CHAPTER 3
Reading Comprehension and Reading Comprehension Problems
A Word-to-Text Integration Perspective
Charles Perfetti, Joseph Z. Stafura, and Suzanne M. Adlof
Comprehension depends on multiple sources of knowledge and the processes that use such knowledge. In addressing reading comprehension problems, research has naturally focused on the higher level processes that are the object of text comprehension researchā making inferences, monitoring comprehension, and so on. In this chapter, we suggest that this approach is incomplete and draw attention to the importance of word-by-word text comprehension processes that integrate a word with the readerās representation of the text.
READING COMPREHENSION PROBLEMS: WHAT ARE THEY?
The new DSM-5 (Diagnostic and Statistical Manual of Mental Disorders-Fifth Edition; American Psychiatric Association, 2013) has a proposed entry for āSpecific Learning Disorderā that specifies a comprehension disorder:
Difficulty understanding the meaning of what is read (e.g., may read text accurately but not understand the sequence, relationships, inferences, or deeper meanings of what is read).
Although having medical and psychiatric specialists defining cognitive and behavioral problems of reading comprehension might be a case of expertise inflation, the proposed DSM is in tune with the reading fieldās view of the problem. However, the proposed DSM entry is a bit wishy-washy in its constraining conditionāspecifically an individual with a specific learning disorder āmay read text accurately.ā Researchers of comprehension deficits insist that the designation ācomprehension deficitā requires accurate (usually described as āadequateā) word reading. Otherwise, they could not be sure that the problem is specifically about comprehension and not word reading.
This āexclusion criterionā is parallel to the standard definition of dyslexia, for which adequate general or nonverbal intelligence has been required. Thus, dyslexia is supposed to be about word-reading difficulties without general cognitive problems; comprehension disability is supposed to be about text comprehension difficulties without word-reading problems.
WHAT IS THE MAJOR CAUSE OF READING COMPREHENSION DISABILITY?
If the major cause of word-reading disability (dyslexia) is in the phonological processing system, what is the major cause of reading comprehension disability? The question is worth posing mainly to emphasize that there is likely to be no single answer. The major candidates, reflecting the emphases from research, would be the following:
- Cognitive system: poor working memory (Goldman & Varma, 1995; Just & Carpenter, 1992); poor cognitive control (Hamm & Hasher, 1992; Hasher & Zacks, 1988; Locascio, Mahone, Eason, & Cutting, 2010).
- General language problems: deficits in oral language (Nation & Snowling, 1997; also, Cain, 2010; Catts, Adlof, & Ellis Weismer, 2006; Nation, Cocksey, Taylor, & Bishop, 2010; Stothard & Hulme, 1995); problems acquiring semantic relationships (Landi & Perfetti, 2007; Nation & Snowling, 1999); specific language disorders often coincidental with dyslexia (Bishop & Snowling, 2004; Catts, 1991; Catts, Adlof, Hogan, & Ellis Weismer, 2005).
- Higher level comprehension problems: failures to make inferences (Oakhill & Garnham, 1988); failures to monitor comprehension (Baker, 1984; Garner, 1980).
However, if we take seriously the idea of specific reading comprehension deficits, we would need to apply the exclusionary standards to them. For example, a general cognitive deficit in working memory might limit comprehension processes and reduce inference making or comprehension monitoring. In such a case, it would be misleading to refer to a deficit in inference making or comprehension monitoring in the usual sense of a deficit. Instead, such difficulties would be manifestations of a more basic cognitive problem. Before one can identify a special disability in inference making (or comprehension monitoring or narrative knowledge), word-reading problems must be excluded, as should problems in memory and spoken language.
This approach becomes clearer when the relevant factors are seen as part of the reading system, rather than in a list. Figure 3.1 shows the reading comprehension system, along with pressure points that potentially influence both knowledge structures and processes in the system. Although processes are illustrated, it is the knowledge components that deserve attention in terms of individual differences. Readers differ in their knowledge of words, their spelling, their pronunciation, their meaning, and the conditions of their useāand in their general knowledge. The basic reading processes (decoding, encoding, retrieval) operate on this knowledge base and may not have as much variability except as affected by the amount of relevant practice they have received. Indeed, the ādefaultā hypothesis for what causes the development of reading skill is relevant knowledge and practice in applying that knowledge in reading. If this sounds too simple, it probably is. But this is the formula for expertise in other domains (Ericsson, Krampe, & Tesch-Romer, 1993), where research has taken the mystery factor out of expertise and supplanted it with explanation based on practice.
Figure 3.1. The comprehension system and its pressure points. The left portion is a simplified, schematic layout of the major knowledge components and processes of the reading comprehension system. The pressure points on the right indicate potential causes of comprehension problems. The connecting lines show hypothesized effects of weaknesses at these pressure points on the major components of the comprehension system. Higher level comprehension processes exert influence on text representations and situation models, whereas word-level pressure points affect the lexical representations that are available for text representations and the situation model. General knowledge, which is omitted from the figure, exerts an influence over most parts of the system. (From Perfetti, C.A. [1999]. Comprehending written language: A blueprint of the reader. In C. Brown & P. Hagoort [Eds.], The neurocognition of language [pp. 167-208]. New York, NY: Oxford University Press.)
The reading field might consider this skill model more seriously. Effective practice at reading text strengthens word knowledge, informational knowledge that supports subsequent reading, and basic and higher level skills in identifying words, retrieving context-relevant word meanings, and making coherence-preserving inferences. The challenge is to make practice effective. Just as component skills in other activities need to be specially tutored (e.g., a backhand in tennis, or chords on the piano), young readers may need help becoming more active readers so that they can monitor their comprehension or draw appropriate inferences. But the real goal, as it is in playing tennis or the piano, is to perform the fully integrated act, in this case the act of reading texts. To perform this act skillfully, much reading is necessary; here, the instructional goal is to provide enough tutoring and scaffolding to support a desire to read (i.e., to practice reading). In this perspective, the goal of trying to find specific comprehension ādeficitsā that afflict large groups of readers seems misplaced.
WORD UNDERSTANDING AND TEXT COMPREHENSION
Practice in reading builds word knowledge, and increased word knowledge supports comprehension skill. Our research has partly focused on these reciprocal ideas:
- Learning words depends on acquiring information about word forms and meanings from word-learning events, and more skilled comprehenders do this better than less skilled comprehenders (Anderson & Freebody, 1981; Balass, Nelson, & Perfetti, 2010; Bolger, Balass, Landen, & Perfetti, 2008; Nagy, Herman, & Anderson, 1985; Perfetti, Wlotko, & Hart, 2005; Stanovich & West, 1989).
- Text comprehension depends on understanding words and integrating their meaning into a mental model of the text; more skilled comprehenders do this better than less skilled comprehenders (Perfetti, Yang, & Schmalhofer, 2008; Yang, Perfetti, & Schmalhofer, 2005, 2007).
The implications of these twin hypotheses for reading comprehension skill have been explored using various methods. However, āon-lineā reading studies are especially important for examining word-learning events and comprehension at a level of detail that can be mapped on to processes and the knowledge stores that they act upon. Here, we focus on the second claim that text comprehension depends on integrating words into a mental model of the text.
Event-Related Potential Studies of Word-to-Text Integration
The idea of mental model-based integration is illustrated here in two slightly different texts. Text (1) begins thus:
(1) While Cathy was riding her bike in the park, dark clouds began to gather, and it started to storm.
Making assumptions about the details of a mental model is tricky. Indeed, the concept of a mental model (Johnson-Laird, 1980), or the more text-specific āsituation modelā (van Dijk & Kintsch, 1983), has been compelling informally, but aside from inferred spatial relations in the earliest mental models described by Johnson-Laird (1981, 1983; also see Zwaan & Radvansky, 1998), specifications have been ad hoc. In the following example, we illustrate one way to represent a situation model, highlighting that its content is in term of referents and events, rather than the literal text.
<SITUATION: Cathy on bike, in the park, dark clouds>
<EVENT: Storm>
As notated above, the model includes two substructures, Situation + Event, to represent the current situation model. The reader has memory access to the situation and can update it through events described in the text. In this case, the situation is updated immediately by the storm event:
<SITUATION: Storm, Cathy on bike in park>
Suppose the next sentence adds to text (1) as shown here:
(1) While Cathy was riding her bike in the park, dark clouds began to gather, and it started to storm. The rain ruined her beautiful sweater.
The text is conceptually coherent. Information from the second sentence can be added to the situation model, introducing a new event, the ruination of a sweater. But comprehension proceeds not only sentence-by-sentence, but also word-by-word. The noun phrase that begins the new sentenceāthe raināis understood immediately in relation to the situation model. It refers to the storm event and can be integrated into the situational representation of the text.
This word-level integration process can be observed through various word-by-word reading measures, including event-related potential (ERP) methods and eye tracking. ERP measures allow for the observation of this integration as it occursāa comprehension process, rather than a comprehension product. Our experiments, using such sentences as (1) above, measured ERPs initiated by the target word, which is rain in the current example. When the target word appeared, the N400 component, an indicator of the fit between the word and its context (Kutas & Federmeier, 2000), was reduced in amplitude, indicating word-to-text integration.
In text (2), which follows the N400 on the word rain, there is a more pronounced negative deflection because it does not fit as easily into its context. There is no antecedent for rain. Equivalently, the situation model contains no referent to which the new event, the rain, can be attached.
(2) When Cathy saw there were no dark clouds in the sky, she took her bike for a ride in the park. The rain that was predicted never occurred.
Experimentally, (2) is a baseline condition against which N400 reductions produced by text manipulations (paraphrase, inference, and explicit mention) are compared. Important is the fact that text (2) is perfectly comprehensible. However, the readerās situation model is different from that in (1)āthere is no storm. So, when the word rain is encountered, there is no rapid integration process that adds the idea of rain to the situation model; instead, a new mental structure is built around the rain and (finally) its nonoccurrence.
We emphasize the sensibility of text (2) because in most research that uses the N400 as an indicator of semantic processing, something that is sensible is compared with something that is not sensible, or anomalous. For example, in a classic N400 study, an ERP is recorded on the sentence-final word when it is sensible, given the preceding context, āThe pizza was too hot to eat,ā as compared with the ERP to the anomalous ātoo hot to drinkā (Kutas & Hillyard, 1980). In these situations, the N400 differences are dramatic and appear amenable to explanations based on expectancy violations (Kutas & Federmeier, 2000; Lau, Almeida, Hines, & Poeppel, 2009). The N400 is assumed to be sensitive to poor fit (failures of expectations to be met, or failures to make sense of what occurs). In our case, however, a comparison is made across sensible texts. The texts differ only in the degree to which they invite an immediate word-to-text integration process. (See Brown & Hagoort, 1993, for N400 interpretations based on postlexical integration processes.) It is difficult to imagine what word a reader might predict across a sentence boundary; nearly any grammatical sentence beginning can continue with coherent ties to the preceding sentence.
The Paraphrase Effect and Comprehension Skill
W...
