You are on page 1of 28

Discourse Processes, 47:641667, 2010 Copyright Taylor & Francis Group, LLC ISSN: 0163-853X print/1532-6950 online DOI:

: 10.1080/01638531003628809

Contributions of Self-Explanation to Comprehension of High- and Low-Cohesion Texts


Yasuhiro Ozuru
Department of Psychology University of Alaska Anchorage

Stephen Briner
Department of Psychology DePaul University

Rachel Best
Department of Physics University of Memphis

Danielle S. McNamara
Department of Psychology University of Memphis

This study examined how the contribution of self-explanation to science text comprehension is affected by the cohesion of a text at a local level. Psychology undergraduates read and self-explained a science text with either low or high local cohesion. Local cohesion was manipulated by the presence or absence of connectives and referential words or phrases that explicitly link successive sentences. After the self-explanation activity, participants answered open-ended comprehension questions about the text. Participants in the high local cohesion condition produced higher quality explanations, including more local bridging
Correspondence concerning this article should be addressed to Yasuhiro Ozuru, Department of Psychology, University of Alaska Anchorage, Social Science Bldg., Room 214B, 3211 Providence Dr., Anchorage, AK 99508-4616. E-mail: afyo1@uaa.alaska.edu

641

642

OZURU, BRINER, BEST, MCNAMARA

self-explanations, than those in the low local cohesion condition. However, these explanations, although higher in quality, did not improve comprehension. Performance on text-based comprehension questions was better in the low local cohesion condition. In addition, the correlation between self-explanation quality and comprehension performance was generally higher in the low local cohesion condition compared to the high local cohesion condition, even after factoring out participants level of topic-relevant knowledge. These data suggest that the contribution of self-explanation to comprehension is larger when the text lacks certain cues that facilitate making connections between successive ideas in a text. Further, the results imply that a key contribution of self-explanation to text comprehension is to induce active inference processes whereby readers ll in conceptual gaps in challenging texts.

Life in modern society requires reading and learning from difcult-to-understand texts, such as science textbooks. Despite the importance of learning from challenging texts, many students struggle to comprehend them adequately. Thus, research on factors that inuence science text comprehension is important for both theoretical and practical reasons. A number of studies have indicated that science text comprehension is inuenced by many factors. For example, reading comprehension can be inuenced by characteristics of the reader (e.g., Voss & Siles, 1996; Walker, 1987), text features (Beck, McKeown, Sinatra, & Loxterman, 1991; McNamara & Kintsch, 1996), and specic tasks or activities that readers carry out during reading (Chi, Bassock, Lewis, Reimann, & Glaser, 1989; van den Broek, Tzeng, Risden, Trabasso, & Basche, 2001). In this study, we are specically interested in how specic types of text cohesion inuence the process of self-explanation. Self-explanation is the process of explaining texts or problems to oneself to better understand the material. Research shows that generating self-explanations improves comprehension and learning (e.g., Chi, de Leeuw, Chiu, & LaVancher, 1994; McNamara, 2004). Yet, there have been relatively few studies conducted to explore the circumstances under which students benet most from selfexplaining texts (however, see Ainsworth & Burcham, 2007). Thus, our general research question is, Under what circumstances will students benet more from self-explaining texts? This research question is theoretically important because the question sheds light on precisely how self-explanation improves comprehension activities. The question is also practically important because it helps scholars and practitioners design effective educational interventions. In this study, we examine whether self-explanation (and its contribution to comprehension) is inuenced by local cohesion, which represents the explicitness of relations between successive sentences in a text.

CONTRIBUTIONS OF SELF-EXPLANATION

643

ROLE OF SELF-EXPLANATION IN TEXT COMPREHENSION Although there are a number of theories and models regarding text comprehension (e.g., Haviland & Clark, 1974; Johnson-Laird & Garnham, 1980; Kintsch, 1988, 1998; McNamara & Magliano, 2009), these models generally agree that the construction of mental models, based on linguistic cues, is a key process in text comprehension. Further, according to many models of text comprehension, a substantial part of text comprehension occurs in the form of relatively automatic activation of preexisting knowledge related to the text content, triggered by textbased input (Adams & Collins, 1979; Anderson, 1994; Kintsch, 1988, 1998; Rumelhart, 1985). Such automatic activation of topic-related knowledge results in a coherent mental representation of the text when a text provides sufcient (but not complete) information about given events or situations, relative to readers topic-relevant knowledge (Kintsch, 1988, 1998; McNamara, Kintsch, Songer, & Kintsch, 1996). However, when a text does not provide sufcient information relative to readers knowledge, readers cannot readily form a coherent mental representation through such an automatic activation. Consequently, the comprehension process driven by the automatic activation of relevant knowledge breaks down. In such cases, the reader must engage in more resource-consuming, follow-up processes. These follow-up processes may involve activating difcult-to-access knowledge or memory for previous sections of the text using effortful, stepby-step reasoning activities (Kintsch, 1993, 2009; Long & Lea, 2005). The execution of follow-up processes makes reading more active in the sense that the direction and extensiveness of the cognitive process is actively controlled by the readers goals instead of being determined by the strength of associative links between textual information and preexisting knowledge (Singer & Remillard, 2004). Subjectively, this type of processing would be experienced by readers as active inferring or reasoning based on partial information in the text (Graesser, Singer, & Trabasso, 1994). Thus, a lack of cohesion in a text should make the reading process more active because active processing is necessary to fully understand a text that contains cohesion gaps. However, evidence indicates that readers often do not spontaneously engage in such active, resource-demanding processes when reading a text. For example, readers do not generate backward causal inferences necessary for comprehension when the text content is unfamiliar (Noordman, Vonk, & Kempff, 1992). Some research also indicates that people often do not realize that a breakdown in comprehension has occurred (Glenberg & Epstein, 1987), suggesting that readers tend to let automatic processing dictate reading processes without actively monitoring the outcome.

644

OZURU, BRINER, BEST, MCNAMARA

Although readers seem not to spontaneously engage in resource-consuming reasoning processes during reading, certain activities or reading practices help the reader to exercise active comprehension processes (McNamara, Ozuru, Best, & OReilly, 2007). Readers make causal inferences about unfamiliar material when the task requires them to do so (Singer, Harkness, & Stewart, 1997). Similarly, skilled readers can increase their comprehension when adjunct questions are asked while reading expository texts (van den Broek et al., 2001). Finally, self-explanation activities may contribute to active processing in a number of different ways, depending on specic circumstances. In this article, we are interested in the ways in which self-explanation engages the reader in resourceconsuming processes oriented toward linking textual information across successive sentences.

COHESION AND TEXT COMPREHENSION Text cohesion represents the degree to which a text is relatively self-contained and does not demand the contribution of the readers topic specic knowledge. Text cohesion is one of the important factors associated with text coherence, which is the extent to which readers mental representations of the text content are meaningfully interconnected (Graesser, McNamara, & Louwerse, 2003; Halliday & Hasan, 1976). To understand an extended passage, readers need to keep track of how individual sentences are related to each other to form a coherent mental model of the text content (Oakhill, Cain, & Bryant, 2003). Texts vary in the extent to which they scaffold readers maintenance of coherence. When reading a highly cohesive text, the majority of information necessary to maintain text coherence is provided by the text itself. On the other hand, reading a less cohesive text requires more effort to ll in informational gaps based on readers preexisting knowledge. There are two types of textual cohesiveness that affects different aspects of coherence: local cohesion and global cohesion (McNamara et al., 1996). Local cohesion varies as a function of the degree of linguistic overlap (e.g., word overlap) or conceptual overlap (e.g., argument overlap) between adjacent sentences, or the presence of specic cues (e.g., connectives) that help readers link the ideas distributed between adjacent sentences in a specic way (Kintsch, 1998). Local cohesion helps readers process adjacent sentences as linguistically meaningful unitsan idea that is consistent with memory-based models of text processing, such as the Resonance Model (e.g., Cook, Halleran, & OBrien, 1998). The Resonance Model postulates that the extent to which newly introduced information (i.e., the current focus of processing) is integrated with the rest of the text to maintain coherence as a function of how well the newly introduced information semantically resonates (e.g., mutually co-

CONTRIBUTIONS OF SELF-EXPLANATION

645

activate) with the information that has already been processed (Cook et al., 1998). Hence, according to this view, texts with higher levels of local cohesion result in more resonance between successive sentences, therefore making the maintenance of local coherence easier based on relatively passive memory-based processing. In contrast, global cohesion helps readers link multiple events, elements, and facts, especially information spread across distant parts of the text (Linderholm et al., 2001). Global cohesion often helps readers relate the text content to the broader context outside the text. Global cohesion can be increased by providing background information about the topic presented in the text, especially at the beginning (Beck et al., 1991), aligning the temporal and causal sequence of the events in the text (Linderholm et al., 2001) or providing headers and topic sentences that offer an overview of the paragraph content (McNamara et al., 1996). Hence, increasing local or global cohesion may contribute to comprehension in different, almost contrasting, ways. Local cohesion is likely to render the comprehension process (or sense of comprehension) easier by facilitating automated process in which link between successive sentences is established (e.g., Cook et al., 1998). In contrast, increasing global cohesion is likely to facilitate more global and extensive comprehension by encouraging extra processing in which readers form extended links between the ideas in the text that are not apparent in low-cohesion texts. According to this ner differentiation of cohesion and its role in text comprehension, the type of text that increases the role of self-explanation should be one that is generally globally cohesive, but lacking local cohesiveness. This type of text provides a large amount of interconnected information in an organized fashion that encourages readers to maintain overall coherence (e.g., across the different sections of the text and between the text content and general knowledge) by linking different parts of text. However, the linking of the basic units of the text (i.e., linking between successive sentences) cannot be accomplished by solely relying on automatic memory-based processing in this type of text. Thus, more active resource-demanding processes, such as self-explanation, should play a particularly important role in comprehension of a globally cohesive, yet locally less cohesive, text. Relatively few studies have explored how specic text features inuence the benet of self-explanation on text comprehension. An exception is a study by Ainsworth and Burcham (2007), which examined the inuence of overall text cohesion (referred to as coherence in their article) and self-explanation on science text comprehension. Ainsworth and Burcham asked college students to read either a minimally cohesive or maximally cohesive biology text either with or without self-explanation. In manipulating text cohesion, Ainsworth and Burcham modied many different aspects of cohesion, such as adding explana-

646

OZURU, BRINER, BEST, MCNAMARA

tions to unpack the meaning of a sentence, topic headers, and manipulating text organization. The authors assessed comprehension using various comprehension questions, such as denition questions, implicit questions, and knowledge inference questions. Students who read a maximally cohesive text learned more, as did learners who self-explained. More important, for this purpose, the study showed no clear interactive effects of text cohesion and self-explanation on text comprehension. One obstacle for interpreting the Ainsworth and Burcham (2007) study is that the authors manipulated the overall cohesion of a text, including both local and global cohesion. Given that local and global cohesion are considered to contribute to comprehension processes in different ways (Gilabert, Martnez, & Vidal-Abarca, 2005; Vidal-Abarca, Martnez, & Gilabert, 2000), it is unclear how the two types of cohesion manipulation affected the benet of selfexplanation. A similar limitation is found in a study of think-aloud protocols (Loxterman, Beck, & McKeown, 1994) in which both local and global cohesion were manipulated. Hence, no study, to our knowledge, has examined the benet of self-explanation when a specic aspect of cohesion is manipulatednamely, local cohesion.

THIS STUDY This study explores how the contribution of self-explanation to the comprehension of a science text is affected by the local cohesion of a text. Specically, we manipulate the explicitness of links between successive sentences, which are marked by linguistic and conceptual overlap and connectives. Our analysis explored how self-explanation and its contribution to science text comprehension are affected by local cohesion through the following three analyses. First, following Ainsworth and Burcham (2007), we examined whether reading a low-cohesion (as opposed to high-cohesion) text affects the generation of self-explanationsin particular, bridging inferences (as measured within selfexplanation protocols; e.g., Chi et al., 1989)which help readers overcome local cohesion gaps in text (McNamara, 2004). Two alternative hypotheses are available. On the one hand, a text low in local cohesion may force the reader to generate bridging inferences because, without them, the reader would experience difculty maintaining local coherence. Alternatively, the presence of cohesion gaps may make bridging inferences more difcult to generate instead of making them easier. In addition, a higher local cohesion text may facilitate linking between successive sentences because of the availability of textual scaffolding (e.g., connectives, larger argument overlap). Second, we examined whether levels of local cohesion affect comprehension as measured by post-reading comprehension questions (i.e., local bridging ques-

CONTRIBUTIONS OF SELF-EXPLANATION

647

tions). Theories of text comprehension (e.g., Kintsch, 1988, 1998), as well as empirical evidence (e.g., Britton & Gulgoz, 1991), postulate that comprehension of ideas expressed in adjacent sentences should be better when reading highcohesion texts. However, if the benet of self-explanation is strong enough to overcome the challenges posed by the conceptual gaps in low-cohesion texts, comprehension in the low-cohesion condition may be better than or equal to performance in the high-cohesion condition. This would be particularly true for readers who produce appropriate self-explanations to overcome the lack of cohesive cues. Such a result would produce an interaction between self-explanation quality and cohesion condition (see the third analysis, which follows). The third and central question is whether the relation between performance on the comprehension questions and the quality of self-explanations produced (e.g., bridging inferences) varies as a function of text cohesion. If self-explanation contributes to performance on comprehension questions, there should be signicant positive correlations between performance on comprehension questions and the quality of self-explanations that are produced, as found in prior research (e.g., Ainsworth & Burcham, 2007; McNamara, 2004; McNamara, OReilly, Best, & Ozuru, 2006). More important, however, if the contribution of selfexplanation to comprehension is inuenced by local cohesion, there should be a corresponding change in correlations between self-explanation quality and performance on comprehension questions as a function of text cohesion. Given that the contribution of self-explanation on comprehension is expected to be larger when reading a low-cohesion text, we expect a higher positive correlation between performance on comprehension questions and self-explanation quality for participants in the low-cohesion condition than the high-cohesion condition. There is some evidence that the production of self-explanations is inuenced by level of reading skill (e.g., Magliano & Millis, 2003; Ozuru, Best, & McNamara, 2004). Further, evidence indicates that effect of text cohesion on comprehension interacts with individual differences, such as reading skill and domain knowledge (e.g., McNamara et al., 1996; OReilly & McNamara, 2006). Hence, we included the individual difference measures as covariates in the analyses.

METHOD Participants Seventy-eight undergraduate students from a university in a large Mid-South city participated in the experiment in exchange for credit for an introductory psychology course.

648

OZURU, BRINER, BEST, MCNAMARA

Materials Text. The text was a 437-word expository passage entitled, Why is There Sex? (see Appendix A). The passage discusses the evolutionary benets and costs of sexual reproduction compared to asexual reproduction. The text was obtained from a chapter on evolutionary psychology in the psychology textbook, Learning and Cognition (Leahey & Harris, 1997). For the low-cohesion condition, participants read the text largely in its original form, with relatively minor changes to decrease local cohesion by replacing nouns with pronouns and removing connectives. For the high-cohesion condition, the text was revised to decrease the extent to which readers needed to make inferences to maintain local coherence. Specic revisions involved (a) adding connectives (e.g., however, because, and therefore), (b) replacing pronouns with specic noun phrases, (c) adding noun(s) to increase argument overlap between adjacent sentences, and (d) adding relational pronouns (e.g., which) to make explicit the relations between phrases in a sentence (see Appendix A). Table 1 provides the key features of the two versions of the text as monitored by Coh-Metrix (Graesser, McNamara, Louwerse, & Cai, 2004). As evident in Table 1, these manipulations effectively increased local cohesion as indicated by an increase in argument overlap between adjacent sentences, the latent semantic analysis cosine (Landauer & Dumais, 1996) between adjacent sentences, and the number of logical and causal connectives. The absence of changes in global cohesion, such as the addition of topic sentences at the beginning of paragraphs, and reorganization of the sentence is evident from the actual text (see Appendix A). In addition, very small changes in word frequency in Table 1 support the notion that there is no signicant change in global cohesion caused by the addition of new information. Although imperfect, average word frequency is sensitive to adding background information and elaboration because these

TABLE 1 Text Features of High- and Low-Cohesion Texts Variable Argument overlap adjacent Latent semantic analysis adjacent Causal connectives Logical connectives Word frequency FleschKincaid grade level Number of words per sentence Number of words Low Cohesion 0.21 0.15 24.3 13.5 2.15 8.17 12.76 370 High Cohesion 0.68 0.28 39.3 27.7 2.07 9.83 14.93 433

CONTRIBUTIONS OF SELF-EXPLANATION

649

manipulations often involve addition of new content words with varying word frequency (e.g., Beck et al., 1991). Seven target sentences were selected for self-explanation, based on the following criteria: (a) Surrounding sentences (including the target) were affected by the cohesion manipulation, (b) comprehension of the sentences was critical for understanding the texts overall meaning, and (c) accurate and deep understanding of the sentences required forming some type of connection with a previous sentence. Most of the target sentences were located at the beginning or the ending of a paragraph. Thus, the target sentences either connected the forthcoming paragraph to the previous paragraph or summed up the paragraphs topic. Text-presentation program. The text was presented using a computer program that presented one sentence at a time based on participants clicking the next button displayed on the screen. The program was designed such that seven target sentences, which participants self-explained, were presented in a red-colored font. Participants were instructed to type their self-explanation for each target sentence into the self-explanation box when it was presented on the screen. After completing each self-explanation, participants were instructed to click a continue button, and were not permitted to view or revise completed self-explanations. We are aware that typing self-explanations as opposed to verbalizing them may have an effect on the production of self-explanations (Hausmann & Chi, 2002; cf. Muoz, Magliano, Sheridan, & McNamara, 2006). Thus, the overall quality of self-explanations observed in this study may not be directly comparable to studies in which participants were required to speak aloud to explain text contents. We used typing methods because we are interested in the effect of typing explanation as a potential training method for a classroom setting. Testing booklets. Two booklets were created for the experiment. Booklet 1 contained the NelsonDenny reading skill assessment. Booklet 2 contained comprehension questions about the science text and background knowledge questions. Booklet 2 was divided into three sections. The rst section contained nine open-ended comprehension questions, the second section contained eight open-ended prior knowledge questions, and the third section contained eight multiple-choice prior knowledge questions. The three sections were separated by a sheet of paper that instructed participants that they could not move back to a previous section once they moved into a new section. Comprehension questions. A systematic method was used to construct the questions. We constructed open-ended questions designed to tap three different levels of comprehension based on guidelines for comprehension question

650

OZURU, BRINER, BEST, MCNAMARA

construction in text processing and education literature (e.g., Anderson, 1982; Goldman & Duran, 1988). A question was classied as text-based if its answer could be provided using information explicitly stated within a given sentence. This type of question should be relatively easy to answer because the relations between the question stem and requested information are explicitly stated within particular sentences. A question was classied as a local bridging inference question if its answer required the integration of information located within adjacent sentence pairs. Answering this type of question requires an understanding of the relations between ideas or concepts across adjacent sentences. Finally, far or global bridging inference questions are similar to local bridging inference questions, but involve the integration of information located more than two sentences apart. Thus, global questions are expected to measure readers understanding of relations between multiple ideas located in a relatively global area of the text. Background knowledge questions. The prior knowledge questions examined participants knowledge relevant to evolutionary psychology, the nature of sexual reproduction, and the immune system. One half of the background knowledge questions were in open-ended format, and the other half were in multiple-choice format. Both formats were used to minimize the effect of question format in the measurement of background knowledge, to diversify the task demands of the questions, and to assess various levels of domain knowledge. The text content targeted by the questions differed across and within multiple-choice and open-ended questions.

Procedure Participants were tested in small groups. They were randomly assigned to a lowcohesion or high-cohesion text condition. First, participants were administered the NelsonDenny reading skills assessment (Book 1) with a time limit of 15 min. Participants then read the text (i.e., low- or high-cohesion versions) presented on a computer screen and self-explained the seven pre-selected target sentences at their own pace. Although the text was presented on a sentence-bysentence basis, participants could look at all of the previously read sentences once they moved on to the next sentence. Participants were asked to read each sentence carefully. Participants were instructed to explain the target sentences, in red font. They were asked to explain what the sentences meant to them to better understand the meaning of the overall text. The instructions also contained an example of a good self-explanation, which integrated ideas contained in separate sentences and prior knowledge to interpret a text. The instructions up to this point did not mention that comprehension questions would follow the reading

CONTRIBUTIONS OF SELF-EXPLANATION

651

and self-explanation activity to avoid participants preparing for a follow-up test during the self-explanation phase of the study. Immediately after reading and self-explaining the text, participants were provided with Booklet 2. We presented the prior knowledge questions at the end of the experiment to avoid exposing participants to topic-related questions that could inuence participants reading behavior or prime high-knowledge readers domain knowledge (see McNamara & Kintsch, 1996). In particular, presenting prior knowledge questions prior to a text reading augments prior knowledge effects by priming high-knowledge readers memory while at the same time discouraging low-knowledge readers. One may argue, in turn, that exposure to the text and comprehension test may prime readers prior knowledge (and, thus, alter performance on the prior knowledge test). Although this may occur, the prior knowledge test is not the focus of this study, and simply serves as a proxy for differences in knowledge that may inuence comprehension. In addition, care was taken to ensure that the texts did not contain answers to the prior knowledge questions. Participants responses to the open-ended questions were independently scored by two raters (Yasuhiro Ozuru and Stephen Briner) by comparing individual responses to answer keys created prior to the experiment. Interrater reliability for open-ended question scoring was greater than 95%. Coding of self-explanation protocols. The quality of self-explanations was analyzed in terms of various dimensions, such as whether the self-explanations contained paraphrases and text- or knowledge-based inferences. Note that this coding process did not involve breaking up explanations into specic units, such as idea units, propositions, or clauses. Instead, the two coders rst identied whether self-explanations contained the key ideas of the target sentences (paraphrase) in the source text. These key ideas are roughly equivalent to idea units of each sentence. We created a list of key ideas for each sentence as a guide for determining whether an explanation contained key information pertaining to the target sentences. We used this method as opposed to breaking up selfexplanations into individual idea units because idea units are often complex, containing sub or sub-sub idea units, making it impractical to break them up in a consistent way. Paraphrases, once identied, were further coded as accurate or inaccurate. Accuracy concerned whether the paraphrase preserved the original meaning of the target sentencesthat is, even when a self-explanation contained phrases originating from a target sentence, the explanation can be coded as inaccurate when the emerging idea involves distortion of an idea conveyed by that sentence. When self-explanations contained ideas that could not be traced to the target sentences (often in addition to the paraphrase), they were coded as inferences. Each meaningful chunk of information that could not be traced to the target sen-

652

OZURU, BRINER, BEST, MCNAMARA

tences was coded as an individual inference. When a self-explanation contained an inference, the source of the inference was identied as one of the following: (a) near bridging in which an inference was based on the preceding sentence; (b) far bridging in which the inference was based on a previous section of the text that was more distant than the previous sentence; (c) general knowledge in which the inference was based on common knowledge, personal experience, or semantic association; and (d) domain knowledge in which the inference was based on knowledge specic to evolution, evolutionary psychology, and other biological knowledge related to evolution, immune system, and reproduction. Hence, a self-explanation could contain any combination of different types of inferences. This approachin particular, differentiating bridging inferences from paraphrasesdiffers from the Ainsworth and Burcham (2007) approach in which reiteration of textual information was classied as a paraphrase whether it was based on a single sentence or multiple sentences. We chose to differentiate paraphrases and bridging inferences because linking ideas located across the sentences often requires greater active processing than merely rephrasing information (McNamara, 2004; Wolfe & Goldman, 2005). When self-explanations contained any type of inference, each inference was further classied in terms of whether it was relevantthat is, whether the inference contributed to the deeper and more coherent comprehension of the target sentence. Finally, when an inference was judged to be relevant, it was classied in terms of accuracythat is, whether the inference was scientically accurate. An example of a coded self-explanation is presented in Appendix B. Two coders initially coded 10 self-explanations together for the purpose of training specic to the text to discuss and resolve discrepancies in coding. Informal counting of the agreement at the training stage conrmed the reliability of the coding. After the training, the coders independently coded approximately 5% of the self-explanations, which were randomly selected. We selected 5% of the items for the reliability analysis because the coders were experienced users of the coding scheme, which had been used to analyze self-explanations pertaining to many different texts. Levels of agreement reported later resembled reliability scores obtained in previous self-explanation analyses in which the same coding scheme was used for different texts (McNamara et al., 2006). Coders agreement with respect to whether self-explanations contained paraphrases, inferences (all types), and the accuracy of paraphrases reached 85% or above. However, agreement concerning the relevance and accuracy of inferencesin particular, general knowledge-based inferenceremained somewhat lower: around 70%. Disagreements were resolved via discussion. After the reliability analysis, the complete self-explanation data were divided in half. Yasuhiro Ozuru and Stephen Briner each independently coded one half of the self-explanations. Coders consulted with each other whenever ambiguities or difculties were encountered in the coding process. The analyses reported in the results are based on the nal coding.

CONTRIBUTIONS OF SELF-EXPLANATION

653

RESULTS The results are reported in terms of the following four aspects of the data analysis. The rst section reports the results of preliminary analyses concerning the individual difference measure scores (reading skill and prior knowledge assessments). The second section reports the results regarding the quality of the self-explanations. The third section reports performance on the comprehension questions. The nal section reports the results regarding the relations between self-explanations and performance on the comprehension questions. Preliminary Analysis: Individual Differences Table 2 presents performance on the individual difference measures as a function of condition. Participants in the high-cohesion condition correctly answered a signicantly higher proportion of multiple-choice prior knowledge questions than participants in the low-cohesion condition. Participants in the high-cohesion group also answered a signicantly higher proportion of the overall prior knowledge questions than the participants in the low-cohesion condition. On the other hand, participants performance on the reading skill assessment did not signicantly differ between the high-cohesion and low-cohesion conditions. Although it is possible that the performance difference on the prior knowledge measure may have resulted from reading the high- or low-cohesion texts that preceded the prior knowledge assessment, we believe the performance difference observed in this study was not caused by the cohesion manipulation of the text; rather, we believe this occurred by chance. Our reasoning is based on studies that used the same procedure (McNamara, 2001; OReilly & McNamara, 2007). No performance differences on prior knowledge measures were reported in these
TABLE 2 Participants Performance on Prior Knowledge and Reading Skill Measures as a Function of Text Cohesion Condition Low Cohesiona Variable Multiple-choice prior knowledge Open-ended prior knowledge Overall prior knowledge Reading skill M 0.47 0.33 0.40 0.53 SD 0.21 0.18 0.16 0.16 High Cohesionb M 0.62 0.40 0.51 0.55 SD 0.21 0.21 0.19 0.19 Statistical Analysis F(1, 77) 9.19 2.90 7.70 0.10 p <.01 .10 <.01 ns

Note. Overall prior knowledge is the average proportional score of the multiple-choice and open-ended questions. a n D 38. b n D 40.

654

OZURU, BRINER, BEST, MCNAMARA

studies in which a prior knowledge assessment was administered at the end of the experiment in which effects of text cohesion were examined using a between-subject design. Hence, a signicant difference exists between high-cohesion and low-cohesion participants level of prior knowledge, although their reading skill appears not to differ. To take into consideration the difference in the level of prior knowledge between the two experimental conditions (high cohesion and low cohesion), we analyzed the data in three different ways. First, we analyzed overall participant data rst as a function of condition (high cohesion vs. low cohesion) using analysis of variance (ANOVA), and then we included participants prior knowledge and reading skill scores as covariates (i.e., analysis of covariance [ANCOVA]) to partial-out the variance associated with individual differences. Third, we repeated the same analyses using data from 48 participants (24 in each condition) after matching participants based on their level of prior knowledge and reading ability. We matched the participants such that variation in the scores for each matched pair did not differ greater than .05 for each proportion of correct scores for the reading skill test and the overall prior knowledge test. The description of the results focuses on the analysis based on the complete sample (ANOVAs and ANCOVAs based on all participants). However, we also report whether the results differed when analyzed with the matched participants. Quality of Self-Explanations Table 3 presents the characteristics of participants self-explanations as a function of cohesion condition. We calculated the frequency of paraphrases, nearbridging, far-bridging, general knowledge-based, and domain knowledge-based inferences that were accurate. We report only the frequency of different types of relevant and accurate self-explanations because we are not interested in occurrence of inaccurate or irrelevant self-explanations (e.g., inaccurate paraphrases, inaccurate inferences) in this study (cf. McNamara, 2004). Six separate one-way ANOVAs and ANCOVAs, using prior knowledge and reading skill as covariates, analyzed the effect of text cohesion on different aspects of self-explanation. The analyses showed that the participants in the highcohesion group produced signicantly more accurate near-bridging inferences (M D 1.60, SD D 0.98) than participants in the low-cohesion group (M D 0.90, SD D 0.87) based on both ANOVAs and ANCOVAs. Participants in the low-cohesion condition tended to produce more paraphrases compared to the high-cohesion condition according to ANOVAs, but the effect disappeared when the covariates were included in the analysis. All other differences between the high- and low-cohesion groups were not signicant (see Table 3). When the participants were matched against each other, this trend remained, with

CONTRIBUTIONS OF SELF-EXPLANATION

655

TABLE 3 Occurrence of Different Types of Explanations Contained in Self-Explanation Protocols as a Function of Text Cohesion Condition Low Cohesion Variable Number of accurate paraphrases Number of accurate near-bridging inferences Number of accurate far-bridging inferences Number of accurate general knowledge-based inferences Number of accurate domain knowledge-based inferences Overall quality: Total number of different types of selfexplanation that occurred M 3.33 0.95 1.33 1.79 0.48 7.76 SD 1.50 0.92 1.48 1.15 0.83 3.64 High Cohesion M 2.50 1.66 1.34 1.83 0.26 7.45 SD 1.44 1.04 1.13 1.53 0.45 3.41

ANOVA F(1, 77) 5.94 10.46 0.01 0.01 2.08 0.01 p .02 .01 ns ns .15 ns

ANCOVA F(1, 75) 2.36 11.13 0.08 0.00 2.55 0.01 p .12 .01 ns ns .12 ns

Note. ANOVA D analysis of variance; ANCOVA D analysis of covariance.

participants in the high-cohesion condition producing signicantly more nearbridging inferences than those in the low-cohesion group, F(1, 43) D 8.37, p D .006. This nding suggests that placing readers in a situation where producing explanations to articulate links between successive sentences to form a coherent understanding (low-cohesion condition) is not, in itself, enough to facilitate the generation of self-explanations linking adjacent sentences. Instead, readers appear to produce more explanations linking adjacent sentence meaning when the text cohesion scaffolds these inferences. Performance on the Comprehension Questions Table 4 presents performance on comprehension questions as a function of condition. We examined whether text cohesion affected comprehension as measured with comprehension questions of varying difculty level (text-based, local bridging, global bridging). A 2 (Cohesion) 3 (Question Difculty) ANOVA revealed a main effect of the level of question difculty, F(2, 154) D 114.80, p < .001 (MSE D 0.04). As one would expect, participants performed better on text-based questions (M D 0.65, SD D 0.30) than local bridging questions (M D 0.48, SD D 0.30), and they also performed better on local bridging questions

656

OZURU, BRINER, BEST, MCNAMARA

TABLE 4 Performance on the Comprehension Questions as a Function of Text Cohesion and Question Difficulty Low Cohesiona Variable Text-based Local Global M 0.71 0.49 0.18 SD 0.31 0.27 0.21 High Cohesionb M 0.58 0.46 0.18 SD 0.27 0.31 0.21

ANOVA F(1, 77) 4.74 0.33 0.10 p <.05 ns ns

ANCOVA F(1, 75) 4.14 0.47 0.37 p <.05 ns ns

Note. ANOVA D analysis of variance; ANCOVA D analysis of covariance. a n D 38. b n D 40.

than global questions (M D 0.18, SD D 0.21; see also Table 4). These effects remained when prior knowledge and reading skill were included as covariates. The main effect of cohesion was not statistically signicant according to both ANOVAs, F(1, 77) D 1.64, p D .5; and ANCOVAs, F(1, 75) D 2.49, p D .12. However, there was a trend indicating that participants performance on the text-based questions was better in the low-cohesion compared to highcohesion condition, as indicated in Table 4. This nding (i.e., the effect specic to text-based questions) has been previously observed with biology texts (e.g., McNamara, 2001; OReilly & McNamara, 2007). Thus, we separately examined the effect of text cohesion for each question type. These analyses indicated that the effect of cohesion was signicant for the performance on text-based question according to both ANOVAs and ANCOVAs (see Table 4). The analysis based on the matched participants showed the same results, indicating that participants performance on text-based questions was better in the low-cohesion condition than the high-cohesion condition, F(1, 44) D 4.02, p D .05. Comprehension Question Performance, Quality of Self-Explanation, and Individual Differences This section reports analyses that explore the possibility that the benet of selfexplanation is larger for a text with low local cohesion. Table 5 presents Pearson correlations between the quality of self-explanation and performance on the comprehension questions as a function of text cohesion and type of comprehension question (text-based, local, and global). Overall, this analysis indicates that performance on comprehension questions was positively correlated with self-explanation quality, and the correlations tended to be higher in the low-cohesion condition (text-based: .629 vs. .332; local: .498 vs. .375),

CONTRIBUTIONS OF SELF-EXPLANATION

657

TABLE 5 Correlations Between Performance on Comprehension Questions and Quality of Self-Explanations as a Function of Question Difficulty and Text Cohesion Question Type and Cohesion Level Text-Based Self-Explanation Paraphrase Near bridging Far bridging General knowledge inference Domain knowledge inference Overall quality of self-explanations *p < .05. **p < .01. Low .447** .477** .458** .304 .243 .629** High .073 .129 .252 .353* .002 .332* Low .480** .224 .399* .161 .281 .498** Local High .035 .284 .226 .403** .110 .375* Global Low .184 .186 .096 .387* .254 .343* High .473** .354* .282 .171 .008 .478**

except for global questions (.343 vs. .478). The difference comparing the correlations between self-explanation quality and performance on comprehension questions is statistically signicant for text-based questions with a one-tailed test, Z D 1.67, p < .05; and marginally signicant with a two-tailed test, p < .1. The difference is not signicant for local and global questions. Overall, it seems that the benet of self-explanation is greater for the low-cohesion text than it is for the high-cohesion text, and this is particularly evident on the text-based questions. We further analyzed the contribution of individual differences, text cohesion, quality of self-explanations, and the interaction between cohesion and self-explanation quality on performance on the comprehension questions using hierarchical regression. Specically, we examined two regression models for performance on each of the three types of comprehension questions. In the rst model, we included a range of predictor variables, such that reading ability and prior knowledge were entered in Block 1, overall quality of self-explanations in Block 2, text cohesion in Block 3, and the interaction between cohesion and self-explanation quality in Block 4. In the second model, the rst two predictor variables (i.e., reading skill and prior knowledge; text cohesion) were the same. However, the predictor variables were altered in Blocks 3 and 4, such that the frequency of near-bridging inferences was used in Block 3, and the interaction between the frequency of near-bridging inferences and text cohesion was used in Block 4. In these analyses, we included an interaction between cohesion and quality of self-explanation (e.g., overall quality of self-explanation or frequency of near-bridging inferences) because the correlational analyses indicated that the contribution of self-explanation to comprehension may differ across the two cohesion conditions.

658

OZURU, BRINER, BEST, MCNAMARA

A predictor variable representing the interaction between cohesion and selfexplanation quality was created by multiplying z scores for self-explanation quality (i.e., the overall quality or frequency of near-bridging inferences) and the contrast code to represent cohesion condition ( 1 for low cohesion and C1 for high cohesion). We included performance on individual difference measures in the rst block so that we were able to analyze the contribution of text cohesion, self-explanation quality, and the interaction between cohesion and selfexplanation quality on comprehension question performance above and beyond effects of individual differences by examining the signicance of R2 associated with each block of predictor variables. Table 6 (Model 1) presents the results of these analyses. With respect to text-based questions, the regression analysis indicated that the main effects of cohesion and self-explanation quality were statistically signicant above and beyond the effects of prior knowledge and reading skill. The results are consistent with the earlier ANCOVAs and correlational analyses. In addition, the analysis indicated that the interaction between cohesion and overall selfexplanation quality is marginally signicant. In addition to Model 1, we also performed a regression (Model 2) using the average frequency of near bridging in the self-explanation protocol. Model 2 showed signicant interaction between cohesion and the frequency of near-bridging inferences, further bolstering the claim that the contribution of self-explanations (in particular, near-bridging inferences) to comprehension differs across low-cohesion and high-cohesion conditions. With respect to local and global bridging questions, only self-explanation quality was a signicant predictor variable. The regression analysis conrmed not only that performance on comprehension questions was related to the overall quality of self-explanations, but also that the contribution of self-explanation to performance on (at least) text-based comprehension questions differed between low-cohesion and high-cohesion conditions. To follow up on the interactive effect of text cohesion and quality of self-explanation on performance on the text-based questions, we divided participants into good and poor self-explainers based on the overall quality of self-explanation using median split, with prior knowledge and reading skill as covariates. We then performed a 2 (Self-Explanation Quality) 2 (Cohesion) ANCOVA on performance on text-based questions, using prior knowledge and reading skill as covariates. Table 7 presents the result of this analysis. The results of ANCOVAs revealed a signicant main effect of cohesion, F(1, 72) D 5.20, p < .05 (MSE D .071); self-explanation quality, F(1, 72) D 15.06, p < .001; and a marginal interaction effect between cohesion and selfexplanation quality, F(1, 72) D 3.02, p D .09. The results were similar with ANOVAs. These results essentially replicate the results of the regression analysis. As can be seen in Table 7, the interaction between cohesion and self-

CONTRIBUTIONS OF SELF-EXPLANATION

659

TABLE 6 Hierarchical Linear Regression on Comprehension Question Performance Variable R2 R 2 F p

Dependent variable: Performance on text-based questions (Model 1) Block Block Block Block 1 2 3 4 (prior knowledge/reading skill) (cohesion) (overall quality of self-explanation) (interaction between cohesion and self-explanation quality) .053/.141 .231* .466* .173 .028 .079 .301 .330 .028 .051 .222 .029 1.080 4.140 23.160 3.110 .35 .05 .00 .08

Dependent variable: Performance on text-based questions (Model 2) Block Block Block Block 1 2 3 4 (prior knowledge/reading skill) (cohesion) (accurate near-bridging inference) (interaction between cohesion and accurate near-bridging inference) .029/.177 .079 .352* .366* .028 .079 .165 .211 .028 .051 .086 .046 1.080 4.140 7.540 4.170 .35 .05 .01 .05

Dependent variable: Performance on local questions (Model 1) Block Block Block Block 1 2 3 4 (prior knowledge/reading skill) (cohesion) (overall quality of self-explanation) (interaction between cohesion and self-explanation quality) .164/.063 .097 .421* .038 .026 .033 .224 .224 .026 .007 .191 .001 1.000 0.052 17.940 0.080 .37 .47 .00 .78

Dependent variable: Performance on local questions (Model 2) Block Block Block Block 1 2 3 4 (prior knowledge/reading skill) (cohesion) (overall quality of self-explanation) (interaction between cohesion and accurate near bridging inference) .121/.114 .219 .287* .029 .026 .033 .106 .106 .026 .007 .073 .000 1.000 0.520 5.960 0.020 .37 .47 .02 .88

Dependent variable: Performance on global questions (Model 1) Block Block Block Block 1 2 3 4 (prior knowledge/reading skill) (cohesion) (overall quality of self-explanation) (interaction between cohesion and self-explanation quality) .233*/.098 .068 .426* .072 .066 .071 .245 .250 .066 .005 .174 .005 2.650 0.370 16.820 0.480 .08 .54 .00 .49

Dependent variable: Performance on global questions (Model 2) Block Block Block Block 1 2 3 4 (prior knowledge/reading skill) (cohesion) (overall quality of self-explanation) (interaction between cohesion and accurate near bridging inference) .186/.161 .260 .309* .088 .066 .071 .158 .161 .066 .005 .087 .003 2.650 0.370 7.600 0.230 .08 .54 .00 .64

*p < .05.

660

OZURU, BRINER, BEST, MCNAMARA

TABLE 7 Performance on Text-Based Questions as Function of Text Cohesion and Self-Explanation Quality (Median Split Based on Overall Self-Explanation Quality) Low Cohesion Variable Poor self-explainer Good self-explainer M 0.54 0.88 SD 0.33 0.16 M 0.52 0.65 High Cohesion SD 0.29 0.25

explanation quality indicates that the effect of self-explanation quality is larger in the low-cohesion condition (.54 vs. .88) than in the high-cohesion condition (.52 vs. .65), suggesting that the benet of self-explanation is larger when the text contains a greater number of conceptual gaps.

DISCUSSION The goal of this study was to examine whether the local cohesion (high or low) of science texts inuences the ways in which readers self-explain and subsequently comprehend these texts. Specically, we hypothesized that the contribution of self-explanation to the comprehension of science text would be larger when reading a lower-cohesion text. The hypothesis is based on the notion that selfexplaining is more critical in understanding a text that has lower local cohesion than one with greater local cohesion because the low-cohesion text lacks explicit cues linking successive ideas to facilitate automatic processes, such as memorybased processing (e.g., Cook et al., 1998). The results consistently indicated that performance on comprehension questions (text-based, local bridging, and global bridging) was signicantly inuenced by the overall quality of readers self-explanationsndings consistent with past research on the contribution of self-explanation to text comprehension (Ainswroth & Burcham, 2007; Chi et al., 1994; McNamara, 2004; McNamara et al., 2006). The inuence of self-explanation quality on comprehension question performance cannot be explained by individual differences (i.e., prior knowledge and reading ability) because the contribution of self-explanation to comprehension question performance was signicant above and beyond the contribution of prior knowledge and reading ability, as indicated by ANCOVAs and hierarchical regressions. The interaction between text cohesion and the quality of self-explanation (and the frequency of near-bridging inferences) on text-based question performance was also observed. This nding supports our main hypothesis that the contri-

CONTRIBUTIONS OF SELF-EXPLANATION

661

bution of self-explanations to comprehension is larger when reading a text with low local cohesion. Specically, the positive contribution of self-explanation on text-based question answering performance was larger for participants in the low-cohesion condition than those in the high-cohesion condition. The nding concerning the interaction contrasts with Ainsworth and Burchams (2007) discovery, whereby no interaction was found between self-explanation and text cohesion in inuencing text comprehension. Although there may be a number of reasons underlying this difference, we suggest that the difference in text manipulation serves as the critical factor. When the text is generally globally cohesive, the contribution of self-explanation to comprehension and learning tends to be larger when the text contains local conceptual gaps (i.e., low local cohesion) because maintaining coherence at this level requires engaging in follow-up processesa process typically facilitated by self-explanation. On the other hand, when other text features, such as global cohesion, are manipulated across the text, increased global cohesion may facilitate more active processing, overshadowing the effect of self-explanation. Nonetheless, this account is speculative at this point. Future studies should attempt to separate the effects of specic text features on the effects of self-explanation on comprehension. In particular, rigorously controlled manipulations of local and global text cohesion and their inuence on different aspects of processing are needed. Although the ndings reported earlier are relatively straightforward and consistent with our predictions, two aspects of the ndings warrant further attention and explanation. First, the nding that the larger number of near-bridging inferences produced by students in the high-cohesion condition suggests that merely increasing task demands in the form of local conceptual gaps between adjacent sentences is not sufcient to facilitate the production of higher quality selfexplanations. Instead, readers appear to make connections between the meaning of successive sentences when it is easier to do so (e.g., when local cohesion is high); or, one may even argue that linkages were already provided to the readers by the text. One theoretical question arising out of this nding concerns the conceptual denition of self-explanation. Our operationalization of self-explanation was based on the assumption that near-bridging inferences observed in the selfexplanation protocols for both the low-cohesion and high-cohesion texts involved linking multiple sentences. However, this positivist denition of self-explanation, in retrospect, may not be appropriate in the sense that it fails to take into consideration the effort undergone to produce the explanations. In other words, we are inclined to believe that a substantial portion of the near-bridging inferences produced by students in the high-cohesion condition were mere verbalizations of the ongoing thought processes directly facilitated by text-based cues that help readers recognize the connection of adjacent sentences in a rather effortless manner (i.e., memory-based processing). In this sense, near bridging in the high-

662

OZURU, BRINER, BEST, MCNAMARA

cohesion condition perhaps reects comprehension that had already occurred, whereas near-bridging inferences in the low-cohesion condition may more likely to represent readers ongoing effort to relate the meanings of adjacent sentences, or acts of explaining, as proposed by Chi et al. (1989). Our data seem to at least partially support the previous speculation. The results of the hierarchical regression on performance on text-based questions indicated signicant interactive effects between cohesion and near-bridging inferences for performance on the text-based questions. This indirectly suggests that performance on text-based questions in the low-cohesion condition was strongly determined by participants producing near-bridging inferences, whereas performance on the same text-based questions in the high-cohesion condition was not as strongly related to the production of near-bridging inferences. Our discovery suggests that near-bridging inferences produced in the two conditions (i.e., low cohesion and high cohesion) may differ in terms of the psychological processes involved. In addition, related to this point about the distinction between self-explanation for comprehension and think-aloud reecting comprehension, we should emphasize that participants in this study were not trained to self-explain difcult texts; they were given the instructions and an example representing a good selfexplanation. The aspect of this experiment leads us to speculate that the ndings may be different (i.e., better self-explanation quality with more frequent bridging for the low-cohesion condition) if students had been trained extensively. This expectation is based on ndings that increased inference generation for lower cohesion text when participants were trained (McNamara, 2004). The second issue deserving attention concerns the interaction between cohesion and self-explanation quality being specic to text-based questions. It is difcult to provide a clear explanation based on the available data. However, similar results have been reported by McNamara (2001), OReilly and McNamara (2007), and Ozuru, Dempsey, and McNamara (2009). They attributed the reason to the difculty level of the text relative to the participants knowledge level. When the text content is difcult, and comprehension tests are memory-based (i.e., when texts are unavailable for review), answering bridging questions often requires remembering unfamiliar information and reasoning with the unfamiliar information. This may render answering bridging questions too difcult for the majority of participants, washing out the subtle differences in performance (oor effects) on bridging questions produced by the experimental manipulation (i.e., cohesion). On the other hand, answering text-based questions only requires remembering unfamiliar content that was explicitly stated within a sentence. This difference in task demands between two types of questions may have contributed to selective sensitivity of the text-based questions to cohesion manipulation. Nonetheless, this account is speculative at best. Future work needs to address this recurrent tendency.

CONTRIBUTIONS OF SELF-EXPLANATION

663

In conclusion, this study provides important information about the contribution of self-explanation to comprehension. First, the ndings conrm and establish the assumption that self-explanation can make a positive contribution to science text comprehension. Further, the study helps us gain insight into a possible reason why self-explanation contributes to comprehension processes. According to the results, one of the ways in which self-explanation contributes to science text comprehension is by facilitating follow-up processing in which readers maintain textual coherence between sentences through resource-demanding inference processes. When such processes are necessary or benecial to building a coherent representation of the text, self-explanation is more likely to lead to better comprehension. When an understanding that meets the readers standard of coherence can be obtained without such processes, self-explanation is unlikely to improve comprehension.

ACKNOWLEDGMENTS This research was supported by funding provided by the Institute of Education Sciences (IES R305G020018-02 and IES R305G040046) to Danielle S. McNamara. Ideas expressed in this material are those of the authors and do not necessarily reect the views of the IES.

REFERENCES
Adams, J., & Collins, A. (1979). A schema-theoretic view of reading. In R. O. Freedle (Ed.), New directions in discourse processing, Vol. 2: Advances in discourse processes (pp. 122). Norwood, NJ: Ablex. Ainsworth, S., & Burcham, S. (2007). The impact of text coherence on learning by self-explanation. Learning and Instruction, 17, 286303. Anderson, R. C. (1982). How to construct achievement tests to assess comprehension. Review of Educational Research, 42, 145170. Anderson, R. C. (1994). Role of the readers schema in comprehension, learning, and memory. In R. Rudell, M. Rudell, & H. Singer (Eds.), Theoretical Models and Processes of Reading (pp. 448468). Newark, DE: IRA. Beck, I. L., McKeown, M. G., Sinatra, G. M., & Loxterman, J. A. (1991). Revising social studies text from a text-processing perspective: Evidence of improved comprehensibility. Reading Research Quarterly, 26, 251276. Britton, B. K., & Gulgoz, S. (1991). Using Kintschs computational model to improve instructional text: Effects of repairing inference calls on recall and cognitive structures. Journal of Educational Psychology, 83, 329345. Chi, M. T. H., Bassok, M., Lewis, M. W., Reimann, P., & Glaser, R. (1989). Self-explanations: How students study and use examples in learning to solve problems. Cognitive Science, 13, 145182. Chi, M. T. H., de Leeuw, N., Chiu, M., & LaVancher, C. (1994). Eliciting self-explanations improves understanding. Cognitive Science, 18, 439477.

664

OZURU, BRINER, BEST, MCNAMARA

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. Gilabert, R., Martnez, G., & Vidal-Abarca, E. (2005). Some good texts are always better: Text revision to foster inferences of readers with high and low prior background knowledge. Learning and Instruction, 15, 4568. Glenberg, A. M., & Epstein, W. (1987). Inexpert calibration of comprehension. Memory & Cognition, 15, 8493. Goldman, S. R., & Duran, R. P. (1988). Answering questions from oceanography texts: Learner, task and text characteristics. Discourse Processes, 11, 373412. Graesser, A. C., McNamara, D. S., & Louwerse, M. M. (2003). What do readers need to learn in order to process coherence relations in narrative and expository text? In A. P. Sweet & C. E. Snow (Eds.), Rethinking reading comprehension (pp. 8298). New York: Guilford. Graesser, A. C., McNamara, D. S., Louwerse, M., & Cai, Z. (2004). Coh-Metrix: Analysis of text on cohesion and language. Behavioral Research Methods, Instruments and Computers, 36, 193202. Graesser, A. C., Singer, M., & Trabasso, T. (1994). Constructing inferences during narrative text comprehension. Psychological Review, 101, 371395. Halliday, M. A. K., & Hasan, R. (1976). Cohesion in English. London: Longman. Hausmann, R. G. M., & Chi, M. T. H. (2002). Can a computer interface support self-explaining? Cognitive Technology, 7, 414. Haviland, S. E., & Clark, H. H. (1974). Whats new? Acquiring new information as a process in comprehension. Journal of Verbal Learning and Verbal Behavior, 13, 512521. Johnson-Laird, P. N., & Garnham, A. (1980). Descriptions and discourse models. Linguistics and Philosophy, 3, 371393. Kintsch, W. (1988). The use of knowledge in discourse processing: A construction integration model. Psychological Review, 95, 163182. Kintsch, W. (1993). Information accretion and reduction in text processing: Inferences. Discourse Processes, 16, 193202. Kintsch, W. (1998). Comprehension: A paradigm for cognition. Cambridge, MA: Cambridge University Press. Kintsch, W. (2009). Learning and constructivism. In S. Tobias & T. M. Duffy (Eds.), Constructivist theory applied to instruction: Success or failure? (pp. 223241). Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Landauer, T. K., & Dumais, S. T. (1996). How come you know so much? From practical problem to theory. In D. Hermann, C. McEvoy, M. Johnson, & P. Hertel (Eds.), Basic and applied memory: Memory in context (pp. 105126). Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Leahey, T. H., & Harris, R. J. (1997). Learning and cognition. Upper Saddle River, NJ: Prentice Hall. Linderholm, T., Everson, M. G., van den Broek, P., Mischinski, M., Crittenden, A., & Samuels, J. (2001). Effects of causal text revision on more or lessskilled readers comprehension of easy and difcult texts. Cognition & Instruction, 18, 525556. Long, D. L., & Lea, R. B. (2005). Have we been searching for meaning in all the wrong places? De-ning the search after meaning principle in comprehension. Discourse Processes, 39, 279 298. Loxterman, J. A., Beck, I. L., & McKeown, M. G. (1994). The effects of thinking aloud during reading on students comprehension of more or less coherent text. Reading Research Quarterly, 29, 353367. Magliano, J. P., & Millis, K. K. (2003). Assessing reading skill with a think-aloud procedure. Cognition & Instruction, 21, 251283. McNamara, D. S. (2001). Reading both high coherence and low coherence texts: Effects of text sequence and prior knowledge. Canadian Journal of Experimental Psychology, 55, 5162.

CONTRIBUTIONS OF SELF-EXPLANATION

665

McNamara, D. S. (2004). SERT: Self-explanation reading training. Discourse Processes, 38, 130. McNamara, D. S., & Kintsch, W. (1996). Learning from text: Effects of prior knowledge and text coherence. Discourse Processes, 22, 247287. McNamara, D. S., Kintsch, E., Songer, N., & Kintsch, W. (1996). Are good texts always better? Interactions of text coherence, background knowledge, and levels of understanding in learning from text. Cognition & Instruction, 14, 143. McNamara, D. S., & Magliano, J. (2009). Towards a comprehensive model of comprehension. Psychology of Learning and Motivation, 51, 297384. McNamara, D. S., OReilly, T., Best, R., & Ozuru, Y. (2006). Improving adolescent students reading comprehension with iSTART. Journal of Educational Computing Research, 34, 147171. McNamara, D. S., Ozuru, Y., Best, R., & OReilly, T. (2007). The 4-pronged comprehension strategy framework. In D. S. McNamara (Ed.), Reading comprehension strategies: Theories, interventions, and technologies (pp. 465496). Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Muoz, B., Magliano, J. P., Sheridan, R., & McNamara, D. S. (2006). Typing versus thinking aloud when reading: Implications for computer-based assessment and training tools. Behavior Research Methods, Instruments, & Computers, 38, 211217. Noordman, L. G., Vonk, W., & Kempff, H. J. (1992). Causal inferences during the reading of expository texts. Journal of Memory and Language, 31, 573590. Oakhill, J. V., Cain, K., & Bryant, P. E. (2003). Dissociation of single-word reading and text comprehension skills. Language and Cognitive Processes, 18, 443468. OReilly, T., & McNamara, D. S. (2007). Reversing the reverse cohesion effect: Good texts can be better for strategic, high-knowledge readers. Discourse Processes, 43, 121152. Ozuru, Y., Best, R., & McNamara, D. S. (2004). Contribution of reading skill to learning from expository texts. In K. Forbus, D. Gentner, & T. Regier (Eds.), Proceedings of the 26th annual meeting of the Cognitive Science Society (pp. 10711076). Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Ozuru, Y., Dempsey, K., & McNamara, D. S. (2009). Prior knowledge, reading skill, and text cohesion in the comprehension of science texts. Learning & Instruction, 19, 228242. Rumelhart, D. (1985). Toward an interactive model of reading. In H. Singer & R. B. Ruddell (Eds.), Theoretical models and processes of reading (3rd ed., pp. 722750). Newark, DE: International Reading Association. Singer, M., Harkness, D., & Stewart, S. T. (1997). Constructing inferences in expository text comprehension. Discourse Processes, 24, 199228. Singer, M., & Remillard, G. (2004). Controlled and automatic inuences in text retrieval. Memory & Cognition, 32, 12231237. Van den Broek, P., Tzeng, Y., Risden, K., Trabasso, T., & Basche, P. (2001). The effects of questioning during and after reading on comprehension at different grades. Journal of Educational Psychology, 93, 521529. Vidal-Abarca, E., Martnez, G., & Gilabert, R. (2000). Two procedures to improve instructional text: Effects on memory and learning. Journal of Educational Psychology, 92, 107116. Voss, J., & Siles, L. (1996). Learning from history text: The interaction of knowledge and comprehension skill with text structure. Cognition & Instruction, 14, 4568. Walker, C. H. (1987). Relative importance of domain knowledge and overall aptitude on acquisition of domain related information. Cognition & Instruction, 4, 2542. Wolfe, M. B., & Goldman, S. R. (2005). Relations between adolescents text processing and reading. Cognition & Instruction, 23, 467502.

666

OZURU, BRINER, BEST, MCNAMARA

APPENDIX A Text Used for Self-Explanation and Reading Comprehension Tasks1 Why Is There Sex? We tend to take the existence of sexual reproduction for granted. However, from an evolutionary standpoint sexual reproduction is a serious puzzle. The simplest and most direct form of reproduction is ssion, which is to reproduce ones genes by creating a copy of oneself. In contrast, sexual reproduction incurs many costs. For example, one has to nd a mate, and has to risk mingling ones genes with another individuals, whose tness is unknown. Consequently, animals have to go to great lengths to nd mates, choose the best mates, and try to guarantee that the offspring are their own. Therefore, sexual reproduction is a costly and risky business. (The cost raises the question as to why sexual reproduction exists, and why it exists in so many species.). This question has been addressed by many theories over the years. The central feature of sexual reproduction that the theories focus on is the fact that sexual reproduction introduces variation into nature. Sex reshufes genes into new combinations. (Therefore, sexual reproduction makes offspring quite different from either parent.) The currently favored account of the evolutionary value of reshufing is the Red Queen theory, which is named after the character in Alice in Wonderland. (According to this theory, the gene reshufing keeps us one step ahead of our most dangerous enemies, including viruses, bacteria, and parasites.) (To better understand how gene reshufing keeps away enemies, imagine owning a home threatened by clever burglars.) You put locks on doors, but the burglars gure out how the locks work and get in. Our immune systems are the locks that our bodies use to repel micro-organisms and parasites. Microorganisms and parasites are the burglars who gure out how to pick the locks of our immune system. Now, if we reproduced by ssion, our offspring would be exact copies of ourselves. (Therefore, the offspring would possess the same locks of the immune system.) Soon, viruses, bacteria, and parasites would evolve into perfect adaptation with our immune system. Thus, we would be defenseless. (However, shufing our genes with sex results in new immune systems.) Therefore, sexual reproduction protects our bodies against the organisms who had adapted themselves to our last generation. These organisms will catch up to us, and we have to keep running away to stay just ahead of them. Hence the term the Red Queen theory.
1 Note. Bolded text D connectives and other linguistic cues added in the high-cohesion text; underlined text D pronoun and other anaphoric expression replaced with noun phrase in the highcohesion text; text in parentheses D target sentences that participants self-explained

CONTRIBUTIONS OF SELF-EXPLANATION

667

The Red Queen turns up in many areas of evolution. Any time organisms interact competitively, the Red Queen may be found. (Therefore, each competitor evolves to adapt to the strategies of the other, who in turn evolves new strategies of defense.)

APPENDIX B Example Coding of Self-Explanation


Target sentence Self-explanation Therefore, the offspring would possess the same locks of the immune system. If offspring were reproduced by ssion, then the locks would always be the same. Therefore, the bacteria would not have to do anything different to infect the offspring since the offspring have the same defenses. Remark (Relevant Part of the SelfExplanation for the Coding) then the locks would always be the same. since the offspring have the same defenses If offspring were reproduced by ssion (see Appendix A for preceding sentence) Therefore, the bacteria would not have to do anything different to infect the offspring

Coding Type Paraphrase Near bridging Far bridging General knowledge Domain knowledge

Coding Accurate paraphrase Accurate near bridging Not present Relevant and accurate general knowledge inference Not present

Copyright of Discourse Processes is the property of Taylor & Francis Ltd and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.

You might also like