Theories of dynamic text comprehension (DTC) aim to capture the integration ... model and behavioral data provide evidence for the validity and necessity of the ...
CURRENT DIRECTIONS IN PSYCHOLOGICAL S CIENCE
Dynamic Text Comprehension An Integrative View of Reading David N. Rapp and Paul van den Broek Department of Educational Psychology, University of Minnesota
ABSTRACT—Reading
is one of the most complex and uniquely human of cognitive activities. Our understanding of the processes and factors involved in text comprehension is quite impressive, but it also is fragmented, with a proliferation of ‘‘mini-theories’’ for specific components that in reality are intertwined and interact with one another. Theories of dynamic text comprehension (DTC) aim to capture the integration of these components. They depict reading comprehension as an ongoing process involving fluctuations in the activation of concepts as the reader proceeds through the text, resulting in a gradually emerging interpretation of the material. Features of texts and characteristics of the reader jointly and interactively affect these fluctuations, influencing and being influenced by the reader’s understanding and memory of what is read. We illustrate the DTC approach by describing one theory, called the Landscape model, and summarize how its simulations match empirical data. We conclude with some implications of the DTC framework for basic and applied reading research.
KEYWORDS—reading; text processing; text comprehension; computational models
Reading is one of the most complex and uniquely human of cognitive activities. Psychological research has greatly enhanced our understanding of the cognitive processes, mental structures, and textual properties that contribute to successful reading by identifying numerous factors that influence comprehension (e.g., the reader’s background knowledge, the difficulty of the text, individual differences in reading skill, and so on). Unfortunately, the success of this research has created problems of its own. First, each factor tends to be studied in relative isolation, resulting in a large number of ‘‘mini-theories’’ about the contribution of individual factors, with little consid-
Address correspondence to David N. Rapp, Department of Educational Psychology, 206 Burton Hall, 178 Pillsbury Drive S.E., University of Minnesota, Minneapolis, MN, 55455; e-mail: rappx009@ umn.edu.
276
eration of how they operate alongside each other and in mutually interactive ways. Second, at times this proliferation of theories has resulted in debates about the validity of one theory over another. We will argue, however, that the different theories may account for complementary and even mutually supportive aspects of reading comprehension. A major reason for the fragmented nature of reading research is that attempts to investigate multiple factors in traditional experimental studies quickly become unwieldy, both in design and interpretability, when considering multidimensional interactions (for example, assessing interactions among readers’ memory and language skills, readers’ prior knowledge, the difficulty of a text, and demographic variables of populations of interest). In this article we exemplify the limitations of focus on mini-theories, and consider how reading research has attempted to address the problem. We introduce a conceptual framework, dynamic text comprehension (DTC), that focuses on multiple factors and their interactions during reading. We illustrate this framework through a particular instance, the Landscape model, which captures a range of empirical data and phenomena in reading comprehension. The convergence of findings from the model and behavioral data provide evidence for the validity and necessity of the DTC framework. We close with a discussion of some theoretical and practical implications of a DTC view. LIMITATIONS OF CURRENT APPROACHES
Examples of the limitations inherent to investigating isolated aspects of reading are not hard to find. One specific example that has received considerable recent interest in the field of text comprehension concerns theoretical accounts of how information is activated (or reactivated) from background knowledge during reading. The research on this issue has resulted in hypotheses that seemingly are in competition but in reality likely describe different aspects of the reading process. Two types of mechanisms have been proposed. From a memory-based perspective, each word, phrase, or concept that a reader processes triggers an automatic spread of activation to other, related words and concepts in memory for the text read so far and background knowledge. In this account, the reader has little or no control
Copyright r 2005 American Psychological Society
Volume 14—Number 5
David N. Rapp and Paul van den Broek
over the information that is activated at any point during reading (Gerrig & McKoon, 1998; O’Brien, Rizzella, Albrecht, & Halleran, 1998). From a constructionist perspective, readers’ goals and strategies play a central role in the activation of information from memory. Readers, according to this view, are described as actively striving to achieve understanding of the text, strategically activating information to satisfy their search for meaning (Graesser, Singer, & Trabasso, 1994). Although both types of processes intuitively seem necessary during comprehension, because they are often studied separately they have traditionally been presented as competing accounts of underlying reading mechanisms. Indeed, the competition between these accounts has, at times, led to acrimonious interchanges among researchers. Only recently have memory-based and constructionist processes been explicitly considered as complementary and perhaps mutually supportive (see Gue´raud & O’Brien, 2005). In fact, theoretical accounts now suggest that a failure to incorporate both mechanisms results in impoverished theories (e.g., Kintsch, 1998; van den Broek, Rapp, & Kendeou, 2005). For example, leaving out more automatic mechanisms fails to explain how multiple, at times even irrelevant meanings are quickly activated during reading (Kintsch, 1998; O’Brien, et al., 1998). Additionally, ignoring strategic components such as the reader’s specific goals or particular demands of the task (e.g., Linderholm & van den Broek, 2002) fails to account for how those irrelevant meanings may ‘‘fall away’’ during comprehension. As we will show, DTC models have the potential to specify how multiple processes, such as the automatic and strategic activation of information from memory, might be combined in a single theoretical framework. A second, more general example of the limitations of minitheories is the fact that some are concerned exclusively with the process of reading, whereas others are concerned with reading products. The former describe how cognitive activity fluctuates during reading, how working-memory limitations or textual features influence such cognitive activity, and so on. Typical methods used to investigate these processes include on-line measures (i.e., measures assessed during moment-by-moment reading, as opposed to afterward) such as reading times, tasks measuring the amount of time it takes to respond to related or unrelated stimuli, and eye movements. In contrast, product-driven research aims to describe the nature of text representations in memory once reading is completed, quantitative and qualitative differences in those representations based on experimenter-manipulated variables (e.g., task requirements or text differences), and so on. Typical methods for investigating these products include off-line measures such as recall, question-answering tasks, and the application of knowledge to novel situations. Exclusive focus on either process or product ignores the obvious fact that the two must be closely connected. First, what happens during reading must somehow be the foundation for what the reader retains afterward, so it is incumbent upon re-
Volume 14—Number 5
searchers to investigate this relationship. Second, and more subtly, readers do not wait until reading is complete to start constructing their mental representations. If interrupted in the middle of a text, readers readily report what the text was about up to that point. Thus, construction of an eventual product is already underway as processes continue to play out. Indeed, in most accounts of on-line processing, it is implicitly assumed that partial representations exist and can influence the on-line processing of subsequent text. As we will see, DTC models make these assumptions explicit by including both process and product, and their interrelations, in a single account. DYNAMIC TEXT COMPREHENSION
DTC addresses the aforementioned limitations by extending research in several ways: It (a) integrates multiple factors (e.g., concept activation, inference construction, individual differences, text properties, characteristics of memory representations) and their interactions in a single framework; (b) attempts to account for the dynamic fluctuations in activation of concepts during moment-by-moment comprehension of the entire text; (c) takes into consideration both the processes and the products of comprehension (and, in some cases, the recursive relation between them); (d) often involves computational simulation of behavioral data to examine these factors and mechanisms (e.g., Goldman & Varma, 1995; Kintsch, 1988; Langston & Trabasso, 1998); and (e) in some cases allows seemingly competing hypotheses to be integrated (e.g., in DTC theories, memory-based and constructionist processes operate interactively). THE LANDSCAPE MODEL
To illustrate the DTC framework, we describe one exemplar, the Landscape model of reading (to access the model, see http:// education.umn.edu/EdPsych/Projects/LandscapeModel/default. html). The Landscape model incorporates multiple cognitive and textual factors influencing comprehension and is intended to capture cognitive activity during reading as well as the mental representation that is gradually constructed over the course of the reading experience. According to the model, a reader proceeds through the text in cycles, with each cycle corresponding to a clause or sentence. From cycle to cycle, concepts fluctuate in activation as a function of four sources: (a) text input in the current cycle, (b) residual information from the preceding cycle, (c) the memory representation constructed for the text read so far, and (d) the reader’s prior knowledge. These fluctuations result in a ‘‘landscape’’ of activations, with concepts waxing and waning in activation over the course of reading. The patterns of activation in the model are determined by an array of reader and text characteristics: attentional or working-memory capacity, amount and content of background knowledge, goals and strategies the reader brings to the task, the organization of the text, and so on. Thus, the model incorporates multiple factors simultaneously.
277
Dynamic Text Comprehension
From the very first reading cycle, the patterns of activation result in a memory representation that is continually updated with each subsequent cycle, eventually leading to a stable representation once reading is completed. Specifically, each individual concept that is activated is added to the representation of the text in memory or, if the concept already was part of the representation, its representation is strengthened. Similarly, coactivation of concepts leads to the generation or strengthening of connections between those concepts. For each of these particular cases, the amount of change in the memory representation is a function of the amount of activation of the concept or concepts and of the existing memory strength. The result is a gradually emerging network representation of the text ideas and their interconnections. The Landscape model is dynamic in several respects. First, it describes the cognitive processes involved in comprehension throughout the duration of the reading of a text, not just at specific points selected to capture the effect of a particular factor. Second, it captures the interactive effects of multiple factors. Third, it posits that at each cycle the memory representation constructed during preceding cycles is a source of activation and, hence, influences subsequent activation patterns; in turn, these cyclical and dynamically fluctuating activations modify the existing representation, resulting in the gradual emergence of a final episodic representation (i.e., accumulated memory for the text experience). Thus, the traditional distinction between process and product is replaced by a recursive interaction between the two. To describe text comprehension, the Landscape model also integrates multiple processes (in addition to multiple factors). For example, it includes memory-based and constructionist processes by postulating two mechanisms. The first is cohort activation: When a concept is activated during reading, other concepts associated with it (its cohort) are also activated. Cohorts either pre-exist in semantic memory (i.e., prior knowledge) or are constructed during reading as concepts concurrently activated during a cycle become associated in the episodic memory representation for the text. The second mechanism is coherencebased retrieval, a strategic mechanism by which information is retrieved with the specific aim of meeting a reader’s standards for coherence. Again, such retrieval can be from the episodic representation constructed thus far or from prior knowledge. The model posits that both cohort activation and coherence-based retrieval take place simultaneously and, hence, that the information retrieved through one mechanism influences the execution of the other mechanism. Thus, the model assigns complementary roles to both types of mechanisms rather than considering them as competing hypotheses. EMPIRICAL TESTS OF THE LANDSCAPE MODEL
This brief overview of the Landscape model illustrates how a DTC framework can overcome the conceptual limitations of
278
mini-theories noted earlier. To be useful and valid, however, any new psychological framework must also be able to account for human performance in ways that its predecessors cannot. We illustrate the utility of DTC using examples from the Landscape model. Comparisons between the Landscape model’s predictions and behavioral data indicate that the model is a valid predictor of reading activity. With regard to the moment-by-moment activation of concepts during the reading of narratives, activation patterns produced by the model are strongly related to patterns of activation obtained from real readers (van den Broek, Young, Tzeng, & Linderholm, 1999). For example, the model can be used to simulate what happens when readers process a text that contains inconsistencies. O’Brien et al. (1998) have reported an impressive array of findings on the circumstances for which readers do or do not detect inconsistencies in texts. When the Landscape model is used to simulate these circumstances, its predictions accurately capture those findings. Additionally, with expository texts, the model has been found to predict the circumstances under which readers notice conflicts between their own misconceptions and correct textual information, as reflected in their reading rates and their verbal descriptions of what they think the text is about as they read (i.e., think-aloud responses; Kendeou & van den Broek, 2005; van den Broek et al., 2005). With regard to reading products, the Landscape model predicts both the probability that different parts of the text will be recalled by actual readers and the order in which they recall that information (van den Broek et al., 2005). Importantly, the predictive power of the Landscape model is greater than that of any of the mini-theories that form its basis. This is reflected in direct comparisons of different components of the model. To return to our earlier example of the competition between memory-based and constructionist views, the relative contribution of each of these types of processes to reading was assessed by selectively removing each from the model. The predictions for memory for expository texts employing a model that included both constructionist and memory-based processes were very strong and were significantly more accurate than the predictions of models in which either component had been removed. An identical pattern of results was observed for memory for narrative texts as well. Thus, the inclusion of both sets of processes not only makes intuitive sense, it also leads to more accurate predictions and to data that align more closely with actual reader performance. IMPLICATIONS FOR THEORY AND PRACTICE
DTC theories provide a framework for considering multiple factors during reading. Moreover they attempt to generalize across different text types, bridging theories, for example, about narrative and expository text comprehension. Such an integrated view is more parsimonious than positing completely separable mechanisms for small subsets of factors. Additionally, including
Volume 14—Number 5
David N. Rapp and Paul van den Broek
multiple factors in any theory or model necessitates identifying exactly how those factors interact. Thus, DTC promotes increased specification when examining such factors, particularly in the implementation of computational models designed to simulate human performance. DTC views, therefore, challenge researchers in the field of discourse processing to expand both the breadth and depth of their theorizing. Although this discussion has focused on the theoretical contributions of DTC, there are practical implications as well. In educational settings, the assessment of students’ reading abilities and the evaluation of traditional interventions often concentrate on the products of reading, through tests that take place after a reading task has been completed. Recently, educational researchers have urged that increased attention be given to the underlying cognitive processes involved in reading and their causal relationships to differences in reading outcomes (e.g., Pearson & Hamm, 2005). Indeed, interventions benefit from a consideration of the activities in which readers engage, or fail to engage, as they proceed through a text. Readers who fail to adequately comprehend what they are reading may engage in qualitatively different processes than readers who succeed do. Thus, interventions are more likely to be effective if they manage to improve the processes that take place during reading. Several reading interventions place an emphasis on such a view, supporting the usefulness of DTC frameworks (e.g., Jenkins, Fuchs, van den Broek, Espin, & Deno, 2003; Linderholm, et al., 2000). DTC provides the theoretical foundation to address these sorts of practical issues.
Recommended Reading Kintsch, W. (1998). (See References) Rapp, D.N., & Taylor, H.A. (2004). Interactive dimensions in the construction of mental representations for text. Journal of Experimental Psychology: Learning, Memory, and Cognition, 30, 988– 1001. van den Broek, P., Rapp, D.N., & Kendeou, P. (2005). (See References)
Acknowledgments—Preparation of this manuscript was supported by the Institute of Education Sciences, Grant R305G040021. REFERENCES Gerrig, R.J., & McKoon, G. (1998). The readiness is all: The functionality of memory-based text processing. Discourse Processes, 26, 67–86.
Volume 14—Number 5
Goldman, S.R., & Varma, S. (1995). CAPping the construction-integration model of discourse comprehension. In C. Weaver, S. Mannes, & C. Fletcher (Eds.), Discourse comprehension: Essays in honor of Walter Kintsch (pp. 337–358). Hillsdale, NJ: LEA. Graesser, A., Singer, M., & Trabasso, T. (1994). Constructing inferences during narrative comprehension. Psychological Review, 101, 371–395. Gue´raud, S., & O’Brien, E.J. (Eds.). (2005). Components of comprehension: A convergence between memory-based processes and explanation-based processes [Special issue]. Discourse Processes, 39(2–3). Jenkins, J.R., Fuchs, L.S., van den Broek, P., Espin, C.L., & Deno, S.L. (2003). Sources of individual differences in reading comprehension and reading fluency. Journal of Educational Psychology, 95, 719–729. Kendeou, P., & van den Broek, P. (2005). The effects of readers’ misconceptions on comprehension of scientific text. Journal of Educational Psychology, 97, 235–245. Kintsch, W. (1988). The use of knowledge in discourse processing: A construction-integration model. Psychological Review, 95, 163–182. Kintsch, W. (1998). Comprehension: A paradigm for cognition. New York, NY: Cambridge University Press. Langston, M.C., & Trabasso, T. (1998). Modeling causal integration and availability of information during comprehension of narrative texts. In H. van Oostendorp & S. Goldman (Eds.), The construction of mental representations during reading (pp. 29–69). Mahwah, NJ: Erlbaum. Linderholm, T., Gaddy, M., van den Broek, P., Mischinski, M., Crittenden, A., & Samuels, J. (2000). Effects of causal text revisions on more- and less-skilled readers’ comprehension of easy and difficult texts. Cognition and Instruction, 18, 525–556. Linderholm, T., & van den Broek, P. (2002). The effects of reading purpose and working memory capacity on the processing of expository text. Journal of Educational Psychology, 94, 778–784. O’Brien, 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, & Cognition, 24, 1200–1210. Pearson, P.D., & Hamm, D.N. (2005). The assessment of reading comprehension: A review of practices-past, present, and future. In S.G. Paris & S.A. Stahl (Eds.), Children’s reading comprehension and assessment (pp. 13–69). Mahwah, NJ: Erlbaum. van den Broek, P., Rapp, D.N., & Kendeou, P. (2005). Integrating memory-based and constructionist processes in accounts of reading comprehension. Discourse Processes, 39, 299–316. van den Broek, P., Young, M., Tzeng, Y., & Linderholm, T. (1999). The landscape model of reading: Inferences and the online construction of a memory representation. In H. van Oostendorp & S.R. Goldman (Eds.), The construction of mental representations during reading (pp. 71–98). Mahwah, NJ: Erlbaum.
279
