Improving Computer Science Education

Djordje M. Kadijevich, Charoula Angeli, Carsten Schulte

• Publisher: Routledge
• Year: 2013
• Print ISBN: 9780415645379
• eBook ISBN: 9781136156472
• Topic: Education
• Subtopic: Education General

Improving Computer Science Education examines suitable theoretical frameworks for conceptualizing teaching and learning computer science. This highly useful book provides numerous examples of practical, "real world" applications of major computer science information topics, such as:

• Spreadsheets
• Databases
• Programming

Each chapter concludes with a section that summarzies recommendations for teacher professional development. Traditionally, computer science education has been skills-focused and disconnected from the reality students face after they leave the classroom. Improving Computer Science Education makes the subject matter useful and meaningful by connecting it explicitly to students' everyday lives.

PART 1
IMPROVING LEARNING

1

TEXT COMPREHENSION IN COMPUTER SCIENCE EDUCATION

Maria Grigoriadou and Alexandra Gasparinatou

Text comprehension

Texts are an important tool for learning. Many students are poor readers or have difficulty understanding textbooks (Snow, 2003). “To optimize learning, should one make the comprehension process as easy as possible, or should one, as many educators insist, ensure that the learner participates actively and intentionally in the process of constructing the meaning of the text?” (Kintsch, 1998). Specifically, should the readers' task be facilitated by improving the comprehensibility of a text or should the readers' active involvement be increased by placing obstacles in their way? In the second case, what sort of obstacles will have beneficial effects on learning and under what conditions? The approach to this question has been the study of characteristics of the text, the characteristics of the individual reader and how these factors affect text comprehension.

A considerable number of empirical studies have been conducted to answer this question. Many of them have demonstrated that readers' background knowledge facilitates and enhances comprehension and learning (McNamara et al., 1996; Gasparinatou & Grigoriadou, 2010). These studies have also shown that readers with greater background knowledge express more interest in the reading material and use more effective reading strategies. Additionally, experts tend to put more effort into learning than novices (Tobias, 1994).

Text comprehension can also be facilitated and enhanced by rewriting poorly written texts to be more cohesive and to provide the reader with all the information needed for a good comprehension (Beyer, 1991; Britton & Gulgoz, 1991; McKeown et al., 1992). Text coherence refers to the extent to which a reader is able to understand the relations between ideas in a text. This is generally dependent on whether these relationships are explicit in the text.

Nevertheless, a cohesive text representation does not always result in better learning. Readers with relevant prior knowledge do not always use that knowledge for learning. They also tend to take the path of least resistance, and if they feel that they easily understand the text they read, they may neglect to activate relevant prior knowledge in order to form links between their knowledge and the new text. Thus, there exists an instructional need to stimulate reader activity (Kintsch, 1998). Consequently, the advantages found for facilitating the reading process by making text more cohesive and the disadvantages demonstrated for facilitating the learning process present contradictory findings.

The construction-integration model

The theoretical framework for this chapter is the construction-integration model for comprehension (Kintsch, 1998). The construction-integration model was an extension of earlier models of comprehension (Kintsch & van Dijk, 1978; van Dijk & Kintsch, 1983), primarily specifying computationally the role of prior knowledge during the comprehension process. According to this model, comprehension arises from an interaction and fusion between the text information and knowledge activated by the learner. The final product of this construction and integration process is referred to as the reader's mental representation of the text.

Levels of understanding

This model distinguishes several different levels in the mental representation of a text that readers construct. The mental representation is a unitary structure, but it is useful to distinguish between certain aspects of that structure. The levels of understanding that are most relevant for the purposes of this chapter are the text base and the situation model.

The text base consists of those elements and relationships that are directly derived from the text itself. To construct the text base, the reader needs syntactic and semantic (lexical) knowledge.

The situation description that a learner constructs on the basis of a text as well asprior knowledge andexperience is called the situation model. A situation model is, therefore, a construction that integrates the textbase and relevant aspects of the learner's knowledge.

The distinction between the micro- and the macro-structure of a text is orthogonal to the text base and situation model distinction. While the micro-structure refers to local text properties, the macro-structure refers to the global organization of text.

A third distinction refers to the quality of each one of these structures. One may have a poor text base (micro or macro), perhaps because the text is poorly written or perhaps because the learner did not encode properly what was there.

According to Kintsch's (1998) model, many factors contribute to learning from text, but prior domain-specific knowledge and the building of a coherent situation model are the driving factors.

The measurement of learning

As the levels of understanding are not separate structures and the situation model, by definition, involves both the text base and long-term memory, a comprehension measure cannot exclusively tap into one level of understanding. Some measures are more indicative of text memory (e.g., recognition, text-based questions, and reproductive recall) whereas other measures are more sensitive to learning (e.g., bridging inference questions, recall elaborations, problem-solving tasks, and keyword sorting tasks).

The former are referred to as text base measures, because all that is required for good performance is a coherent text base understanding. The latter are referred to as situation model measures, because to perform well on them, the reader must form a well-integrated situation model of the text during the comprehension process (Kintsch, 1998; McNamara et al., 1996).

Text cohesion

The degree to which the concepts, ideas, and relationships with a text are explicit has been referred to as text cohesion, whereas the effect of text cohesion on readers' comprehension has been referred to as text coherence (Graesser, McNamara, & Louwerse, 2003; McNamara et al., 1996). Text coherence refers to the extent to which a reader is able to understand the relations between ideas in a text and this is generally dependent on whether these relationships are explicit in the text. Text cohesion is one of the important dimensions along which text varies. It is an objective feature of texts, an important factor to determine text coherence, which is a subjective psychological state of a reader (Graesser et al., 2003).

Research in text comprehension

In the domain of history, Britton and Gulgoz (1991) found that participants who read the cohesive texts had better recall of the material, and developed a better situation model than participants who read less cohesive texts. Voss and Ney Silfies (1996) found that learning from an expanded text was related to reading-comprehension skill, whereas learning from an unexpanded text was a function of prior knowledge. Vidal-Abarca, Martinez, and Gilabert (2000) found that cohesive texts helped learners to generate a deeper understanding of the texts. The positive impact of cohesive text has been replicated by Linderholm et al. (2000) and also by Gilabert, Martinez, and Vidal-Abarca (2005).

In the domain of biology, McNamara et al. (1996) found that readers who know little about the domain of the text benefit from a maximally cohesive text, whereas high-knowledge readers benefit from a minimally cohesive text. Similarly, McNamara and Kintsch (1996) found that less cohesive text was more helpful for high-knowledge readers when they were asked questions, which revealed their understanding of the material (e.g., keyword sorting problem or open-ended questions asked after a delay). McNamara (2001) found that the lowcohesion advantage for learners of high-knowledge manifested at the text base model level. Ozuru, Dempesey, and McNamara (2009) found that (a) reading a high-cohesion text improved text-based comprehension, (b) overall comprehension was positively correlated with participants' prior knowledge, and (c) skilled participants gained more from high-cohesion text. In the domain of physics, Boscolo and Mason (2003) found that high-knowledge readers benefited from a minimally cohesive text.

In the domain of computer science, Beyer (1991) used a computer manual as his learning material. He revised the original manual by making its macro-structure explicit. The revised text proved to be significantly improved compared with the original version, but the improvement was restricted to problem-solving tasks. There is a lack of studies concerning learning from computer science texts. Computer science texts differ from those in social and natural sciences due to the following reasons (ACM & IEEE, 2008):

• Computer science texts are complex depending on factors mainly inherent in the texts. Much of their content is abstract and technical, far removed from everyday experience.

• Texts in computer science require students to utilize concepts from many different fields. All computer science students must learn to integrate theory and practice, to recognize the importance of abstraction, and to appreciate the value of good engineering design.

• Computer science texts assist students to develop a high level of understanding systems as a whole.

• Computer science texts must help students to encounter many recurring themes such as abstraction, complexity, and evolutionary change. They must also assist them to encounter principles (i.e., those associated with caching such as the principle of locality), with sharing a common resource, and so on.

Consequently, the way in which cohesion manipulations influence the comprehension and consequently the learning of computer science texts (e.g., computer networks texts) may differ from that of social and natural science texts. For the reasons mentioned above, we conducted two studies. In our first study (Gasparinatou & Grigoriadou, 2010), we investigated the effects of background knowledge on learning from high- and low-cohesion texts in the domain of computer science. The comprehension of 58 undergraduate students was examined in the domain of local network topologies using four versions of a text, orthogonally varying local and global cohesion. Participants' comprehension was examined through free-recall measure, text-based, elaborative-inference, bridging-inference and problem-solving questions, and a sorting task. The results showed that students with low and high background knowledge performed better with a high- and a low-cohesion text respectively, which implies adjusting text cohesion level to students' background knowledge.

In our second study (Gasparinatou & Grigoriadou, 2011a), we examined whether high-knowledge readers in computer science benefit from a text of low cohesion. Undergraduate students (n = 65) read one of four versions of a text concerning local network topologies, orthogonally varying local and global cohesion. Participants' comprehension was examined through free-recall measure, text-based, bridging-inference, elaborative-inference, problem-solving questions, and a sorting task. The results indicated that high-knowledge readers benefited from the low-cohesion text. The interaction of text cohesion and knowledge was reliable for the sorting activity, for elaborative-inference and for problem-solving questions. Although high-knowledge readers performed better in text-based and in bridging-inference questions with the low-cohesion text, the interaction of text cohesion and knowledge was not reliable. The results suggest a more complex view of when and for whom textual cohesion affects comprehension and consequently learning in computer science. These results also support the hypothesis that a text that requires gap-filling inferences is beneficial for learning in computer science.

Supporting comprehension with personalized learning environments

There is a growing literature of studies focusing on assisting comprehension through personalized learning environments. In the early 1990s, the system Point&Query (P&Q), a hypertext/hypermedia system, was developed (Graesser, Langston, & Bagget, 1993). Students learned entirely by asking questions and interpreting answers to questions. In order to ask a question, the learner would point to a hot spot on the display by clicking a mouse. Then a list of questions would be presented. Thus, the learner could ask a question very easily, by two quick clicks of a mouse. On the average, a learner ends up asking 120 questions per hour, which is approximately 700 times the rate of questions in the classroom. The learner also is exposed to good questions because high-quality questions are presented on the menu of question options. Evaluations of the P&Q software asking revealed, however, that it is not sufficient to simply expo...

Table of contents

• Cover
• Half Title
• Full Title
• Copyright
• CONTENTS
• Foreword
• Preface
• Acknowledgements
• Contributors
• Part 1 Improving learning
• Part 2 Methodological perspectives
• Part 3 Improving teaching
• About the contributors
• Subject index