Printer Friendly

Enhancing hypertext documents to support learning from text.

Many researchers in human-computer interaction have noted that a potentially serious drawback with using hypertext as a document format is that users may easily become "lost in hyperspace" knowing neither where one is in a document nor how to get to where one would like to be [1-4]. Some have suggested that the situation is like being lost in a forest without a trail map.

Perhaps because of this analogy, interface researchers have proposed a seemingly intuitive solution: Provide a map to the reader. Such maps consist of labelled nodes and directed links, with the current location made salient by a different color or other technique. However, there have been no direct empirical evaluations of the effectiveness of such maps.

We have previously suggested constraining the topology of the hypertext network as a method for decreasing user navigation problems [5]. The purpose of the study reported here was to compare the effectiveness of the two approaches, constraining the hypertext structure and providing maps. The most critical findings are reported here; a full research report is currently in preparation.


Overview. In this study, 80 subjects read a tutorial about operating a new state-of the-art JVC hi-fi Video Cassette Recorder. The tutorial was presented on a Macintosh IIci using Hypercard. Subjects viewed one of five versions of the tutorial; all five versions contained the same content (approximately 2,300 words of text with three simple figures). Four of the versions were created by crossing two independent variables, (a) constrained network structure vs. free network structure, and (b) accompanying map vs. no map. A fifth control version was a standard "linear" format. The design is shown in Figure 1.

Subjects read one of the versions, then were asked to summarize the tutorial and answer a series of question probes. (Question probes are questions that are systematically designed to elicit a body of knowledge regarding "what, why, how" and similar types of knowledge [6].) Subjects were then asked to solve two problems using the hi-fi VCR to perform particular tasks, and to answer a questionnaire.

Materials (Implementation of Independent Variables). To develop the five versions, we first used knowledge engineering procedures to develop a semantic network of the instructional information [6]. We next transformed the network into a linear text version and two different hypertext structures--a free network structure and a constrained structure. We then developed maps to accompany each hypertext version.

The four hypertext versions are briefly described below:

1. Free Network Structure, No Map. For this version, we

used a cluster coherence algorithm [5] to divide the

semantic network into hypertext frames. This

algorithm determines what information to bind together

into hypertext nodes and what links to include in

the system, with a goal of retaining conceptual

coherence and minimizing the number of links. The

resultant network contained 26 nodes and 40 links

(see Figure 2).

2. Free Network Structure, Map. For this version, we

simply added a map, which is essentially the map

shown in Figure 2. Subjects could access the map

from any screen by clicking on a button with a

N/S/E/W star icon and the words "To the Map." In

addition, the subject's position was made salient by

reverse coloring of the node (white on black).

3. Constrained Structure, No Map. For this version, we

first identified the "top superordinate information"

in the semantic network. This information was

considered central and critical to understanding the

material. We wrote the information as four top-level

frames for the hypertext document. We networked

the remaining material to each of the top-level

frames using the cluster coherence algorithm. Figure

3 shows the structure of the constrained hypertext


Some nodes were repeated so that the hypertext

networks were independent for each top-level

frame. Subjects were required to go through the top

four frames, with hypertext buttons allowing them

access to the network attached to them. At any time

they could return "home" to the frame they had left

via a hypertext button at the top of the screen.

4. Constrained Structure, Map. The fourth version

consisted of the hypertext system described in version

3, but the subjects also had access to a map at any

time. Each map was designed like the one shown in

Figure 1, except that it depicted only the local

network that the subject was in, not the other three.

Subjects and Procedure. Subjects were 80 male and female undergraduate students at the University of Idaho who participated individually for course credit in Psychology 100. Each subject read a short introduction to the tutorial and the use of hypertext buttons. They were asked to read the tutorial at their own pace. After finishing, subjects were asked to perform four tasks:

1. Write a summary of the tutorial.

2. Answer 39 question probes covering the basics of

the material in the tutorial.

3. Solve two problems using the hi-fi VCR,

demonstrating basic understanding of the text as well as

some inferencing.

4. Answer a questionnaire with open-ended and rating

scales, eliciting their reaction to the tutorial. We also collected data concerning the order and time of subjects' progression through the document frames. At this time, we have evaluated most of the data from tasks 2, 3, and 4.


Question Probes. Because many question probes had more than one correct answer (e.g., "What are the ALC settings?" had three correct answers), we scored subject answers to the 39 question probes for gist, using a checksheet of the correct items (102 total). Thus, question probe scores represent the total number of correct answers out of 102.

Mean scores for the analysis are given in Table 1. Analysis of Variance (ANOVA) reveled that the linear version resulted in more correct answers than any of the 4 hypertext versions; F(1,78) = 25.96, p < 0.001. In addition, none of the hypertext versions was significantly different from any other. That is, memory for details was not affected by the topological constraint or provision of maps.
Mean number of correct answers to
question probes
        Linear   Constrained   Free Network
No Map   34.9       19.7           22.3
Map                 24.3           22.4

Problem Solving. Subjects' problem-solving performance was scored on a 5-point scale for each of the 2 problems. For the analysis, scores for the two problems were combined. Table 2 shows the mean scores out of 10 for the 5 groups.
Mean number of points on two
problems combined
        Linear   Constrained   Free Network
No Map   7.19       6.50           5.56
Map                 7.06           5.50

A one-way ANOVA for the 5 groups showed no significant effect. However, when the map and no-map conditions were combined, the linear and constrained group subjects performed significantly better than the free-network subjects, F(2,77) = 4.0, p < 0.015. The linear and the constrained network subjects did not differ significantly.

Consistent with this, a 2 x 2 factorial ANOVA for only the hypertext groups showed a significant effect of topology constraint, F(1,60) = 4.67, p < 0.05; however, there was no effect of map, nor an interaction.

Subjective Ratings. The subjects were asked to rate the hypertext system on a number of 9-point scales, including how much they liked the system, whether it detracted from learning the material, how frequently they used the map, and so forth.

Regarding subjects' liking of the system, a three-way ANOVA found a marginal effect with mean ratings of 5.31 linear, 6.68 constrained, and 5.92 free network, F(2,77) = 2.54, p = 0.08. Further analyses showed that liking for the constrained system was significantly higher than for the linear, F(1,46) = 4.45, p < 0.05, and marginally higher than for the free network, F(1,62) = 2.5, p = 0.11. A 2 x 2 ANOVA for the hypertext systems showed no differences.

Regarding map use, subjects varied widely on whether they used the map and whether it was helpful. In general, more of the free-network subjects than the constrained-network subjects said that they needed the map (means were 6.94 vs 4.69, where 9 = strongly needed map) and that they used the map (5.44 vs. 4.63, where 9 = extremely frequently).


Overall, the results reveal a pattern of data where subjects learning from a linear format document obtained the most factual information (undoubtedly because they were "encouraged" to read all of the screens). However, in practice, subjects learning from the constrained-hypertext system did as well at actual problem solving as did subjects who read the linear version. On the other hand, the free-network subjects performed significantly worse on problem solving.

Subjective ratings showed that subjects marginally preferred the constrained-hypertext topology. Supporting this, we found that when asked for open-ended reactions to their tutorial system (before answering the rating scale items), the comments from subjects in the free-network group were typically negative (total of 13 positive and 36 negative for all subjects), as compared with the constrained system, which elicited more positive comments (total of 34 positive and 13 negative).

Perhaps the most interesting and provocative result is that provision of a map had no significant effects for any of the analyses. Viewed another way, the problems engendered by the free-network system were not ameliorated by the provision of a map. Subjects in the free-network/map condition still had substantive difficulties with the system; 3 out of the 16 said (without prompting) that they got stuck. Five other subjects said they found the system either very frustrating or difficult to use.

In summary, we can conclude that if it is critical that students learn the details of material in a document, the instructor cannot rely on the student to hypertext through the information and acquire it; linear formats should probably be retained. On the other hand, a hybrid system with a constrained structure seems to be a good alternative when only more global overall learning is required. The hybrid-system students were able to navigate through the material and acquire the basic understanding needed for subsequent problem solving. Their liking of the system was higher than that of students using standard linear formats. With this system, performance and subjective reactions were positive regardless of whether a map was provided or not.


[1]R. J. Glushko, "Transforming Text into Hypertext for a Compact Disc Encyclopedia," in CHI 89 Proceedings (New York: Association for Computing Machinery, 1989), 293-298. [2]A. Nicol, "Interface Design for Hyperdata: Models, Maps, and Cues," in Proceedings of the 32nd Annual Meeting of the Human Factors Society (Santa Monica, CA: Human Factors Society, 1988), 308-312. [3]F. Valdez, M. Chignell, and B. Glenn, "Browsing Models for Hypermedia Databases," in Proceedings of the 32nd Annual Meeting of the Human Factors Society (Santa Monica, CA: Human Factors Society, 1988), 318-322. [4]]T. Whalen and A. Patrick, "Conversational Hypertext: Information Access Through Natural Language Dialogues with Computers," in CHI 89 Proceedings, (New York: Association for Computing Machinery, 1989), 289-292. [5]S. E. Gordon and V. Lewis, "Knowledge Engineering for Hypertext Instructional Systems," Proceedings of the 34th Annual Meeting of the Human Factors Society (Santa Monica, CA: Human Factors Society, 1990), 1412-1416. [6]S. E. Gordon and R. T. Gill, "Knowledge Acquisition with Question Probes and Conceptual Graph Structures," in Questions and Information Systems, T. Lauer, E. Peacock, and A. Graesser, eds. (Hillsdale, NJ: Lawrence Erlbaum Associates, in press).
COPYRIGHT 1992 Society for Technical Communication
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1992 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Charney, Davida
Publication:Technical Communication
Article Type:Column
Date:May 1, 1992
Previous Article:Communication management.
Next Article:Customer satisfaction measurement.

Terms of use | Privacy policy | Copyright © 2022 Farlex, Inc. | Feedback | For webmasters |