Interest in "learning" has fluctuated more widely in the past decades than women's fashions. If my memory serves me right, I did not hear a single lecture or discussion on learning as a graduate student in Vienna in the 1950s. When I came to America, however, learning was almost the sole topic where I went to graduate school, and I quickly succumbed to the fascination of habit strength and drive–reward interactions. However, I had hardly graduated when all that became passé, and learning, together with my beloved Markov models, disappeared into history's storage closet. The students at Colorado during the past 10 years have heard about as much about "learning" as I had years ago at Vienna. What has happened? Has America caught up with Europe? Or have we relapsed into the dark ages?

This project involves developing a reading comprehension curriculum based on the Construction-Integration model (Kintsch, 1998) and implementing a pilot efficacy study in middle-school classrooms. The curriculum (BRAVO) explicitly defines the cognitive processes involved in skilled reading, teaching students how to establish local and global text coherence and use background knowledge to create a mental model – the prerequisite to deep, meaningful learning. A crucial and unique component is the use of sequenced texts related to a single over-arching topic, enabling students to use their expanding prior knowledge while reading complex expository texts. BRAVO, consistent with Common Core State Standards, bridges the divide between grade-school narratives and secondary-level textbooks.

LSA is a machine learning method that constructs a map of meaning that permits one to calculate the semantic similarity between words and texts. We describe an educational application of LSA that provides immediate, individualized content feedback to middle school students writing summaries.

This article explores the role of self-regulation in strategies that readers use to decide the order in which to read the different sections of a hypertext. This study explored 3 main strategies for link selection based on (a) link screen position, (b) link interest, and (c) the semantic relation of a link with the section just read. This study followed Winne's (1995, 2001) model of self-regulated learning to try to explain why some readers select hyperlinks based on strategies that lead to lower levels of comprehension (i.e., screen position and personal interest). Results from 2 studies revealed that readers with low prior knowledge base their decisions on what to read next on a default screen position or on link interest more often if they are instructed to set a low learning goal, if they regularly use shallow learning strategies (e.g., memorizing), or if they are poor at calibrating their comprehension. Readers' link selection strategies mediated the effect of the self-regulation variables studied on comprehension.

There is a need for a tutor that coaches students through the process of composing a written summary. Summarizing is increasingly valued by teachers, as well as by school accountability experts, not only as an important skill in its own right, but as a means of promoting deeper levels of comprehension and learning. How to teach kids to summarize is the problem. Conveying the underlying strategies is the easy part, but like most skills, one learns by doing, and in this case it takes a lot of practice. Few schools have the resources to guide individual students through multiple drafts and revisions of their summaries or essays. Summary Street® was designed to address this problem by giving students lots of opportunity to summarize informational text guided by individualized feedback. As a true offspring of LSA, feedback from Summary Street® is directed at the content of the writing. It tells the writer whether the summary conveys enough informa-tion about each of the main topics of a text; whether the summary is of the appropriate length; and whether it contains repeated information or information that is not highly relevant to the overall topic. It also warns the writer if too much of the material has been lifted directly from the source text and, of course, it provides a spell check. However, Summary Street®does not provide feedback on other problems with the mechanics of writing, such as sentence structure and punctuation, nor does it assess style, organization, and the truth value of the content.

This article begins with a brief overview of the psychological research on the pedagogical efficacy of questions and describes how recent research has led to a reevaluation of the role of questions within the framework of a theoretical model of discourse comprehension processes and knowledge representation. A computer-based tutor for reading comprehension under development by our research team is described. The tutor uses an interactive books interface to present a structured sequence of questions to elicit summaries from young readers. The students receive feedback on content coverage by means of Latent Semantic Analysis (LSA). The questions in the tutor are based on the principled guidelines presented in the final section of this article.

Having students express their understanding of difficult, new material in their own words is an effective method to deepen their comprehension and learning. Summary Street® is a computer tutor that offers a supportive context for students to practice this activity by means of summary writing, guiding them through successive cycles of revising with feedback on the content of their writing. Automatic evaluation of the content of student summaries is enabled by Latent Semantic Analysis (LSA). This article describes an experimental study of the comprehension and writing tutor, in which 8th-grade students practiced summary writing over a 4-week period, either with or without the guidance of the tutor. Students using Summary Street® scored significantly higher on an independent comprehension test than the control group for test items that tapped gist level comprehension. Their summaries were also judged to be significantly superior in blind scoring on several measures of writing quality. Students of low-to-moderate achievement levels benefitted most from the tool.

Summary Street is educational software based on latent semantic analysis (LSA), a computer method for representing the content of texts. The classroom trial described here demonstrates the power of LSA to support an educational goal by providing automatic feedback on the content of students' summaries. Summary Street provides this feedback in an easy-to-grasp, graphic display that helps students to improve their writing across multiple cycles of writing and revision on their own before receiving a teacher's final evaluation. The software thus has the potential to provide students with extensive writing practice without increasing the teacher's workload. In classroom trials 6th-grade students not only wrote better summaries when receiving content-based feedback from Summary Street, but also spent more than twice as long engaged in the writing task. Specifically, their summaries were characterized by a more balanced coverage of the content than summaries composed without this feedback. Greater improvement in content scores was observed with texts that were difficult to summarize. Classroom implementation of Summary Street is discussed, including suggestions for instructional activities beyond summary writing.

This paper describes a series of classroom trials during which we developed Summary Street, an educational software system that uses Latent Semantic Analysis to support writing and revision activities. Summary Street provides various kinds of feedback, primarily about whether a student summary adequately covers important source content and fulfills other requirements, such as length. The feedback allows students to engage in extensive, independent practice in writing and revising without placing excessive demands on teachers for feedback. We first discuss the underlying educational rationale, then present some results of the trials conducted with the system. We describe the collaborative process among researchers and teachers which enabled the development of a viable and supportive educational tool and its integration into classroom instruction.

The ability of 20 normally achieving and 20 learning-disabled 8th- and 9th-grade readers to comprehend and interpret 2 fairly long and complex narratives, describing the emotional reactions of characters to realistic situations, was compared. The pattern of recall across story categories was similar for both groups. However, the learning-disabled readers not only recalled less overall than the normal readers, they were also less successful at differentiating levels of importance in the macrostructure of the stories. All students included less of the information needed to understand the characters' interactions in the more difficult story than in the easier story. Although normal readers could supply this information when directly probed for it, learning-disabled students were less successful in this respect, suggesting serious weaknesses in their ability to construct an appropriate situation model. Implications for the instruction of learning-disabled students are discussed.

Two experiments, theoretically motivated by the construction-integration model of text comprehension (W. Kintsch, 1988), investigated the role of text coherence in the comprehension of science texts. In Experiment 1, junior high school students' comprehension of one of three versions of a biology text was examined via free recall, written questions, and a key-word sorting task. This study demonstrates advantages for globally coherent text and for more explanatory text. In Experiment 2, interactions among local and global text coherence, readers' background knowledge, and levels of understanding were examined. Using the same methods as in Experiment 1, we examined students' comprehension of one of four versions of a text, orthogonally varying local and global coherence. We found that readers who know little about the domain of the text benefit from a coherent text, whereas high-knowledge readers benefit from a minimally coherent text. We argue that the poorly written text forces the knowledgeable readers to engage in compensatory processing to infer unstated relations in the text. These findings, however, depended on the level of understanding, text base or situational, being measured by the three comprehension tasks. Whereas the free-recall measure and text-based questions primarily tapped readers' superficial understanding of the text, the inference questions, problem-solving questions, and sorting task relied on a situational understanding of the text. This study provides evidence that the rewards to be gained from active processing are primarily at the level of the situation model rather than at the superficial level of text-base understanding.

While learning from text can be considerably improved by making the text more coherent at both the local and global level for low-knowledge readers, readers with adequate background knowledge can actually benefit when a text is not fully explicit and contains coherence gaps that they can fill in on the basis of their own knowledge (McNamara, E. Kintsch, Songer & W. Kintsch, in press; McNamara & W. Kintsch, in preparation). The effects obtained in these studies may be attributed to the fact that a fully explicit text induces in high-knowledge readers an illusory feeling of knowing. In consequence, these readers do not process the text at a deep enough level to ensure learning. This hypothesis is rested in an experiment in which readers are asked to comment on their understanding after every sentence of a text, thereby forcing all readers to be more active processors. It is shown that under these circumstances both high- and low-knowledge readers learn best with a high-coherence text. We conclude that making a text more coherent is beneficial for all readers, irrespective of their background knowledge, if active processing can be induced.

This study explored how students' mental representations of an expository text and the inferences they generated varied as a function of text difficulty and of differences in the task. Ninety-six students from Grades 6 and 10 and college were asked to write summaries of an expository text and then to answer orally several probe questions about the content. Reading difficulty was systematically manipulated at the microstructure and macrostructure processing levels. The results support the prediction of qualitative changes in the way the meaning is represented by different age groups in different text conditions. These are related to the amount and kinds of inferential processes on which the summaries were based. Interestingly, college students generalized the content more in summarizing texts with poor macrostructure than in summarizing texts with good macrostructure. That more macropropositional statements occurred in responding to the probe questions than in the summaries could be explained in terms of the different retrieval conditions that prevailed. Some educational implications of these findings are discussed.

Sixteen 4-year-olds and sixteen 6-year-olds were shown four picture stories consisting of 15 to 18 pictures without text. The stories were well structured, consisting of two or more causally and temporally related episodes. The children were asked to describe each picture, and, after seeing all the pictures of a story, to recall the story without pictures. The pictures were either presented in their normal order or in scrambled order. The data analysis concentrated upon the comparison between the responses in the normal condition when the children were telling a story and in the scrambled condition, when they were merely responding to the pictures as such without the story context. The results showed that even the 4-year-olds, but especially the older children, were interpreting the pictures as stories in the normal condition and that their knowledge about stories, i.e., the story schema, determined the nature of their responses. Even in the scrambled condition the 6-year-olds tried to make sense of the pictures in terms of a story by making inferences, attributing thoughts and emotions to the characters, and using narrative conventions, while the 4-year-olds often reverted to a simple labeling strategy. In recall all of these trends were emphasized. Those parts of the descriptions that were best integrated into a story were recalled best, while nonintegrated descriptions tended to be forgotten.