representation to human recall (van den Broek, Kendeou, Sung,
& Chen, 2003; van den Broek et al., 2002). The models prediction for recall of
propositions, based on several factors including the accumulated node strength of
concepts over the course or reading, was strongly related to the frequency of recall
of participants. In a related set of simulations, the Landscape Model was used to
simulate comprehension of scientific texts designed to refute students misconcep-
tions. The simulations showed that the Landscape Model accurately predicted the
circumstances under which misconceptions and correct conceptualizations are si-
multaneously activated, and are most likely to lead to revisions of prior knowledge
(Kendeou & van den Broek, 2004).
Differences in Background Knowledge
We have simulated the effects of readers prior knowledge on text comprehension.
There is considerable evidence that readers prior knowledge, particularly incor-
rect prior knowledge (i.e., misconceptions), affects comprehension both online
and off-line (Hynd & Alvermann, 1989; Kendeou, Rapp, & van den Broek, in
press; Kendeou & van den Broek, in press). We captured these effects by conduct-
ing two simulations and comparing them to human recall for readers with and
without misconceptions for text topics. In these simulations, we included prior
knowledge in the model by means of activation vectors established prior to read-
ing. The two simulations predictions for recall were significantly related to the
frequency of recall of human participants with and without misconceptions, re-
spectively. Moreover, the correlations between contrasting model simulations and
readers (e.g., the model without misconceptions and readers with misconceptions)
were substantially smaller. In addition, differences between the two simulations
identified differences in human recall.
308 VAN DEN BROEK, RAPP, KENDEOU
Reading Goals
Previous research has shown that reading processes systematically vary as a func-
tion of reading purpose. For example, when reading to study, readers are more
likely to engage in processes aimed at establishing coherence than when reading
for entertainment (e.g., Linderholm&van den Broek, 2002; Lorch et al., 1993; van
den Broek et al., 2001). Linderholm et al. (2004) conducted simulations for each
reading purpose by varying the models input parameters to capture differences in
readers standards for coherence. The simulations demonstrated that reading for
study resulted in more overall activation, and activation of more concepts, than did
reading for entertainment. In particular, the texts main ideas received consider-
ably more activation in the study simulation than in the entertainment simulation.
Comparisons to human data showed that the model captured readers adjustment
of inferential processes as a function of reading goal.
Inconsistency Detection
The Landscape Model has successfully captured the processes by which readers
detect inconsistencies in texts. Linderholm et al. (2004) conducted simulations of
the texts previously used by OBrien and colleagues in several of their inconsis-
tency detection studies. In the Landscape Model, the likelihood of detecting an in-
consistency depends on the extent to which the initial text information (which con-
tradicts information in later statements) received activation in previous reading
cycles, how strongly it was connected to other text concepts, and the extent to
which the later information semantically overlapped with the earlier, inconsistent
text information (van den Broek et al., 1999). This simulation accurately predicted
the patterns of reading times reported by Rizzella and OBrien (1996).
Relative Contribution of Cohort Activation
and Coherence-Based Processes
The Landscape Model assumes a dynamic interaction between memory-based and
constructionist processes. To determine whether both are necessary for the
models prediction of reading comprehension, we conducted a set of three simula-
tions of expository text reading, manipulating the contribution of cohort activation
and coherence-based retrieval across the simulations. The first simulation included
both cohort activation and coherence-based retrieval (consisting of the models de-
fault focus on referential and causal coherence). In this simulation, the models
prediction for recall of propositions, based on the accumulated node strength over
the course of reading, was strongly related to the frequency of recall of partici-
pants. (r = .70). When cohort activation was removed from the modelbut coher-
ence-based processes were retainedits predictive power decreased significantly
INTEGRATING READING PROCESSES 309
(r = .60). When coherence-based processes were removedbut cohort activation
was retainedthe models predictive power decreased even more (r = .50).
In a second set of simulations, we replicated these findings for narrative texts,
with remarkably similar correlations. In this second set we also included a simula-
tion in which both cohort activation and coherence-based processes were elimi-
nated. In this simulation, predictive power decreased even further. The results of
these two sets of simulations showthat both cohort activation and coherence-based
retrieval were necessary components in the model for capturing human reading
performance.
In summary, the Landscape Model has been used to capture various reading
situations, text types, and behavioral measures of reading comprehension (e.g.,
recall and latency data). The strong correspondence between predicted and ob-
served data indicates that the Landscape Model has considerable psychological
validity. The architecture of the model, and in particular the dynamic interaction
between memory-based and constructionist processing, unifies reading compre-
hension into a single conceptual framework that has high predictive power. In-
deed, in cases in which this dynamic interaction is impaired (as demonstrated
by the final sets of simulations described earlier), the models predictive power
is significantly reduced.
DISCUSSION
We have outlined an interactive viewthat combines autonomous and passive mem-
ory-based processes with constructionist processes that are strategic and goal-di-
rected. We hope to have demonstrated that these two sets of processes can be con-
ceived of as components of a dynamically interacting system. An interactive view
reconciles apparent differences between memory-based and constructionist views
by implementing them in a single conceptual framework. This view is similar in
spirit to conceptual frameworks such as the ConstructionIntegration model
(Kintsch, 1988, 1998), in which memory-based spread of activation (construction)
isfollowedbyacoherence-basedinferencegenerationphase(integration), andother
two-stage models designed to capture the complementary nature of memory-based
and constructionist models (e.g., Long et al. 1996; Richards & Singer, 2001). Evi-
dence for such a two-stage process comes not only from behavioral and computa-
tional studies, as described previously, but also from neuropsychological research
(e.g., Beeman, 1998; Long & Baynes, 2002; Mason & Just, 2004; Virtue, van den
Broek, &Linderholm, inpress). The Landscape Model makes explicit howsuchun-
derlying two-stage processes mutually influence each other and how together they
functionallyleadtothegradual emergenceof amemoryrepresentationof thetext.
An implication of the interactive view of reading comprehension is that any
kind of inference (e.g., spatial, causal, referential, thematic, character-based,
310 VAN DEN BROEK, RAPP, KENDEOU
moral) can in principle be generated by virtue of either memory-based or mem-
ory-based-followed-by-constructionist processes, given the right combination of
text and reader variables. Consequently, attempts to rigidly assign particular types
of inferences to one set of processes or another are not fruitful and, indeed, mis-
guided. If the information required for making a particular inference is readily
available through memory-based processes, there is no need for constructionist
processes. Of course, certain types of inferences, such as thematic and other ab-
stract inferences, tend to require simultaneous activation of information that is not
immediately available through memory-based processes and, hence, tend to in-
volve more constructionist processes. The point is that both situations are possible.
For example, if by happy circumstance or clever design on the part of the author the
required information becomes available through memory-based spread of activa-
tion, then that inference too will be generated for free (McKoon & Ratcliff,
1992).
Although memory-based and constructionist processes often operate in concert
with each other, they may, in fact, conflict. One illustration of this situation is the
basis for the inconsistency paradigm: Comprehension may be moving along fine if
it were not for the problem that the earlier, inconsistent information becomes acti-
vated, causing conflict and slow down in reading. A second, more dramatic illus-
tration is the finding that memory-based processes may activate information that
results in conflict but in actuality is not relevant. For example, when two characters
are described, information about one character may spill over to a second character
via resonance. When one character is described later in the text, information about
both characters is activated (Cook et al., 1998; Long & Chong, 2001), potentially
resulting in interference.
The integration of memory-based and constructionist processes into a single
theoretical (and computational) framework enriches both views. Accounts of read-
ing that focus on memory-based processes have limited themselves to issues con-
cerning the activation of concepts during reading, leaving comprehension pro-
cesses per se out of their purview. Conversely, constructionist accounts that focus
on coherence-based and integrative processes tend to assume that activation occurs
but do not explicitly specify how this may occur. To address the limits of these ac-
counts, we have argued that they complement each other.
One reason that memory-based and constructionist accounts may have been
considered in opposition to each other is that it is tempting to equate mem-
ory-based processes with bottom-up processing and constructionist processes with
top-down processing. Bottom-up views traditionally describe stimulus-driven pro-
cesses that do not necessitate higher order cognition. Top-down views traditionally
rely on background knowledge and existing memory representations to mediate
the interpretations of stimuli. We contend that these terms, in their pure form, are
not adequate to describe either process, but have often been used as such. For ex-
ample, memory-based views are not direct analogs of bottom-up processing.
INTEGRATING READING PROCESSES 311
These views argue that information in memory is activated to a large degree by ele-
ments of a text, thereby assigning an important role to pre-existing memory repre-
sentations, including organized knowledge such as scripts and schemas. Thus,
memory-based views to some degree require top-down information (e.g., concepts
in prior knowledge that will be activated broadly) to account for relevant activation
processes. In a similar sense, constructionist views are not direct analogs of
top-down processing. In text comprehension, expectations or preferences on the
part of the reader must interact with the actual words being perceived (e.g., Rapp et
al., 2001); top-down processing is affected directly by bottom-up processing.
Thus, both memory-based and constructionist views rely on a combination of bot-
tom-up and top-down processing. Thinking about them as either completely bot-
tom-up or completely top-down fails to capture the wide array of reader processes
that operate during naturalistic text experiences.
The integration of memory-based and constructionist processes highlights the
importance of investigating the exact nature of their interaction. For example, al-
though we have described the two types of processes as operating in sequence, they
may in fact overlap and alternate (van den Broek et al., 1999). If the information
activated via memory-based processes is insufficient to satisfy the readers stan-
dards, this may lead to an increase in constructionist processing. These processes
may involve selective focus on a subset of activated concepts that, in turn, may trig-
ger a second wave of memory-based spread of activation. Of course, even in this
case, the initial activation of this dynamic process is likely driven by mem-
ory-based activation. Alternatively, it is conceivable that the constructionist pro-
cesses evoke a strategic search of memory that is different from cohort activation
(cf. Long &Lea, 2005/this issue). Afinal possibility is that an increase in construc-
tionist processes involves a loosening of the constraints on activation from mem-
ory-based processes, resulting in broader availability of information (we thank an
anonymous reviewer for these suggestions). An important question for future re-
search is to determine the boundary conditions and the exact mechanisms of inter-
action between memory-based and constructionist processes. Continued research
on the nature of integrated memory-based and constructionist processes will be
necessary to outline how memory is constructed, accessed, and retrieved during
readingspontaneously as well as strategically.
CONCLUSION
Models of text processing have described a variety of underlying mechanisms that
attempt to account for the ways in which people read, understand, and remember
texts. To outline these processes, a variety of explanations have been invoked, in-
volving both passive and strategic knowledge activation. We contend that only by
considering the dynamic interactions of these processes, rather than the viability of
312 VAN DEN BROEK, RAPP, KENDEOU
each in isolation, can we propose a generalizable theory that appropriately ac-
counts for naturalistic text comprehension. Our simulations using the Landscape
Model represent one attempt at illustrating the importance and validity of these in-
teractions.
ACKNOWLEDGMENTS
This research was supported by the Center for Cognitive Sciences at the University
of Minnesota through Grant No. HD07151 from the National Institute of Child
Health and Human Development, by the Guy Bond Endowment for Reading and
Literacy, by a Golestan fellowship at the Netherlands Institute for Advanced Study
in the Humanities and Social Sciences to Paul van den Broek, by a Faculty Sum-
mer Research Fellowship from the Office of the Dean of the Graduate School of
the University of Minnesota to David N. Rapp, and by an Eva O. Miller Fellowship
to Panayiota Kendeou. We thank Robert F. Lorch, Jr. for his comments on an ear-
lier version of this article.
REFERENCES
Albrecht, J. E., & OBrien, E. J. (1993). Updating a mental model: Maintaining both local and global
coherence. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19, 10611069.
Beeman, M. (1998). Coarse semantic coding and discourse comprehension. In M. Beeman & C.
Chiarello (Eds.), Getting it right: The cognitive neuroscience of right hemisphere language compre-
hension (pp. 225284). Mahwah, NJ: Lawrence Erlbaum Associates, Inc.
Collins, A. M., & Loftus, E. F. (1975). A spreading-activation theory of semantic processing. Psycho-
logical Review, 82, 407428.
Collins, A. M., & Quillian, M. R. (1969). Retrieval time from semantic memory. Journal of Verbal
Learning and Verbal Behavior, 8, 240247.
Cook, A. E., Halleran, J. G., & OBrien, E. J. (1998). What is readily available? A memory-based view
of text processing. Discourse Processes, 26, 109129.
Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing: A framework for memory research.
Journal of Verbal Learning and Verbal Behavior, 11, 671684.
Estes, W. K. (1972). An associative basis for coding and organization in memory. In A. W. Melton &E.
Martin (Eds.), Coding processes in human memory (pp. 161190). Washington, DC: Winston.
Gerrig, R. J. (1993). Experiencing narrative worlds. New Haven, CT: Yale University Press.
Gerrig, R. J., & McKoon, G. (1998). The readiness is all: The functionality of memory-based text pro-
cessing. Discourse Processes, 26, 6786.
Goldman, S. R., & Saul, E. U. (1990). Flexibility in text processing: A strategy competition model.
Learning and Individual Differences, 2, 181219.
Graesser, A., Singer, M., & Trabasso, T. (1994). Constructing inferences during narrative comprehen-
sion. Psychological Review, 101, 371395.
Hynd, C., & Alvermann, D. E. (1989). Overcoming misconceptions in science: An on-line study of
prior knowledge activation. Reading Research and Instruction, 28, 1226.
INTEGRATING READING PROCESSES 313
Kendeou, P., Rapp, D. N., & van den Broek, P. (in press). The influence of readers prior knowledge on
text comprehension and learning from text. In Progress in Education. Nova Science.
Kendeou, P., & van den Broek, P. (2004, August). Comprehension of refutation texts: A simulation us-
ing the Landscape Model. Paper presented at the Fourteenth Annual Meeting of the Society for Text
and Discourse, Chicago.
Kendeou, P., & van den Broek, P. (in press). The effects of readers misconceptions on comprehension
of scientific text. Journal of Educational Psychology.
Kintsch, W. (1988). The role of knowledge in discourse comprehension: A constructionintegration
model. Psychological Review, 95, 163182.
Kintsch, W. (1998). Comprehension: A paradigm for cognition. Cambridge, England: Cambridge Uni-
versity Press.
Linderholm, T., & van den Broek, P. (2002). The effects of reading purpose and working memory ca-
pacity on the processing of expository text. Journal of Educational Psychology, 94, 778784.
Linderholm, T., Virtue, S., Tzeng, Y., & van den Broek, P. (2004). Fluctuations in the availability of in-
formation during reading: Capturing cognitive processes using the Landscape Model. Discourse
Processes, 37, 165186.
Lorch, R. E. J., Lorch, E. P., &Klusewitz, M. A. (1993). College students conditional knowledge about
reading. Journal of Educational Psychology, 85, 239252.
Long, D. L., &Baynes, K. (2002). Discourse representation in the two cerebral hemispheres. Journal of
Cognitive Neuroscience, 14, 228242.
Long, D. L., & Chong, J. L. (2001). Comprehension skill and global coherence: A paradoxical picture
of poor comprehenders abilities. Journal of Experimental Psychology: Learning, Memory, and Cog-
nition, 27, 14241429.
Long, D. L., & Lea, R. B. (2005). Have we been searching for meaning in all the wrong places? De-
fining the search after meaning principle in comprehension. Discourse Processes, 39, 279298.
Long, D. L., Seely, M. R., & Oppy, B. J. (1996). The availability of causal information during reading.
Discourse Processes, 22, 145170.
Lucas, M. M., Tanenhaus, M. K., &Carlson, G. N. (1990). Levels of representation in the interpretation
of anaphoric reference and instrument inference. Memory and Cognition, 18, 611631.
Mason, R. A., &Just, M. A. (2004). Howthe brain processes causal inferences in text: Atheoretical ac-
count of generation and integration component processes utilizing both cerebral hemispheres. Psy-
chological Science, 15, 17.
McKoon, G., Gerrig, R. J., &Greene, S. B. (1996). Pronoun resolution without pronouns: Some conse-
quences of memory-based text processing. Journal of Experimental Psychology: Learning, Memory,
and Cognition, 22, 919932.
McKoon, G., & Ratcliff, R. (1992). Inference during reading. Psychological Review, 99, 440466.
McKoon, G., & Ratcliff, R. (1995). The minimalist hypothesis: Directions for research. In C. A.
Weaver, S. Mannes, & C. R. Fletcher (Eds.), Discourse comprehension: Essays in honor of Walter
Kintsch (pp. 97116). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.
McNamara, D. S., & Kintsch, W. (1996). Learning from texts: Effects of prior knowledge and text co-
herence. Discourse Processes, 22, 247288.
Myers, J. L., &OBrien, E. J. (1998). Accessing the discourse representation during reading. Discourse
Processes, 26, 131157.
Myers, J. L., OBrien, E. J., Albrecht, J. E., & Mason, R. A. (1994). Maintaining global coherence dur-
ing reading. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20, 876886.
Narvaez, D., van den Broek, P., & Ruiz, A. B. (1999). The influence of reading purpose on inference
generation and comprehension in reading. Journal of Educational Psychology, 91, 488496.
OBrien, E. J. (1995). Automatic components of discourse comprehension. In E. P. Lorch & E. J.
OBrien (Eds.), Sources of coherence in reading (pp. 159176). Hillsdale, NJ: Lawrence Erlbaum
Associates, Inc.
314 VAN DEN BROEK, RAPP, KENDEOU
OBrien, E. J., & Albrecht, J. E. (1991). The role of context in accessing antecedents in text. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 18, 777784.
OBrien, E. J., Albrecht, J. E., Hakala, C. M., & Rizzella, M. L. (1995). Activation and suppression of
antecedents during reinstatement. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 21, 626634.
OBrien, E. J., & Myers, J. L. (1999). Text comprehension: A view from the bottom-up. In S. R.
Goldman, A. C. Graesser, & P. van den Broek (Eds.), Narrative comprehension, causality, and co-
herence: Essays in honor of TomTrabasso (pp. 3553). Mahwah, NJ: Lawrence ErlbaumAssociates,
Inc.
OBrien, E. J., Rizzella, M. L., Albrecht, J. E., & Halleran, J. G. (1998). Updating a situation model: A
memory-based text processing view. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 24, 12001210.
Rapp, D. N., & Gerrig, R. J. (2002). Readers reality-driven and plot-driven analyses in narrative com-
prehension. Memory and Cognition, 30, 779788.
Rapp, D. N., Gerrig, R. J., & Prentice, D. A. (2001). Readers trait-based models of characters in narra-
tive comprehension. Journal of Memory and Language, 45, 737750.
Rapp, D. N., &Taylor, H. A. (2004). Interactive dimensions inthe constructionof mental representations
for text. Journal of Experimental Psychology: Learning, Memory, and Cognition, 30, 9881001.
Richards, E., & Singer, M. (2001). Representation of complex goal structures in narrative comprehen-
sion. Discourse Processes, 31, 111135.
Rizzella, M. L., & OBrien, E. J. (1996). Accessing global causes during reading. Journal of Experi-
mental Psychology: Learning, Memory, and Cognition, 22, 12081218.
Singer, M., Graesser, A. C., &Trabasso, T. (1994). Minimal or global inference during reading. Journal
of Memory and Language, 33, 421441.
Smith, E. E., Shoben, E. J., & Rips, L. J. (1974). Structure and process in semantic memory: A featural
model for semantic decisions. Psychological Review, 81, 214241.
Suh, S., & Trabasso, T. (1993). Inferences during reading: Converging evidence from discourse analy-
sis, talk-aloud protocols, and recognition priming. Journal of Memory and Language, 32, 279300.
Sundermeier, B. A., van den Broek, P., &Zwaan, R. A. (in press). Causal coherence and the availability
of locations and objects during narrative comprehension. Memory and Cognition.
Turnipseed, J. (2003). Baghdad express. New York: Penguin.
van den Broek, P., Fletcher, C. R., & Risden, K. (1993). Investigations of inferential processes in read-
ing: A theoretical and methodological integration. Discourse Processes, 16, 169180.
van den Broek, P., Kendeou, P., Sung, Y. C., & Chen, M. (2003, June). Comprehension and memory of
science texts: Asimulation using the Landscape Model. Paper presented at the Annual Meeting of the
Society for Text and Discourse, Madrid, Spain.
van den Broek, P., Lorch, R. E. J., Linderholm, T., &Gustafson, M. (2001). The effects of readers goals
on inference generation and memory for texts. Memory and Cognition, 29, 10811087.
van den Broek, P., OBrien, E. J., Halleran, J. G., & Kendeou, P. (2004). Memory-based and explana-
tory processes during reading. Manuscript in preparation.
van den Broek, P., Risden, K., &Husebye-Hartmann, E. (1995). The role of readers standards of coher-
ence in the generation of inferences during reading. In E. P. Lorch & E. J. OBrien (Eds.), Sources of
coherence in reading (pp. 353374). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.
van den Broek, P., Risden, K., Fletcher, C. R., & Thurlow, R. (1996). A landscape view of reading:
Fluctuating patterns of activation and the construction of a stable memory representation. In B. K.
Britton & A. C. Graesser (Eds.), Models of understanding text (pp. 165187). Hillsdale, NJ: Law-
rence Erlbaum Associates, Inc.
van den Broek, P., Rohleder, L., &Narvaez, D. (1996). Causal inferences in the comprehension of liter-
ary texts. In R. J. Kreuz &M. S. MacNealy (Eds.), Empirical approaches to literature and aesthetics
(pp. 179200). Norwood, NJ: Ablex.
INTEGRATING READING PROCESSES 315
van den Broek, P., Virtue, S., Everson, M., Tzeng, Y., & Sung, Y. C. (2002). Comprehension and mem-
ory of science texts: Inferential processes and the construction of a mental representation. In J. Otero,
J. A. Leon, & A. C. Graesser (Eds.), The psychology of science text comprehension (pp. 131154).
Mahwah, NJ: Lawrence Erlbaum Associates, Inc.
van den Broek, P., Young, M., Tzeng, Y., & Linderholm, T. (1999). The landscape model of reading. In
H. van Oostendorp &S. R. Goldman (Eds.), The construction of mental representations during read-
ing (pp. 7198). Mahwah, NJ: Lawrence Erlbaum Associates, Inc.
Virtue, S., van den Broek, P., &Linderholm, T. (in press). Hemispheric processing of causal inferences.
Memory & Cognition.
Zwaan, R. A. (1994). Effect of genre expectations on text comprehension. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 20, 920933.
316 VAN DEN BROEK, RAPP, KENDEOU

Das könnte Ihnen auch gefallen