HUMAN FACTORS, 1988, 30(4), 473-486.

Copyright (c) 1988, The Human Factors Society, Inc. All rights reserved.

Reader-Controlled Computerized Presentation of Text

PAUL MUTER, RICHARD S. KRUK, MARY ANNE BUTTIGIEG, and T. JIN KANG, Department of Psychology, University of Toronto, Toronto, Ontario, Canada

Address correspondence to Paul Muter, Department of Psychology, University of Toronto, Toronto, Ont., M5S 3G3, Canada, muter@psych.utoronto.ca.

ABSTRACT. In five experiments new methods of computer-supported reading were introduced and tested. Self-pacing was permitted in rapid serial visual presentation (RSVP), in which words are presented sequentially at a fixed locus. In Experiments 2 through 5, regressions were also allowed in RSVP. Larger regressions yielded slower reading. Regressions back to the beginning of the sentence were more frequent than regressions back two words. There was no difference in reading speed or comprehension caused by the nature, analogue (mouse), or digital (keystroke) of the control over reading speed, but there was a greater tendency to change speed in the analogue condition than in the digital condition. In Experiment 5 subjects read for over two hours in an RSVP condition with self-pacing and regressions or in a normal page condition; subjective reports of discomfort were not different in the two conditions, but reading speed in the RSVP condition was approximately half that in the page condition. Overall these results indicate that permitting reader control in RSVP is feasible but that permitting regressions sometimes results in slow reading.

INTRODUCTION

Computer technology provides the potential for improved ways of reading. What is the best method of presenting continuous text via computers? What are the best methods under various circumstances?

It has often been demonstrated that text displayed in normal page format is read substantially slower from current computer screens than from the printed page, whether the task is normal reading (Kruk and Muter, 1984; Muter, Latremouille, Treurniet, and Beam, 1982) or proofreading (Gould and Grischkowsky, 1984; Wright and Lickorish, 1983). Centuries of selection and decades of research and development have gone into improving the presentation of text on the printed page. In designing text for computers it is common to adapt principles from the printed page (Tinker, 1963), but what is optimal for the printed page may not be optimal for computer screens (Kolers, Duchnicky, and Ferguson, 1981).

A key difference between the printed page and the computer is that the computer allows dynamic presentation of text. Among other advantages, dynamic displays may attract attention better than static displays (Wright, 1987). Three dynamic presentation methods are vertical scrolling, Times Square, and rapid serial visual presentation. Evidence suggests that vertical scrolling, which is often used in movie credits, may be nonoptimal (Kolers et at., 1981). Similarly, the available evidence suggests that the Times Square method (horizontal right-to-left scrolling, as seen on many electronic billboards) is not efficient with regard to reading speed and comprehension (Granaas, McKay, Laham, Hurt, and Juola, 1984; Sekey and Tietz, 1982).

Perhaps the most widely tested dynamic technique is rapid serial visual presentation, or RSVP (Forster, 1970; Juola, Ward, and McNamara, 1982; Potter, Kroll, and Harris, 1980), whereby words are presented sequentially at a fixed locus. Thus the distribution of the text is temporal instead of spatial. (Consistent with Potter, 1984, and others, we will use the synecdoche "RSVP" even when the rate is not rapid.)

RSVP is potentially useful for at least six reasons. First, when optimal parameters are established, RSVP may be an efficient way to present continuous text in general (Cocklin, Ward, Chen, and Juola, 1984), primarily because with RSVP there is no need to expend cognitive processing capacity on controlling eye movements. Second, RSVP may be the best method of presenting text when display space is limited (e.g., on cramped consoles, multiple-window displays, or wristwatch terminals). Third, RSVP may be useful in teaching reading, especially to some dyslexics (Potter, 1984). Fourth, RSVP is an effective technique for searching for particular items (scanning) or for other tasks in which every word must be fixated (Potter et al., 1980). In the age of information explosion scanning is becoming increasingly important. Fifth, RSVP may be particularly advantageous for readers with impaired peripheral vision (Williamson, Muter, and Kruk, 1986). And sixth, RSVP can be an important tool in studying cognitive processes (Potter, 1984).

Research to date has indicated that reading from a computer screen is approximately as efficient with RSVP as with normal page format (Juola et al., 1982; Masson, 1983). Typically this research has utilized computer-paced presentation of text - the text is presented at a fixed rate and the dependent measure is comprehension score. In some applications of RSVP the reader will prefer selfpacing of presentation and the ability to perform regressions. (Regressions typically make up a substantial proportion of eye movements in normal reading - see Crowder, 1982; Just and Carpenter, 1980.) Two questions arise: Is RSVP with regression capability and self-pacing feasible? How should the reader exert control over pacing and regressions? The present experiments introduce and test techniques for self-pacing and regressions in RSVP.

EXPERIMENT 1

In Experiment 1 subjects read passages of text from a computer screen. There were two conditions - the page condition, in which text was presented in a normal format, and the RSVP condition. In contrast to previously published research, the pace of reading was under the control of the subject. Reading speed and comprehension were the dependent measures of primary interest.

Method

Subjects. In response to campus advertisements 30 university students volunteered to participate as subjects in the experiment. Some subjects participated for course credit and some were paid.

Materials and apparatus. Subjects read 20 passages from Reader's Digest, a subset of the materials used by Masson (1983). The mean length of the passages was 124 words. Ten short-answer questions were prepared for each passage (Williamson, 1985). To establish a baseline response rate, 20 students who had not read the passages attempted to answer the questions. The mean score was 2.3%.

Text was displayed on a 30.5 cm (diagonal) green monochrome video monitor. Display of text was controlled by an Apple 11 Plus microcomputer with an upper- and lowercase character generator and a clock. The character set was designed on a 5 x 7 dot matrix. Horizontally there were approximately two characters per cm. Characters were light on a dark background. Subjects sat approximately 50 cm from the screen. Text in the page condition was displayed with a maximum of 39 characters per line and a maximum of 20 lines per page. Each passage was short enough to take up no more than one screenful in the page condition. In the RSVP condition words were presented one at a time, centered, on the twelfth line of the 20-line screen.

Procedure. A random half of the subjects were given the following instructions:

In this experiment you will be reading passages of prose. Please try to read with as much comprehension and speed as possible.

In one condition the screen will be filled with a passage. As soon as you have finished reading the passage, press any key.

In another condition words will be presented one at a time in the center of the screen. To speed up the rate of presentation by 10%, hit the right key; to slow down the rate of presentation by 10%, hit the left key. You may hit these keys as many times as you like. The speed at the end of a passage will be maintained at the beginning of the next passage in this condition until you hit a key.


The instructions for the other half of the subjects mentioned slowing down first and speeding up second.

Each subject experienced 20 trials - one for each passage. Each pair of trials included one trial in each condition (page and RSVP) in random order. Each subject was given a different randomized order of the 20 passages.

Subjects initiated the presentation of text by a keypress. In the page condition the entire passage appeared on the screen, where it remained until the subject pressed a key again. RSVP passages were preceded by a 2-s display of a fixation point located in the center of the screen. Subjects controlled the rate of presentation by pressing keys in the upper right-hand corner of the keyboard - one key increased the speed by 10% and another key decreased the speed by 10%. At the end of each sentence there was a pause of 500 ms. (A pause of this duration benefits RSVP reading, perhaps because it provides time for postsentence processing; Cocklin et al., 1984; Masson, 1983). The 10 short-answer questions were presented on a Macintosh computer. The order of the comprehension questions for each passage was randomized for each subject. Subjects had 10 s to write down the answer to each question.

In the RSVP condition the starting rate was approximately 150 words per minute (wpm) for odd-numbered subjects and approximately 390 wpm for even-numbered subjects. The rate was not readjusted on each trial - it was the same at the beginning of RSVP trial n + 1 as it had been at the end of RSVP trial n for a particular subject.

Reading rates were measured by dividing the number of words by the reading time.

Results and Discussion

The results of primary interest are presented in Figures 1 and 2. Performance in RSVP with self-pacing was essentially identical to page format performance when initial reading rates were similar.

It is clear that in the RSVP condition the starting rate had a substantial effect on how subjects chose to trade off speed and comprehension. Even-numbered subjects maintained a reading speed in RSVP faster than any other condition, even after 10 passages (Figure 1), but paid a price by having poorer comprehension (Figure 2). In the absence of an index that suitably combines speed and comprehension it is indeterminate whether performance overall was better or worse in the RSVP condition than in the page condition.

Odd-numbered subjects performed approximately equally well in the RSVP and page conditions with regard to both speed and comprehension. The mean reading speed of both even- and odd-numbered subjects in the page condition was similar at the outset to the mean reading speed for odd-numbered subjects in the RSVP condition and remained so throughout the 10 passages. The same was true of the comprehension scores.

Analyses of variance supported the aforementioned conclusions. The interaction between condition (RSVP or page) and starting rate (390 wpm or 150 wpm) was significant for both reading speed, F(1,28) = 13.89, p < 0.001, and comprehension, F(1,28) = 9.71, p < 0.01.

For even-numbered subjects the mean number of keypresses per subject was 15.6 for the slow-down key and 4.7 for the speed-up key. Standard deviations were 12.0 and 6.7 keypresses, respectively. For odd-numbered subjects the means were 8.4 for the slowdown key (SD = 6.1) and 5.3 for the speed-up key (SD = 4.7). (Although the RSVP rate generally increased for odd-numbered subjects, it does not follow that there were more speed-up presses than slow-down presses - it depends on when the keys were pressed.)

In sum, the results of Experiment 1 suggest that under RSVP conditions, self-pacing did not disrupt performance. In past studies performance in RSVP without self-pacing has been comparable to performance in page format (Juola et al., 1982; Masson, 1983). In the present experiment speed and comprehension in RSVP with self-pacing (but without regressions) were as high as in page format reading. This is somewhat surprising in that the requirement to control speed should drain some cognitive processing capacity. Perhaps a compensating factor is that the level of arousal or motivation is higher in the present self-paced task than in the relatively passive computer-paced RSVP task.

EXPERIMENT 2

In normal reading a substantial proportion of eye movements are regressions (Crowder, 1982; Just and Carpenter, 1980). Presumably readers using RSVP will often prefer the ability to make regressions. In Experiment 2 the impact of permitting regressions in RSVP reading was assessed and the effect on performance of regressions of different sizes - two, four, or eight words - was determined.

Method

The method in Experiment 2 was the same as in Experiment 1, with the following exceptions.

Twenty-four subjects participated. The page condition was not included. Four RSVP conditions were tested: with no regressions or with regressions of two, four, or eight words. The subject simply pressed a specified key with the left hand to obtain a regression.

A regression of size s was defined as follows: If the key was pressed during presentation of word w, all words from w - s + 1 up to w, inclusive, were re-presented. For example, for a two-word regression, word w - 1 and word w were re-presented. Before re-presentation there was a 0.5-s pause during which the screen was blank. The duration of the original presentation of word w was unaffected.

Each subject completed five trials in one condition followed by five in another condition, and so on. All 24 possible orders of the four conditions were used, one for each subject. At the beginning of each block of five trials in a condition the presentation rate was set at approximately 390 wpm. The rate within each condition was not readjusted on each trial - the rate at the beginning of trial n + 1 was the same as at the end of trial n. As in Experiment 1, subjects could increase or decrease the rate of presentation by 10% at any time.

Instructions included the following:

In order to comprehend well, sometimes it is desirable to look back, or regress. If you hit the key labeled 1, some words will be represented. If you hit this key while in the "Backup 8 Condition," the eight previous words will be re-presented (in the same order). If you hit this key while in the "Backup 4 Condition," the four previous words will be re-presented, and so on. In the "No Backup Condition," pressing this key will have no effect. Before each passage you will be told what condition you will be in.

You may hit any of the three keys as many times as you like.


Results and Discussion

The data in Figure 3 suggest that there was no difference in reading speed between the no-regression condition and the two-word regression condition but that larger regressions tended to produce lower reading speeds. According to the analysis of variance, the effects of both condition, F(3,69) = 11.49, p < 0.001, and trials, F(4,92) = 50.01, p < 0.001, were reliable. Reading speed declined over trials either because of fatigue or because subjects were still approaching their preferred speed after five trials. There was also a significant interaction, F(12,276) = 6.82, p < 0.001 - condition apparently had a greater effect in the first trial.

Comprehension was not affected by condition, F(3,69) < 1, and improved over trials, F(4,92) = 17.74, p < 0.001. (Again, it is unknown whether overall performance declined or improved over trials because comprehension improved when speed declined.)

The mean numbers of keypresses by condition are presented in Table 1. The pattern of results obtained suggests that subjects may have been more inclined to press the regression button when it produced a larger effect. Note that there were some spurious presses in the no-regression condition.

It is important to realize that, a priori, regressions do not necessarily decrease overall reading speed. Reading speed is a function of the number of times all three keys are pressed. If a regression facilitates comprehension to a sufficient extent, later reading speed could be enhanced, possibly yielding a faster overall speed.

The results of Experiment 2 suggest that permitting fixed-size regressions in RSVP can have a negative effect on reading speed. This may have been because of the lack of flexibility of the regressions and the fact that linguistic features were not taken into account.

EXPERIMENT 3

In a further attempt to approximate normal reading, more than one kind of regression was simultaneously allowed in Experiment 3. Two-word regressions and regressions to the beginning of a major linguistic unit - the present sentence - were possible at any time.

Method

The method of Experiment 3 was the same as that of Experiment 2 with the following exceptions.

There were only two conditions - a regression condition and a no-regression condition. In the regression condition subjects could regress either two words or back to the beginning of the sentence by hitting either of two keys in the upper left-hand corner of the keyboard. A regression size of two was selected because the two-word condition produced the best performance of any regression condition in Experiment 2.

The 24 subjects completed 10 trials in one condition followed by 10 trials in the other condition. Half of the subjects experienced the regression condition first and half, the no-regression condition. At the beginning of each block of 10 trials the presentation rate was set at approximately 390 wpm. The rate was not readjusted on each trial within each block of 10 trials. As in Experiments 1 and 2, subjects could increase or decrease the rate of presentation by 10% at any time.

Instructions for the regression condition included the following:

In order to comprehend well, sometimes it is desirable to look back, or regress. If you hit the key labeled 1, text will be re-presented starting at the beginning of the present sentence. If you hit the key labeled 2, the two preceding words will be re-presented.

Results and Discussion

The data in Figure 4 suggest a slight decrement in reading speed when the two kinds of regressions were simultaneously possible, but this trend did not reach statistical significance, F(1,23) = 2.88. Similarly, these regressions had no effect on comprehension, F(1,23) < 1. As in Experiment 2, trials had reliable effects in opposite directions on reading speed, F(9,207) = 64.78, p < 0.001, and comprehension, F(9,207) = 12.39, p < 0.001.

Means and standard deviations of the number of keypresses are presented in Table 2. There was a greater use of sentence regressions than two-word regressions, t(23) = 6.98, p < 0.001. Perhaps subjects prefer larger regressions as suggested in Experiment 2, or perhaps subjects prefer regressions back to the beginning of a meaningful linguistic unit as opposed to regressions of an arbitrary number of words that may or may not lead back to the beginning of a meaningful linguistic unit.

EXPERIMENT 4

Reading speed is a continuous variable, but in Experiments 1-3 input devices were discrete (keys). Experimental evidence suggests that mouse control is superior to key control for many computer tasks (Card, Moran, and Newell, 1983). In Experiment 4 control over reading speed was either analogue (via a mouse) or digital (via keys, as in Experiments 1-3). Regressions were retained in Experiments 4 and 5 because in our opinion regression capability will be desired by readers. Regressions remained under keyboard control.

Method

The method of Experiment 4 was the same as in Experiment 3 with the following exceptions.

Subjects. Twenty-four university students participated as subjects.

Apparatus. For the first time a Macintosh Plus microcomputer was used to display the text and comprehension questions. The screen dimensions were 16.5 X 10 cm. Screen resolution was 512 X 342 pixels. The 12-point characters were proportionally spaced (within words) and were presented in black on a white background. The maximum matrix size was 9 X 12 pixels. Horizontally there were approximately 16 cm of text per line and 4.7 characters per cm. There was a maximum of 19 lines per page.

Design and procedure. The independent variables were the method of control for selfpacing - keyboard or mouse - and the trial numbers, 1 through 8. These two variables were fully crossed within subjects. The order of the pacing methods was counterbalanced across subjects. Odd-numbered subjects used the key controls for eight passages followed by the mouse control for the remaining eight passages; the even-numbered subjects followed the reverse order.

For the keyboard condition subjects were told that the presentation rate could be adjusted by using two keys in the upper righthand corner of the keyboard. As in earlier experiments, one key increased the rate of presentation by 10% for each press and the other decreased the rate by 10%.

For the mouse condition subjects were told to use the mouse with their right hands to control the rate at which the words were presented. Moving the mouse to the right caused the presentation rate to increase and moving the mouse to the left caused the presentation rate to decrease.

The obtained speed in words per minute for a given mouse position was

(max - min)p/w + min

where max is the maximum possible rate (380), min is the minimum possible rate (60), p is the current position of the mouse in pixels, and w is the total width of the screen in pixels (512).

As in Experiment 3, two types of regressions were permitted - two-word regressions and regressions to the beginning of the sentence. Regressions were allowed in both the keyboard and mouse conditions. As in the earlier experiments, the two types of regressions were controlled by two keys in the upper left-hand corner of the keyboard.

Sixteen passages were chosen randomly for each subject from the set of 20 used in Experiments 1-3.

Subjects began presentation of the passages by clicking the mouse. In the first mouse-controlled trial a small square was displayed on the screen indicating the mouse position that corresponded to the initial presentation rate for the text. To start the block of mouse-controlled trials the experimenter clicked the mouse in the square. A fixation point was displayed in the middle of the screen before the text was presented and after a brief pause the text appeared.

The initial presentation rate was approximately 300 wpm for the first passage in both the mouse-controlled and keyboard-controlled blocks of trials. After the first passage the presentation rate for trial n + 1 was equal to the rate at which the subject finished trial n.

Results and Discussion

The reading speed results are presented in Figure 5. The interaction between condition and trial did not approach significance, F(7,161) < 1, nor did the main effect of condition, F(1,23) < 1. The effect of trials was reliable with regard to both reading speed, F(7,161) = 6.23, p < 0.001, and comprehension, F(7,161) = 5.53, p < 0.001.

The data in Table 3 suggest that there was a tendency to increase or decrease speed more in the mouse condition than in the keyboard condition, t(23) = 2.36, p < 0.05.

The unexpected finding that mouse control over speed produced no better performance than keyboard control might have been attributable to the homogeneity of the passages. The nature of the input device may be important only if a substantial amount of changing of speeds is required. A second possibility is that there was insufficient information or feedback in the mouse condition. In the keyboard condition subjects knew that each keypress changed the speed by 10%. The observed tendency to change speed more often in the mouse condition than in the keyboard condition is consistent with this idea. The use of a spring-loaded joystick instead of a mouse could solve the problem, as feedback would be provided in the form of muscle tension caused by the resistance of the joystick.

EXPERIMENT 5

In the future people may read novels, journals, and other large bodies of text from computer screens (Duchnicky and Kolers, 1983; Wright, 1987). In Experiment 5 each subject read text presented either in RSVP with selfpacing and regressions or in page format for more than two hours in one session.

Method

Subjects. In exchange for their participation, 32 university students were paid $20.00.

Materials and apparatus. Forty-seven short stories were taken from The Complete Works of Saki (Munro, 1976). The stories were arbitrarily divided into two sets. This reading material was previously used by Muter et al. (1982).

The apparatus was the same as in Experiment 4. Speed of presentation was controlled by the mouse.

Design and procedure. The independent variables were presentation method, page format or RSVP, and set number (1 or 2). Presentation method was a between-subjects variable and set number was a within-subjects variable. The dependent measures were reading speed, comprehension, and subjective estimates of dizziness, headache, nausea, fatigue, eyestrain, and desire to read more.

Subjects, assigned alternately to the page or RSVP condition, were asked to read printed instructions. The instructions stated that the subject was to read with as much speed and comprehension as possible for 70 min (Set 1), after which he or she could take a 5-min break. After the break the subject was required to read using the same method for another 70 min (Set 2). Subjects could take short breaks at any time, though they were asked to keep breaks to a minimum because break times were included when computing reading rates. Subjects read for 3 min in the appropriate condition for practice.

The RSVP condition included self-pacing (via mouse) and two kinds of regressions - two-word regressions and sentence regressions. In addition to the usual instructions, subjects were told how to stop the presentation and thereby take a short break ("press the period key"); text presentation could be resumed by pressing any key.

When reading via page format subjects were instructed to press the space bar after they had read the contents of the screen to continue to the next screenful. Sentences did not cross page boundaries. At the end of each set subjects were asked to indicate to the experimenter the last word they had read.

The initial presentation rate in the RSVP condition was approximately 120 wpm. This rate was chosen because it was an approximation to the subjects' final mean reading speed via RSVP in Experiment 4. The maximum possible rate was approximately 450 wpm and the minimum possible rate was approximately 50 wpm. Sentences and story titles were followed by a 500-ms pause and each story was followed by a 3-s pause.

In the page condition the format of the text presented on the screen was similar to the format of the text in The Complete Works of Saki. For example, text that was centered or indented in the book was centered or indented on the screen. There was no hyphenation, however, and the number of words per line was different from the number in the book.

Each subject was left alone in the room during the reading periods. The end of each 70-min period was signaled by a beep from the computer. In the page condition the experimenter counted the number of words that the subject had read on the final screen. This total was then added to the total reported by the computer program (the program excluded the current screen when calculating the number of words presented).

After the second 70-min reading period, subjects were given 15 min to answer 30 short-answer comprehension questions, three on each of the first five stories in each of the two sets. (Only three questions were asked per story to allow as much time as possible for reading.) In addition, subjects were given a questionnaire that asked for subjective judgments - on a seven-point scale - of dizziness, headache, nausea, fatigue, and eyestrain both before and after completing the experiment. Subjects also judged their desire to read more.

Results and Discussion

Reading was approximately twice as fast in the page condition as in the RSVP condition in both Set 1 and Set 2 (Table 4). This effect was highly reliable, F(1,30) = 36.60, p < 0.001. As in Muter et al. (1982), reading was significantly faster in Set 2 than in Set 1, F(1,30) = 17.49, p < 0.001, and there was no interaction between presentation method and set, F(1,30) < 1.

With regard to comprehension, there were no effects of presentation method, F(1,30) < 1; set, F(1,30) = 1.75; or the interaction, F(1,30) = 1.31. Mean comprehension scores were between 4 and 5 out of a possible score of 9. Because the slowest reader read only three stories per set, only questions on these stories were included in the results. The comprehension tests, therefore, may have been less sensitive than in the earlier experiments, and the possibility of a differential speed-accuracy trade-off in Experiment 5 cannot be ruled out - subjects may have chosen to sacrifice speed for comprehension in the RSVP condition more than in the page condition.

From Experiment 1 we know that starting rate can influence later reading rate, and the results in Experiment 5 may have been affected by the relatively low starting rate of 120 wpm in the RSVP condition. However, this was the final rate in the previous experiment with similar subjects. Furthermore, the reading materials in Experiment 5 were more difficult than those in Experiment 4. If anything, a more reasonable starting rate in Experiment 5 would have been less than 120 wpm.

Keypress data are given in Table 5. Once again sentence regressions were preferred to two-word regressions, t(15) = 2.14, p < 0.05.

As measured by subjective reports (see Table 6), there was some buildup of discomfort - particularly fatigue and eyestrain - but there were no reliable differences between the two conditions. (This pattern is similar to that obtained by Muter et al. [1982] in comparing extended reading from a book to extended reading from a computer screen.) There was little desire to read more in either condition, and the difference was not significant.

GENERAL DISCUSSION

Techniques of permitting reader control in computerized text presentation were tested in five experiments. Results suggest that under various circumstances, permitting self-pacing and regressions in RSVP is feasible. In some conditions regressions had a negative effect on reading speed, but otherwise performance and comfort were not adversely affected by the presence of reader control.

In Experiment 1 RSVP with self-pacing was compared with page format reading. In the RSVP condition starting rate had a substantial effect on how subjects chose to trade off speed and comprehension. Subjects who were given a fast starting rate maintained a reading speed in RSVP faster than in any other condition. They paid a price by having poorer comprehension. Subjects who started at a slower rate - comparable to the starting rate in the page condition - performed equally well in the two conditions.

In Experiment 2 subjects had regression capability in RSVP in addition to self-pacing and were able to regress two, four, or eight words depending on the condition. The larger the regression, the slower the reading. In Experiment 3 subjects had the option of two kinds of regression simultaneously; however, no reliable effect on reading speed or comprehension was found. Subjects executed more sentence regressions than two-word regressions. Experiment 4 assessed input devices; there was no difference in reading speed or comprehension caused by the nature of the control over reading speed, but subjects exhibited a greater tendency to change speeds in the mouse condition than in the keyboard condition.

In Experiment 5 subjects read for 140 min in a page condition or in an RSVP condition with self-pacing and regressions. Questionnaire results indicated a buildup of discomfort, but this buildup was not different in the two conditions. Reading speed was approximately twice as fast in the page condition as in the RSVP condition.

Several interpretations can be made of subjects' slow reading in the RSVP condition in Experiment 5. First, the materials were more difficult than in the earlier experiments. Buttigieg and Muter (1987) found that reading speed in RSVP is more strongly affected by passage difficulty than is reading speed in page format. Second, the greater length of the reading session may have caused readers to trade off comprehension and speed differently in the unfamiliar condition in Experiment 5. Third, more effort may be required to read text in the unfamiliar RSVP condition than in page format. This difference in effort may be negligible when reading short texts but may accumulate during extended reading. Fourth, there is evidence that perceptual processes are different in RSVP than in page reading (Buttigieg and Muter, 1987), and these may exert an effect on reading speed in an extended session.

Regressions in the RSVP conditions were different in several ways from normal regressions in page format. Regressions in RSVP may be more disorienting than in normal reading in which the spatial layout of the text gives the reader a clear idea of where he or she regressed to in relation to other parts of the text. Furthermore, in the present experiments the repeated words were presented at the usual rate, and the reader did not have the option of skimming through all or part of the reviewed material or of proceeding extremely slowly. Finally, regression size was constrained. Normally regression size is completely under the control of the reader and allows for linguistic units such as sentences and phrases to be taken into account. More powerful computer programs with automatic parsers (Granaas, 1985) may solve some of these problems.

Permitting regressions had a negative effect on RSVP reading in the present experiments under some conditions. It is possible that regressions in general have a detrimental effect on reading performance - determining when to regress may consume an undue amount of cognitive processing capacity, and the comprehension gained by a regression may not compensate for the time lost. Some reading training programs recommend no regressions at all (Carver, 1971). It may be impossible to test the effect of regressions in normal reading because of the difficulty of exerting experimental control. In any event it is probable that in some applications readers of dynamic text will want to have regression capability.

Duchaestel has argued that "textual display is and will remain an art" (1982, p. 167). We submit that the experimental method can also be brought to bear with profit and that the present experiments provide an example. With more research to optimize various aspects of computer-supported text presentation, we may see an important increase in the bandwidth of the communications channel from the computer to the reader.

ACKNOWLEDGMENTS

This work was supported by grants U0149 and G1779 from the Natural Sciences and Engineering Research Council of Canada. We thank Susanne Ruschka, Miriam Strzinar, and Linda Tilley for assistance, and Ken Dion, Norm Slamecka, Ian Spence, and especially Nancy Williamson for helpful comments.

REFERENCES

Buttigieg, M. A., and Muter, P. (1987). Rapid serial visual presentation and word-shape. In Proceedings of the Fourth Mid-Central Ergonomics/Human Factors Conference (pp. 46-52). Urbana: University of Illinois.

Card, S. K., Moran, T. P., and Newell, A. (1983). The psychology of human-computer interaction. Hillsdale, NJ: Erlbaum.

Carver, R. P. (1971). Sense and nonsense in speed reading. Silver Spring, MD: Revrac.

Cocklin, T. G., Ward, N. J., Chen, H.-C., and Juola, J. F. (1984). Factors influencing readability of rapidly presented text segments. Memory and Cognition, 12, 431-442.

Crowder, R. G. (1982) The psychology of reading. New York: Oxford University Press.

Duchaestel, P. C. (1982). Textual display techniques. In D. H. Jonassen (Ed.), The technology of text (pp. 167-191). Englewood Cliffs, NJ: Educational Technology Publications.

Duchnicky, R. L., and Kolers, P. A. (1983). Readability of text scrolled on video display terminals as a function of window size. Human Factors, 25, 683-692.

Forster, K. I. (1970). Visual perception of rapidly presented word sequences of varying complexity. Perception and Psychophysics, 8, 215-221.

Gould, J. D., and Grischkowsky, N. (1984). Doing the same work with hard copy and with cathode-ray tube (CRT) computer terminals. Human Factors, 26, 323-337.

Granaas, M. (1985). Simple, applied text parsing. Behavior Research Methods, Instruments, and Computers, 17, 209-216.

Granaas, M. M., McKay, T. D., Laham, R. D., Hurt, L. D., and Juola, J. F. (1984). Reading moving text on a CRT screen. Human Factors, 26, 97-104.

Juola, J. F., Ward, N. J., and McNamara, T. (1982). Visual search and reading of rapid serial presentations of letter strings, words, and text. Journal of Experimental Psychology: General, 111, 208-227.

Just, M. A., and Carpenter, P. A. (1980). A theory of reading: From eye fixations to comprehension. Psychological Review, 87, 329-354.

Kolers, P. A., Duchnicky, R. L., and Ferguson, D. C. (1981). Eye movement measurement of readability of CRT displays. Human Factors, 23, 517-527.

Kruk, R. S., and Muter, P. (1984). Reading of continuous text on video screens. Human Factors, 26, 339-345.

Masson, M. E. J. (1983). Conceptual processing of text during skimming and rapid sequential reading. Memory and Cognition, 11, 262-274.

Munro, H. H. (1976). The complete works of Saki. Garden City, NY: Doubleday.

Muter, P., Latremouille, S. A., Treurniet, W. C., and Beam, P. (1982). Extended reading of continuous text on television screens. Human Factors, 24, 501-508.

Potter, M.C. (1984). Rapid serial visual presentation (RSVP): A method for studying language processing. In D. E. Kieras and M. A. Just (Eds.), New methods in reading comprehension research (pp. 91-118). Hillsdale, NJ: Erlbaum.

Potter, M. C., Kroll, J. F., and Harris, C. (1980). Comprehension and memory in rapid sequential reading. In R. S. Nickerson (Ed.), Attention and performance VIII (pp. 395-418). Hillsdale, NJ: Erlbaum.

Sekey, A., and Tietz, J. (1982). Text display by "saccadic scrolling." Visible Language, XVI, 63-76.

Tinker, M. A. (1963). Legibility of print. Ames: Iowa State University Press.

Williamson, N. L. (1985). Rapid serial visual presentation: Some theoretical and applied implications. Unpublished master's thesis, University of Toronto.

Williamson, N. L., Muter, P., and Kruk, R. S. (1986). Computerized presentation of text for the visually handicapped. In E. Hjelmquist and L.-G. Nilsson (Eds.), Communication and handicap: Aspects of psychological compensation and technical aids (pp. 115-125). Amsterdam: Elsevier Science Publishers.

Wright, P. (1987). Reading and writing for electronic journals. In B. K. Britten and S. M. Glynn (Eds.), Executive control processes in reading (pp. 23-55). Hillsdale, NJ: Erlbaum.

Wright, P., and Lickorish, A. (1983). Proofreading texts on screen and paper. Behaviour and Information Technology, 2, 227-235.