Computer-Adapted Text Presentation to Increase

Comprehension and Active Processing

Dr. Robert Clifford, University of South Florida

Introduction

Most instructors assume that maximum coherence in course materials is best for all students, but the level of text coherence can actually be manipulated to affect the learning process and comprehension, sometimes to the advantage of students already possessing prior domain knowledge. Many elements of a passage can enhance coherence, among them various types of inferences in the text, global organization clues, logical topic sentences, and so on. We might ask, then, how students respond to different levels of text coherence? Does it affect their comprehension or active processing, depending on their amount of prior knowledge? And if this is true, how can we provide students with the type of text most appropriate to their level of expertise? In this essay I will first present a brief overview of the research on this topic. Then, I will show an example of how I have attempted to integrate this concept into my Adaptive Tutor for lower level music students.

Research on Text Coherence

Britton and Gülgöz (1991) adapted passages of text according to three principles used in a computer program by Miller and Kintsch (1980). This program is designed to identify areas of text that lack coherence. First, it identifies where inferences in the text are implicit rather than explicit. Second, it identifies the linking words so they can be repeated in the next sentence. Finally, it arranges sentences so the reader reads old information (from the last sentence) first before going on to the new information in the sentence. As one might expect, the authors found that when these changes were implemented, coherence was improved and readers were better able to perform recall or summarizing tasks.

Van den Broek (1994) identifies several types of causal inferences in text. Forward inferences anticipate the causes of future events in the text, while backward inferences serve to connect an event to its antecedents. One type of backward inference, the elaborative inference, is of particular interest to us here. This is where a reader accesses background information to make connections in the text. According to van den Broek, readers encode these types of inferences in their mental representations of the text. When the inferences are made explicit, coherence is increased and readers are better able to complete tasks of recall or summarizing tasks.

McNamara, Kintsch, Songer, and Kintsch (1996) suggest, however, that in some cases a poorly-written text might actually be beneficial to certain students. According to the authors, when readers work a little harder to figure out a text, the result is a deeper understanding of the subject. Implicit here, of course, is that the reader has the necessary background information to figure out the poorly-written text. They are not suggesting that the novice learner would be better served by an incomprehensible text, but rather that when one spends more time figuring out a passage of text, the meaning is more deeply ingrained in the reader’s mind.

McNamara et al. use the construction-integration model of Walter Kintsch (1988) as a point of departure. According to this model, readers form two types of structure within their mental representations. The text base contains only the facts contained in the text. It is usually organized as the author organizes it, and the information contained there is most useful for simple tasks like recall or summarizing. The other part of the mental representation is the situation model. This model integrates the information in the text with the reader’s prior domain knowledge into a meaningful representation, organized according to the reader’s own perspective. The situation model is where deep learning occurs, and the information in it is useful for complex tasks such as problem solving, or applying the information to novel situations.

Here is a brief example of one way that McNamara et al. manipulate text coherence. Two sentences are connected by the word “therefore,” a clue to how they relate.

The blood cannot get rid of enough carbon dioxide through the lungs. Therefore, the blood becomes purplish.

To lessen the coherence, the authors leave out the connective “Therefore” between the sentences and thus it seems as if they could be two unrelated sentences on the same topic. According to the authors, the knowledgeable reader will supply the appropriate connective, while the less knowledgeable reader will not. When the reader with prior domain knowledge is forced to figure out such a passage, the result is a deeper understanding of the material, since the reader engages the situation model to construct a unique mental representation of the topic. The coherence of a text can also be altered by global organization: logical subheadings, a clear sequence of topic sentences, and bullets or other means of helping the reader keep topics organized. On the local level, any implicit references can be made explicit, explanatory phrases can be added, and definitions can be inserted into the prose, all to maximize clarity and coherence.

Relevance to Music Pedagogy

Much of what we teach our students, especially in music theory, is characterized by various levels of complexity. For example, simple tasks might include memorization of terms, recalling something from a text, or reciting a list of voice-leading guidelines. Complex tasks could include solving a voice leading problem with information not explicitly contained in the text, applying information to novel situations, using information to compose a piece, or performing an analysis. One specific, and common, example: students can usually recognize a non-harmonic tone in an isolated example (simple), but they have difficulty transferring that knowledge to an actual musical example that embellishes the model (complex). According to McNamara et al., if the students with at least some prior knowledge have to work a little harder to figure out these non-harmonic tones, they will be better able to apply them in complex tasks like analysis, since they become part of the situation model in their mental representation.

These theories are interesting and have important implications for learning in all fields, but their application is not without problems or challenges. First, if the amount of prior domain knowledge is important to the student’s ability to build representations in the situation model, how does one actually determine what the student already knows? The experiments I’ve mentioned, especially those by McNamara et al., test the participants in some way to determine prior knowledge in the desired topic area. If one makes some type of determination, how does the instructor provide an appropriate level of text coherence for each student? Finally, if we increase active processing for the knowledgeable student, are we then shortchanging the novice, who would also benefit from increased active processing?

Adaptive Tutor

The solution to the problem of supplying these different levels of coherence is an adaptive text presentation such as one finds with adaptive hypertext or hypermedia. In this medium, one can present text that changes according to the desired situation, in this case the level of expertise of the user. For my Adaptive Tutor, I use three levels—Review, Basic, and Advanced—that are chosen by the user according to their own assessment of their abilities. The Advanced level features text that is minimally coherent, along with musical examples completely without annotations. This is intended for the student with a certain amount of prior knowledge. By working to figure out the text and musical example, the student builds the mental representation necessary to solve musical problems such as a voice leading problem, or perhaps a composition where the new concepts must be applied.

In the Basic level, the text is mostly as it is in the minimally coherent version with a few clarifications here and there. Appropriate subheadings have been added and the paragraphs are broken up to aid the reader’s global organization. There are also annotations on the musical examples to aid in relating them to the information contained in the text. Finally, in the Review level, all text and subheadings are designed for maximum coherence. Definitions are inserted in the prose, there are subheadings for global organization, the musical examples have ample annotations, and so on. The users of this version, those with little or no prior knowledge, would be expected to be able to use the information in the text mostly for recall or summarizing tasks.

Here is an example of the minimally coherent text and musical example from the voice leading lesson.

Each root progression has models of voice leading. Major and minor chords in progressions of a perfect fourth always share one common tone. In both ascending and descending fourth progressions it is possible to keep it in the same voice. This is the common tone resolution model.

I purposely omitted certain connectives so that the sentences are a bit vague. For example, sentences one and two mention root progressions, but the first mentions voice leading while the second does not, instead talking about the common tone found in such progressions. The knowledgeable reader will be able to infer that the two sentences are related. Similarly, in the third sentence, I mention keeping “it” in the same voice, but it is not clear whether I am referring to the common tone from the previous sentence or the perfect fourth from sentence three. Finally, the last sentence mentions the common tone resolution model, but never really says that it is a type of voice leading. The musical example has no annotations at all.

The medium coherence version is similar, but notice that now subheadings have been added (I have shown the added text with underlining).

Chord Resolution Models

Each root progression has models to determine voice leading. In the following examples, both keyboard and SATB style are illustrated.

Ascending and Descending Progressions by Fourth: Major and minor chords in root progressions of a perfect fourth always share one common tone. In both ascending and descending fourth progressions it is possible to keep it in the same voice. This is the common tone resolution model.

The passage is broken into two smaller chunks to aid global organization, and the first paragraph mentions the types of textures that will be illustrated. There are some annotations on the example. Finally, here is the same passage revised for maximum coherence.

Chord Resolution Models

Each root progression has a limited number of chord resolution models to determine voice leading between adjacent chords. In the following examples, both keyboard style and SATB style (Soprano, Alto, Tenor, Bass) are illustrated.

Ascending and Descending Progressions by Fourth: Major and minor chords used in root progressions of a perfect fourth always share one common tone. In both ascending and descending fourth progressions it is possible to keep the common tone in the same melody line, or voice. This type of voice leading is called the common tone resolution model, which is shown here with both descending and ascending root progressions. In this example, the common tone is G, and it remains in the same voice in each model.

Now the reader learns immediately that the models determine voice leading between adjacent chords. The reference to SATB is explained in the text, as is the reference to voice (defined as a melody line), and the resolution model is now identified as a type of voice leading. The text orients the student to the musical example, which now shows annotations for chord names.

I use Javascript and HTML to fashion this web-based tutor. The user enters his or her skill level on the first page, which sets the appropriate variable. When each subsequent page loads, I test for that variable to find which level the user has chosen. Then, the “document.write” command places the appropriate text on the page. There are several ways to achieve this adaptability. A basic text can be altered by adding words or phrases in appropriate locations in the text. Or, when many changes are called for, the command can be used to write an entire passage to the page. The musical examples are also chosen according to the Level variable that the user sets when entering the program.

Conclusions

The Adaptive Tutor has been tested with students in a music fundamentals class, but I have only recently added this level of text manipulation, so it remains at this time a work in progress. The research seems to indicate that this type of manipulation can improve active processing and thus the type of deep understanding that is so beneficial to our students. In doing so, however, we should be careful not to ignore the less knowledgeable student, who can also benefit from active processing, perhaps from different types of activities. Since it would be impossible to present different levels of coherence in paper texts, it is left to some sort of computerized presentation, where content can be altered dynamically. With music students, one of the most common problems I encounter is the inability of students to apply abstract concepts or models to actual music through analysis or composing. Hopefully this type of approach will provide a means by which instructors can adapt to the variety of students they teach.

Reference List

Britton, Bruce K., and Sami Gülgöz. 1991. Using Kintsch’s Computational Model to Improve Instructional Text: Effects of Repairing Inference Calls on Recall and Cognitive Structures. Journal of Educational Psychology 83, no. 3: 329-345.

Kintsch, Walter. 1988. The Role of knowledge in Discourse Comprehension: A Construction- Integration Model. Psychological Review 95, no. 2: 163-182.

Mannes, S. and Walter Kintsch. 1987. Knowledge Organization and Text Organization. Cognition and Instruction 4: 91-115.

McNamara, Danielle S., E. Kintsch, N. B. Songer, and W. Kintsch. 1996. Are Good Texts Always Better? Interactions of Text Coherence, Background Knowledge, and Levels of Understanding in Learning From Text. Cognition and Instruction 14, no. 1: 1-43.

Miller, J. R., and Walter Kintsch. 1980. Readability and Recall of Short Prose Passages: A Theoretical Analysis. Journal of Experimental Psychology: Human Learning and Memory 6: 335-354.

van den Broek, Paul. 1994. Chapter 16: Comprehension and Memory of Narrative Texts. In Handbook of Psycholinguistics. ed. Morton Ann Gernsbacher. San Diego: Academic Press: 539-588.