![]()
p.5
Part I
Overviews
![]()
p.7
1
Reading Comprehension Theories
A View from the Top Down
Panayiota Kendeou
UNIVERSITY OF MINNESOTA
Edward J. O’Brien
UNIVERSITY OF NEW HAMPSHIRE
The Nature of Reading Comprehension
Comprehension is complex and multidimensional. It depends on the execution and integration of many processes (for reviews, see Alexander & The Disciplined Reading and Learning Research Laboratory, 2012; Kendeou & Trevors, 2012; McNamara & Magliano, 2009; RAND, 2002). To understand a sentence, for example, one must visually process the words, identify their phonological, orthographic, and semantic representations, and connect words using rules of syntax to form an understanding of the underlying meaning. In this chapter, we do not focus on any of the factors that help readers at these lower levels of the comprehension process (Perfetti & Stafura, 2014). Instead, we focus on comprehension processes beyond individual words and sentences. These comprehension processes help readers understand and integrate meaning across sentences, make use of relevant background knowledge, generate inferences, identify the discourse structure, and take into consideration the authors’ goals and motives (Graesser, 2015). The end product is a mental representation that reflects an overall meaning of the extended written discourse. For all of these comprehension processes to be successful, many interacting factors play a role, such as reader characteristics, text properties, and the demands of the task at hand (Kintsch, 1998; RAND, 2002; Rapp & van den Broek, 2005). These factors individually and jointly influence reading processes and products. Achieving a basic level of comprehension that reflects the intended meaning of a simple text, however, is not sufficient. Individuals must also be able to analyze, synthesize, and evaluate information within and across texts (Alexander & DRLRL, 2012; NAEP, 2013). These even higher-level processes result in deeper comprehension (Graesser, 2015) and learning from texts (Goldman & Pellegrino, 2015).
p.8
Discourse scientists have recently called for a focus on these sorts of higher-level processes (Kendeou, McMaster, & Christ, 2016; McNamara, Jacovina, & Allen, 2016) that support deeper comprehension of the texts we read (Graesser, 2015), and thus deep learning (Goldman & Pellegrino, 2015). These calls have implications for shifting the focus of the discourse processes field from the investigation of passive to strategic processes and from simple to complex products. In this chapter, we discuss the implications for embracing this shift in focus in the context of theories and models of reading comprehension. We also provide an overview of the main challenges this shift in focus poses for our existing theoretical frameworks, methods, and definitions.
Embracing Complexity: A Theory of Reading Comprehension?
The complexity of reading comprehension demands a theory that describes the cognitive and linguistic processes involved, as well as their respective interactions across development. Recently, Perfetti and Stafura (2014) argued that “There is no theory of reading, because reading has too many components for a single theory” (p. 22). This statement begs the question: Does the complexity of reading truly limit the development of a comprehensive theory of reading comprehension? We agree with Perfetti and Stafura that for a phenomenon as complex as reading comprehension it is unlikely to develop a single, comprehensive theory in a traditional sense. But if we cannot develop a comprehensive theory of reading comprehension, what is the alternative? One alternative is the development of multiple theories, each focusing on a separate component of reading. This approach has been fruitful for the development of theories for lower-level processes, such as the reading and learning of words (see Ehri, 2014 for a review). A second alternative is the development of a ‘good-enough’ approximation of a theory of reading comprehension. In fact, discourse scientists have put forth several models and frameworks that have attempted to do just that. These models and frameworks are concerned with the mental representation the reader constructs in the process of understanding entire texts. As such, collectively they consist of a good theoretical approximation of many component processes that would be needed to develop a comprehensive model of reading comprehension.
Reviewing current models of reading comprehension processes is beyond the scope of this chapter, and excellent reviews can be found elsewhere (e.g., McNamara & Magliano, 2009). We would like, however, to highlight the focus and main claims of several of these models. To this end, we will briefly review the prominent models in the extant literature that deal with the comprehension of single texts, because the comprehension of multiple texts will be touched upon in other sections in this volume. These single-text models include the Construction-Integration model (Kintsch & van Dijk, 1978; Kintsch, 1988), the Structure Building model (Gernsbacher, 1990), the Resonance model (Myers & O’Brien, 1998; O’Brien & Myers, 1999), the Event-Indexing model (Zwaan, Magliano, & Graesser, 1995), the Causal Network model (Trabasso, van den Broek, & Suh, 1989), the Constructionist model (Graesser, Singer, & Trabasso, 1994), and the Landscape model (van den Broek, Young, Tzeng, & Linderholm, 1999). Each of these models taps into different aspects of reading comprehension processes and, collectively, they provide a complete view of reading comprehension processes.
The Construction-Integration model (Kintsch, 1988) posits that comprehension involves two phases – construction and integration. Construction is the activation of information in the text along with the reader’s related prior knowledge, whereas integration is the spreading of the activation. The result of these two phases is the construction of a mental representation of what the text is about (i.e., a situation model). In the Structure Building model (Gernsbacher, 1990) comprehension involves laying a foundation structure at the beginning of reading a text that serves to map subsequent text information. When subsequent text information maps well on this structure, it gets integrated into the evolving mental representation, and when it does not the reader shifts and lays a foundation for another structure. Importantly, Gernsbacher proposed that these processes (i.e., laying the foundation, mapping, and shifting) are common across modalities (text, discourse, picture, video). The Resonance model (Myers & O’Brien, 1998; O’Brien & Myers, 1999) describes factors that influence the activation of information from the text and the reader’s prior knowledge during reading. The model has been employed to show that memory-based retrieval mechanisms are sufficient to explain much of what is involved in the activation of information that is used to construct inferences during reading. It has also provided an excellent account of how readers gain access to earlier portions of a text (e.g., Albrecht & O’Brien, 1995; O’Brien, 1995; Peracchi & O’Brien, 2004). The Event-Indexing model (Zwaan et al., 1995) specifies five dimensions of coherence during situation model construction that readers are sensitive to and monitor during reading: time, space, causality, motivation, and agents. The Causal Network model (Trabasso et al., 1989) describes situation model construction through the lens of causal reasoning, and outlines four types of causal relations (enabling, psychological, motivational, and physical) that facilitate the construction of causal networks during reading. The Constructionist model (Graesser et al., 1994) focuses on the deliberate, strategic processes that can result in deep comprehension. The Landscape model (van den Broek et al., 1999) posits comprehension as the fluctuation of concept activation that is driven by text factors as well as the reader’s criteria for comprehension or standards of coherence.
p.9
To date, we contend that the best approximation to a theory of reading comprehension is the Construction-Integration (CI) model. The CI model (Kintsch & van Dijk, 1978; Kintsch, 1988; van Dijk & Kintsch, 1983) describes the iterative processes in mapping current discourse input to the prior discourse context and background knowledge. Construction refers to the activation of the information in the text and background knowledge. There are four potential sources of activation during the construction phase: the current text input (e.g., a proposition), the prior sentence, background knowledge, and prior text. As this information is activated, it is connected into a network of propositions and concepts. Integration refers to the continuous spread of activation within this network with the level of activation of individual propositions fluctuating until activation settles. Eventually, over iterations, the level of activation of individual propositions within the network stabilizes. Activation sources from the construction process are iteratively integrated within the limits set by working memory. This iterative process prunes the network so that only those concepts and ideas that are connected to many others stay in the network; less connected concepts lose activation and drop from the model. This pruning process tends to maintain concepts and ideas that are relevant to the current situation model while reducing activation of irrelevant concepts and ideas. With every new cycle this process repeats and, at the completion of reading, the result is a situation model, namely a mental representation of what the text is about.
The notion of a situation model proposed in the context of the CI model is central in the discourse literature because it is considered the final product of reading comprehension, whereas its construction involves the processes of reading comprehension. Despite its sophistication and status as the best approximation of what the reader takes away from a completed comprehension process, situation models are insufficient to account for what is described as deeper comprehension (Graesser, 2015). Deeper comprehension also involves higher-order thinking (McNamara et al., 2016) such as analyzing, synthesizing, and evaluating information. No current model, including the CI model, can sufficiently explain this level of comprehension. An excellent example of the distinction between understanding at the level of the situation model and understanding what goes beyond the situation model is demonstrated in Goldman’s recent work on the comprehension of literary texts (e.g., narratives, folk tales, poems, science fiction), as exemplified by the differences between a ‘literal stance’ and an ‘interpretive stance’ (Goldman, McCarthy, & Burkett, 2015). Specifically, in the context of a literal stance, namely an orientation toward what the text ‘says,’ readers develop an understanding of what the text is about akin to the situation model level. However, in the context of an interpretive stance, namely an orientation toward what the text ‘means’ (e.g., comprehending the intended meaning of a fable or parable), readers develop an understanding abstracted from the specific situation. For example, an interpretive understanding of the fable about “the tortoise and the hare” requires far more than the development of a situation model of a race between a tortoise and a hare; it requires a much deeper interpretive understanding (i.e., slow and steady wins the race). This level of understanding constitutes deeper comprehension.
p.10
Greater precision in the conceptualization and prediction of deeper comprehension is, in our view, a reasonable target for the development of a comprehensive theory of reading comprehension. Engaging in further theory development in the field of discourse processes is important not only for the field itself, but also for its position in current and future policy initiatives aimed at increasing reading proficiency and learning (Goldman & Pellegrino, 2015; Graesser, 2015; Kendeou et al., 2016). McNamara and Magliano (2009) noted that such a comprehensive theory can only be viable if it can generate testable predictions and be computationally implemented. In doing so, there needs to be a greater specification of reading comprehension processes and products across a range of situations, readers, and tasks. We turn to this issue next.
From Passive to Strategic Processes
Given the increased focus on deeper comprehension, we believe that the field of discourse processes needs to move toward greater specification and approximation of deeper comprehension as a product and the processes involved in it. Inevitably, this also means moving toward greater specification and approximation of strategic processing, as strategic processing and deeper comprehension are closely intertwined. That is, strategic processing is often necessary for deeper comprehension to occur (Graesser, 2007). We do not deny, however, that in situations in which we are dealing with a highly-skilled reader with expertise in a particular domain, deeper comprehension could also occur with relatively few, if any, strategic processes (Ericsson & Kintsch, 1995). But, in general, we contend that as we increase the focus on deeper comprehension, it requires that we increasingly shift our focus away from passive processes and more toward strategic processes.
Strategic processes are intentional, voluntary, and generally fully conscious; they are also more amenable to self-regulation and tend to rely more on serial processing (Schneider & Shiffrin, 1977; Sternberg, 1999). These characteristics make strategic processes more dependent on attentional resources and subject to voluntary control. In contrast, passive processes are effortless, unintentional, and generally outside conscious awareness; they are also difficult to stop or regulate and typically rely on a parallel type of processing (Schneider & Shiffrin, 1977; Sternberg, 1999). These characteristics make passive processes relatively independent of attentional resources and not subject to voluntary control.
Despite these definitions, trying to define with any degree of precision exactly what constitutes passive and strategic processing during reading comprehension is an almost impossible task because the line that divides passive from strategic processes is thin. Not only is it thin, but it is always fluctuating and shifting as a function of the reader, the text, and the task at hand. For example, the process of generating inferences that can be strategic for some individuals (mediated by expertise; Ericsson & Kintsch, 1995), texts (mediated by background knowledge demands; McNamara & Kintsch, 1996), tasks (Cerdán, Gilabert, & Vidal-Abarca, 2011; Rouet, Vidal-Abarca, Erboul, & Millogo, 2001), instructions (McCrudden, Magliano, & Schraw, 20...
