Citation
An electrophysiological and behavioral analysis of cognitive load during reading

Material Information

Title:
An electrophysiological and behavioral analysis of cognitive load during reading
Creator:
Raney, Gary Ernest
Copyright Date:
1990
Language:
English

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright Gary Ernest. Raney. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Resource Identifier:
AHN8819 ( LTUF )
23587964 ( OCLC )
0024308899 ( ALEPH )

Downloads

This item has the following downloads:


Full Text
AN ELECTROPHYSIOLOGICAL AND BEHAVIORAL ANALYSIS
OF COGNITIVE LOAD DURING READING
By
GARY ERNEST RANEY

A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA

1990




ACKNOWLE DGMENTS
I would like to thank the members of my committee, Ira Fischler, Jeff Farrar, Richard Griggs, Keith White, and Don Childers, for their help and guidance with this dissertation. A special note of thanks needs to be given to my advisor, Ira Fischler. Ira has helped me to learn how to conduct research, he has allowed me to pursue my own interests, he has provided guidance and support whenever I needed it, and as importantly, he has treated me as a friend as well as a student. Additional thanks need to be given to my parents, who are a constant source of support. An immeasurable amount of thanks needs to be given to my wife, Jill, for helping me make this possible. I give her my thanks and my never-ending love.




TABLE OF CONTENTS
page
ACKNOWLEDGMENTS ......................................... ii
ABSTRACT ................................................. v
CHAPTERS
1 INTRODUCTION ....................................... 1
Overview of Dissertation ........................... 3
The Secondary Task Procedure ....................... 4
Reaction Time Studies of Cognitive Load ............ 6
Event-Related Potential Studies of Cognitive Load.14
Description of ERPs ............................ 14
ERPs and Cognitive Load ........................ 15
2 EXPERIMENT 1 ...................................... 19
Introduction ...................................... 19
Method ............................................ 26
Subjects ....................................... 26
Materials and Apparatus ........................ 26
Procedure ...................................... 31
Results ........................................... 36
Comprehension Quiz Data ........................ 36
ERP Data ....................................... 36
Reaction Time Data ............................. 48
Discussion ........................................ 49
Experiment 1B ..................................... 53
Results and Discussion ............................ 55
Experiment 1C ..................................... 56
Results and Discussion ............................ 57
General Discussion ................................ 59
3 EXPERIMENT 2 ...................................... 68
Introduction ...................................... 68
Method ............................................ 77
Subjects ............. 77
Materials, Apparatus and Procedure ............. 77
Results ........................................... 80
Recall Data .................................... 80
ERP Data ....................................... 83
Discussion ........................................ 92
iii




Experiment 2B ......
Results and Discussion .............
Recall Data .....................
ERP Data ........................
Experiment 2C ......................
Results and Discussion .............
General Discussion .................
4 SUMMARY AND CONCLUSIONS ............
Text Difficulty ....................
Text Structure .....................
Future Research ....................
APPENDIX A PASSAGES USED FOR EXPERIMENTS
APPENDIX B SENTENCES USED IN THE READING
APPENDIX C PASSAGES USED FOR EXPERIMENTS
AND 2C .......................

..94
..95
..95
..96
.100 .100 .103
.106

............
............
............
............
............
............
............

. . . . . . . .
. . . . . . . 1
.............. 113
1 AND 1C ..... 117 SPAN TEST .... 120 1B, 2, 2B, .............. 125

.129 .135

REFERENCES ............................................
BIOGRAPHICAL SKETCH ...................................




Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
An Electrophysiological and Behavioral Analysis
of Cognitive Load During Reading By
Gary Ernest Raney
August, 1990
Chairman: Dr. Ira Fischler
Major Department: Psychology
The effects of text difficulty and structure on cognitive load during reading were examined using a secondary task procedure. The Nl-P2 component of the event-related potential and reaction time to secondary auditory probes were used as measures of load. In three experiments, subjects read a text twice while probes were presented. Experiment 1 found larger Nl-P2 responses to probes during the second than first reading for high "reading span" subjects, even when no response to the probes was required. When shorter passages were read in Experiment 1B, larger Nl-P2 amplitude during the second reading occurred for high and low reading span groups. In Experiment 1, probe detection latencies were measured for other subjects. Detection times were longer during the second reading. Experiment 1C showed the pattern of detection times was not due to a lack of response




competition between the reading and probe detection task. The Nl-P2 data indicate reduced cognitive load during the second reading, while detection times indicate the opposite. It was concluded that Nl-P2 amplitude and probe detection times indicate different aspects of cognitive load. The Nl-P2 reflects early occurring, lower-level processes, while detection time reflects later occurring, higher-level processes.
The levels effect, which refers to the greater recall of high- versus low-importance information after reading, was examined in three additional studies. It was hypothesized that more resources are devoted to encoding high-importance information, which should lead to increased processing load. In Experiment 2, no levels effect was found in free recall protocols or in the Nl-P2 response to probes embedded in high- or low-importance information. Experiment 2B found that when subject strategies led them to focus on the main points, Nl-P2 amplitude was smaller when processing high-importance material, but only for high reading span subjects. Experiment 2C found no difference in detection times for probes embedded in high- or lowimportance information. It was concluded that if the task is structured in a manner that does not limit subjects' strategies and allows them to freely allocate their resources, additional resources may be devoted to the processing of high-importance material.




CHAPTER 1
INTRODUCTION
Reading is an extremely complex and important skill
which is used by most individuals every day of their adult lives. The importance of reading is reflected in the amount of research conducted to try to understand this skill. One important area of research concerns the relation between the amount of cognitive effort put forth during reading and subsequent memory and comprehension of the material being read. The notion of cognitive effort may be thought of in terms of an analogy to physical effort. Just as a heavy weight demands more muscular exertion to be lifted than does a light weight, a passage of text with many complex or difficult ideas may demand more mental effort to process than does a relatively simple passage. The amount of mental effort required is commonly referred to as the cognitive load.
Cognitive load has been used as an umbrella term to describe the types and amount of mental muscle, or what cognitive psychologists call resources or cognitive capacity, required to perform a task such as reading. These resources include attention, capacity, effort, storage and retrieval processes, and processing strategies, to name a few (e.g., Britton & Tesser, 1982; Kahneman, 1973; Navon & Gopher, 1979). The central theme




in the area of reading research is that different types of text material, as well as different sections within a text, may require different amounts of one's cognitive capacity. Increased capacity requirements during reading are taken as an indicator of greater cognitive load (Britton & Tesser, 1982; Britton & Price, 1981).
Within this dissertation, the concept of cognitive
load will be used to examine theories of reading and text comprehension. Two issues will be addressed. The first concerns the effect of text difficulty on cognitive load. The general idea is that difficult texts should require greater effort to process than relatively easy texts. If true, then a measure of cognitive load should reflect this difference. Prior research addressing this issue is inconclusive.
The second issue addressed is known as the "levels effect." The levels effect refers to the fact that when individuals recall a text, they tend to recall a greater proportion of the main ideas than supporting ideas or details. Some theories predict that the main ideas of a text are processed differently, and possibly to a greater extent, than supporting ideas and details (e.g., Kintsch & van Dijk, 1978; Meyer, 1975). It is appropriate to ask if this differential processing requires more attention or effort. If more attention is required, then a measure of cognitive load should reflect this extra processing. The research in this area is also inconclusive.




There is an abundance of literature addressing the
effect of text difficulty on cognitive load and the levels effect. However, all of the studies are based on behavioral measures, such as recall tests and reaction time to detect a secondary event while reading. The present studies will make use of these behavioral measures, but will also use an electrophysiological measure of cognitive load based on the Event-Related Potential (ERP). It is hoped that by using converging measures that some clarity will be brought to the research on cognitive load and reading.
Overview of Dissertation
The remainder of this chapter will be organized as
follows. First, since the majority of research described in the following review, as well as each of the studies performed for this dissertation, were based on a version of the secondary task procedure, a brief review of this methodology will be given. Second, research involving the effect of text difficulty on cognitive load will be examined. In Chapter 2 I will introduce the method used in the first experiments and describe their purpose. Three studies which examine the relation between text difficulty and cognitive load will then be presented. The literature on the levels effect will be reviewed in the introduction to Chapter 3. Three studies which examine the effect of informational importance (the levels effeC*) on cognitive load will then be presented. Chapter 4




provides a summary of the primary results and final conclusions.
The Secondary Task Procedure
The secondary task procedure, which is also called a dual task procedure, has been frequently used to measure the cognitive load or capacity requirements to perform a task. In the secondary task procedure a primary task is performed concurrently with a secondary task. The assumption underlying the secondary task method is that individuals have a limited amount of mental resources, and the primary and secondary tasks will compete for these limited resources (Kahneman, 1973; Navon & Gopher, 1979). Kahneman (1973), Kerr (1973), and Wickens (1983) each provide detailed reviews of the secondary task method.
Kahneman (1973) originally proposed that there was a single pool of resources from which the primary and secondary tasks drew. Navon and Gopher (1979) subsequently proposed a multiple resource view. They suggested the existence of separate pools of resources for different processes, although the total amount of resources is still limited. For example, there may be separate pools of perceptual resources, cognitive resources, and response resources. The multiple resource view allows different secondary tasks to have different levels of competition for resources with the primary task.
Regardless of whether a single or multiple resource
view is taken, the logic of the method is the same; as the




cognitive demands of the primary task increase, there will be less cognitive capacity available for performing the secondary task. When the resources available for completing the secondary task are reduced, performance on the secondary tasks should decrease. Decreased performance on the secondary task is taken as an indicator of increased demands or cognitive load in the primary task. Multiple resource models also require that the primary and secondary tasks compete for common resources. Thus, if the primary and secondary task do not overlap in any manner, changes in the primary task should not affect performance on the secondary task (Wickens, 1976).
The most common secondary task used in studies of reading is secondary probe detection. In this task, subjects read a text (the primary task) while a series of unrelated secondary probes are presented for the subject to respond to (the secondary task). The probes are usually short duration (100 msec or less) tones or light flashes. Both behavioral and Event-Related Potential (ERP) measures of performance have made use of this methodology. In behavioral studies, reaction time to detect the probes is recorded. Increased reaction time to detect a probe is taken as indicating greater cognitive demands in the primary task. In electrophysiological studies, the ERP response to the probes is recorded. A smaller amplitude ERP response indicates greater cognitive demands by the primary task. Different "components" of




the ERP may be sensitive to different aspects of cognitive load. This will be discussed in greater detail later in this dissertation.
Reaction Time Studies of Cognitive Load
The secondary probe task has been extensively used to study changes in cognitive load during reading (see below). The working hypothesis of these studies was that as the demands of the reading task increase, reaction time to detect a secondary probe should also increase. Britton and Price (1981) validated this assumption by examining the performance operating characteristic of this procedure. Performance operating characteristics describe the pattern of performance changes in one task as the characteristics of a second task are varied. Britton and Price instructed subjects to focus different amounts of attention on reading (the primary task) and probe detection (the secondary task). Their results indicated that as more attention was given to the reading task, probe detection latencies increased. They concluded that the probe detection task appears to be sensitive to the amount of attentional capacity given to a primary reading task.
Easy vs. Difficult Text. In a series of studies by Bruce Britton and his colleagues, subjects read passages that differed in difficulty while performing the secondary probe detection task. In Britton, Westbrook, and Holdredge (1978) and Britton, Zeigler, and Westbrook




(1980) texts were short passages which ranged in difficulty from primary (easy) to college (difficult) level. In Britton, Holdredge, Curry, and Westbrook (1979) a common set of passages was presented with a title, which made them easy to understand, or without a title, which made them difficult to understand. Thus, in these three studies semantic difficulty was manipulated. Surprisingly, the results of each study showed longer probe reaction times during the reading of easy text than during the reading of difficult text. Based on the premises of the secondary task technique, the authors concluded that easy text filled cognitive capacity more than difficult text.
Britton and his colleagues have also used secondary
task performance to measure cognitive load when structural aspects of a text are manipulated. In Britton, Glynn, Meyer, and Penland (1982), subjects read passages composed of simple or complex vocabularies and syntax while holding passage content (meaning) constant. They also presented passages which did or did not contain cues to signal the importance of ideas in the text (e.g., "therefore," "in contrast"). Reaction time to the probes was slower for text with complex syntax and for text without signals. That is, reaction time was longer for the difficult conditions. In contrast, in Britton et al.'s prior studies probe response times were faster for difficult text than for easy text.




The results of Britton Glynn, Meyer, and Penland
(1982), which showed faster probe responses when reading easy text, appears to be in conflict with Britton et al.'s (1978, 1979, 1980) earlier studies, which found just the opposite. One explanation proposed by Britton, Glynn, Meyer, and Penland (1982) to explain this difference was referred to as the cognitive contents hypothesis. According to this hypothesis, the amount of capacity used when reading is a function of the amount of information activated. As more information is activated by the primary task, more capacity is used, which leaves less capacity for the secondary task. Since easy text is more readily understood, more propositions (from the text and memory) may be activated. This would leave less capacity to respond to the probes, producing longer reaction times.
A study by Britton and Tesser (1982) provides another explanation which they called the prior knowledge hypothesis. They suggested that prior knowledge uses some of the individual's limited cognitive capacity just as performing the task uses capacity. Thus, as the amount of prior knowledge is increased, there will be less capacity available for performance of the secondary task. To test this, subjects read passages with and without prior knowledge of the material. Secondary probe detection times were longer when prior knowledge was given. This supports the hypothesis that prior knowledge uses cognitive capacity.




The prior knowledge hypothesis can be used to explain the apparent inconsistency between Britton et al.'s early (1978, 1979, 1980) and later (Britton et al., 1982) work. In Britton, Holdredge, Curry, and Westbrook (1979), difficulty was manipulated by including (easy) or excluding (difficult) the title to the story. Since the passages were constructed to make the topic difficult to perceive without a title, little prior knowledge could be employed when no title was given. This would lead to less capacity being used in the difficult condition. In Britton, Westbrook, and Holdredge (1978) and Britton, Zeigler, and Westbrook (1980), difficulty was defined by the predictability of the upcoming material using the dloze procedure. They suggest that since easy texts were more predictable than difficult texts, more prior knowledge could be applied to the easy texts. This would lead to easy texts filling more capacity. In Britton, Glynn, Meyer, and Penland (1982) semantic content was held constant while syntactic complexity was varied. In this case increased complexity did require more capacity. Since semantic content was held constant for easy and difficult conditions, equal amounts of prior knowledge could be instantiated. Thus, the difficult syntax would simply add to the processing capacity requirements.
Inhoff and Flemming (1989) suggest another
explanation which is based on a methodological problem in Britton et al.'s studies. Namely, subjects may have




reread portions of the texts in order to understand the material, which could reduce cognitive load. Since in Britton et al.'s studies probes were applied on a random basis, probes could occur after rereading a segment of a text. This "second reading" may have made the portion of a text following a rereading easier to understand, and, consequently, reduce cognitive load. If a probe occurred after a rereading, it would reflect an artificially reduced cognitive load.
There is evidence that rereading makes a text easier to process. Levy and her coworkers (Levy & Begin, 1984; Levy, Newell, Snyder, & Timmins, 1986; Levy & Kirsner, 1989) have analyzed the changes in processing which occur during repeated readings of a text. In a series of studies based on proofreading tasks, they have shown that processing efficiency increases at both the perceptual level, such as converting print to words, and the cognitive level, such as comprehending the general meaning of a passage, during a second reading of a text. Therefore, if in Britton et al.'s studies more rereadings occurred for the difficult texts, then cognitive load would be reduced and probe reaction times could be decreased. Research has shown that more regressions (rereadings) to earlier parts of a text do occur when reading difficult text than when reading easy text (Inhoff, 1983; Just & Carpenter, 1987). Furthermore, if more rereadings occurred for difficult text, then reading




times should be greater for difficult than for easy text. In each of Britton et al.'s studies reading time was greater for the difficult texts, supporting the possibility of increased rereadings.
As in prior studies, Inhoff and Flemming (1989)
presented subjects with easy and difficult passages to read for comprehension while performing the secondary probe detection task. Passage difficulty was determined by having a group of subjects rate the general difficulty of a passage. There was an important procedural difference between this study and the earlier work by Britton et al. In Britton et al.'s studies, an entire passage was presented for the subjects to read in a normal manner. Inhoff and Flemming (1989) used a self-paced, single word presentation format. In this procedure, only one word at a time was seen and subjects pressed a key to make successive words appear. This eliminated the possibility of rereading prior parts of a text.
The results of Inhoff and Flemming's (1989) study
showed faster probe reaction times for easy text than for difficult text. These results are in direct opposition to Britton et al.'s early work (1978, 1979, 1980). In order to further test the idea that rereading decreases cognitive load, a mini-experiment (N = 5) was conducted in which subjects read a passage twice. Four of the subjects had shorter probe reaction time during the second reading. For the fifth subject there was no difference. All




subjects read slower during the first reading. This is consistent with the idea that rereading may reduce the difficulty of a text and suggests that this possibility needs to be controlled.
Inhoff and Flemming (1989) discussed their results in terms of a modified cognitive contents hypothesis. According to the contents hypothesis, propositions from the text and memory are activated during reading. If difficult text activated more propositions, then more capacity would be used and responses would be slowed. Recall that Britton, Glynn, Meyer, and Penland (1982) suggested that easy text activates more propositions. Research by Kintsch and van Dijk (1978), however, supports the idea that difficult text may lead to the activation and search of more propositions than easy text. This supports Inhoff and Flemming's view of the cognitive contents hypothesis.
The prior knowledge hypothesis was not addressed by Inhoff and Flemming (1989), but may be applied to their study. There are some differences between Inhoff and Flemming's materials and Britton et al.'s materials. First, Inhoff and Flemmning defined difficulty by having subjects rate the overall difficulty of the passage. Britton et al. (1978, 1980) defined difficulty in terms of dloze probability (word predictability) and by the presence or absence of a title (i.e., knowing or not knowing the topic) in Britton et al. (1979). As




previously mentioned, Britton et al.'s easy passages may have instantiated more prior knowledge. For Inhoff and Flemming's results to be consistent with the prior knowledge hypothesis, it would have to assumed that the difficult passages led to the use of more prior knowledge than the easy passages. The amount of prior knowledge is, of course, an unknown, but the origin of their materials raises this possibility.
Inhoff and Flemming's passages were excerpts from out-of-print introductory psychology textbooks, weekly news magazines, and newspapers. Passages dealt with topics such as an experiment at a hospital in Wisconsin and brain regulatory mechanisms. The topics of both the easy and difficult texts may be considered somewhat unfamiliar, making it possible that no more prior knowledge could be applied to the easy than difficult passages. If this was true, then the added semantic and syntactic complexity of the difficult passages would lead to increased cognitive load.
It should be noted that explanations based on
differences in the materials are post hoc. It is also possible that these changes do not matter, and that the important difference is methodological. In Britton et al.'s studies the passages were read in a normal fashion while in Inhoff and Flemming's study the passages were read one word at a time. Thus, the opposite pattern of results may be in part due to the manner in which the




materials were presented. Furthermore, the prior knowledge hypothesis is inconsistent with the results of Inhoff and Flemming's rereading study, which found faster probe response during the second reading of a passage.
In summary, there is evidence that easy text fills more cognitive capacity than difficult text (Britton et al. 1978, 1979, 1980), and vice versa (Britton et al., 1982; Inhoff and Flemming, 1989). Two explanations were proposed to explain this pattern of conflicting results. The first is the possibility of rereading prior material. The second is the interaction of prior knowledge with the semantic and syntactic aspects of the reading material. Future research needs to control for each of these possibilities.
Event-Related Potential Studies of Cognitive Load Description of ERPs
Event-related potentials are a measure of the
electrical activity generated by a collection of neurons. The ERP is described in terms of the peaks and troughs of the waveform, which are referred to as components. The latency and polarity of a component are then used to describe the ERP wave. For example, a negative (N) going wave which occurs at least 100 msec after stimulus onset is labeled an N100. Components are sometimes combined if an event typically influences the components together. For example, the NlOO-P200 (abbreviated Nl-P2) is a negative followed by a positive wave.




Early components (typically 100 msec or less after stimulus onset) are called exogenous components and primarily reflect physical aspects of a stimulus, such as intensity. Later components are called endogenous components and primarily reflect cognitive aspects of a stimulus, such as whether a word makes sense within a sentence (Hillyard & Woods, 1979).
Components are also distinguished by their
sensitivity to experimental manipulations and their distribution across the scalp. The NI-P2 has been shown to be responsive to a subject's attentional focus within a task and is maximal (in amplitude) at the vertex (Naatanen, 1988; Picton & Hillyard, 1975). The P300 is sensitive to rare or unique events and some memory processes and is maximal at central posterior locations (Gopher & Donchin, 1986). The experiments performed for this dissertation were designed around the Nl-P2 component of the ERP. Therefore, the following review will focus on this component.
ERPs and Cognitive Load
Event-related brain potentials have been used as an indicator of cognitive load, although they have not been applied directly to the study of reading. The typical ERP study of cognitive load uses the secondary probe method in the same manner as behavioral (reaction time) studies. A primary advantage of ERPs is that they do not require an overt response to be recorded, although both behavioral




and electrophysiological (ERP) measures are usually gathered. A number of properties of ERPs may be evaluated. The most common are peak amplitude and latency of specific ERP components.
As with behavioral versions of the secondary task method, the sensitivity of the ERP to variations in the attentional demands of the primary and secondary tasks needs to be demonstrated. Picton and Hillyard (1974) were the first to show that attending to a series of stimuli (tones) enhanced the amplitude of the Nl-P2 relative to nonattended stimuli. Schwent and Hillyard (1975), Hink, van Voorhis, Hillyard, and Smith (1977), and Hink, Hillyard, and Benson (1978) further showed the sensitivity of the Nl-P2 to selective attention using variations of a dichotic listening paradigm. In these studies, subjects were told to attend to or ignore stimuli in a specified sensory channel (e.g., left, right, or both). The authors found that the amplitude of the N1 reflects the distribution of attentional resources among competing inputs. They proposed a relation between the informational load on a subject and the size of the N1 enhancement for attended versus non-attended information. Each of the above studies found the effect to be maximal at the vertex.
Note that some of the prior studies focused on the N1 component only. Depending on the methodology used, the N1 and P2 can be separated (Picton, Hillyard, Krauz, &




Galambos, 1974). Subsequent work has verified the relation between selective attention and the amplitude of both the N1 and P2 components of the ERP (Hillyard, 1985; Naatanen & Michie, 1979; Nakahara & Ikeda, 1987; Parasuraman, 1978, 1980). These studies consistently support the conclusion that the amplitude of the NI-P2 may serve as an index of the amount of cognitive resources devoted to processing a channel of information. It should be noted that the neural process responsible for enhanced N1 amplitude is still under debate (Hillyard, 1981; Naatanen, 1988; Naatanen & Michie, 1979), but the fact that amplitude changes do occur is not in question.
The concept of competing channels of inputs used in the prior studies can be applied to the secondary task procedure. In the secondary task there are two channels (tasks) which are competing for attentional resources. If attention is focused on the primary channel (task), then the amplitude of the NI-P2 response to probes in a secondary channel should vary depending on the amount of "residual" attention allocated the secondary task.
In summary, there is evidence that NI-P2 responses to a secondary probe reduce in amplitude as more attentional resources are devoted to the primary task. The research has not, however, directly addressed the issue of reading. Other research reviewed by Fischler and Raney (in press) has shown the ERP to be a sensitive measure of lexical, semantic, and syntactic factors in sentence processing.




18
Thus, there is reason to believe that ERPs may provide a method for evaluating cognitive demands during text processing.




CHAPTER 2
EXPERIMENT 1
Introduction
Experiment 1 was designed to explore the effects of text difficulty on cognitive load using both behavioral (reaction time) and electrophysiological (ERP) measures. There were two primary purposes for Experiment 1. The first was to try and clarify the relation between text difficulty and cognitive load. The second was to determine if ERPs could be used as a nonintrusive measure of cognitive load.
The procedure used may be called a read-reread
paradigm. Subjects read a text, and then immediately reread the same text. The first reading served as the difficult condition and the second as the easy condition. The passages were presented one word at a time in a common location on a display. This technique is known as the single-word Rapid-Visual-Serial-Presentation (RSVP) technique (Potter, 1984). While reading, subjects were presented with a series of secondary auditory probes. Two response conditions were included. In the first, both reaction time to detect the probe and ERP responses to the probe were measured. In the second condition, only ERP responses were measured.




To use a rereading paradigm, it must be demonstrated that the first reading is more cognitively demanding than the second reading. Recall that Levy and her colleagues (Levy & Begin, 1984; Levy, Newell, Snyder, & Timmins, 1986; Levy & Kirsner, 1989) have shown that both perceptual and cognitive processes become more efficient during a second reading of a text. Britton, Glynn, and Smith (1985) also describe the effects of rereading on processing load. They suggest that the output products of a first reading, such as the development of a schema for interpreting the material, will be available during the second reading. Access of text-related information should be facilitated, and cognitive load should be reduced. Also recall that Inhoff and Flemming (1989) had subjects reread a text while performing a probe detection task. They found shorter probe detection times during the second reading, which is suggestive of reduced cognitive load. Furthermore, reading times for a second reading of a text are shorter (Inhoff, 1983, Just & Carpenter, 1987). Thus, there is strong support for the contention that a second reading requires reduced effort.
Individual Differences in Cognitive Load. To explore the possibility of individual differences in cognitive load, subjects were grouped based on their individual reading span scores (Daneman and Carpenter, 1980). In the reading span test, subjects read sentences and are asked to remember the final word of each sentence. Reading span




is taken as the maximum number of sentences a subject can read and then accurately remember their final words.
The reading span test was proposed as a measure which taxes both the processing and storage functions of working memory. Daneman and Carpenter (1980) suggested that one aspect of individual differences in reading ability may be working memory capacity. A good reader may have more efficient reading processes, which would leave more capacity available for memory. In contrast, poor readers may have less efficient processes, which would leave less capacity for memory (Daneman, 1987). The sensitivity to both processing and storage operations makes this measure particularly useful in the present study. While reading, the reader must simultaneously perform processing and storage operations. Thus, reading span taps two of the primary processes which may contribute to cognitive load.
Reading span has been shown to be significantly
related to traditional measures of reading ability, such as verbal SAT scores and to performance on reading comprehension tests (Daneman and Carpenter, 1980). In addition, individuals with high reading spans have been found to be better at resolving inconsistencies within sentences (Daneman and Carpenter, 1983), and constructing the meaning of an unfamiliar word from context (Daneman & Green, 1986).
Evaluation of the Method. There are a number of
benefits to the rereading paradigm and RSVP. First, since




the same passages are used in the easy and difficult condition, perfect control is obtained over syntactic, semantic, and vocabulary differences between easy and difficult texts. Second, using the RSVP method eliminates the possibility of regressions (rereading earlier parts of a text). Third, since the rate of presentation is constant, overall cognitive load for the difficult conditions cannot be decreased by increasing the reading time. This is important since Britton, Muth, and Glynn (1986) showed that when reading time is limited, increased effort will be allocated to difficult portions of a text. When reading time is not limited, subjects will slow down, effectively reducing the load at that point.
The use of ERPs as a measure provides the ability to address a primary criticism of the secondary task procedure; that simultaneously performing a primary and secondary task changes the normal processing demands of each task (Kahneman, 1973; Kerr, 1973; Wickens, 1983). To reduce the intrusiveness of a secondary task, it has been suggested that the task be based on discrete events which compete only for processing resources (i.e., no common response demands) and occur on a small proportion of trials during the primary task (Brown, 1978; Kerr, 1973). ERPs to secondary probes meet these requirements. In contrast with behavioral measures, ERPs provide a method for measuring behavior without any overt response to the secondary probe, which eliminates the secondary task




entirely, and makes the probe potentially less intrusive. This also eliminates competition for response resources and leaves competition only for processing resources. ERPS could thus offer a more valid measure of cognitive load than behavioral measures (Wickens, 1983).
It should be noted that most ERP studies have
required subjects to make an overt response to the probe; thus the competition for response resources is still a problem. One possible drawback of not requiring a response to the probes is that the probes may be ignored, which may reduce the effect of cognitive load on the ERP measure. This issue is examined in the present study by including ERP conditions with and without a response to the probes.
The restrictions suggested by Brown (1978) and Kerr (1973) do not invalidate the usefulness of behavioral measures. Consistent with their suggestions, a response to a secondary probe is a discrete event and occurs during a small proportion of trials. But since a response is required, there still may be competition for response resources. What is important is whether the primary and secondary tasks have common response demands (e.g., both require a response with the same hand). This is not the case in the previously discussed studies of reading. Furthermore, Britton, Piha, Davis, and Wehausen (1978) have shown that the addition of a secondary probe detection response does not lead to reduced performance on




the reading task. This implies that the response to the probes does not significantly disrupt reading, and is therefore a useful measure of cognitive load.
Hypotheses. The rereading paradigm provides a method for testing the cognitive contents hypothesis proposed by Inhoff and Flemming (1989) and the prior knowledge hypothesis proposed by Britton and Tesser (1982). The primary assumption of each of these hypotheses is that when less cognitive resources are required by the reading task, more attentional capacity should be available to respond to the probes (either overtly or covertly). According to the cognitive contents hypothesis, the second reading should require less capacity; therefore, more capacity should be available to respond to the probes. According to the prior knowledge hypothesis, the second reading should actually consume more capacity. since the first reading makes the passage familiar, a larger amount of prior knowledge should be invoked during the second reading. This would leave less resources available to respond to the probes. Thus, the cognitive contents hypothesis and the prior knowledge hypothesis lead to two different sets of predictions.
The following should occur if the cognitive contents hypothesis is correct. First, reaction time to detect the probes should be longer during the first reading of a passage. Second, the amplitude of the Nl-P2 ERP response to the probes should be smaller during the first reading.




Past research has found the effect to be largest at the vertex, and a similar pattern is expected here. Third, each of these effects should be mediated by an interaction with reading span. Specifically, if high reading span subjects are able to comprehend and remember more of a passage during the first reading, then the second reading will be needed to a lesser degree. In other words, the reduction in cognitive load will be greater for the high reading span group.
In contrast, the prior knowledge hypothesis predicts the following. First, reaction time to detect the probes should be faster during the first reading of a passage. Second, the amplitude of the N1-P2 ERP response to the probes should be larger during the first reading. The effect should be largest at the vertex. Third, a weak prediction is made for the main effects to interact with reading span. For an interaction to occur, it must be assumed that the high reading span group has more prior knowledge to apply than the low reading span group. This assumption is tenuous, at best. Therefore the interaction with reading span may not be as reliable as predicted by the cognitive contents hypothesis, which bases its prediction solely on memory capacity.
For both the cognitive contents and prior knowledge hypothesis, a main effect of reading span might be predicted. Since the high reading span individuals tend to be better readers (Daneman, 1987), the task may be




easier for them. This represents a between-subjects difference; thus the effect might not be strong. Furthermore, there is little evidence from past research regarding between groups differences for normal subjects in overall amplitude of the ERP.
Method
Subjects
Forty University of Florida undergraduates (15 men and 25 women) participated as part of an introductory psychology course requirement. All subjects reported being native English speakers and having normal or corrected-to-normal vision. Data from eight additional subjects were not included due to unacceptably high electrode impedances, technical problems, or experimenter error.
Materials and Apparatus
Stimulus Presentation System. An IBM PC AT
controlled the stimulus presentation. The text passages were displayed on a Samsung high resolution monitor (amber characters on a black background) which was connected to the monochrome output of a Color Graphics Adaptor (CGA) on the PC. The monitor was located 2/3 of a meter directly in front of the subject. While subjects read each passage, they heard a series of auditory probe tones. Probes were presented using the electronic circuit described by Raney and Shuman (1989), which was triggered by the PC. Probes were played through a 2 1/4 inch




speaker located approximately 1/2 meter behind the subjects.
Text Passages. Twelve passages were taken from GRE (Martinson & Crocetti, 1987), SAT (The College Board, 1988; Gruber, 1988), and ACT (Martinson, Fazzone, Haynes & Haynes, 1988) study manuals for use as stimuli in Experiment 1. These manuals were chosen since they contain well-written passages which have been standardized at an appropriate level of difficulty for college students. The passages were modified in three ways. First, passage length was adjusted, if necessary, to be between 418 and 437 words. Second, very rare or uncommon words were replaced with a simpler term or a short phrase which conveyed the same meaning. Third, an attempt was made to replace words with 12 or more letters with shorter length words or with a phrase which conveyed the same meaning. Words which could not be easily replaced were presented in two parts, that is, hyphenated (e.g., anthropologist was presented as "anthro-" "pologist"). The second and third modifications were needed to make the passages more readable within the single-word presentation method (described below).
The 12 passages were divided into two groups of six
(Group A and Group B). Subjects were alternately assigned to read passages from Group A or B. Two sets of passages were used in an attempt to increase the generalizability of the results, although passage group was not treated as




a factor in the experiment. After reading a text once or twice (described below) subjects completed a fivequestion, short answer quiz over the material. A sample passage and its corresponding quiz is presented in Appendix A. The source of each passage used in this experiment is also presented in Appendix A.
While subjects read each passage, a series of 10
auditory probe tones was presented. The probes were 100 msec, 1000 Hz, 60 dB tones. The intensity of the tones was determined using a Radio Shack, hand held sound level meter (model 33-2050). Intensity was measured by placing the meter in the same position as a subject's head would be during the experiment, and then reading peak intensity using the "All weighting scale.
The location of the probes within each passage was determined in the following manner. Each passage was divided into 10 approximately equal length sections. Within each section, probe location was randomly determined for each of the twelve passages based on the following five restrictions. First, probes were not presented during the first 10 words of a passage. Second, probes were presented only during the display of a content word (hereafter referred to as a target word). Third, probes were not presented during the first word of a sentence. Fourth, probe locations were different for the first and second reading of a passage (described below). Fifth, probes were not presented during the display of




either part of hyphenated words. All probes were presented 100 msec after the onset of a target word and the same probe locations were used for each subject.
Reading Span Test. The reading span test was
composed of 80, unrelated English sentences which varied in length from 9 to 13 words. Each sentence was typed on a 4 x 6 inch (10 x 15 cm) index card. The sentences were arranged into groups of 2, 3, 4, 5, and 6 sentences. There were four sets of each group size. Blank cards were inserted between each group to serve as recall markers (described below). Sixty of the sentences were created for the experiment, and 20 were taken from Daneman and Green (1986, Experiment 2). These 20 sentences were shortened in length to match the initial 60. The sentences used in the reading span test are presented in Appendix B. The final 20 sentences are those which were abstracted from Daneman and Green (1986).
EEG and RT Recording System. EEG and RT data
collection were controlled by the PC, which was equipped with a Scientific Solutions (Tecmar) Labmaster board. The Labmaster performed the analog-to-digital conversion of the EEG data, collection of the RT data, and the timing of the stimulus presentation during data collection. During the experiment, subjects were seated in a sound attenuated, electrically shielded booth. A circulation fan produced a background noise of approximately 50 dB.




EEG was recorded from CZ, F3, F4, and a left and
right hemisphere temporal-parietal location (which will be referred to as LS and RS, respectively) according to the International 10/20 System (Jasper, 1958). LS and RS refer to the area near the end of the sylvian fissure on the left (LS) and right (RS) hemisphere. Specifically, these points are located midway between T3 and P3 (for LS) and between T4 and P4 (for RS). The region around LS is more classically known as Wernicke's area (Geschwind, 1979). Each electrode was referenced to linked mastoids. Eye blinks were monitored using electrodes placed supraorbitally and behind the external canthus of the right eye. Eye blink artifacts in the EEG signal were corrected using the procedure described by Gratton, Coles, and Donchin (1983).
The EEG signals were amplified using Coulbourn
Hi-Gain Bioamplifiers (Model S75-01) with a bandpass of
0.1 to 40 Hz and a 60 Hz notch filter. The amplified EEG signal was sampled at a rate of 125 Hz per recording site and stored on disk for subsequent analysis. Beckman Ag/AgCl Biopotential skin electrodes and Beckman Electrolyte gel were used in all cases. Electrode impedances never exceeded 5 Kohms and were measured both pre- and post-test.
EEG data were recorded beginning 80 msec prior to the display of a target word and continued for 920 msec. This period encompassed the 80 msec prior to a target word




through the 40 msec after the offset of the word following the target. EEG data were collected only during the presentation of target words and words immediately following targets. The timing of the data collection period can be seen in Figure 1.
Probe detection responses were made using a Microsoft Mouse. The output of the mouse was input as a data channel into the Labmaster board and sampled in the same manner as the EEG data. Consequently, the temporal resolution of the probe detection data (i.e., reaction time) is 8 msec. The probe detection data were stored on disk for later analysis.
Procedure
All subjects were initially given the reading span test. In this test, subjects read sentences out loud in groups of 2, 3, 4, 5, or 6, and in this order. After the group size reached 6, the group size switched backed to 2 and again increased. This was repeated through four sets of each group size. The subject's task was to remember the last word of each sentence in a group. Subjects were instructed to turn over the index card after reading a sentence so that the sentence would no longer be visible, and then to begin reading the next sentence. This was continued until a blank card was reached, which served as a cue to verbally recall the sentence final words from the preceding group of sentences. Subjects were instructed to read at their own rate and not to take long pauses between




sentences. They were also instructed to try and recall the words in the order in which they were presented, but if they could not remember the order of presentation, they were encouraged to recall the words in any order. They were further told that it was better to recall many words out of order, than only a few words in proper order.
The average of the subjects three largest recalls was taken as a measure of their reading span. For example, if a subject correctly recalled groups of 4, 4, 3, and 4, then their score was 4. Prior to analysis, subjects were grouped based on their reading span scores. Those with a reading span score of 4 or greater were classified as high reading span. The remaining subjects were classified as low reading span.
Following the reading span test, subjects were
prepared for the EEG recording. Once this was completed subjects were seated in a recording booth and the main experiment began. Each subject was shown a set of passages to read, and following each passage they were tested for comprehension. The passages were presented one word at a time at a rate of 120 words-per-minute. Each word was displayed in the center of the monitor for 300 msec. The monitor screen was then cleared, and, following a 200 msec inter-word-interval, the next word was presented. An additional 500 msec delay was added after the last word of each sentence. Eye movement research has shown that short pauses are usually taken between




sentences (Just & Carpenter, 1987). The additional 500 msec delay between sentences was added to simulate this pause and make the task more natural.
To manipulate cognitive load, subjects read some of the texts twice in succession. The first reading served as the high cognitive load condition (i.e., difficult) and the second reading served as the low cognitive load condition (i.e, easy). A 5 to 10 second break was given between the first and second reading. While subjects were reading, they heard a series of auditory probes. Subjects were alternately assigned to either the response or noresponse condition. Those in the response condition pressed a key on the mouse as quickly as possible when they heard a probe. Those in the no-response group made no response to the probe. In all other respects, the procedures for the two groups were identical.
All subjects were read the following instructions prior to beginning the experiment.
"In this part of the study you will be presented with 6
short passages to read. Please read the passages
silently. Each passage only takes a few minutes to
read. The passages will be presented one word at a time
on the monitor in front of you. Each word will be
presented briefly in the center of the screen and then
removed before the next word is displayed. The words will appear and disappear quickly, so you will need to concentrate when reading the words. After reading each




passage you will take a five question, short answer quiz
over the material. You have ten minutes to complete each quiz. For half the passages, you will read the passages once and then take the quiz. For the other
half, you will read the passages twice and then take the
quiz. After each passage a message will appear
indicating that the passage is complete. At this point
I will tell you whether you will read the passage a
second time. I will do this using the intercom located
under the monitor. While you are reading, a series of
"beeps," or tones will be presented using a small
speaker located behind you. The beeps will occur at
different points within each passage."
Those in the no-response condition were then told:
"You do not need to make any response when you hear a
beep. You do not need to remember where the beeps
occurred. I want to emphasize that your primary task is
to read and comprehend the passages."
Those in the response condition were then told:
"Every time you hear a beep press the left key on the
mouse as quickly as possible. You do not need to
remember where the beeps occurred. Please try and
respond as quickly as possible to the beeps. However, I want to emphasize that your primary task is to read and
comprehend the passages."
Subjects were given a short demonstration passage
which contained six probes. The demonstration passage was




a paraphrase of the previous instructions. Subjects in the response group were instructed to respond to the probes in the demonstration in order to practice the response task.
Subjects were alternately assigned to read passages
from Group A or Group B. Although subjects were told they would read six passages, they actually read only four. Either the second or third passage presented was randomly chosen to be read only once. This was done to confirm the expectation that some of the texts would be read only once, which insured that subjects concentrated during the first reading. The first passage read by each subject was controlled so that every passage would be presented first approximately an equal number of times. Passage order for the second, third, and fourth passages was predetermined randomly for each subject. Subjects in the response and no-response group were matched so that pairs of subjects received the same passages in the same order. The subjects were allowed to take breaks, if desired, between each passage.
After finishing the experiment electrode impedances were again checked. Subjects were interviewed regarding what strategies they used to complete the task and were then debriefed. The entire procedure lasted no more than
2 1/2 hours.




Results
Comprehension Quiz Data
The quizzes were scored by an undergraduate research assistant who was blind regarding the condition of each subjects' quizzes. Each quiz answer was given a score ranging from zero to three based on the following criteria. A score of zero was given if no answer was provided or if all information in an answer was incorrect. A score of one was given if only a short phrase was provided as an answer. Answers in this group may have contained incorrect information or be based on information not provided in the text. A score of two was given if the answer was correct and longer than a short phrase and contained some detail or explanation. A score of three was given to correct answers with details and explanations using wording from the text. A few of the questions could be completely answered using a short phrase.
In this experiment, three passages were read twice
before taking a quiz, but only one passage was read only a single time before taking a quiz. This could lead to an inflated variance in the single reading condition, since the average score would be based on fewer quizzes. To control for this, only one quiz from a double reading was used as comparison data. Recall that either the second or third passage presented was read only once. If the second passage was read only once, then the third passage was used as the double reading comparison. If the third




passage was read only once, then the second passage was used as the double reading comparison. This equalized the amount of data included in the analysis and had the added benefit of equating the number of passages completed before quizzes in the single versus double reading condition.
Data from the quizzes were analyzed using a 2
(reading span; high or low) x 2 (presentation; 1st reading and 2nd reading) mixed ANOVA, with reading span being the between-subjects factor. Presentation was significant, F(1,38) = 12.68, p < .001. The mean scores for quizzes taken after one versus two readings was 8.4 and 10.2, respectively, indicating that performance on the quizzes improved after a second reading. Neither reading span nor the reading span x presentation interaction reached significance (both F's < 1.0). ERP Data
In all analyses based on ERP data, Nl-P2 amplitude was quantified as the difference between the minimum voltage occurring between 108 and 180 msec after probe onset (Ni), and the maximum voltage occurring between 160 and 245 msec after probe onset (P2). Visual inspection of the average ERPs indicated that the latency of the Nl-P2 responses fell within this range for each subject. The Nl-P2 in this and subsequent analyses refers to the Nl-P2 response to the secondary auditory probes. For ease of reference, this will simply be referred to as the Nl-P2.




Amplitudes for Nl-P2 responses in all analyses are given in microvolts (uv).
The N1-P2 amplitude data was subjected to a 2
(reading span; high or low) x 2 (response; yes or no) x 2 (presentation; ist reading and 2nd reading) x 5 (site; CZ, F3, F4, LS, and RS) mixed ANOVA with reading span and response being between-subjects factors. The main effect of site was significant, F(4,144) = 63.68, p < .0001. Tukey's test (at p < .05) indicated that the mean amplitude N1-P2 for CZ (14.99 uv) was significantly larger than that of F4 (10.19 uv) and F3 (9.98 uv), which did not significantly differ. Each of these sites had a significantly larger mean NI-P2 than RS (7.89 uv) and LS (7.44), which did not significantly differ.
The reading span x presentation interaction was marginally significant, F(1,36) = 3.87, p < .057. Overall, NI-P2 amplitude for the high reading span group was smaller during the first (9.75 uv) than the second (10.87 uv) reading. This pattern was slightly reversed for the low reading span group (10.08 uv for the first reading and 9.7 uv for the second). However, differences between the first and second reading for either group were not significant. No other effects reached significance (all F's < 1.0).
ERP waves for the response and no-response groups,
separated by reading span, reading presentation, and site are presented in Figure 2-1. Four features are




noticeable. Initially there is a small negative followed by positive response to the target word (T), which begins around 100 msec after target onset. A large Nl-P2 to the probes (P) then dominates the waveforms. The N1 component reaches its minimum around 120 msec after probe onset, and the P2 component reaches its maximum around 200 msec after probe onset. For the response groups (panels A and B), there is a large, later positive component which peaks at approximately 350 msec after probe onset. This component represents the additional processing associated with the probe detection response. This component is substantially reduced in the no-response groups (panels C and D). Lastly, a response to the word following the target (T+1) begins around 100 msec after word onset.
Note that there is a high degree of similarity
between the waveshapes for the response groups (Panels A and B) and for the no-response groups (Panels B and C). Further, for both response conditions, the waves are highly similar until the point beginning around 250 msec after probe onset, at which time the response group shows the additional positive component associated with the response demand. This high degree of consistency within and between the groups demonstrates the reliability of the ERP measure. Also note that the pattern of the waveshapes are similar for each site, but differ primarily in amplitude.




Figure 2-1. ERP waveforms for Experiment 1 separated by site, reading span, and response group. For each panel, the solid line represents the ERP during the first reading and the dashed line represents the ERP during the second reading. Onset times for the target word (T), probe (P), and the word following the target (T+1) are indicated on the abscissa.
Panel A represents the high reading span, response group. Panel B represents the low reading span, response group. Panel C represents the high reading span, no-response group. Panel D represents the low reading span, no-response group.




41
i i i II I
10.
Site = CZ A
10
I I I I i I I IiI
10 .
-10 -0
0
C Site = CZ C
10
-10
I I I II I I I
Site CZ D
T P T+1
- 10 I Ii II I I Ii I
0 100 300 500 700 900
Time (msec)




III II I I
Site = F3 A
10 10
-10
S i I I I I
Site = F3 B
10 10
0
> -10
0
L
SSite iF3 c
Fi0
0l
-10
i i i t i ii I
Site = F3 D
10
T P T+I
-10 1I I Ii 1 1 7
0 100 300 500 700 900
Time (msec)
Figure 2-1--continued




43
1 I IIII IIII
Site = F4 A
10
0-
-10
I I I ItI I I
Site = F4 B
10
U)
0
- -t
oI I I I t I I
to
-10
Site = F4 D
10
T P T+l
- 10 I I i I I I I
0 100 300 500 700 900
Time (msec)
Figure 2-1--continued




I I I l

I I I I

Site = LS B
I 1 I I 1 I

I I I I I I I I I C
Site = LS C
l I i 1 l I- I I

Site = LSI Site = LS

T PF
I Ii I

I I I I I I I I
D
T+I
I I I I I I I I

0 100 300 500
Time (msec)
Figure 2-1--continued

I I
Site = LS

A

700

900

I I I i

I I I

I

. . . . i m I J I

I I I J I

I I I I I I




45
1 I i I 1 I I I I
Site = RS A
10
-10
I I I I I
Site = RS B
I0
O
0
-10
0
-10
Site RS D
10 0
T P T+I
- 10 I Ii I I I Ii I I
0 100 300 500 700 900
Time (msec)

Fiqure 2-1--continued




Since there was an apriori expectation of differences between sites, the Nl-P2 data were reanalyzed separately for each site. Thus, the following results were based on 2 (reading span; high or low) x 2 (response; yes or no) x
2 (presentation; 1st reading and 2nd reading) mixed ANOVAs.
For CZ, the only effect to reach significance was the expected reading span x presentation interaction, F(1,36) = 5.05, p < .031 (all other F's < 1.0). Simple effects analyses of this interaction showed a significantly larger Nl-P2 during the second reading than during the first reading for the high reading span group, F(1,36) =5.64, p < .025, but not for the low span group, F(1,36) =1.29, p > .05. The means for this interaction are presented in Table 2-1.
For F3, none of the effects reached significance at the p < .05 level of significance. The reading span x presentation interaction approached significance, F (1,36) = 3.32, p < .08 (all other F's < 1.0). The pattern of results for this interaction was the same as for CZ, but comparisons between first and second readings were not significant for either reading span group (see Table 2-1).
For F4, none of the effects reached significance at the p < .05 level of significance (reading span x presentation interaction, F(1,36) = 2.31, p < .15; all other F's < 1.0). (See Table 2-1.)




Table 2-1
Mean NI-P2 Amplitude (uv) for Hiah

and Low Readina Span

Groups During the First and Second Reading of a Passage

High Reading Span
Reading
First Second

14.13 9.28 10.08 6.68 8.01

16.26 10.36 10.68 8.22 7.98

Low Reading Span
Reading
First Second

15.15 10.31 10.21 7.92 8.00

14.44 9.66 9.28 8.21 7.77

For LS, the main effect of presentation reached significance, F(1,36) = 10.17, p < .003. Mean NI-P2 amplitude was smaller during the first (7.3 uv) than second (8.22 uv) reading. This effect was mediated by a reading span x presentation interaction, F(1,36) = 4.66, p < .038. Simple effects tests indicated that, for the high reading span group, NI-P2 amplitude was smaller during the first than second reading F(1,36) = 14.29, p < .001. N1-P2 amplitude for the low reading span group did not significantly differ between the first and second reading (F < 1.0). (See Table 2-1.)
For RS, none of the effects reached significance (all F's < 1.0). (See Table 2-1.)

S ite




Summary of ERP Results. When all recording sites were combined, there was a nonsignificant trend towards the predicted reading span x presentation interaction. When recording sites were examined individually, CZ and LS had significant reading span x presentation interactions, and this approached significance for F3. Past research has found attentional effects on NI-P2 amplitude to occur primarily at CZ. In this study, LS was just as sensitive to changes in attentional demands as CZ, even though the overall amplitude at LS was smaller than CZ. At least for the high reading span group, NI-P2 amplitude increased during a second reading of a passage at both CZ and LS. Importantly, there was not a main effect of response, nor any interaction with response, at any recording site. This indicates that an overt response to the probes is not needed to obtain the expected NI-P2 difference. Reaction Time Data
Reaction time analyses are based only on the 20
subjects in the response group. The reaction time data were analyzed using a 2 (reading span; high or low) x 2 (presentation; ist and 2nd reading) mixed ANOVA, with reading span as the between-subjects factor. Only the main effect of presentation was significant, F(1,18) = 10.71, p < .005. Mean reaction time was faster during the first than second reading (see Table 2-2). No other effects reached significance (for reading span, F(1,18) =
1.59, p < .21; for the interaction, F(1,18) < 1.0).




Table 2-2
Mean Response Times (msec) to Detect Probes and Standard Deviations (in parentheses) for High and Low Reading Span Groups During the First and Second Reading of a Passage
Reading
Reading Span First Second Mean
High 410 (68) 430 (75) 420 (70)
Low 450 (62) 467 (62) 459 (66)
Mean 430 (67) 449 (74)
Discussion
The main finding from the ERP data was an increase in Nl-P2 amplitude during the second reading of a passage for the high reading span group. This supports the predictions made by the cognitive contents hypothesis. According to the cognitive contents hypothesis, fewer resources should be required during the second reading. This would leave more resources available to respond to the probes, which would lead to a larger Nl-P2 response during the second reading. Thus, the ERP does appear to reflect the reduction in resources needed during the second reading by the high reading span group. Note that there was an interaction between reading span and presentation for the ERP data but not for the recall (quiz) data. This suggests that the ERP measure may be more sensitive than the recall measure to individual differences in processing load.




The interaction between reading span and presentation was expected, but the low reading span group showed no difference at all, which was unexepected. One possible explanation for the lack of difference in the low reading span group is based on reading ability. Prior research has shown a correlation between reading span and reading comprehension test scores of around .55 (Daneman, 1987). It is possible that for the low reading span group both readings of a text were difficult, in the sense that capacity requirements were heavily taxed. That is, low reading span subjects may have had to use maximum capacity to comprehend and store the information in a passage.
The idea of using maximum capacity is analogous to a ceiling effect. The amount of capacity used by the low reading span group may have reached its ceiling during each reading. If this is true, then presenting shorter passages, which tax capacity less heavily, should not lead to a ceiling effect. This would allow an increase in Nl-P2 amplitude during the second reading for both reading span groups.
Support for the idea of a ceiling effect on capacity is also present in the quiz data. Given that reading span is a test of memory capacity and reading comprehension ability, quiz scores should be lower for the low reading span group than for the high reading span group. Since this was not the case, it is tempting to conclude that the low reading span group allocated maximum capacity to the




reading task in order to store as much information as possible. The concept of a ceiling effect on capacity is addressed in an additional study (Experiment 1B) presented in the next section of this chapter.
A second important finding from the ERP data was the lack of any main effect or interaction with response condition. The pattern of results for those subjects who overtly responded to the probes was identical to those who made no response to the probes. It should be stressed that subjects in the no-response group were not asked to make any type of response to or decision about the probes. In fact, self-reports by subjects indicated that they ignored the probes after hearing them a few times. This suggests the ERP can serve as a measure of cognitive load in the absence of a response. This does, of course, make the ERP measure very nonintrusive. Furthermore, this may allow ERPs to be used when an overt response is not compatible with the primary task.
The main finding based on the reaction time data was longer probe detection times during the first reading of a passage. This result is consistent with the prior knowledge hypothesis, which states that the first reading led to the instantiation of a greater amount of prior knowledge during the second reading. Thus, more capacity is used during the second reading, which leaves fewer resources available to respond to the probes. As a result, reaction times increase during the second reading.




There is another reason why probe responses may not have been faster during the second reading; the primary and secondary tasks may not have competed for common resources. This reason does not exclude the explanation proposed by the prior knowledge hypothesis, but may be independent of it. The idea of competition for resources is based on the views of multiple resource theories, which state that two tasks will interfere with each other only if they compete for common resources (Gopher & Navon, 1980; Navon & Gopher, 1979; Wickens, 1983). To determine if common resources are used, the component processes of a task need to be determined. In terms of the reading task, the major components are perceptual and cognitive, such as converting print to words and then determining their meaning. In terms of the secondary task, there are also some perceptual and cognitive demands, since subjects have to perceive and recognize the probe, although these demands are probably minimal. In addition, the secondary task has a large response component, which is the primary component of the secondary task (Wickens, 1983). Thus, the dominant components of the reading task and the secondary task do not overlap. If this rough task analysis is correct, then there is little reason to expect any interference between the reading and secondary probe response tasks, since their primary demands may not compete for common resources (Wickens, 1976).




If this "lack of response competition" hypothesis is correct, it may help to explain the opposite results of the present study and that of Inhoff and Flemming (1989). Recall that Inhoff and Flemming conducted a mini-study in which subjects read a passage twice while responding to secondary probes. They found shorter reaction times during the second reading. In their study, subjects made successive words appear by pressing a key. In the present study, words were automatically presented at a predeterimined rate. This methodological difference could substantially change the demands of the primary task. In addition to perceptual and cognitive demands, Inhoff and Flemming's primary task also had response demands. Thus, both the primary and secondary tasks competed for common resources. This response competition hypothesis was tested in an additional study (Experiment 1C) which will be presented in a following section of this chapter.
Experiment lB
The purpose of Experiment lB was to test the idea of a ceiling effect on capacity. To review, reductions in Nl-P2 amplitude during a second reading were found for the high reading span group, but not the low span group. It was proposed that the low reading span group may have had to devote maximal capacity to the task during each reading, therefore no reduction in load was found.
To test this hypothesis, the procedure used in
Experiment 1 was repeated using shorter text passages,




which should place less demands on cognitive capacity. That is, there is simply less information to process and store in memory. This reduction in capacity demands should allow low reading span subjects to remain below their maximal capacity, which would lead to reduced cognitive load during the second reading of a passage.
Ten subjects participated in this study. Six of the subjects were paid ten dollars for their participation. The remaining four participated as part of a class requirement, and were from the same subject pool used in Experiment 1. All subjects were undergraduate students from the University of Florida.
The procedure, apparatus, and instructions were identical to Experiment 1 except for the following changes. First, each subject read six short passages which were between 160 and 178 words in length. A sample passage is presented in Appendix C. All the passages were read twice in succession and a free recall was produced after the second reading of each passage. A free recall was used so that subjects would not be limited regarding what could be recalled. During each reading of a passage, eight secondary probes were presented. Since the response and no-response conditions of Experiment 1 produced similar results, only the no-response condition was performed here. Therefore, only ERP data were recorded.




Results and Discussion
The NI-P2 amplitude data were analyzed using a 2
(reading span) x 2 (presentation) x 5 (site) mixed ANOVA, with reading span as the between-subjects factor. The main effect of site was again significant, F(4,32) = 21.35, p < .001. Tukey's test (at p < .05) indicated that the mean Nl-P2 amplitude of CZ (15.2 uv) was significantly greater than the mean amplitude of F3 (10.02 uv), F4 (9.74 uv), and RS (7.67 uv), which did not differ. Mean amplitude for each of these sites was larger than that of LS (6.32 uv). The main effect of presentation was also significant, F(1,8) = 5.25, p = .05. Mean NI-P2 amplitude was smaller during the first (9.21 uv) than second (10.36 uv) reading.
When the sites were analyzed individually, there were no significant effects for F3, F4, or RS. For CZ, the only effect to approach significance was presentation, F(1,8) = 3.06, p < .11. Mean amplitude NI-P2 at CZ during the first reading was 14.55 uv, and 15.81 uv during the second reading. For LS, presentation was significant, F(1,8) = 8.31, p < .02. Mean amplitude NI-P2 at LS during the first reading was 5.75 uv, and 6.89 uv during the second reading. The general pattern of results at CZ and LS replicate the results from Experiment 1. That is, NI-P2 amplitude increased during the second reading.
Importantly, there were no reading span x
presentation interactions in either the overall ANOVA, or




when sites were analyzed individually. As predicted, using shorter passages lead to an increase in N1-P2 amplitude from the first to second reading for both reading span groups. This supports the conclusion that, in Experiment 1, there may have been a ceiling effect on capacity. Specifically, the low reading span group may have been devoting full capacity to both the first and second readings when long texts were being read.
Experiment 1C
The purpose of Experiment 1C was to test the "lack of response competition" hypothesis. To review, it was proposed that since the reading task contains no response requirements, there should be no competition for resources with the secondary probe response, which is comprised primarily of response demands. To test the "lack of response competition" hypothesis, a small study was conducted which added response demands to the reading task.
Experiment 1C replicated the self-paced procedure used by Inhoff and Flemming (1989). That is, subjects pressed a key to display successive words while simultaneously responding to probes. If this additional response component in the primary task leads to interference with the secondary probe response, then this would help explain the discrepancy between the reaction time and ERP data in the present study. This would also help explain the difference between Britton et al.'s




(1978, 1979, 1980) studies, which show faster probe response when reading difficult text, and Inhoff and Flemming's (1989) study, which shows just the opposite. Recall that in Britton et al.'s studies, subject read text presented in a normal manner, thus there was no response component to the primary reading task.
Ten subjects read the same text passages presented in Experiment 1. All subjects participated for class credit and were from the same subject pool used in Experiment 1. The experiment was conducted using a Compaq Deskpro 386 equipped with a VGA monitor. Subjects controlled the display of the words by pressing the spacebar on the computer keyboard. Each press of the spacebar erased the current word and displayed the next word. While reading, subjects responded to the auditory probes by pressing the zero key on the numeric keypad of the keyboard. Reaction time to press the zero key was measured. The instructions were the same as used in the response condition of Experiment 1 except that information regarding ERP recording was deleted, and an explanation about pressing the space bar to display words was added. Subjects were given one practice passage to familiarize them with the procedure. All other aspects of the experiment were the same as Experiment 1.
Results and Discussion
The reaction time data were analyzed in the same
manner as Experiment 1. Once again, presentation was the




only significant factor, F(1,8) = 7.61, p < .025 (all other F's < 1.0). Mean reaction times were faster during the first than second reading (see Table 2-3). Table 2-3
Mean Response Times (msec) and Standard Deviations (in parentheses) to Detect Probes for High and Low Reading Span Groups During the First and Second Reading of a Passage for Experiment 1C
Reading
Reading Span First Second Mean
High 380 (18) 400 (15) 390 (19)
Low 444 (160) 468 (137) 456 (141)
Mean 412 (112) 434 (98)
Overall, this pattern of results is essentially identical to the reaction time data of Experiment 1. Taken together, the results of Experiments 1 and 1C strongly argue against the "lack of response competition" hypothesis. Consequently, this lends support to the prior-knowledge hypothesis. As previously stated, the first reading may have led to the use of more prior knowledge during the second reading. If knowledge consumes capacity, then the additional capacity requirements could have led to slower probe detection times during the second reading.




General Discussion
Individually, the results of the ERP and reaction time data are quite straightforward. The ERP data from Experiment 1 and 1B indicate a reduction in cognitive load during the second reading of a text. This supports the cognitive contents hypothesis. The reaction time data from Experiments 1 and 1C indicate increased cognitive load during the second reading of a text. This supports the prior knowledge hypothesis. Experiment 1C ruled out the possibility that lack of response competition between the reading and probe detection tasks could lead to slower reaction times during the second reading.
There is one more reason why reaction times could
have been longer during the second reading; subjects may have simply given less attention to the probe response during the second reading of a passage. This is consistent with a study by Britton and Price (1981), which found longer reaction time to detect a probe when less attention was focused on the secondary task. Since the quiz scores for both reading span groups were higher after a second reading, there is evidence that the subjects did attend to the passages during the second readings. As Inhoff and Flemming (1989) note, even small changes in the amount of attention given to the primary task could lead to substantial changes in probe detection times. Unfortunately, for the rereading method used here, this explanation leads to the same predictions as the prior




knowledge hypothesis, and therefore they can not be easily distinguished. Furthermore, these two explanations are not mutually exclusive. In fact, consumption of capacity by prior knowledge could lead to reduced attention to the probe response task. If capacity was filled by prior knowledge, then there may simply be fewer attentional resources available to devote to the secondary task.
For any of the above hypotheses to explain the
contradictory pattern of results from the ERP and reaction time data, an additional assumption needs to be made; this is that ERPs and reaction time may measure different aspects of cognitive load. If ERPs and reaction times were reflecting the same processes, then the two measures should not have divergent patterns of results. For example, if less attention was given to the probes during the second reading, then reaction times should be slowed, and a smaller Nl-P2 response should occur. This is, of course, the opposite of what happened to the Nl-P2. if the ERP and reaction time measures are reflecting different cognitive processes, then this would explain the apparent contradiction in their results.
The temporal aspects of the ERP and reaction time
measures lend credence to this idea. Specifically, the Nl response to the probe begins less than 200 msec after a target word's onset and the P2 begins less than 300 msec after target onset. In contrast, probe detection responses averaged around 520 msec after target onset.




Thus, there is a large amount of processing which may take place between the time of the Nl-P2 response and the probe detection response. This additional processing may influence the reaction time measure. To examine this explanation, a closer look at previous research involving the processing requirements during the first and second reading of a text is needed.
As mentioned earlier, Levy and her colleagues (Levy, 1983; Levy & Begin, 1984; Levy, Newell, Snyder, & Timmins, 1986; Levy & Kirsner, 1989) have studied processing changes during repeated readings of a text. In these studies various types of proofreading tasks were used to study changes in letter-, word- and text-level processing. Letter-level processes refer to the basic perceptual processes involved in individual letter analysis and recognition. word-level processing refers to the processes required to perceive and comprehend a single word. Text-level processes refer to those processes involved in integrating the meaning of a word into its context. The typical procedure involved proofreading a the text for errors across multiple readings. By using different types of errors, the efficiency of processing at the letter-, word-, and text-level could be examined.
Levy (1983) had subject proofread for spelling errors during a first (unfamiliar) and second (familiar) reading. The text contained different errors during each reading. She found that subjects detected more spelling errors in




familiar texts than in unfamiliar texts. This improvement was maintained even when the order of the words was scrambled during the first and second reading. Familiar passages were also read faster, thus there was no possibility of a speed-accuracy tradeoff. Levy concluded that the improved error detection was the result of more efficient letter- and word-level processing.
Levy and Begin (1984) replicated Levy's (1983)
results and also explored higher-order proofreading tasks. In one of Levy and Begin's experiments (their Experiment 3), subjects proofread for misspellings in passages which contained semantic inconsistencies or no inconsistencies. Inconsistency was manipulated using "garden path" sentences. These sentences contained ambiguous words which could be interpreted in two ways (consistent or inconsistent with the meaning of the passage), depending on the meaning biased by a prior sentence. These word meaning errors could only be detected if the semantic properties of the text were fully analyzed. They found that subjects were sensitive to the meaning of a text even on repeated readings, that is, word meaning errors were detected during each reading. In addition, detection of spelling errors decreased when a text contained an inconsistent phrase, and this disruption was larger for familiar texts. Levy and Begin suggested that when a semantic inconsistency was encountered, additional resources were allocated to the higher-level semantic




analysis. This left fewer resources available for the proofreading task, which produced the decrement in spelling error detection. Levy and Begin also concluded that overall improvements in spelling error detection across readings were due to more efficient processing at the word level.
Levy and Begin (1984) interpreted their results in
terms of resource allocation. They suggest that resources are shared between lower-level processes involved in analyzing individual words and higher-level processes involved in comprehending the overall meaning of a text. Resources are allocated to the lower- and higher-level processes as needed. Thus, if higher-level processes become more demanding, fewer resources will be allocated to the performance of lower-level functions. This is reflected in the larger decrease in word-level error detection for familiar than for unfamiliar passages when semantic inconsistencies are present in a text. Levy and Begin state that "meaning distortions recruit resources to the semantic processor, leaving fewer resources for the proofreading task, thus resulting in a loss in proofreading performance" (p 631).
Levy, Newell, Snyder, & Timmins (1986) extended Levy and Begin's (1984) work. They examined proofreading performance during four consecutive readings of a text. They also demonstrated more efficient processing of familiar texts at lower- and higher-levels of analysis.




However, their study places some constraints on the concepts of processing efficiency and resource allocation. A few results are particularly relevant. First, they noted improved reading speed across presentations, suggesting that lower-level processes became increasingly more efficient. Second, detection of non-word errors, which were relatively easy to detect, did not increase with multiple presentations, whereas the detection of word errors (meaning errors), which were more difficult to detect, did increase. Only when error detection is difficult were additional resources allocated to the error detection task. They use this latter finding to argue that increased accuracy of error detection is not an automatic result of reprocessing, and is dependent on the subject's goals and the task demands. They suggest that "processing efficiency is best viewed in terms of processes becoming faster and less resource demanding, so that more attention is available for strategic allocations within the task" (p. 477).
There are two important conclusions to be noted in Levy et al.'s work. First, they show that different aspects of reading may become more efficient when a text is reread. Second, increased efficiency of processing leads to the availability of more resources, which the reader can allocate to the relevant demands of the task. These results can be used to help explain the findings of the present study. In the present study, the goal is to




remember as much material as possible so that the quiz questions may be correctly answered. Therefore, if the lower-level processes are more efficient during the second reading, subjects may allocate more resources to higher-level processes such as processing the meaning of the material and storing the material in memory. This would lead to increased cognitive load for the higherlevel processes during the second reading.
This brings us back to the possibility that ERPs and reaction times measure different aspects of cognitive load. If the ERP measure is reflecting early, lower-level stages of analysis, such as word recognition, lexical access, and some semantic analysis, then these processes may indeed be more efficient during the second reading of a passage. This would lead to increased N1-P2 amplitude during the second reading of a passage. Probe response times may be more sensitive to later, higher-level stages of analysis, such as text-level comprehension and memory processes. If more resources were allocated to the important memory demands of the task during the second reading, then reaction time should be slower during a second reading since the memory load is increased. Note that this explanation does not require the assumption of prior knowledge activation. Increased reaction time is reflecting increased memory load, but the contents of memory do not need to be prior knowledge. In this study,




the additional information in memory may simply be additional information from the text.
If ERPs and reaction time reflect different
processes, then the pattern of results is explainable. Levy et al. (1983, 1984, 1986, 1989) provide strong evidence that lower- and higher-levels of processing each become more efficient during multiple readings of a text. Importantly, this increased efficiency allows resources to be reallocated based on the demands of the task. Thus, in the present study, fewer resources are needed for lowerlevel analysis during the second reading. This is reflected by the reduced Nl-P2 amplitude. In contrast, increased processing efficiency allows additional resources to be allocated to the memorial demands of the task. This leads to increased probe detection times during the second reading.
In summary, the studies described in this chapter
support the conclusion that the Nl-P2 response can be used as a measure of certain aspects of cognitive load. Experiment 1 found increased Nl-P2 amplitude during a second reading for the high reading span group. Experiment lB replicated Experiment 1, but used shorter passages. When this was done, increased Nl-P2 amplitude during the second reading was found for both reading span groups. Experiment 1 also showed slower probe reaction times during the second reading. Experiment 1C replicated this effect and ruled out the possibility that a lack of




response competition between the reading and probe detection tasks could have been responsible for the slower responses during the second reading. The contrasting results of the ERP and reaction time data were explained in terms of each being sensitive to different aspects of the reading task. It was suggested that the Nl-P2 is more sensitive to lower-level aspects of the task, while the reaction time measure is more sensitive to higher-level aspects of the task.




CHAPTER 3
EXPERIMENT 2
Introduction
All of the studies described so far have investigated overall levels of cognitive load while reading. The secondary task methodology has also been used to examine changes in load within a text. One topic that has received much attention is the "levels effect." The levels effect refers to the fact that when information from a text is recalled, more superordinate (gist-level or high-importance) information is usually recalled than subordinate (supporting detail or low-importance) information (Kintsch & Van Dijk, 1978; Meyer, 1975; Miller, 1985). This effect is described as reflecting the hierarchical organization of most texts. Specifically, information high in the hierarchy (superordinate) is more important for understanding a text than information low in the hierarchy (subordinate).
The present study examines why superordinate
information is more frequently recalled than subordinate. One possibility that has been proposed is that additional resources are allocated to the processing of highimportance information (Britton, Meyer, Simpson, Holdredge, & Curry, 1979; Cirilo & Foss, 1980). If additional resources are allocated, then cognitive load




should be increased when processing high-importance information. Measures of cognitive load can be applied to test this notion of processing resources.
A number of hypotheses have been put forth to explain the levels effect (e.g., Britton, Meyer, Simpson, Holdredge, & Curry, 1979; Britton, Meyer, Hodge, & Glynn, 1980; Britton, Muth, & Glynn, 1986; Cirilo, 1981; and Cirilo & Foss, 1980;). These hypotheses may be divided into two general classes; those that interpret the levels effect as primarily the result of processes which occur during encoding, and those that interpret the effect as primarily the result of processes which occur during retrieval.
Hypotheses emphasizing encoding of text are also
referred to as selection hypotheses. Selection hypotheses imply that the reader selects important (superordinate) portions of a text for differential (extra or unique) processing (Britton, Meyer, Simpson, Holdredge, & Curry, 1979; Cirilo & Foss, 1980). Thus, the levels effect occurs because extra processing is performed on the hierarchically important information, which increases the likelihood of the information being stored. This extra processing requires more capacity and should increase the cognitive load during encoding of this material. This hypothesis is consistent with related work which shows increased recall of material when more effort (defined as increased comprehension difficulty) is applied to its




processing (Tyler, Hertel, McCallum, & Ellis, 1979; Walker, Jones, & Mar, 1983). Britton, Meyer, Hodge, and Glynn (1980) also describe encoding differences as potentially increasing the number of retrieval paths in memory. This would also enhance subsequent recall.
Not all theories assume the reader is "directly" in control of this resource allocation process. For example, according to the flexible allocation hypothesis, the reader can flexibly allocate different amounts of attention to different parts of the text. In contrast, according to the cognitive contents hypothesis (a derivative of the previously discussed cognitive contents hypothesis), the amount of attention allocated to a portion of a text is a function of the text itself (i.e., its structure). Thus, the flexible allocation hypothesis describes the reader as actively changing the amount of attention allocated to elements of a text, whereas the contents hypothesis assumes the learner's allocation of resources is passively influenced by the text (Britton, Glynn, Meyer, & Penland, 1982; Britton & Price, 1981; Navon & Gopher, 1979).
Retrieval hypotheses suggest that the levels effect is due to processes which occur when information is recalled. During encoding, high- and low-level information (in the text's content structure) are both equally likely to be stored. Information high in the text structure is described as being more accessible than




low-level information and is therefore more likely to be recalled (Britton, Meyer, Hodge, & Glynn, 1980; Britton, Meyer, Simpson, Holdredge, & Curry, 1979; Meyer, 1975). The reason for this is that high-level information is stored in superordinate positions within one's memory structure, which make them easier to access (Meyer, 1975).
Selection and retrieval hypotheses lead to different predictions concerning reading time and cognitive load during encoding. Selection hypotheses predict increased reading time and increased cognitive load for highimportance information. Retrieval hypotheses predict no differences. There is support for both of these outcomes.
In Britton, Meyer, Hodge, and Glynn (1980), subjects were presented with texts to read and reading time was measured. The texts were constructed so that a paragraph which was hierarchically important in one passage was of low importance in another passage. This controlled for the effects of syntax, vocabulary, and complexity of the target information. There were no differences in reading time for high- and low-importance information. After reading, subjects were presented with retrieval cues to try and reduce the levels effect. Retrieval cues almost eliminated the levels effect. This implies that both high- and low-importance information were stored, and that differences in recall were more a function of retrieval than encoding differences.




Britton, Simpson, Hoidredge, and Curry, (1979)
conducted a similar study in which subjects read while performing a secondary probe detection task. The texts again contained high- and low-importance target paragraphs. Reading time and probe reaction time were not affected by the relative importance of the target paragraph. Thus, no support was found for encoding differences.
On the other hand, Cirilo and Foss (1980) presented evidence suggesting encoding differences. In their study, subjects read pairs of stories with common target sentences which were of high or low importance to the passage. Reading times were slower when the target sentence was of high importance. This suggests more processing time was given to the high-importance information.
Britton, Muth, and Glynn (1986) extended the results of Cirilo and Foss (1980) in three experiments. The first was identical to Cirilo and Foss' (1980) study. Results again indicated longer reading times for target sentences when they were of greater importance. In their second experiment the reading time of the passages was limited by displaying the passages using a rapid-serial-visualpresentation format (RSVP). Short phrases were presented sequentially at a predetermined rate. The levels effect in recall was maintained even under conditions of limited exposure time. Their third experiment again used the RSVP




format, but subjects also performed a secondary probe detection task. Responses to the probes were slower when reading important target sentences, which provides evidence supporting increased capacity demands when encoding high-importance information. The authors conclude that when reading at their own pace, subjects will allot more time to important information. When exposure time is limited, the same result may be accomplished by allocating increased effort to the processing of the important information.
The results of Britton, Muth, and Glynn (1986)
provide an explanation regarding the lack of differences in cognitive load when reading high- and low-importance information found by Britton, Simpson, Holdredge, and Curry (1979). In that study, subjects read the texts at their own pace. Thus, cognitive load may have been reduced by increasing the time spent on the important material. A similar conclusion regarding the distribution of load over time is discussed by Britton, Westbrook, and Holdredge (1978). Britton, Meyer, Hodge, and Glynn (1980) note that in their study, as well as in Britton, Simpson, Holdredge, and Curry's (1979) study, reading time was measured over an entire target paragraph. They suggest that selective attention may be more readily applied to shorter segments of text, such as sentences, than to paragraph size segments. The targets used by Cirilo and




Foss (1980), and subsequently by Britton, Muth, and Glynn (1986) were, indeed, short sentences.
To summarize, the relative importance of encoding and retrieval processes in producing the levels effect remains unclear, but evidence favors the conclusion that both encoding and retrieval processes effect recall. Studies which find support of encoding have used small segments of text as targets. This suggests selective attention may be limited to, or most effectively applied to small portions of text. Furthermore, when reading time is restricted, cognitive load appears to increase when important information is processed. Future studies need to control for the effects of reading time. In addition, small segments should be used as targets so that the cognitive demands within specific portions of a text may be measured.
The purpose of Experiment 2 was to investigate the effect of within-text differences in cognitive load as a function of the hierarchical importance of information. Specifically, Experiment 2 was used to examine the levels effect and to try and determine if there are encoding differences between high- and low-importance information and whether these differences are reflected in measures of cognitive load.
In this experiment, subjects read a series of texts which had been analyzed using the method described by Meyer (1975) to determine the hierarchical importance of




information in each text passage. While reading, subjects were presented with a series of secondary auditory probes. The probes were systematically presented at points representing high- or low-importance information. The texts were presented using the RSVP method described in Experiment 1. After reading each text, subjects produced a written, free recall of the passage. Subjects were again grouped into high and low reading span.
Since Experiment 1 showed an identical pattern of
results for the response and no-response conditions, the response condition was not included in this experiment. Thus, no reaction time data were collected. The data being analyzed are the amplitudes of the N1-P2 responses to probes. If more capacity is consumed during the processing of high-importance information, then probes embedded in high-importance material should be associated with smaller N1-P2 responses.
Evaluation of the Method. The basic format of
measuring the processing requirements of high- or lowimportance information within a text has previously been used to examine the levels effect. The main presentation difference between this and prior research is that a single-word RSVP presentation format is used. The smallest segment of text presented in an RSVP format has been phrases or short sentences (e.g., Cirilo & Foss, 1980; Britton, Muth, & Glynn, 1986). In these studies the levels effect still occurred (as indexed by recall




measures). This demonstrates that the RSVP method can be used to examine within-text variations in cognitive load.
The benefits of this method are similar to those of Experiment 1. The RSVP method prevents rereading of difficult material, and cognitive load cannot be reduced by increasing reading time. Again, the ERP provides a nonintrusive measure of cognitive load.
Hypotheses. The procedure provides a method for testing the predictions of selection hypotheses. The assumption of this experiment is the same as Experiment 1. That is, as the reading task demands more resources, less will be available to respond to the probes. According to selection hypotheses, processing load should be greater when reading high-importance information than when reading low-importance information. Therefore, less capacity will be available to respond to the probes when reading highimportance information.
Selection hypotheses lead to the following
predictions. First, the Nl-P2 response to probes embedded in high-importance information should be smaller than the Nl-P2 response to probes embedded in low-importance information. This effect should be largest at the vertex. Second, these effects should be mediated by an interaction with reading span. Past research has shown that good readers are more sensitive to the structure of a passage than poor readers (Meyer, 1975; Meyer, 1985). Therefore the high reading span group should show a larger levels




effect in both free recall measures and the Nl-P2 response. Third, the high reading span group should recall a greater amount of text than the low reading span group.
Method
Subi ects
Twenty subjects (10 men and 10 women) from the same subject pool as Experiment 1 participated. Data from 2 additional subjects were not included due to unacceptably high electrode impedances.
Materials. Apparatus and Procedure
The same stimulus presentation and EEG recording
system used in Experiment 1 were used again here. In the present experiment, subjects read six passages, each passage being read once. After reading each passage, subjects produced a written, free recall of the material. Subjects were allowed five minutes for their recall. Instructions stated to recall everything that could be remembered about a passage. All other aspects of the procedure and instructions were identical to Experiment 1.
Text Passages. Six passages were used as stimuli in Experiment 2. The passages were between 160 and 178 words in length. While reading each passage, a series of 8 auditory probes were presented. Two of the passages (#13 and #14) and their content structures (described below) were taken from Raney, Dunn, & Rust (1989). Another two passages (#15 and #18) and their content structures were




shortened versions of passages from Meyer (1975). The content structures of the final two passages were derived by the experimenter. A sample passage and its content structure is presented in Appendix C. The source of each passage is also presented in Appendix C.
Probe position was restricted in three ways. First, probes were not presented during the first 8 words of a passage. Second, probes were not presented during the first word of a sentence. Lastly, half of the probes were embedded within points representing high-importance information, while the other half were embedded at points representing low-importance information. Meyer's (1975) system was used to determine the relative importance of information in a text. This system is summarized below.
Meyer's (1975) system leads to the production of a
hierarchically arranged tree structure, called a semantic content structure. The content structure is composed of words from the text, specified case roles, and a set of labels which define the relationships between propositions. Each of these elements is called an idea unit.
The content structure illustrates the pattern of
subordination of ideas in a passage. The high-level or superordinate ideas are supported by related subordinate ideas within the passage's structure. The high-level ideas generally represent the main ideas of the passage,




whereas the lower-level ideas describe or expand upon the higher-level ideas.
Meyer's system also labels the relations among ideas. There are two categories of relations: rhetorical predicates (also called rhetorical relations) and role relations. Rhetorical predicates are found at the top levels (high in the hierarchy) of the content structure and show how their subordinate ideas are related. In other words, rhetorical relations represent the global structure of the text. For example, the sample passage presented in Appendix C (Text #14) describes the use of chemical pesticides. Line 5 in its content structure is a covariance rhetorical relation. The covariance relation describes a cause and effect relation. In this case, the use of CHEMICAL PESTICIDES (line 6) causes HARM (line 48) to the ENVIRONMENT (line 52).
Role relations are typically used to relate lexical predicates together. Lexical predicates are words from the text which subsume other propositions as their arguments. Role relations classify how lexical predicates are related to their arguments. In the Chemical Pesticides passage, the lexical predicate IS REQUIRED (line 8) is related to the phrase AMERICAN COMMERCIAL GROWERS (line 12) by the role relation agent. Agent describes the instigator of an action. In this example the action is the USE OF CHEMICAL PESTICIDES (line 6).




When analyzed, the passages had either 4, 5, or 6
levels of subordination in their content structures. The number of levels is dependent on the structure of an individual passage. For passages with 4 levels, the "top" two levels were combined to represent high-importance information, while the lower two levels represented lowimportance information. For passages with 5 levels, the top three were used as high-importance and the lower two as low-importance. For passages with 6 levels, the top three were used as high-importance and the lower three as low-importance. After deriving a content structure for each passage, probes were systematically placed within points representing high- or low-importance information.
Results
Recall Data
The free recall protocols were scored in accordance with Meyer's (1975) system by the experimenter. The experimenter was blind regarding the condition of each subjects' protocols at the time of scoring. Each protocol was scored for the presence of idea units based on the semantic content structure of the passage. Content units were scored as present if the words, or paraphrased words from a passage were present in the protocol. If a content unit was recalled in its proper relationship to other information in the text, then the corresponding relationship unit was also scored as present. Since there was not an equal number of high- and low-importance idea




units in a given passage, the number of high- and low-importance units recalled was converted to a percentage. The average percentage of high- and lowimportance information recalled was used as data in the following analysis.
The free recall data were analyzed using a 2 (reading span; high or low) x 2 (importance; high and low) mixed ANOVA, with reading span being the between-subjects factor. The main effect of reading span was significant, F(1,18) = 16.81, p < .0007. As expected, the mean percentage of idea units recalled by the high reading span group was greater than the low reading span group (see Table 3-1). Neither the main effect of importance nor the reading span x importance interaction reached significance, [F(1,18) = 3.42, p < .082; and F(1,18)
2.13, p < .162]. The pattern of recall is, however, worth noting since for the high reading span group the amount of high-importance information recalled was lower than the amount of low-importance information recalled. This did not occur for the low reading span group. This pattern of recall data is exactly the opposite of what was expected. That is, it was expected that more high- than lowimportance information would be recalled.
One explanation for the lack of a levels effect in the present study may be due to instructions and subject strategies. Subjects were instructed to recall everything they could from the passage. A common question asked by




subjects was if they were to try and recall every word from a passage. If this question was asked, they were told that if they remembered every word in the passage, then that is what they should recall. This may have overly stressed verbatim recall. Table 3-1
Mean Percentage Recall and Standard Deviations (in Parentheses) of High- and Low-Importance Idea Units for the High and Low Reading Span Groups
Importance
Reading Span High Low Mean
High 43.8 (6.6) 52.5 (11.1) 48.2 (9.9)
Low 35 (7.5) 36 (10.5) 35.5 (8-9)
Mean 39.4 (8.1) 44.3 (13.5)
After completing the study, subjects were interviewed regarding how they approached the task and what strategies they used to try and remember the material. The most common approach mentioned was to simply try and get all the information possible, since they would have to recall it later. Many subjects said that they concentrated on examples and details, such as numbers and dates, since they thought these would be important to recall. This type of information would typically be low-level information in the content structure. A few subjects actually said they didn't really try and get the "meaning"




of the texts, but just tried to remember everything possible. Thus, it may be that subjects were led into a somewhat abnormal reading strategy by the instructions and the nature of the task. Meyer (1985) also reports that the levels effect can be altered based on changes in subject strategies and task demands.
Four of the passages used in this study have been used in prior research. Dunn (1985) has shown a levels effect in free recall for similar versions of the Nuclear Breeder Reactor (Text #15), Loss of body Water (Text #13), and Chemical Pesticides (Text #14) passages (see Appendix C). Meyer (1975) has shown a levels effect in free recall for longer versions of the Nuclear Breeder Reactor (Text #15) and Supertanker (Text #18) passages (see Appendix C). This indicates that the lack of a levels effect in the present study is not due to the use of atypical passages. ERP Data
The Nl-P2 data analyzed in this and subsequent
analyses is the amplitude of the Nl-P2 response to probes embedded in high- or low-importance information. For the sake of textual simplicity, this will be referred to simply as the Nl-P2 amplitude for high- and low-importance information. Nl-P2 amplitude was defined in the same manner as Experiment 1.
Figure 3-1 presents the ERP traces for Experiment 2. As in the ERPs from Experiment 1, the traces show a small negative followed by a positive wave beginning around 100




msec after target (T) onset; a large NI-P2 to the probes
(P) beginning around 200 msec after target onset; and a response the to word following the target (T+l) beginning around 100 msec after its onset.
The NI-P2 amplitude data was analyzed using a 2
(reading span) x 2 (importance) x 5 (site) mixed ANOVA, with reading span being the between-subjects factor. The main effect of site was significant, F(4,72) = 50.94, p < .0001. Tukey's test (at p < .05) indicated that the mean amplitude NI-P2 at Cz (12.8 uv) was significantly larger than that of F4 (9.99 uv) and F3(9.8 uv), which did not differ. Each of these sites had significantly larger mean NI-P2 amplitudes than RS(7.2 uv) or LS (6.4 uv), which did not significantly differ.
There was also a significant reading span x
importance interaction, F(4,18) = 5.53, p < .03. The mean amplitude NI-P2 for the high reading span group was 9.5 uv for the high- and 8.6 uv low-importance material. Means for the low reading span group were 9.1 uv for the highand 9.9 uv for the low-importance material. However, none of the simple effects comparisons for this interaction were significant (all F's < 1.0). The main effect of reading span and all other interactions did not reach significance in the overall analysis.




Figure 3-1. ERP waveforms for Experiment 2 separated by site and reading span. For each panel, the solid line represents the ERP during the reading of high-importance information and the dashed line represents the ERP during the reading of low-importance information. onset times of the target word (T), probe (P), and the word following the target (T+1) are indicated on the abscissa. Panel A represents the high reading span group. Panel B represents the low reading span group.




I I
Site = CZ

AI I I I
A

I I I I I I ___ I I I -

I I
Site = CZ

I I I I 1I I I B
B

T P
0I 10 I0
0 100

T+l

300
Time

500

I I I

700

900

(msec)

. . i




AI I I I I I
A

I I I I
Site = F3

T P
0 100

I I I

T+I

300
Time

500

I I B
B
I I I 7

700

900

(msec)

Figure 3-1--continued

I I
Site = F3

' I

i i i l I I




I I
Site = F4

I I
Site = F4

T P
I I I II

100

I I I I I I I I
A
I I I I I I I I

I I I I i

T+I
I I I I I

300 Time

500

1 I 1
B
I I I

700

900

(msec)

Figure 3-1--continued

I I I I I I I I I

.. . . . I I I

I I

I I I




AI I I I I I I
Site LSA

I I I I I I I I I I

I I
Site = LS

T P
I Ii Ii

BI I I I I I I I
B

I I I l

I I I

300 500
Time (msec)

700

Figure 3-1--continued

0 100

900

I I I I I I i & I I I I &

Site = LS




I I I I I 1 1 1

I I
Site = RS

- ---

_ I I

I I
Site = RS

B

T P I l li I

I Ii Ii I I I

0 100

I I I I

300 Time

500

700

900

(msec)

Figure 3-1--continued

A

l l l




When the sites were analyzed individually, none of the effects reached significance for F3, LS, or RS. For CZ, the only effect to reach significance was the reading span x importance interaction, F(1,18) = 6.83, p < .018. Simple effect tests indicated that the only significant difference was between the high reading span, high importance condition, and the low reading span, high importance condition, F(1,18) = 5.96, p < .03 (see Table 3-2). Thus, the only significant difference between conditions for CZ was between reading span groups for the high importance information. There were no differences based on importance for either reading span group.
Table 3-2
Mean Nl-P2 Amplitude (uv) for Probes Embedded in High- and Low-Importance Information for High and Low Reading Span Groups
High Reading Span Low Reading Span
Importance Importance
Site High Low High Low
CZ 13.8 12.3 11.9 13.2
F4 10.1 9.1 9.7 11.1
For F4, the reading span x importance interaction was significant, F(1,18) = 5.4, p < .033. Simple effects tests indicated that the only significant difference was




between the high reading span, low importance condition, and the low reading span, low importance condition, F(1,18) = 5.96, p < .03 (see Table 3-2). Thus, the only difference for site LS was between reading span groups. Once again, there were no differences based on importance.
Summary of ERP Results. When all recording sites
were combined, there was a significant interaction between reading span and importance. The high span group had a slightly higher amplitude N1-P2 for high-importance information than for low-importance information. This pattern was reversed for low reading span group. However, simple effects comparisons between levels of importance or reading span were not significant. When the sites were analyzed individually, the only significant effects were reading span x importance interactions for sites CZ and F4. Simple effects tests showed that none of these effects resulted from differences based on the importance of the material. The pattern of average amplitude for each site replicates Experiment 1. Specifically, amplitude was largest at CZ, next largest at F3 and F4, and smallest at RS and LS.
Discussion
Overall, Experiment 2 provides no evidence supporting selection hypotheses. The prediction of a smaller amplitude N1-P2 response to probes presented during the processing of high-importance information was not confirmed for either reading span group. This suggests




that the high-importance information was not selected for differential processing. Based on the ERP data alone, the conclusion that selection hypotheses are incorrect seems warranted.
Three findings limit the validity of the conclusion that selection hypotheses are incorrect. First, the recall data showed no evidence of a standard levels effect. Specifically, less high-importance information was recalled than low-importance information. Thus, there is no evidence suggesting that extra processing may have been performed on the high-importance information. If this is true, then there is no reason to expect the ERP data to reflect a levels effect.
The second reason to doubt the previous conclusion is based on subject strategies and the demands of the task. Recall that subject's self-reports indicated that no preferential treatment was given to high-importance information. Subjects indicated that they tried to retain as much information as possible, regardless of its relative importance to the passage. This may have been partly a function of passage length. The passages were fairly short, and subjects reported that remembering most of the material was not difficult. Also, the instructions given to subjects emphasized that they should recall everything from a passage that they could. Thus, they were instructed not to recall or focus only on the main ideas.




The third reason for doubting the prior conclusion is that a pilot study using a similar procedure did indeed show evidence of a levels effect on the ERP data. That is, Nl-P2 amplitude was smaller for probes embedded within high-importance information. The only difference between the pilot study and Experiment 2 was that in the pilot subjects read each passage twice before the free recall. Subjects' self-reports indicated that this procedural change led to a substantial change in strategies. Specifically, subjects reported that they focused on the main ideas during the first reading, and tried to integrate the details during the second reading. This was reflected in the Nl-P2 amplitude data. Nl-P2 amplitude was smaller for high-importance information, but only during the first reading. To verify this result, the procedure used in the pilot study was replicated, and is reported below.
Experiment 2B
The purpose of Experiment 2B was to replicate the
pilot study described above. Ten subjects participated. The materials, apparatus, and procedure were identical to Experiment 2, except that subjects read each passage twice prior to the free recall. Probes were presented at different locations during the first and second readings. The important prediction is a reading span x presentation x importance interaction. Specifically, Nl-P2 amplitude should be smaller for probes embedded in high-importance




Full Text

PAGE 1

AN ELECTROPHYSIOLOGICAL AND BEHAVIORAL ANALYSIS OF COGNITIVE LOAD DURING READING By GARY ERNEST RANEY A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 1990

PAGE 2

ACKNOWLEDGMENTS I would like to thank the members of my committee, Ira Fischler, Jeff Farrar, Richard Griggs, Keith White, and Don Childers, for their help and guidance with this dissertation. A special note of thanks needs to be given to my advisor, Ira Fischler. Ira has helped me to learn how to conduct research, he has allowed me to pursue my own interests, he has provided guidance and support whenever I needed it, and as importantly, he has treated me as a friend as well as a student. Additional thanks need to be given to my parents, who are a constant source of support. An immeasurable amount of thanks needs to be given to my wife, Jill, for helping me make this possible. I give her my thanks and my never-ending love.

PAGE 3

TABLE OF CONTENTS page ACKNOWLEDGMENTS a ABSTRACT v CHAPTERS 1 INTRODUCTION Overview of Dissertation 3 The Secondary Task Procedure 4 Reaction Time Studies of Cognitive Load 6 Event-Related Potential Studies of Cognitive Load. 14 Description of ERPs 14 ERPs and Cognitive Load 15 2 EXPERIMENT 1 19 Introduction 19 Method 2 6 Subjects 26 Materials and Apparatus 26 Procedure 31 Results 36 Comprehension Quiz Data 3 6 ERP Data .... .36 Reaction Time Data 4 8 Discussion 49 Experiment IB 53 Results and Discussion 55 Experiment 1C 56 Results and Discussion 57 General Discussion 59 3 EXPERIMENT 2 68 Introduction 68 Method 7 7 Subjects 77 Materials, Apparatus and Procedure 77 Results 80 Recall Data 80 ERP Data 83 Discussion 92

PAGE 4

Experiment 2B 94 Results and Discussion 95 Recall Data 95 ERP Data 9 6 Experiment 2C 100 Results and Discussion 100 General Discussion 103 4 SUMMARY AND CONCLUSIONS 106 Text Difficulty 107 Text Structure Ill Future Research 113 APPENDIX A PASSAGES USED FOR EXPERIMENTS 1 AND 1C 117 APPENDIX B SENTENCES USED IN THE READING SPAN TEST.... 12 APPENDIX C PASSAGES USED FOR EXPERIMENTS IB, 2, 2B, AND 2C 125 REFERENCES 129 BIOGRAPHICAL SKETCH 13 5

PAGE 5

Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy An Electrophysiological and Behavioral Analysis of Cognitive Load During Reading By Gary Ernest Raney August, 1990 Chairman: Dr. Ira Fischler Major Department: Psychology The effects of text difficulty and structure on cognitive load during reading were examined using a secondary task procedure. The N1-P2 component of the event-related potential and reaction time to secondary auditory probes were used as measures of load. In three experiments, subjects read a text twice while probes were presented. Experiment 1 found larger N1-P2 responses to probes during the second than first reading for high "reading span" subjects, even when no response to the probes was required. When shorter passages were read in Experiment IB, larger N1-P2 amplitude during the second reading occurred for high and low reading span groups. In Experiment 1, probe detection latencies were measured for other subjects. Detection times were longer during the second reading. Experiment 1C showed the pattern of detection times was not due to a lack of response

PAGE 6

competition between the reading and probe detection task. The N1-P2 data indicate reduced cognitive load during the second reading, while detection times indicate the opposite. It was concluded that N1-P2 amplitude and probe detection times indicate different aspects of cognitive load. The N1-P2 reflects early occurring, lower-level processes, while detection time reflects later occurring, higher-level processes. The levels effect which refers to the greater recall of highversus low-importance information after reading, was examined in three additional studies. It was hypothesized that more resources are devoted to encoding high-importance information, which should lead to increased processing load. In Experiment 2, no levels effect was found in free recall protocols or in the N1-P2 response to probes embedded in highor low-importance information. Experiment 2B found that when subject strategies led them to focus on the main points, N1-P2 amplitude was smaller when processing high-importance material, but only for high reading span subjects. Experiment 2C found no difference in detection times for probes embedded in highor lowimportance information. It was concluded that if the task is structured in a manner that does not limit subjects' strategies and allows them to freely allocate their resources, additional resources may be devoted to the processing of high-importance material.

PAGE 7

CHAPTER 1 INTRODUCTION Reading is an extremely complex and important skill which is used by most individuals every day of their adult lives. The importance of reading is reflected in the amount of research conducted to try to understand this skill. One important area of research concerns the relation between the amount of cognitive effort put forth during reading and subsequent memory and comprehension of the material being read. The notion of cognitive effort may be thought of in terms of an analogy to physical effort. Just as a heavy weight demands more muscular exertion to be lifted than does a light weight, a passage of text with many complex or difficult ideas may demand more mental effort to process than does a relatively simple passage. The amount of mental effort required is commonly referred to as the cognitive load. Cognitive load has been used as an umbrella term to describe the types and amount of mental muscle, or what cognitive psychologists call resources or cognitive capacity, required to perform a task such as reading. These resources include attention, capacity, effort, storage and retrieval processes, and processing strategies, to name a few (e.g., Britton & Tesser, 1982; Kahneman, 1973; Navon & Gopher, 1979). The central theme 1

PAGE 8

in the area of reading research is that different types of text material, as well as different sections within a text, may require different amounts of one's cognitive capacity. Increased capacity requirements during reading are taken as an indicator of greater cognitive load (Britton & Tesser, 1982; Britton & Price, 1981). Within this dissertation, the concept of cognitive load will be used to examine theories of reading and text comprehension. Two issues will be addressed. The first concerns the effect of text difficulty on cognitive load. The general idea is that difficult texts should require greater effort to process than relatively easy texts. If true, then a measure of cognitive load should reflect this difference. Prior research addressing this issue is inconclusive. The second issue addressed is known as the "levels effect." The levels effect refers to the fact that when individuals recall a text, they tend to recall a greater proportion of the main ideas than supporting ideas or details. Some theories predict that the main ideas of a text are processed differently, and possibly to a greater extent, than supporting ideas and details (e.g., Kintsch & van Dijk, 1978; Meyer, 1975). It is appropriate to ask if this differential processing requires more attention or effort. If more attention is required, then a measure of cognitive load should reflect this extra processing. The research in this area is also inconclusive.

PAGE 9

There is an abundance of literature addressing the effect of text difficulty on cognitive load and the levels effect. However, all of the studies are based on behavioral measures, such as recall tests and reaction time to detect a secondary event while reading. The present studies will make use of these behavioral measures, but will also use an electrophysiological measure of cognitive load based on the Event-Related Potential (ERP) It is hoped that by using converging measures that some clarity will be brought to the research on cognitive load and reading. Overview of Dissertation The remainder of this chapter will be organized as follows. First, since the majority of research described in the following review, as well as each of the studies performed for this dissertation, were based on a version of the secondary task procedure, a brief review of this methodology will be given. Second, research involving the effect of text difficulty on cognitive load will be examined. In Chapter 2 I will introduce the method used in the first experiments and describe their purpose. Three studies which examine the relation between text difficulty and cognitive load will then be presented. The literature on the levels effect will be reviewed in the introduction to Chapter 3. Three studies which examine the effect of informational importance (the levels effect ) on cognitive load will then be presented. Chapter 4

PAGE 10

provides a summary of the primary results and final conclusions. The Secondary Task Procedure The secondary task procedure, which is also called a dual task procedure, has been frequently used to measure the cognitive load or capacity requirements to perform a task. In the secondary task procedure a primary task is performed concurrently with a secondary task. The assumption underlying the secondary task method is that individuals have a limited amount of mental resources, and the primary and secondary tasks will compete for these limited resources (Kahneman, 1973; Navon & Gopher, 1979). Kahneman (1973), Kerr (1973), and Wickens (1983) each provide detailed reviews of the secondary task method. Kahneman (197 3) originally proposed that there was a single pool of resources from which the primary and secondary tasks drew. Navon and Gopher (1979) subsequently proposed a multiple resource view. They suggested the existence of separate pools of resources for different processes, although the total amount of resources is still limited. For example, there may be separate pools of perceptual resources, cognitive resources, and response resources. The multiple resource view allows different secondary tasks to have different levels of competition for resources with the primary task. Regardless of whether a single or multiple resource view is taken, the logic of the method is the same; as the

PAGE 11

cognitive demands of the primary task increase, there will be less cognitive capacity available for performing the secondary task. When the resources available for completing the secondary task are reduced, performance on the secondary tasks should decrease. Decreased performance on the secondary task is taken as an indicator of increased demands or cognitive load in the primary task. Multiple resource models also reguire that the primary and secondary tasks compete for common resources. Thus, if the primary and secondary task do not overlap in any manner, changes in the primary task should not affect performance on the secondary task (Wickens, 197 6) The most common secondary task used in studies of reading is secondary probe detection. In this task, subjects read a text (the primary task) while a series of unrelated secondary probes are presented for the subject to respond to (the secondary task) The probes are usually short duration (100 msec or less) tones or light flashes. Both behavioral and Event-Related Potential (ERP) measures of performance have made use of this methodology. In behavioral studies, reaction time to detect the probes is recorded. Increased reaction time to detect a probe is taken as indicating greater cognitive demands in the primary task. In electrophysiological studies, the ERP response to the probes is recorded. A smaller amplitude ERP response indicates greater cognitive demands by the primary task. Different "components" of

PAGE 12

6 the ERP may be sensitive to different aspects of cognitive load. This will be discussed in greater detail later in this dissertation. Reaction Time Studies of Cognitive Load The secondary probe task has been extensively used to study changes in cognitive load during reading (see below) The working hypothesis of these studies was that as the demands of the reading task increase, reaction time to detect a secondary probe should also increase. Britton and Price (1981) validated this assumption by examining the performance operating characteristic of this procedure. Performance operating characteristics describe the pattern of performance changes in one task as the characteristics of a second task are varied. Britton and Price instructed subjects to focus different amounts of attention on reading (the primary task) and probe detection (the secondary task) Their results indicated that as more attention was given to the reading task, probe detection latencies increased. They concluded that the probe detection task appears to be sensitive to the amount of attentional capacity given to a primary reading task. Easy vs. Difficult Text In a series of studies by Bruce Britton and his colleagues, subjects read passages that differed in difficulty while performing the secondary probe detection task. In Britton, Westbrook, and Holdredge (1978) and Britton, Zeigler, and Westbrook

PAGE 13

(1980) texts were short passages which ranged in difficulty from primary (easy) to college (difficult) level. In Britton, Holdredge, Curry, and Westbrook (1979) a common set of passages was presented with a title, which made them easy to understand, or without a title, which made them difficult to understand. Thus, in these three studies semantic difficulty was manipulated. Surprisingly, the results of each study showed longer probe reaction times during the reading of easy text than during the reading of difficult text. Based on the premises of the secondary task technigue, the authors concluded that easy text filled cognitive capacity more than difficult text. Britton and his colleagues have also used secondary task performance to measure cognitive load when structural aspects of a text are manipulated. In Britton, Glynn, Meyer, and Penland (1982) subjects read passages composed of simple or complex vocabularies and syntax while holding passage content (meaning) constant. They also presented passages which did or did not contain cues to signal the importance of ideas in the text (e.g., "therefore," "in contrast") Reaction time to the probes was slower for text with complex syntax and for text without signals. That is, reaction time was longer for the difficult conditions. In contrast, in Britton et al.'s prior studies probe response times were faster for difficult text than for easy text.

PAGE 14

8 The results of Britton Glynn, Meyer, and Penland (1982), which showed faster probe responses when reading easy text, appears to be in conflict with Britton et al.'s (1978, 1979, 1980) earlier studies, which found just the opposite. One explanation proposed by Britton, Glynn, Meyer, and Penland (1982) to explain this difference was referred to as the cognitive contents hypothesis. According to this hypothesis, the amount of capacity used when reading is a function of the amount of information activated. As more information is activated by the primary task, more capacity is used, which leaves less capacity for the secondary task. Since easy text is more readily understood, more propositions (from the text and memory) may be activated. This would leave less capacity to respond to the probes, producing longer reaction times. A study by Britton and Tesser (1982) provides another explanation which they called the prior knowledge hypothesis. They suggested that prior knowledge uses some of the individual's limited cognitive capacity just as performing the task uses capacity. Thus, as the amount of prior knowledge is increased, there will be less capacity available for performance of the secondary task. To test this, subjects read passages with and without prior knowledge of the material. Secondary probe detection times were longer when prior knowledge was given. This supports the hypothesis that prior knowledge uses cognitive capacity.

PAGE 15

The prior knowledge hypothesis can be used to explain the apparent inconsistency between Britton et al.'s early (1978, 1979, 1980) and later (Britton et al 1982) work. In Britton, Holdredge, Curry, and Westbrook (1979), difficulty was manipulated by including (easy) or excluding (difficult) the title to the story. Since the passages were constructed to make the topic difficult to perceive without a title, little prior knowledge could be employed when no title was given. This would lead to less capacity being used in the difficult condition. In Britton, Westbrook, and Holdredge (1978) and Britton, Zeigler, and Westbrook (1980), difficulty was defined by the predictability of the upcoming material using the cloze procedure. They suggest that since easy texts were more predictable than difficult texts, more prior knowledge could be applied to the easy texts. This would lead to easy texts filling more capacity. In Britton, Glynn, Meyer, and Penland (1982) semantic content was held constant while syntactic complexity was varied. In this case increased complexity did require more capacity. Since semantic content was held constant for easy and difficult conditions, equal amounts of prior knowledge could be instantiated. Thus, the difficult syntax would simply add to the processing capacity requirements. Inhoff and Flemming (1989) suggest another explanation which is based on a methodological problem in Britton et al.'s studies. Namely, subjects may have

PAGE 16

10 reread portions of the texts in order to understand the material, which could reduce cognitive load. Since in Britton et al.'s studies probes were applied on a random basis, probes could occur after rereading a segment of a text. This "second reading" may have made the portion of a text following a rereading easier to understand, and, consequently, reduce cognitive load. If a probe occurred after a rereading, it would reflect an artificially reduced cognitive load. There is evidence that rereading makes a text easier to process. Levy and her coworkers (Levy & Begin, 1984; Levy, Newell, Snyder, & Timmins, 1986; Levy & Kirsner, 1989) have analyzed the changes in processing which occur during repeated readings of a text. In a series of studies based on proofreading tasks, they have shown that processing efficiency increases at both the perceptual level, such as converting print to words, and the cognitive level, such as comprehending the general meaning of a passage, during a second reading of a text. Therefore, if in Britton et al.'s studies more rereadings occurred for the difficult texts, then cognitive load would be reduced and probe reaction times could be decreased. Research has shown that more regressions (rereadings) to earlier parts of a text do occur when reading difficult text than when reading easy text (Inhoff, 1983; Just & Carpenter, 1987). Furthermore, if more rereadings occurred for difficult text, then reading

PAGE 17

11 times should be greater for difficult than for easy text. In each of Britton et al.'s studies reading time was greater for the difficult texts, supporting the possibility of increased rereadings. As in prior studies, Inhoff and Flemming (1989) presented subjects with easy and difficult passages to read for comprehension while performing the secondary probe detection task. Passage difficulty was determined by having a group of subjects rate the general difficulty of a passage. There was an important procedural difference between this study and the earlier work by Britton et al In Britton et al.'s studies, an entire passage was presented for the subjects to read in a normal manner. Inhoff and Flemming (1989) used a self-paced, single word presentation format. In this procedure, only one word at a time was seen and subjects pressed a key to make successive words appear. This eliminated the possibility of rereading prior parts of a text. The results of Inhoff and Flemming 's (1989) study showed faster probe reaction times for easy text than for difficult text. These results are in direct opposition to Britton et al.'s early work (1978, 1979, 1980). In order to further test the idea that rereading decreases cognitive load, a mini-experiment (N = 5) was conducted in which subjects read a passage twice. Four of the subjects had shorter probe reaction time during the second reading. For the fifth subject there was no difference. All

PAGE 18

12 subjects read slower during the first reading. This is consistent with the idea that rereading may reduce the difficulty of a text and suggests that this possibility needs to be controlled. Inhoff and Flemming (1989) discussed their results in terms of a modified cognitive contents hypothesis. According to the contents hypothesis, propositions from the text and memory are activated during reading. If difficult text activated more propositions, then more capacity would be used and responses would be slowed. Recall that Britton, Glynn, Meyer, and Penland (1982) suggested that easy text activates more propositions. Research by Kintsch and van Dijk (1978) however, supports the idea that difficult text may lead to the activation and search of more propositions than easy text. This supports Inhoff and Flemming 's view of the cognitive contents hypothesis. The prior knowledge hypothesis was not addressed by Inhoff and Flemming (1989) but may be applied to their study. There are some differences between Inhoff and Flemming's materials and Britton et al.'s materials. First, Inhoff and Flemming defined difficulty by having subjects rate the overall difficulty of the passage. Britton et al. (1978, 1980) defined difficulty in terms of cloze probability (word predictability) and by the presence or absence of a title (i.e., knowing or not knowing the topic) in Britton et al (1979). As

PAGE 19

13 previously mentioned, Britton et al.'s easy passages may have instantiated more prior knowledge. For Inhoff and Flemming's results to be consistent with the prior knowledge hypothesis, it would have to assumed that the difficult passages led to the use of more prior knowledge than the easy passages. The amount of prior knowledge is, of course, an unknown, but the origin of their materials raises this possibility. Inhoff and Flemming's passages were excerpts from out-of-print introductory psychology textbooks, weekly news magazines, and newspapers. Passages dealt with topics such as an experiment at a hospital in Wisconsin and brain regulatory mechanisms. The topics of both the easy and difficult texts may be considered somewhat unfamiliar, making it possible that no more prior knowledge could be applied to the easy than difficult passages. If this was true, then the added semantic and syntactic complexity of the difficult passages would lead to increased cognitive load. It should be noted that explanations based on differences in the materials are post hoc. It is also possible that these changes do not matter, and that the important difference is methodological. In Britton et al.'s studies the passages were read in a normal fashion while in Inhoff and Flemming's study the passages were read one word at a time. Thus, the opposite pattern of results may be in part due to the manner in which the

PAGE 20

14 materials were presented. Furthermore, the prior knowledge hypothesis is inconsistent with the results of Inhoff and Flemming's rereading study, which found faster probe response during the second reading of a passage. In summary, there is evidence that easy text fills more cognitive capacity than difficult text (Britton et al. 1978, 1979, 1980), and vice versa (Britton et al., 1982; Inhoff and Flemming, 1989). Two explanations were proposed to explain this pattern of conflicting results. The first is the possibility of rereading prior material. The second is the interaction of prior knowledge with the semantic and syntactic aspects of the reading material. Future research needs to control for each of these possibilities. Event-Related Potential Studies of Cognitive Load Description of ERPs Event-related potentials are a measure of the electrical activity generated by a collection of neurons. The ERP is described in terms of the peaks and troughs of the waveform, which are referred to as components. The latency and polarity of a component are then used to describe the ERP wave. For example, a negative (N) going wave which occurs at least 100 msec after stimulus onset is labeled an N100. Components are sometimes combined if an event typically influences the components together. For example, the N100-P200 (abbreviated N1-P2) is a negative followed by a positive wave.

PAGE 21

15 Early components (typically 100 msec or less after stimulus onset) are called exogenous components and primarily reflect physical aspects of a stimulus, such as intensity. Later components are called endogenous components and primarily reflect cognitive aspects of a stimulus, such as whether a word makes sense within a sentence (Hillyard & Woods, 1979) Components are also distinguished by their sensitivity to experimental manipulations and their distribution across the scalp. The N1-P2 has been shown to be responsive to a subject's attentional focus within a task and is maximal (in amplitude) at the vertex (Naatanen, 1988; Picton & Hillyard, 1975). The P300 is sensitive to rare or unigue events and some memory processes and is maximal at central posterior locations (Gopher & Donchin, 1986) The experiments performed for this dissertation were designed around the N1-P2 component of the ERP. Therefore, the following review will focus on this component. ERPs and Cognitive Load Event-related brain potentials have been used as an indicator of cognitive load, although they have not been applied directly to the study of reading. The typical ERP study of cognitive load uses the secondary probe method in the same manner as behavioral (reaction time) studies. A primary advantage of ERPs is that they do not require an overt response to be recorded, although both behavioral

PAGE 22

16 and electrophysiological (ERP) measures are usually gathered. A number of properties of ERPs may be evaluated. The most common are peak amplitude and latency of specific ERP components. As with behavioral versions of the secondary task method, the sensitivity of the ERP to variations in the attentional demands of the primary and secondary tasks needs to be demonstrated. Picton and Hillyard (1974) were the first to show that attending to a series of stimuli (tones) enhanced the amplitude of the N1-P2 relative to nonattended stimuli. Schwent and Hillyard (1975) Hink, van Voorhis, Hillyard, and Smith (1977), and Hink, Hillyard, and Benson (1978) further showed the sensitivity of the N1-P2 to selective attention using variations of a dichotic listening paradigm. In these studies, subjects were told to attend to or ignore stimuli in a specified sensory channel (e.g., left, right, or both). The authors found that the amplitude of the Nl reflects the distribution of attentional resources among competing inputs. They proposed a relation between the informational load on a subject and the size of the Nl enhancement for attended versus non-attended information. Each of the above studies found the effect to be maximal at the vertex. Note that some of the prior studies focused on the Nl component only. Depending on the methodology used, the Nl and P2 can be separated (Picton, Hillyard, Krauz &

PAGE 23

17 Galambos, 1974) Subsequent work has verified the relation between selective attention and the amplitude of both the Nl and P2 components of the ERP (Hillyard, 1985; Naatanen & Michie, 1979; Nakahara & Ikeda, 1987; Parasuraman, 1978, 1980). These studies consistently support the conclusion that the amplitude of the N1-P2 may serve as an index of the amount of cognitive resources devoted to processing a channel of information. It should be noted that the neural process responsible for enhanced Nl amplitude is still under debate (Hillyard, 1981; Naatanen, 1988; Naatanen & Michie, 1979), but the fact that amplitude changes do occur is not in question. The concept of competing channels of inputs used in the prior studies can be applied to the secondary task procedure. In the secondary task there are two channels (tasks) which are competing for attentional resources. If attention is focused on the primary channel (task) then the amplitude of the N1-P2 response to probes in a secondary channel should vary depending on the amount of "residual" attention allocated the secondary task. In summary, there is evidence that N1-P2 responses to a secondary probe reduce in amplitude as more attentional resources are devoted to the primary task. The research has not, however, directly addressed the issue of reading. Other research reviewed by Fischler and Raney (in press) has shown the ERP to be a sensitive measure of lexical, semantic, and syntactic factors in sentence processing.

PAGE 24

18 Thus, there is reason to believe that ERPs may provide a method for evaluating cognitive demands during text processing.

PAGE 25

CHAPTER 2 EXPERIMENT 1 Introduction Experiment 1 was designed to explore the effects of text difficulty on cognitive load using both behavioral (reaction time) and electrophysiological (ERP) measures. There were two primary purposes for Experiment 1. The first was to try and clarify the relation between text difficulty and cognitive load. The second was to determine if ERPs could be used as a nonintrusive measure of cognitive load. The procedure used may be called a read-reread paradigm. Subjects read a text, and then immediately reread the same text. The first reading served as the difficult condition and the second as the easy condition. The passages were presented one word at a time in a common location on a display. This technique is known as the single-word Rapid-Visual-Serial-Presentation (RSVP) technique (Potter, 1984) While reading, subjects were presented with a series of secondary auditory probes. Two response conditions were included. In the first, both reaction time to detect the probe and ERP responses to the probe were measured. In the second condition, only ERP responses were measured. 19

PAGE 26

20 To use a rereading paradigm, it must be demonstrated that the first reading is more cognitively demanding than the second reading. Recall that Levy and her colleagues (Levy & Begin, 1984; Levy, Newell, Snyder, & Timmins, 1986; Levy & Kirsner, 1989) have shown that both perceptual and cognitive processes become more efficient during a second reading of a text. Britton, Glynn, and Smith (1985) also describe the effects of rereading on processing load. They suggest that the output products of a first reading, such as the development of a schema for interpreting the material, will be available during the second reading. Access of text-related information should be facilitated, and cognitive load should be reduced. Also recall that Inhoff and Flemming (1989) had subjects reread a text while performing a probe detection task. They found shorter probe detection times during the second reading, which is suggestive of reduced cognitive load. Furthermore, reading times for a second reading of a text are shorter (Inhoff, 1983, Just & Carpenter, 1987). Thus, there is strong support for the contention that a second reading reguires reduced effort. Individual Differences in Cognitive Load To explore the possibility of individual differences in cognitive load, subjects were grouped based on their individual reading span scores (Daneman and Carpenter, 1980) In the reading span test, subjects read sentences and are asked to remember the final word of each sentence. Reading span

PAGE 27

21 is taken as the maximum number of sentences a subject can read and then accurately remember their final words. The reading span test was proposed as a measure which taxes both the processing and storage functions of working memory. Daneman and Carpenter (1980) suggested that one aspect of individual differences in reading ability may be working memory capacity. A good reader may have more efficient reading processes, which would leave more capacity available for memory. In contrast, poor readers may have less efficient processes, which would leave less capacity for memory (Daneman, 1987) The sensitivity to both processing and storage operations makes this measure particularly useful in the present study. While reading, the reader must simultaneously perform processing and storage operations. Thus, reading span taps two of the primary processes which may contribute to cognitive load. Reading span has been shown to be significantly related to traditional measures of reading ability, such as verbal SAT scores and to performance on reading comprehension tests (Daneman and Carpenter, 1980) In addition, individuals with high reading spans have been found to be better at resolving inconsistencies within sentences (Daneman and Carpenter, 1983) and constructing the meaning of an unfamiliar word from context (Daneman & Green, 1986) Evaluation of the Method There are a number of benefits to the rereading paradigm and RSVP. First, since

PAGE 28

22 the same passages are used in the easy and difficult condition, perfect control is obtained over syntactic, semantic, and vocabulary differences between easy and difficult texts. Second, using the RSVP method eliminates the possibility of regressions (rereading earlier parts of a text) Third, since the rate of presentation is constant, overall cognitive load for the difficult conditions cannot be decreased by increasing the reading time. This is important since Britton, Muth, and Glynn (1986) showed that when reading time is limited, increased effort will be allocated to difficult portions of a text. When reading time is not limited, subjects will slow down, effectively reducing the load at that point. The use of ERPs as a measure provides the ability to address a primary criticism of the secondary task procedure; that simultaneously performing a primary and secondary task changes the normal processing demands of each task (Kahneman, 1973; Kerr, 1973; Wickens, 1983). To reduce the intrusiveness of a secondary task, it has been suggested that the task be based on discrete events which compete only for processing resources (i.e., no common response demands) and occur on a small proportion of trials during the primary task (Brown, 1978; Kerr, 1973). ERPs to secondary probes meet these requirements. In contrast with behavioral measures, ERPs provide a method for measuring behavior without any overt response to the secondary probe, which eliminates the secondary task

PAGE 29

23 entirely, and makes the probe potentially less intrusive. This also eliminates competition for response resources and leaves competition only for processing resources. ERPS could thus offer a more valid measure of cognitive load than behavioral measures (Wickens, 1983) It should be noted that most ERP studies have required subjects to make an overt response to the probe; thus the competition for response resources is still a problem. One possible drawback of not requiring a response to the probes is that the probes may be ignored, which may reduce the effect of cognitive load on the ERP measure. This issue is examined in the present study by including ERP conditions with and without a response to the probes. The restrictions suggested by Brown (1978) and Kerr (1973) do not invalidate the usefulness of behavioral measures. Consistent with their suggestions, a response to a secondary probe is a discrete event and occurs during a small proportion of trials. But since a response is required, there still may be competition for response resources. What is important is whether the primary and secondary tasks have common response demands (e.g., both require a response with the same hand) This is not the case in the previously discussed studies of reading. Furthermore, Britton, Piha, Davis, and Wehausen (1978) have shown that the addition of a secondary probe detection response does not lead to reduced performance on

PAGE 30

24 the reading task. This implies that the response to the probes does not significantly disrupt reading, and is therefore a useful measure of cognitive load. Hypotheses The rereading paradigm provides a method for testing the cognitive contents hypothesis proposed by Inhoff and Flemming (1989) and the prior knowledge hypothesis proposed by Britton and Tesser (1982) The primary assumption of each of these hypotheses is that when less cognitive resources are reguired by the reading task, more attentional capacity should be available to respond to the probes (either overtly or covertly) According to the cognitive contents hypothesis, the second reading should require less capacity; therefore, more capacity should be available to respond to the probes. According to the prior knowledge hypothesis, the second reading should actually consume more capacity. Since the first reading makes the passage familiar, a larger amount of prior knowledge should be invoked during the second reading. This would leave less resources available to respond to the probes. Thus, the cognitive contents hypothesis and the prior knowledge hypothesis lead to two different sets of predictions. The following should occur if the cognitive contents hypothesis is correct. First, reaction time to detect the probes should be longer during the first reading of a passage. Second, the amplitude of the N1-P2 ERP response to the probes should be smaller during the first reading.

PAGE 31

25 Past research has found the effect to be largest at the vertex, and a similar pattern is expected here. Third, each of these effects should be mediated by an interaction with reading span. Specifically, if high reading span subjects are able to comprehend and remember more of a passage during the first reading, then the second reading will be needed to a lesser degree. In other words, the reduction in cognitive load will be greater for the high reading span group. In contrast, the prior knowledge hypothesis predicts the following. First, reaction time to detect the probes should be faster during the first reading of a passage. Second, the amplitude of the N1-P2 ERP response to the probes should be larger during the first reading. The effect should be largest at the vertex. Third, a weak prediction is made for the main effects to interact with reading span. For an interaction to occur, it must be assumed that the high reading span group has more prior knowledge to apply than the low reading span group. This assumption is tenuous, at best. Therefore the interaction with reading span may not be as reliable as predicted by the cognitive contents hypothesis, which bases its prediction solely on memory capacity. For both the cognitive contents and prior knowledge hypothesis, a main effect of reading span might be predicted. Since the high reading span individuals tend to be better readers (Daneman, 1987) the task may be

PAGE 32

26 easier for them. This represents a between-subjects difference; thus the effect might not be strong. Furthermore, there is little evidence from past research regarding between groups differences for normal subjects in overall amplitude of the ERP. Method Subjects Forty University of Florida undergraduates (15 men and 25 women) participated as part of an introductory psychology course reguirement. All subjects reported being native English speakers and having normal or corrected-to-normal vision. Data from eight additional subjects were not included due to unacceptably high electrode impedances, technical problems, or experimenter error. Materials and Apparatus Stimulus Presentation System An IBM PC AT controlled the stimulus presentation. The text passages were displayed on a Samsung high resolution monitor (amber characters on a black background) which was connected to the monochrome output of a Color Graphics Adaptor (CGA) on the PC. The monitor was located 2/3 of a meter directly in front of the subject. While subjects read each passage, they heard a series of auditory probe tones. Probes were presented using the electronic circuit described by Raney and Shuman (1989) which was triggered by the PC. Probes were played through a 2 1/4 inch

PAGE 33

27 speaker located approximately 1/2 meter behind the subjects. Text Passages Twelve passages were taken from GRE (Martinson & Crocetti, 1987) SAT (The College Board, 1988; Gruber, 1988), and ACT (Martinson, Fazzone, Haynes & Haynes, 1988) study manuals for use as stimuli in Experiment 1. These manuals were chosen since they contain well-written passages which have been standardized at an appropriate level of difficulty for college students. The passages were modified in three ways. First, passage length was adjusted, if necessary, to be between 418 and 4 37 words. Second, very rare or uncommon words were replaced with a simpler term or a short phrase which conveyed the same meaning. Third, an attempt was made to replace words with 12 or more letters with shorter length words or with a phrase which conveyed the same meaning. Words which could not be easily replaced were presented in two parts, that is, hyphenated (e.g., anthropologist was presented as "anthro-" "pologist") The second and third modifications were needed to make the passages more readable within the single-word presentation method (described below) The 12 passages were divided into two groups of six (Group A and Group B) Subjects were alternately assigned to read passages from Group A or B. Two sets of passages were used in an attempt to increase the generalizability of the results, although passage group was not treated as

PAGE 34

28 a factor in the experiment. After reading a text once or twice (described below) subjects completed a fivequestion, short answer quiz over the material. A sample passage and its corresponding quiz is presented in Appendix A. The source of each passage used in this experiment is also presented in Appendix A. While subjects read each passage, a series of 10 auditory probe tones was presented. The probes were 100 msec, 1000 Hz, 60 dB tones. The intensity of the tones was determined using a Radio Shack, hand held sound level meter (model 33-2050) Intensity was measured by placing the meter in the same position as a subject's head would be during the experiment, and then reading peak intensity using the "A" weighting scale. The location of the probes within each passage was determined in the following manner. Each passage was divided into 10 approximately equal length sections. Within each section, probe location was randomly determined for each of the twelve passages based on the following five restrictions. First, probes were not presented during the first 10 words of a passage. Second, probes were presented only during the display of a content word (hereafter referred to as a target word) Third, probes were not presented during the first word of a sentence. Fourth, probe locations were different for the first and second reading of a passage (described below) Fifth, probes were not presented during the display of

PAGE 35

29 either part of hyphenated words. All probes were presented 100 msec after the onset of a target word and the same probe locations were used for each subject. Reading Span Test The reading span test was composed of 80, unrelated English sentences which varied in length from 9 to 13 words. Each sentence was typed on a 4 x 6 inch (10 x 15 cm) index card. The sentences were arranged into groups of 2, 3, 4, 5, and 6 sentences. There were four sets of each group size. Blank cards were inserted between each group to serve as recall markers (described below) Sixty of the sentences were created for the experiment, and 2 were taken from Daneman and Green (1986, Experiment 2) These 20 sentences were shortened in length to match the initial 60. The sentences used in the reading span test are presented in Appendix B. The final 2 sentences are those which were abstracted from Daneman and Green (1986) EEG and RT Recording System EEG and RT data collection were controlled by the PC, which was equipped with a Scientific Solutions (Tecmar) Labmaster board. The Labmaster performed the analog-to-digital conversion of the EEG data, collection of the RT data, and the timing of the stimulus presentation during data collection. During the experiment, subjects were seated in a sound attenuated, electrically shielded booth. A circulation fan produced a background noise of approximately 50 dB.

PAGE 36

30 EEG was recorded from CZ F3 F4 and a left and right hemisphere temporal-parietal location (which will be referred to as LS and RS, respectively) according to the International 10/20 System (Jasper, 1958) LS and RS refer to the area near the end of the sylvian fissure on the left (LS) and right (RS) hemisphere. Specifically, these points are located midway between T3 and P3 (for LS) and between T4 and P4 (for RS) The region around LS is more classically known as Wernicke's area (Geschwind, 1979). Each electrode was referenced to linked mastoids. Eye blinks were monitored using electrodes placed supraorbitally and behind the external canthus of the right eye. Eye blink artifacts in the EEG signal were corrected using the procedure described by Gratton, Coles, and Donchin (1983) The EEG signals were amplified using Coulbourn Hi-Gain Bioamplif iers (Model S75-01) with a bandpass of 0.1 to 40 Hz and a 60 Hz notch filter. The amplified EEG signal was sampled at a rate of 125 Hz per recording site and stored on disk for subsequent analysis. Beckraan Ag/AgCl Biopotential skin electrodes and Beckman Electrolyte gel were used in all cases. Electrode impedances never exceeded 5 Kohms and were measured both preand post-test. EEG data were recorded beginning 80 msec prior to the display of a target word and continued for 920 msec. This period encompassed the 80 msec prior to a target word

PAGE 37

31 through the 4 msec after the offset of the word following the target. EEG data were collected only during the presentation of target words and words immediately following targets. The timing of the data collection period can be seen in Figure 1. Probe detection responses were made using a Microsoft Mouse. The output of the mouse was input as a data channel into the Labmaster board and sampled in the same manner as the EEG data. Conseguently the temporal resolution of the probe detection data (i.e., reaction time) is 8 msec. The probe detection data were stored on disk for later analysis. Procedure All subjects were initially given the reading span test. In this test, subjects read sentences out loud in groups of 2, 3, 4, 5, or 6 and in this order. After the group size reached 6, the group size switched backed to 2 and again increased. This was repeated through four sets of each group size. The subject's task was to remember the last word of each sentence in a group. Subjects were instructed to turn over the index card after reading a sentence so that the sentence would no longer be visible, and then to begin reading the next sentence. This was continued until a blank card was reached, which served as a cue to verbally recall the sentence final words from the preceding group of sentences. Subjects were instructed to read at their own rate and not to take long pauses between

PAGE 38

32 sentences. They were also instructed to try and recall the words in the order in which they were presented, but if they could not remember the order of presentation, they were encouraged to recall the words in any order. They were further told that it was better to recall many words out of order, than only a few words in proper order. The average of the subjects three largest recalls was taken as a measure of their reading span. For example, if a subject correctly recalled groups of 4, 4, 3, and 4, then their score was 4. Prior to analysis, subjects were grouped based on their reading span scores. Those with a reading span score of 4 or greater were classified as high reading span. The remaining subjects were classified as low reading span. Following the reading span test, subjects were prepared for the EEG recording. Once this was completed subjects were seated in a recording booth and the main experiment began. Each subject was shown a set of passages to read, and following each passage they were tested for comprehension. The passages were presented one word at a time at a rate of 120 words-per-minute. Each word was displayed in the center of the monitor for 300 msec. The monitor screen was then cleared, and, following a 200 msec inter-word-interval, the next word was presented. An additional 500 msec delay was added after the last word of each sentence. Eye movement research has shown that short pauses are usually taken between

PAGE 39

33 sentences (Just & Carpenter, 1987) The additional 500 msec delay between sentences was added to simulate this pause and make the task more natural To manipulate cognitive load, subjects read some of the texts twice in succession. The first reading served as the high cognitive load condition (i.e., difficult) and the second reading served as the low cognitive load condition (i.e, easy) A 5 to 10 second break was given between the first and second reading. While subjects were reading, they heard a series of auditory probes. Subjects were alternately assigned to either the response or noresponse condition. Those in the response condition pressed a key on the mouse as quickly as possible when they heard a probe. Those in the no-response group made no response to the probe. In all other respects, the procedures for the two groups were identical. All subjects were read the following instructions prior to beginning the experiment. "In this part of the study you will be presented with 6 short passages to read. Please read the passages silently. Each passage only takes a few minutes to read. The passages will be presented one word at a time on the monitor in front of you. Each word will be presented briefly in the center of the screen and then removed before the next word is displayed. The words will appear and disappear quickly, so you will need to concentrate when reading the words. After reading each

PAGE 40

34 passage you will take a five question, short answer quiz over the material. You have ten minutes to complete each quiz. For half the passages, you will read the passages once and then take the quiz. For the other half, you will read the passages twice and then take the quiz. After each passage a message will appear indicating that the passage is complete. At this point I will tell you whether you will read the passage a second time. I will do this using the intercom located under the monitor. While you are reading, a series of "beeps," or tones will be presented using a small speaker located behind you. The beeps will occur at different points within each passage." Those in the no-response condition were then told: "You do not need to make any response when you hear a beep. You do not need to remember where the beeps occurred. I want to emphasize that your primary task is to read and comprehend the passages." Those in the response condition were then told: "Every time you hear a beep press the left key on the mouse as quickly as possible. You do not need to remember where the beeps occurred. Please try and respond as quickly as possible to the beeps. However, I want to emphasize that your primary task is to read and comprehend the passages." Subjects were given a short demonstration passage which contained six probes. The demonstration passage was

PAGE 41

35 a paraphrase of the previous instructions. Subjects in the response group were instructed to respond to the probes in the demonstration in order to practice the response task. Subjects were alternately assigned to read passages from Group A or Group B. Although subjects were told they would read six passages, they actually read only four. Either the second or third passage presented was randomly chosen to be read only once. This was done to confirm the expectation that some of the texts would be read only once, which insured that subjects concentrated during the first reading. The first passage read by each subject was controlled so that every passage would be presented first approximately an equal number of times. Passage order for the second, third, and fourth passages was predetermined randomly for each subject. Subjects in the response and no-response group were matched so that pairs of subjects received the same passages in the same order. The subjects were allowed to take breaks, if desired, between each passage. After finishing the experiment electrode impedances were again checked. Subjects were interviewed regarding what strategies they used to complete the task and were then debriefed. The entire procedure lasted no more than 2 1/2 hours.

PAGE 42

36 Results Comprehension Quiz Data The quizzes were scored by an undergraduate research assistant who was blind regarding the condition of each subjects' quizzes. Each quiz answer was given a score ranging from zero to three based on the following criteria. A score of zero was given if no answer was provided or if all information in an answer was incorrect. A score of one was given if only a short phrase was provided as an answer. Answers in this group may have contained incorrect information or be based on information not provided in the text. A score of two was given if the answer was correct and longer than a short phrase and contained some detail or explanation. A score of three was given to correct answers with details and explanations using wording from the text. A few of the questions could be completely answered using a short phrase. In this experiment, three passages were read twice before taking a quiz, but only one passage was read only a single time before taking a quiz. This could lead to an inflated variance in the single reading condition, since the average score would be based on fewer quizzes. To control for this, only one quiz from a double reading was used as comparison data. Recall that either the second or third passage presented was read only once. If the second passage was read only once, then the third passage was used as the double reading comparison. If the third

PAGE 43

37 passage was read only once, then the second passage was used as the double reading comparison. This egualized the amount of data included in the analysis and had the added benefit of eguating the number of passages completed before guizzes in the single versus double reading condition. Data from the guizzes were analyzed using a 2 (reading span; high or low) x 2 (presentation; 1st reading and 2nd reading) mixed ANOVA, with reading span being the between-subjects factor. Presentation was significant, F(l,38) = 12.68, p < .001. The mean scores for guizzes taken after one versus two readings was 8.4 and 10.2, respectively, indicating that performance on the guizzes improved after a second reading. Neither reading span nor the reading span x presentation interaction reached significance (both F's < 1.0). ERP Data In all analyses based on ERP data, N1-P2 amplitude was guantified as the difference between the minimum voltage occurring between 108 and 180 msec after probe onset (Nl) and the maximum voltage occurring between 160 and 245 msec after probe onset (P2) Visual inspection of the average ERPs indicated that the latency of the N1-P2 responses fell within this range for each subject. The N1-P2 in this and subseguent analyses refers to the N1-P2 response to the secondary auditory probes. For ease of reference, this will simply be referred to as the N1-P2.

PAGE 44

38 Amplitudes for N1-P2 responses in all analyses are given in microvolts (uv) The N1-P2 amplitude data was subjected to a 2 (reading span; high or low) x 2 (response; yes or no) x 2 (presentation; 1st reading and 2nd reading) x 5 (site; CZ F3, F4, LS, and RS) mixed ANOVA with reading span and response being between-subjects factors. The main effect of site was significant, F(4,144) = 63.68, p < .0001. Tukey's test (at p < .05) indicated that the mean amplitude N1-P2 for CZ (14.99 uv) was significantly larger than that of F4 (10.19 uv) and F3 (9.98 uv) which did not significantly differ. Each of these sites had a significantly larger mean N1-P2 than RS (7.89 uv) and LS (7.44), which did not significantly differ. The reading span x presentation interaction was marginally significant, F(l,36) = 3.87, p < .057. Overall, N1-P2 amplitude for the high reading span group was smaller during the first (9.75 uv) than the second (10.87 uv) reading. This pattern was slightly reversed for the low reading span group (10.08 uv for the first reading and 9.7 uv for the second). However, differences between the first and second reading for either group were not significant. No other effects reached significance (all F's < l.o) ERP waves for the response and no-response groups, separated by reading span, reading presentation, and site are presented in Figure 2-1. Four features are

PAGE 45

39 noticeable. Initially there is a small negative followed by positive response to the target word (T) which begins around 100 msec after target onset. A large N1-P2 to the probes (P) then dominates the waveforms. The Nl component reaches its minimum around 12 msec after probe onset, and the P2 component reaches its maximum around 2 00 msec after probe onset. For the response groups (panels A and B) there is a large, later positive component which peaks at approximately 3 50 msec after probe onset. This component represents the additional processing associated with the probe detection response. This component is substantially reduced in the no-response groups (panels C and D) Lastly, a response to the word following the target (T+l) begins around 100 msec after word onset. Note that there is a high degree of similarity between the waveshapes for the response groups (Panels A and B) and for the no-response groups (Panels B and C) Further, for both response conditions, the waves are highly similar until the point beginning around 250 msec after probe onset, at which time the response group shows the additional positive component associated with the response demand. This high degree of consistency within and between the groups demonstrates the reliability of the ERP measure. Also note that the pattern of the waveshapes are similar for each site, but differ primarily in amplitude.

PAGE 46

Figure 2-1 ERP waveforms for Experiment 1 separated bysite, reading span, and response group. For each panel, the solid line represents the ERP during the first reading and the dashed line represents the ERP during the second reading. Onset times for the target word (T) probe (P) and the word following the target (T+l) are indicated on the abscissa. Panel A represents the high reading span, response group. Panel B represents the low reading span, response group. Panel C represents the high reading span, no-response group. Panel D represents the low reading span, no-response group.

PAGE 47

41 if) ~5 > -10 o u ^ 10 Site = CZ -10 Site = CZ T P -10 T+l J I I I J L 100 300 500 700 Time (msec) 900

PAGE 48

42 10 10 U) ~o > -10 o L (J 3 10 10 10 i i r Site = F3 Site = F3 Site = F3 Site = F3 i r i r T P j u u L T+l J I I I 100 300 500 700 Time (msec) Figure 2-1 — continued 900

PAGE 49

43 -10 if) ~o > -10 o L_ U "*> io -10 1 r Site = F4 Site = F4 Site = F4 Site = F4 i r A 100 300 500 700 Time (msec) Figure 2-1 — continued 900

PAGE 50

44 10 (D "o > -10 o £_ u ^ 10 -10 i i r Site = LS Site = LS Site = LS Site = LS T P i r A T+l 100 300 500 700 900 Time (msec) Figure 2-1 — continued

PAGE 51

45 10 -10 10 o -10 o u 3 10 Site = RS Site = RS Site = RS Site = RS o T P J L T+l J L 100 300 500 700 Time (msec) 900 Figure 2-1 — continued

PAGE 52

46 Since there was an apriori expectation of differences between sites, the N1-P2 data were reanalyzed separately for each site. Thus, the following results were based on 2 (reading span; high or low) x 2 (response; yes or no) x 2 (presentation; 1st reading and 2nd reading) mixed ANOVAs For CZ, the only effect to reach significance was the expected reading span x presentation interaction, F(l,36) = 5.05, p < .031 (all other F's < 1.0). Simple effects analyses of this interaction showed a significantly larger N1-P2 during the second reading than during the first reading for the high reading span group, F(l,36) = 5.64, p < .025, but not for the low span group, F(l,36) = 1.29, p > .05. The means for this interaction are presented in Table 2-1. For F3 none of the effects reached significance at the p < .05 level of significance. The reading span x presentation interaction approached significance, F(l,36) = 3.32, p < .08 (all other F's < 1.0). The pattern of results for this interaction was the same as for CZ but comparisons between first and second readings were not significant for either reading span group (see Table 2-1) For F4 none of the effects reached significance at the p < .05 level of significance (reading span x presentation interaction, F(l,36) = 2.31, p < .15; all other F's < 1.0). (See Table 2-1.)

PAGE 53

47 Table 2-1 Mean N1-P2 Amplitude (uv) for High and Low Reading Span Groups During the First and Second Reading of a Passage High Reading Span Low Reading Span Reading Reading Site First Second First Second cz

PAGE 54

48 Summary of ERP Results When all recording sites were combined, there was a nonsignificant trend towards the predicted reading span x presentation interaction. When recording sites were examined individually, CZ and LS had significant reading span x presentation interactions, and this approached significance for F3 Past research has found attentional effects on N1-P2 amplitude to occur primarily at CZ In this study, LS was just as sensitive to changes in attentional demands as CZ even though the overall amplitude at LS was smaller than CZ At least for the high reading span group, N1-P2 amplitude increased during a second reading of a passage at both CZ and LS Importantly, there was not a main effect of response, nor any interaction with response, at any recording site. This indicates that an overt response to the probes is not needed to obtain the expected N1-P2 difference. Reaction Time Data Reaction time analyses are based only on the 20 subjects in the response group. The reaction time data were analyzed using a 2 (reading span; high or low) x 2 (presentation; 1st and 2nd reading) mixed ANOVA, with reading span as the between-subjects factor. Only the main effect of presentation was significant, F(l,18) = 10.71, p < .005. Mean reaction time was faster during the first than second reading (see Table 2-2) No other effects reached significance (for reading span, F(l,18) = 1.59, p < .21; for the interaction, F(l,18) < 1.0).

PAGE 55

49 Table 2-2 Mean Response Times (msec) to Detect Probes and Standard Deviations fin parentheses) for High and Low Reading Span Groups During the First and Second Reading of a Passage Reading Reading Span First Second Mean High

PAGE 56

50 The interaction between reading span and presentation was expected, but the low reading span group showed no difference at all, which was unexepected. One possible explanation for the lack of difference in the low reading span group is based on reading ability. Prior research has shown a correlation between reading span and reading comprehension test scores of around .55 (Daneman, 1987). It is possible that for the low reading span group both readings of a text were difficult, in the sense that capacity requirements were heavily taxed. That is, low reading span subjects may have had to use maximum capacity to comprehend and store the information in a passage. The idea of using maximum capacity is analogous to a ceiling effect. The amount of capacity used by the low reading span group may have reached its ceiling during each reading. If this is true, then presenting shorter passages, which tax capacity less heavily, should not lead to a ceiling effect. This would allow an increase in N1-P2 amplitude during the second reading for both reading span groups. Support for the idea of a ceiling effect on capacity is also present in the quiz data. Given that reading span is a test of memory capacity and reading comprehension ability, quiz scores should be lower for the low reading span group than for the high reading span group. Since this was not the case, it is tempting to conclude that the low reading span group allocated maximum capacity to the

PAGE 57

51 reading task in order to store as much information as possible. The concept of a ceiling effect on capacity is addressed in an additional study (Experiment IB) presented in the next section of this chapter. A second important finding from the ERP data was the lack of any main effect or interaction with response condition. The pattern of results for those subjects who overtly responded to the probes was identical to those who made no response to the probes. It should be stressed that subjects in the no-response group were not asked to make any type of response to or decision about the probes. In fact, self-reports by subjects indicated that they ignored the probes after hearing them a few times. This suggests the ERP can serve as a measure of cognitive load in the absence of a response. This does, of course, make the ERP measure very nonintrusive. Furthermore, this may allow ERPs to be used when an overt response is not compatible with the primary task. The main finding based on the reaction time data was longer probe detection times during the first reading of a passage. This result is consistent with the prior knowledge hypothesis, which states that the first reading led to the instantiation of a greater amount of prior knowledge during the second reading. Thus, more capacity is used during the second reading, which leaves fewer resources available to respond to the probes. As a result, reaction times increase during the second reading.

PAGE 58

52 There is another reason why probe responses may not have been faster during the second reading; the primary and secondary tasks may not have competed for common resources. This reason does not exclude the explanation proposed by the prior knowledge hypothesis, but may be independent of it. The idea of competition for resources is based on the views of multiple resource theories, which state that two tasks will interfere with each other only if they compete for common resources (Gopher & Navon, 1980; Navon & Gopher, 1979; Wickens, 1983). To determine if common resources are used, the component processes of a task need to be determined. In terms of the reading task, the major components are perceptual and cognitive, such as converting print to words and then determining their meaning. In terms of the secondary task, there are also some perceptual and cognitive demands, since subjects have to perceive and recognize the probe, although these demands are probably minimal. In addition, the secondary task has a large response component, which is the primary component of the secondary task (Wickens, 1983) Thus, the dominant components of the reading task and the secondary task do not overlap. If this rough task analysis is correct, then there is little reason to expect any interference between the reading and secondary probe response tasks, since their primary demands may not compete for common resources (Wickens, 1976)

PAGE 59

53 If this "lack of response competition" hypothesis is correct, it may help to explain the opposite results of the present study and that of Inhoff and Flemming (198 9) Recall that Inhoff and Flemming conducted a mini-study in which subjects read a passage twice while responding to secondary probes. They found shorter reaction times during the second reading. In their study, subjects made successive words appear by pressing a key. In the present study, words were automatically presented at a predeterimined rate. This methodological difference could substantially change the demands of the primary task. In addition to perceptual and cognitive demands, Inhoff and Flemming 's primary task also had response demands. Thus, both the primary and secondary tasks competed for common resources. This response competition hypothesis was tested in an additional study (Experiment 1C) which will be presented in a following section of this chapter. Experiment IB The purpose of Experiment IB was to test the idea of a ceiling effect on capacity. To review, reductions in N1-P2 amplitude during a second reading were found for the high reading span group, but not the low span group. It was proposed that the low reading span group may have had to devote maximal capacity to the task during each reading, therefore no reduction in load was found. To test this hypothesis, the procedure used in Experiment 1 was repeated using shorter text passages,

PAGE 60

54 which should place less demands on cognitive capacity. That is, there is simply less information to process and store in memory. This reduction in capacity demands should allow low reading span subjects to remain below their maximal capacity, which would lead to reduced cognitive load during the second reading of a passage. Ten subjects participated in this study. Six of the subjects were paid ten dollars for their participation. The remaining four participated as part of a class requirement, and were from the same subject pool used in Experiment 1. All subjects were undergraduate students from the University of Florida. The procedure, apparatus, and instructions were identical to Experiment 1 except for the following changes. First, each subject read six short passages which were between 160 and 178 words in length. A sample passage is presented in Appendix C. All the passages were read twice in succession and a free recall was produced after the second reading of each passage. A free recall was used so that subjects would not be limited regarding what could be recalled. During each reading of a passage, eight secondary probes were presented. Since the response and no-response conditions of Experiment 1 produced similar results, only the no-response condition was performed here. Therefore, only ERP data were recorded.

PAGE 61

55 Results and Discussion The N1-P2 amplitude data were analyzed using a 2 (reading span) x 2 (presentation) x 5 (site) mixed ANOVA, with reading span as the between-subjects factor. The main effect of site was again significant, F(4,32) = 21.35, p < .001. Tukey's test (at p < .05) indicated that the mean N1-P2 amplitude of CZ (15.2 uv) was significantly greater than the mean amplitude of F3 (10.02 uv) F4 (9.74 uv) and RS (7.67 uv) which did not differ. Mean amplitude for each of these sites was larger than that of LS (6.32 uv) The main effect of presentation was also significant, F(l,8) = 5.25, p = .05. Mean N1-P2 amplitude was smaller during the first (9.21 uv) than second (10.36 uv) reading. When the sites were analyzed individually, there were no significant effects for F3 F4 or RS. For CZ the only effect to approach significance was presentation, F(l,8) = 3.06, p < .11. Mean amplitude N1-P2 at CZ during the first reading was 14.55 uv, and 15.81 uv during the second reading. For LS, presentation was significant, F(l,8) = 8.31, p < .02. Mean amplitude N1-P2 at LS during the first reading was 5.75 uv, and 6.89 uv during the second reading. The general pattern of results at CZ and LS replicate the results from Experiment 1. That is, N1-P2 amplitude increased during the second reading. Importantly, there were no reading span x presentation interactions in either the overall ANOVA, or

PAGE 62

56 when sites were analyzed individually. As predicted, using shorter passages lead to an increase in N1-P2 amplitude from the first to second reading for both reading span groups. This supports the conclusion that, in Experiment 1, there may have been a ceiling effect on capacity. Specifically, the low reading span group may have been devoting full capacity to both the first and second readings when long texts were being read. Experiment 1C The purpose of Experiment 1C was to test the "lack of response competition" hypothesis. To review, it was proposed that since the reading task contains no response reguirements, there should be no competition for resources with the secondary probe response, which is comprised primarily of response demands. To test the "lack of response competition" hypothesis, a small study was conducted which added response demands to the reading task. Experiment 1C replicated the self -paced procedure used by Inhoff and Flemming (1989) That is, subjects pressed a key to display successive words while simultaneously responding to probes. If this additional response component in the primary task leads to interference with the secondary probe response, then this would help explain the discrepancy between the reaction time and ERP data in the present study. This would also help explain the difference between Britton et al.'s

PAGE 63

57 (1978, 1979, 1980) studies, which show faster probe response when reading difficult text, and Inhoff and Flemming's (1989) study, which shows just the opposite. Recall that in Britton et al.'s studies, subject read text presented in a normal manner, thus there was no response component to the primary reading task. Ten subjects read the same text passages presented in Experiment 1. All subjects participated for class credit and were from the same subject pool used in Experiment 1. The experiment was conducted using a Compaq Deskpro 386 equipped with a VGA monitor. Subjects controlled the display of the words by pressing the spacebar on the computer keyboard. Each press of the spacebar erased the current word and displayed the next word. While reading, subjects responded to the auditory probes by pressing the zero key on the numeric keypad of the keyboard. Reaction time to press the zero key was measured. The instructions were the same as used in the response condition of Experiment 1 except that information regarding ERP recording was deleted, and an explanation about pressing the space bar to display words was added. Subjects were given one practice passage to familiarize them with the procedure. All other aspects of the experiment were the same as Experiment 1. Results and Discussion The reaction time data were analyzed in the same manner as Experiment 1. Once again, presentation was the

PAGE 64

58 only significant factor, F(l,8) = 7.61, p < .025 (all other F's < 1.0). Mean reaction times were faster during the first than second reading (see Table 2-3) Table 2-3 Mean Response Times (msec) and Standard Deviations (in parentheses) to Detect Probes for High and Low Reading Span Groups During the First and Second Reading of a Passage for Experiment 1C Reading Reading Span First Second Mean High

PAGE 65

59 General Discussion Individually, the results of the ERP and reaction time data are quite straightforward. The ERP data from Experiment 1 and IB indicate a reduction in cognitive load during the second reading of a text. This supports the cognitive contents hypothesis. The reaction time data from Experiments 1 and 1C indicate increased cognitive load during the second reading of a text. This supports the prior knowledge hypothesis. Experiment 1C ruled out the possibility that lack of response competition between the reading and probe detection tasks could lead to slower reaction times during the second reading. There is one more reason why reaction times could have been longer during the second reading; subjects may have simply given less attention to the probe response during the second reading of a passage. This is consistent with a study by Britton and Price (1981), which found longer reaction time to detect a probe when less attention was focused on the secondary task. Since the quiz scores for both reading span groups were higher after a second reading, there is evidence that the subjects did attend to the passages during the second readings. As Inhoff and Flemming (1989) note, even small changes in the amount of attention given to the primary task could lead to substantial changes in probe detection times. Unfortunately, for the rereading method used here, this explanation leads to the same predictions as the prior

PAGE 66

60 knowledge hypothesis, and therefore they can not be easily distinguished. Furthermore, these two explanations are not mutually exclusive. In fact, consumption of capacity by prior knowledge could lead to reduced attention to the probe response task. If capacity was filled by prior knowledge, then there may simply be fewer attentional resources available to devote to the secondary task. For any of the above hypotheses to explain the contradictory pattern of results from the ERP and reaction time data, an additional assumption needs to be made; this is that ERPs and reaction time may measure different aspects of cognitive load. If ERPs and reaction times were reflecting the same processes, then the two measures should not have divergent patterns of results. For example, if less attention was given to the probes during the second reading, then reaction times should be slowed, and a smaller N1-P2 response should occur. This is, of course, the opposite of what happened to the N1-P2. If the ERP and reaction time measures are reflecting different cognitive processes, then this would explain the apparent contradiction in their results. The temporal aspects of the ERP and reaction time measures lend credence to this idea. Specifically, the Nl response to the probe begins less than 200 msec after a target word's onset and the P2 begins less than 3 00 msec after target onset. In contrast, probe detection responses averaged around 52 msec after target onset.

PAGE 67

61 Thus, there is a large amount of processing which may take place between the time of the N1-P2 response and the probe detection response. This additional processing may influence the reaction time measure. To examine this explanation, a closer look at previous research involving the processing reguirements during the first and second reading of a text is needed. As mentioned earlier, Levy and her colleagues (Levy, 1983; Levy & Begin, 1984; Levy, Newell, Snyder, & Timmins, 1986; Levy & Kirsner, 1989) have studied processing changes during repeated readings of a text. In these studies various types of proofreading tasks were used to study changes in letter-, wordand text-level processing. Letter-level processes refer to the basic perceptual processes involved in individual letter analysis and recognition. Word-level processing refers to the processes reguired to perceive and comprehend a single word. Text-level processes refer to those processes involved in integrating the meaning of a word into its context. The typical procedure involved proofreading a the text for errors across multiple readings. By using different types of errors, the efficiency of processing at the letter-, word-, and text-level could be examined. Levy (1983) had subject proofread for spelling errors during a first (unfamiliar) and second (familiar) reading. The text contained different errors during each reading. She found that subjects detected more spelling errors in

PAGE 68

62 familiar texts than in unfamiliar texts. This improvement was maintained even when the order of the words was scrambled during the first and second reading. Familiar passages were also read faster, thus there was no possibility of a speed-accuracy tradeoff. Levy concluded that the improved error detection was the result of more efficient letterand word-level processing. Levy and Begin (1984) replicated Levy's (1983) results and also explored higher-order proofreading tasks. In one of Levy and Begin 's experiments (their Experiment 3) subjects proofread for misspellings in passages which contained semantic inconsistencies or no inconsistencies. Inconsistency was manipulated using "garden path" sentences. These sentences contained ambiguous words which could be interpreted in two ways (consistent or inconsistent with the meaning of the passage) depending on the meaning biased by a prior sentence. These word meaning errors could only be detected if the semantic properties of the text were fully analyzed. They found that subjects were sensitive to the meaning of a text even on repeated readings, that is, word meaning errors were detected during each reading. In addition, detection of spelling errors decreased when a text contained an inconsistent phrase, and this disruption was larger for familiar texts. Levy and Begin suggested that when a semantic inconsistency was encountered, additional resources were allocated to the higher-level semantic

PAGE 69

63 analysis. This left fewer resources available for the proofreading task, which produced the decrement in spelling error detection. Levy and Begin also concluded that overall improvements in spelling error detection across readings were due to more efficient processing at the word level Levy and Begin (1984) interpreted their results in terms of resource allocation. They suggest that resources are shared between lower-level processes involved in analyzing individual words and higher-level processes involved in comprehending the overall meaning of a text. Resources are allocated to the lowerand higher-level processes as needed. Thus, if higher-level processes become more demanding, fewer resources will be allocated to the performance of lower-level functions. This is reflected in the larger decrease in word-level error detection for familiar than for unfamiliar passages when semantic inconsistencies are present in a text. Levy and Begin state that "meaning distortions recruit resources to the semantic processor, leaving fewer resources for the proofreading task, thus resulting in a loss in proofreading performance" (p 631) Levy, Newell, Snyder, & Timmins (198 6) extended Levy and Begin's (1984) work. They examined proofreading performance during four consecutive readings of a text. They also demonstrated more efficient processing of familiar texts at lowerand higher-levels of analysis.

PAGE 70

64 However, their study places some constraints on the concepts of processing efficiency and resource allocation. A few results are particularly relevant. First, they noted improved reading speed across presentations, suggesting that lower-level processes became increasingly more efficient. Second, detection of non-word errors, which were relatively easy to detect, did not increase with multiple presentations, whereas the detection of word errors (meaning errors) which were more difficult to detect, did increase. Only when error detection is difficult were additional resources allocated to the error detection task. They use this latter finding to argue that increased accuracy of error detection is not an automatic result of reprocessing, and is dependent on the subject's goals and the task demands. They suggest that "processing efficiency is best viewed in terms of processes becoming faster and less resource demanding, so that more attention is available for strategic allocations within the task" (p. 477) There are two important conclusions to be noted in Levy et al.'s work. First, they show that different aspects of reading may become more efficient when a text is reread. Second, increased efficiency of processing leads to the availability of more resources, which the reader can allocate to the relevant demands of the task. These results can be used to help explain the findings of the present study. In the present study, the goal is to

PAGE 71

65 remember as much material as possible so that the quiz questions may be correctly answered. Therefore, if the lower-level processes are more efficient durinq the second readinq, subjects may allocate more resources to hiqher-level processes such as processinq the meaninq of the material and storinq the material in memory. This would lead to increased coqnitive load for the hiqherlevel processes durinq the second readinq. This brinqs us back to the possibility that ERPs and reaction times measure different aspects of coqnitive load. If the ERP measure is reflectinq early, lower-level staqes of analysis, such as word recoqnition, lexical access, and some semantic analysis, then these processes may indeed be more efficient durinq the second readinq of a passaqe. This would lead to increased N1-P2 amplitude durinq the second readinq of a passaqe. Probe response times may be more sensitive to later, hiqher-level staqes of analysis, such as text-level comprehension and memory processes. If more resources were allocated to the important memory demands of the task durinq the second readinq, then reaction time should be slower durinq a second readinq since the memory load is increased. Note that this explanation does not require the assumption of prior knowledqe activation. Increased reaction time is reflectinq increased memory load, but the contents of memory do not need to be prior knowledqe. In this study,

PAGE 72

66 the additional information in memory may simply be additional information from the text. If ERPs and reaction time reflect different processes, then the pattern of results is explainable. Levy et al. (1983, 1984, 1986, 1989) provide strong evidence that lowerand higher-levels of processing each become more efficient during multiple readings of a text. Importantly, this increased efficiency allows resources to be reallocated based on the demands of the task. Thus, in the present study, fewer resources are needed for lowerlevel analysis during the second reading. This is reflected by the reduced N1-P2 amplitude. In contrast, increased processing efficiency allows additional resources to be allocated to the memorial demands of the task. This leads to increased probe detection times during the second reading. In summary, the studies described in this chapter support the conclusion that the N1-P2 response can be used as a measure of certain aspects of cognitive load. Experiment 1 found increased N1-P2 amplitude during a second reading for the high reading span group. Experiment IB replicated Experiment 1, but used shorter passages. When this was done, increased N1-P2 amplitude during the second reading was found for both reading span groups. Experiment 1 also showed slower probe reaction times during the second reading. Experiment 1C replicated this effect and ruled out the possibility that a lack of

PAGE 73

67 response competition between the reading and probe detection tasks could have been responsible for the slower responses during the second reading. The contrasting results of the ERP and reaction time data were explained in terms of each being sensitive to different aspects of the reading task. It was suggested that the N1-P2 is more sensitive to lower-level aspects of the task, while the reaction time measure is more sensitive to higher-level aspects of the task.

PAGE 74

CHAPTER 3 EXPERIMENT 2 Introduction All of the studies described so far have investigated overall levels of cognitive load while reading. The secondary task methodology has also been used to examine changes in load within a text. One topic that has received much attention is the "levels effect." The levels effect refers to the fact that when information from a text is recalled, more superordinate (gist-level or high-importance) information is usually recalled than subordinate (supporting detail or low-importance) information (Kintsch & Van Dijk, 1978; Meyer, 1975; Miller, 1985) This effect is described as reflecting the hierarchical organization of most texts. Specifically, information high in the hierarchy (superordinate) is more important for understanding a text than information low in the hierarchy (subordinate) The present study examines why superordinate information is more freguently recalled than subordinate. One possibility that has been proposed is that additional resources are allocated to the processing of highimportance information (Britton, Meyer, Simpson, Holdredge, & Curry, 1979; Cirilo & Foss, 1980). If additional resources are allocated, then cognitive load 68

PAGE 75

69 should be increased when processing highimportance information. Measures of cognitive load can be applied to test this notion of processing resources. A number of hypotheses have been put forth to explain the levels effect (e.g., Britton, Meyer, Simpson, Holdredge, & Curry, 1979; Britton, Meyer, Hodge, & Glynn, 1980; Britton, Muth, & Glynn, 1986; Cirilo, 1981; and Cirilo & Foss, 1980;). These hypotheses may be divided into two general classes; those that interpret the levels effect as primarily the result of processes which occur during encoding, and those that interpret the effect as primarily the result of processes which occur during retrieval Hypotheses emphasizing encoding of text are also referred to as selection hypotheses. Selection hypotheses imply that the reader selects important (superordinate) portions of a text for differential (extra or unique) processing (Britton, Meyer, Simpson, Holdredge, & Curry, 1979; Cirilo & Foss, 1980). Thus, the levels effect occurs because extra processing is performed on the hierarchically important information, which increases the likelihood of the information being stored. This extra processing requires more capacity and should increase the cognitive load during encoding of this material. This hypothesis is consistent with related work which shows increased recall of material when more effort (defined as increased comprehension difficulty) is applied to its

PAGE 76

70 processing (Tyler, Hertel, McCallum, & Ellis, 1979; Walker, Jones, & Mar, 1983). Britton, Meyer, Hodge, and Glynn (1980) also describe encoding differences as potentially increasing the number of retrieval paths in memory. This would also enhance subseguent recall. Not all theories assume the reader is "directly" in control of this resource allocation process. For example, according to the flexible allocation hypothesis, the reader can flexibly allocate different amounts of attention to different parts of the text. In contrast, according to the cognitive contents hypothesis (a derivative of the previously discussed cognitive contents hypothesis) the amount of attention allocated to a portion of a text is a function of the text itself (i.e., its structure) Thus, the flexible allocation hypothesis describes the reader as actively changing the amount of attention allocated to elements of a text, whereas the contents hypothesis assumes the learner's allocation of resources is passively influenced by the text (Britton, Glynn, Meyer, & Penland, 1982; Britton & Price, 1981; Navon & Gopher, 1979) Retrieval hypotheses suggest that the levels effect is due to processes which occur when information is recalled. During encoding, highand low-level information (in the text's content structure) are both equally likely to be stored. Information high in the text structure is described as being more accessible than

PAGE 77

71 low-level information and is therefore more likely to be recalled (Britton, Meyer, Hodge, & Glynn, 1980; Britton, Meyer, Simpson, Holdredge, & Curry, 1979; Meyer, 1975). The reason for this is that high-level information is stored in superordinate positions within one's memory structure, which make them easier to access (Meyer, 197 5) Selection and retrieval hypotheses lead to different predictions concerning reading time and cognitive load during encoding. Selection hypotheses predict increased reading time and increased cognitive load for highimportance information. Retrieval hypotheses predict no differences. There is support for both of these outcomes. In Britton, Meyer, Hodge, and Glynn (1980) subjects were presented with texts to read and reading time was measured. The texts were constructed so that a paragraph which was hierarchically important in one passage was of low importance in another passage. This controlled for the effects of syntax, vocabulary, and complexity of the target information. There were no differences in reading time for highand lowimportance information. After reading, subjects were presented with retrieval cues to try and reduce the levels effect. Retrieval cues almost eliminated the levels effect. This implies that both highand lowimportance information were stored, and that differences in recall were more a function of retrieval than encoding differences.

PAGE 78

72 Britton, Simpson, Holdredge, and Curry, (1979) conducted a similar study in which subjects read while performing a secondary probe detection task. The texts again contained highand low-importance target paragraphs. Reading time and probe reaction time were not affected by the relative importance of the target paragraph. Thus, no support was found for encoding differences. On the other hand, Cirilo and Foss (1980) presented evidence suggesting encoding differences. In their study, subjects read pairs of stories with common target sentences which were of high or low importance to the passage. Reading times were slower when the target sentence was of high importance. This suggests more processing time was given to the high-importance information. Britton, Muth, and Glynn (1986) extended the results of Cirilo and Foss (1980) in three experiments. The first was identical to Cirilo and Foss' (1980) study. Results again indicated longer reading times for target sentences when they were of greater importance. In their second experiment the reading time of the passages was limited by displaying the passages using a rapid-serial-visualpresentation format (RSVP) Short phrases were presented sequentially at a predetermined rate. The levels effect in recall was maintained even under conditions of limited exposure time. Their third experiment again used the RSVP

PAGE 79

73 format, but subjects also performed a secondary probe detection task. Responses to the probes were slower when reading important target sentences, which provides evidence supporting increased capacity demands when encoding high-importance information. The authors conclude that when reading at their own pace, subjects will allot more time to important information. When exposure time is limited, the same result may be accomplished by allocating increased effort to the processing of the important information. The results of Britton, Muth, and Glynn (1986) provide an explanation regarding the lack of differences in cognitive load when reading highand low-importance information found by Britton, Simpson, Holdredge, and Curry (1979) In that study, subjects read the texts at their own pace. Thus, cognitive load may have been reduced by increasing the time spent on the important material. A similar conclusion regarding the distribution of load over time is discussed by Britton, Westbrook, and Holdredge (1978) Britton, Meyer, Hodge, and Glynn (1980) note that in their study, as well as in Britton, Simpson, Holdredge, and Curry's (1979) study, reading time was measured over an entire target paragraph. They suggest that selective attention may be more readily applied to shorter segments of text, such as sentences, than to paragraph size segments. The targets used by Cirilo and

PAGE 80

74 Foss (1980) and subsequently by Britton, Muth, and Glynn (1986) were, indeed, short sentences. To summarize, the relative importance of encoding and retrieval processes in producing the levels effect remains unclear, but evidence favors the conclusion that both encoding and retrieval processes effect recall. Studies which find support of encoding have used small segments of text as targets. This suggests selective attention may be limited to, or most effectively applied to small portions of text. Furthermore, when reading time is restricted, cognitive load appears to increase when important information is processed. Future studies need to control for the effects of reading time. In addition, small segments should be used as targets so that the cognitive demands within specific portions of a text may be measured. The purpose of Experiment 2 was to investigate the effect of within-text differences in cognitive load as a function of the hierarchical importance of information. Specifically, Experiment 2 was used to examine the levels effect and to try and determine if there are encoding differences between highand low-importance information and whether these differences are reflected in measures of cognitive load. In this experiment, subjects read a series of texts which had been analyzed using the method described by Meyer (1975) to determine the hierarchical importance of

PAGE 81

75 information in each text passage. While reading, subjects were presented with a series of secondary auditory probes. The probes were systematically presented at points representing highor low-importance information. The texts were presented using the RSVP method described in Experiment 1. After reading each text, subjects produced a written, free recall of the passage. Subjects were again grouped into high and low reading span. Since Experiment 1 showed an identical pattern of results for the response and no-response conditions, the response condition was not included in this experiment. Thus, no reaction time data were collected. The data being analyzed are the amplitudes of the N1-P2 responses to probes. If more capacity is consumed during the processing of high-importance information, then probes embedded in high-importance material should be associated with smaller N1-P2 responses. Evaluation of the Method The basic format of measuring the processing reguirements of highor lowimportance information within a text has previously been used to examine the levels effect. The main presentation difference between this and prior research is that a single-word RSVP presentation format is used. The smallest segment of text presented in an RSVP format has been phrases or short sentences (e.g., Cirilo & Foss, 1980; Britton, Muth, & Glynn, 1986). In these studies the levels effect still occurred (as indexed by recall

PAGE 82

76 measures) This demonstrates that the RSVP method can be used to examine within-text variations in cognitive load. The benefits of this method are similar to those of Experiment 1. The RSVP method prevents rereading of difficult material, and cognitive load cannot be reduced by increasing reading time. Again, the ERP provides a nonintrusive measure of cognitive load. Hypotheses The procedure provides a method for testing the predictions of selection hypotheses. The assumption of this experiment is the same as Experiment 1. That is, as the reading task demands more resources, less will be available to respond to the probes. According to selection hypotheses, processing load should be greater when reading high-importance information than when reading lowimportance information. Therefore, less capacity will be available to respond to the probes when reading highimportance information. Selection hypotheses lead to the following predictions. First, the N1-P2 response to probes embedded in high-importance information should be smaller than the N1-P2 response to probes embedded in low-importance information. This effect should be largest at the vertex. Second, these effects should be mediated by an interaction with reading span. Past research has shown that good readers are more sensitive to the structure of a passage than poor readers (Meyer, 1975; Meyer, 1985) Therefore the high reading span group should show a larger levels

PAGE 83

77 effect in both free recall measures and the N1-P2 response. Third, the high reading span group should recall a greater amount of text than the low reading span group Method Subjects Twenty subjects (10 men and 10 women) from the same subject pool as Experiment 1 participated. Data from 2 additional subjects were not included due to unacceptably high electrode impedances. Materials, Apparatus and Procedure The same stimulus presentation and EEG recording system used in Experiment 1 were used again here. In the present experiment, subjects read six passages, each passage being read once. After reading each passage, subjects produced a written, free recall of the material. Subjects were allowed five minutes for their recall. Instructions stated to recall everything that could be remembered about a passage. All other aspects of the procedure and instructions were identical to Experiment 1. Text Passages Six passages were used as stimuli in Experiment 2. The passages were between 160 and 178 words in length. While reading each passage, a series of 8 auditory probes were presented. Two of the passages (#13 and #14) and their content structures (described below) were taken from Raney, Dunn, & Rust (1989) Another two passages (#15 and #18) and their content structures were

PAGE 84

78 shortened versions of passages from Meyer (1975) The content structures of the final two passages were derived by the experimenter. A sample passage and its content structure is presented in Appendix C. The source of each passage is also presented in Appendix C. Probe position was restricted in three ways. First, probes were not presented during the first 8 words of a passage. Second, probes were not presented during the first word of a sentence. Lastly, half of the probes were embedded within points representing high-importance information, while the other half were embedded at points representing low-importance information. Meyer's (1975) system was used to determine the relative importance of information in a text. This system is summarized below. Meyer's (197 5) system leads to the production of a hierarchically arranged tree structure, called a semantic content structure. The content structure is composed of words from the text, specified case roles, and a set of labels which define the relationships between propositions. Each of these elements is called an idea unit. The content structure illustrates the pattern of subordination of ideas in a passage. The high-level or superordinate ideas are supported by related subordinate ideas within the passage's structure. The high-level ideas generally represent the main ideas of the passage,

PAGE 85

79 whereas the lower-level ideas describe or expand upon the higher-level ideas. Meyer's system also labels the relations among ideas. There are two categories of relations: rhetorical predicates (also called rhetorical relations) and role relations. Rhetorical predicates are found at the top levels (high in the hierarchy) of the content structure and show how their subordinate ideas are related. In other words, rhetorical relations represent the global structure of the text. For example, the sample passage presented in Appendix C (Text #14) describes the use of chemical pesticides. Line 5 in its content structure is a covariance rhetorical relation. The covariance relation describes a cause and effect relation. In this case, the use of CHEMICAL PESTICIDES (line 6) causes HARM (line 48) to the ENVIRONMENT (line 52) Role relations are typically used to relate lexical predicates together. Lexical predicates are words from the text which subsume other propositions as their arguments. Role relations classify how lexical predicates are related to their arguments. In the Chemical Pesticides passage, the lexical predicate IS REQUIRED (line 8) is related to the phrase AMERICAN COMMERCIAL GROWERS (line 12) by the role relation agent Agent describes the instigator of an action. In this example the action is the USE OF CHEMICAL PESTICIDES (line 6)

PAGE 86

80 When analyzed, the passages had either 4, 5, or 6 levels of subordination in their content structures. The number of levels is dependent on the structure of an individual passage. For passages with 4 levels, the "top" two levels were combined to represent high-importance information, while the lower two levels represented lowimportance information. For passages with 5 levels, the top three were used as high-importance and the lower two as low-importance. For passages with 6 levels, the top three were used as high-importance and the lower three as low-importance. After deriving a content structure for each passage, probes were systematically placed within points representing highor low-importance information. Results Recall Data The free recall protocols were scored in accordance with Meyer's (1975) system by the experimenter. The experimenter was blind regarding the condition of each subjects' protocols at the time of scoring. Each protocol was scored for the presence of idea units based on the semantic content structure of the passage. Content units were scored as present if the words, or paraphrased words from a passage were present in the protocol. If a content unit was recalled in its proper relationship to other information in the text, then the corresponding relationship unit was also scored as present. Since there was not an egual number of highand low-importance idea

PAGE 87

81 units in a given passage, the number of highand lowimportance units recalled was converted to a percentage. The average percentage of highand lowimportance information recalled was used as data in the following analysis. The free recall data were analyzed using a 2 (reading span; high or low) x 2 (importance; high and low) mixed ANOVA, with reading span being the between-subjects factor. The main effect of reading span was significant, F(l,18) = 16.81, p < .0007. As expected, the mean percentage of idea units recalled by the high reading span group was greater than the low reading span group (see Table 3-1) Neither the main effect of importance nor the reading span x importance interaction reached significance, [F(l,18) = 3.42, p < .082; and F(l,18) = 2.13, p < .162]. The pattern of recall is, however, worth noting since for the high reading span group the amount of high-importance information recalled was lower than the amount of low-importance information recalled. This did not occur for the low reading span group. This pattern of recall data is exactly the opposite of what was expected. That is, it was expected that more highthan lowimportance information would be recalled. One explanation for the lack of a levels effect in the present study may be due to instructions and subject strategies. Subjects were instructed to recall everything they could from the passage. A common question asked by

PAGE 88

82 subjects was if they were to try and recall every word from a passage. If this question was asked, they were told that if they remembered every word in the passage, then that is what they should recall. This may have overly stressed verbatim recall. Table 3-1 Mean Percentage Recall and Standard Deviations (in Parentheses) of Highand Low-Importance Idea Units for the High and Low Reading Span Groups Importance Reading Span High Low Mean High

PAGE 89

83 of the texts, but just tried to remember everything possible. Thus, it may be that subjects were led into a somewhat abnormal reading strategy by the instructions and the nature of the task. Meyer (1985) also reports that the levels effect can be altered based on changes in subject strategies and task demands. Four of the passages used in this study have been used in prior research. Dunn (1985) has shown a levels effect in free recall for similar versions of the Nuclear Breeder Reactor (Text #15), Loss of body Water (Text #13), and Chemical Pesticides (Text #14) passages (see Appendix C) Meyer (1975) has shown a levels effect in free recall for longer versions of the Nuclear Breeder Reactor (Text #15) and Supertanker (Text #18) passages (see Appendix C) This indicates that the lack of a levels effect in the present study is not due to the use of atypical passages. ERP Data The N1-P2 data analyzed in this and subsequent analyses is the amplitude of the N1-P2 response to probes embedded in highor low-importance information. For the sake of textual simplicity, this will be referred to simply as the N1-P2 amplitude for highand low-importance information. N1-P2 amplitude was defined in the same manner as Experiment 1. Figure 3-1 presents the ERP traces for Experiment 2. As in the ERPs from Experiment 1, the traces show a small negative followed by a positive wave beginning around 100

PAGE 90

84 msec after target (T) onset; a large N1-P2 to the probes (P) beginning around 200 msec after target onset; and a response the to word following the target (T+l) beginning around 100 msec after its onset. The N1-P2 amplitude data was analyzed using a 2 (reading span) x 2 (importance) x 5 (site) mixed ANOVA, with reading span being the between-subjects factor. The main effect of site was significant, F(4,72) = 50.94, p < .0001. Tukey's test (at p < .05) indicated that the mean amplitude N1-P2 at Cz (12.8 uv) was significantly larger than that of F4 (9.99 uv) and F3(9.8 uv) which did not differ. Each of these sites had significantly larger mean N1-P2 amplitudes than RS(7.2 uv) or LS (6.4 uv) which did not significantly differ. There was also a significant reading span x importance interaction, F(4,18) = 5.53, p < .03. The mean amplitude N1-P2 for the high reading span group was 9 5 uv for the highand 8 6 uv low-importance material. Means for the low reading span group were 9 1 uv for the highand 9.9 uv for the low-importance material. However, none of the simple effects comparisons for this interaction were significant (all F's < 1.0). The main effect of reading span and all other interactions did not reach significance in the overall analysis.

PAGE 91

Figure 3-1 ERP waveforms for Experiment 2 separated by site and reading span. For each panel, the solid line represents the ERP during the reading of high-importance information and the dashed line represents the ERP during the reading of low-importance information. Onset times of the target word (T) probe (P) and the word following the target (T+l) are indicated on the abscissa. Panel A represents the high reading span group. Panel B represents the low reading span group.

PAGE 92

86 If) O > -10 o u 10 •10

PAGE 93

87 if) ~o > -10 o u ^ 10 Site = F3 Site = F3 i r T P J LL i i i r A T+l J L 100 300 500 700 Time (msec) 900 Figure 3-1 — continued

PAGE 94

88 U) ~o > -10 o c_ u ^> 10 Site = F4 Site = F4 T P i i r T+l J Ll J L i r A 100 300 500 700 Time (msec) 900 Figure 3-1 — continued

PAGE 95

89

PAGE 96

90

PAGE 97

91 When the sites were analyzed individually, none of the effects reached significance for F3 LS or RS For CZ, the only effect to reach significance was the reading span x importance interaction, F(l,18) = 6.83, p < .018. Simple effect tests indicated that the only significant difference was between the high reading span, high importance condition, and the low reading span, high importance condition, F(l,18) = 5.96, p < .03 (see Table 3-2) Thus, the only significant difference between conditions for CZ was between reading span groups for the high importance information. There were no differences based on importance for either reading span group. Table 3-2 Mean N1-P2 Amplitude (uv) for Probes Embedded in Highand Low-Importance Information for High and Low Reading Span Groups High Reading Span Low Reading Span Importance Importance Site High Low High Low CZ

PAGE 98

92 between the high reading span, low importance condition, and the low reading span, low importance condition, F(l,18) = 5.96, p < .03 (see Table 3-2). Thus, the only difference for site LS was between reading span groups. Once again, there were no differences based on importance. Summary of ERP Results When all recording sites were combined, there was a significant interaction between reading span and importance. The high span group had a slightly higher amplitude N1-P2 for high-importance information than for low-importance information. This pattern was reversed for low reading span group. However, simple effects comparisons between levels of importance or reading span were not significant. When the sites were analyzed individually, the only significant effects were reading span x importance interactions for sites CZ and F4. Simple effects tests showed that none of these effects resulted from differences based on the importance of the material. The pattern of average amplitude for each site replicates Experiment 1. Specifically, amplitude was largest at CZ next largest at F3 and F4 and smallest at RS and LS Discussion Overall, Experiment 2 provides no evidence supporting selection hypotheses. The prediction of a smaller amplitude N1-P2 response to probes presented during the processing of high-importance information was not confirmed for either reading span group. This suggests

PAGE 99

93 that the high-importance information was not selected for differential processing. Based on the ERP data alone, the conclusion that selection hypotheses are incorrect seems warranted. Three findings limit the validity of the conclusion that selection hypotheses are incorrect. First, the recall data showed no evidence of a standard levels effect. Specifically, less high-importance information was recalled than lowimportance information. Thus, there is no evidence suggesting that extra processing may have been performed on the high-importance information. If this is true, then there is no reason to expect the ERP data to reflect a levels effect. The second reason to doubt the previous conclusion is based on subject strategies and the demands of the task. Recall that subject's self-reports indicated that no preferential treatment was given to high-importance information. Subjects indicated that they tried to retain as much information as possible, regardless of its relative importance to the passage. This may have been partly a function of passage length. The passages were fairly short, and subjects reported that remembering most of the material was not difficult. Also, the instructions given to subjects emphasized that they should recall everything from a passage that they could. Thus, they were instructed not to recall or focus only on the main ideas.

PAGE 100

94 The third reason for doubting the prior conclusion is that a pilot study using a similar procedure did indeed show evidence of a levels effect on the ERP data. That is, N1-P2 amplitude was smaller for probes embedded within high-importance information. The only difference between the pilot study and Experiment 2 was that in the pilot subjects read each passage twice before the free recall. Subjects' self-reports indicated that this procedural change led to a substantial change in strategies. Specifically, subjects reported that they focused on the main ideas during the first reading, and tried to integrate the details during the second reading. This was reflected in the N1-P2 amplitude data. N1-P2 amplitude was smaller for high-importance information, but only during the first reading. To verify this result, the procedure used in the pilot study was replicated, and is reported below. Experiment 2B The purpose of Experiment 2B was to replicate the pilot study described above. Ten subjects participated. The materials, apparatus, and procedure were identical to Experiment 2, except that subjects read each passage twice prior to the free recall. Probes were presented at different locations during the first and second readings. The important prediction is a reading span x presentation x importance interaction. Specifically, N1-P2 amplitude should be smaller for probes embedded in high-importance

PAGE 101

95 information during the first reading, and this effect should be larger for the high reading span group. Some aspects of the data from this study were presented earlier as Experiment IB, in Chapter 2. Since each passage was read twice in this study, it was also possible to use the data as a replication of Experiment 1. The same data were reanalyzed here, including both importance (high and low) and presentation (first and second reading) as factors in the analyses of the N1-P2 data. Conseguently any effects which were reported as significant in Experiment IB will also be significant in the following analyses. Therefore, only effects involving the additional factor of importance will be discussed. Results and Discussion Recall Data The free recall data were scored in the same manner as Experiment 2. The recall data were analyzed using a 2 (reading span) x 2 (importance) mixed ANOVA, with reading span being the between-subjects factor. The only effect to reach significance was importance, F(l,8) = 7.31, p < .027. As occurred in Experiment 2, more lowimportance than high-importance information was recalled (see Table 3-3) In terms of a levels effect, this needs to be interpreted with caution, since in this study two readings preceded the free recall. Because of this, it is impossible to know the proportion of highand lowimportance information that was learned during the first

PAGE 102

96 High

PAGE 103

97 noted in the results of Experiment IB, there were main effects for site and presentation. Since these are not of primary interest here, they will not be reviewed. There was also a reading span x presentation x importance x site interaction. To break down this 4-way interaction, the sites were analyzed separately. When this was done, the only sites to show a significant effect involving importance were F3 and LS. For F3 the predicted reading span x presentation x importance interaction was significant F3, F(l,8) = 8.31, p < .021. Analysis of this interaction indicated a significant simple interaction of presentation and importance for the high reading span group, F(l,8) = 12.19, p < .001, but not for the low reading span group (F < 1.0). When this interaction was further analyzed, N1-P2 amplitude for the high reading span group was found to be smaller for the high-importance than for the lowimportance information during the first reading only, F(l,8) = 16.67, p < .001. Mean N1-P2 amplitudes are presented in Table 3-3. For site LS, there was a main effect of presentation, as described in Chapter 2. There also was a significant reading span x importance interaction, F(l,8) = 6.74, p < .032. For the high reading span group, mean N1-P2 amplitude was greater during the presentation of lowimportance information (7.02 uv) than high-importance information (5.75 uv) although this difference was not

PAGE 104

98 Table 3-3 Mean N1-P2 Amplitude (uv) for Probes Embedded in Highand Low-Importance Information for High and Low Reading Span Groups During the First and Second Reading of a Passage High Reading Span Low Reading Span

PAGE 105

99 Since there was an apriori expectation of a reading span x importance interaction for the first reading, an additional analysis for each site was performed using the data from the first reading only. Analyses showed no significant effects for F4 and RS The expected reading span x importance interaction approached significance for CZ [F(l,8) = 3.78, p < .089] and F3 [F(l,8) = 4.52, p < .066]. Although the interaction was not statistically significant, for both CZ and F3 all 5 of the high reading span subjects showed smaller amplitude N1-P2 responses to high-importance information during the first reading. The marginal significance is most likely due to the small sample size. This interaction was, however, significant for LS, F(l,8) = 10.79, p < .011. For each of these sites, N1-P2 amplitude was smaller during the presentation of highthan low-importance information, but only for the high reading span group. Thus, the pattern of data indicates a trend towards a smaller N1-P2 response to probes embedded in high-importance information, but this seems to occur only during the first reading and only for the high reading span group. Figure 3-2 presents a representative ERP trace for this experiment. To summarize, analysis of the ERP data showed the expected interaction between reading span, presentation, and importance. Analyses based on each site revealed significant differences in N1-P2 amplitude based on highand low-importance for the high reading span group only.

PAGE 106

100 Additional evidence for a reading span x importance interaction was found when the first reading was examined separately for sites CZ F3 and LS As predicted, when there was a difference in N1-P2 amplitude between probes embedded in highand low-importance information, it only occurred during the first reading. Experiment 2C One guestion which was not addressed in Experiment 2 or 2B is whether probe detection times would reflect a levels effect during reading. To test this, Experiment 2C repeated the procedure of Experiment 2 but used probe detection latencies as the measure of load. Ten subjects participated. The apparatus, procedure, and instructions were identical to Experiment 2. Results and Discussion The reaction time data were analyzed using a 2 (reading span) x 2 (importance) mixed ANOVA, with reading span being the between-subjects factor. Neither of the main effects or the interaction were significant (all F's < 1.96, p >.2). There were no differences in reaction time as a function of importance or reading span (see Table 3-4) Thus, the reaction time data show no levels effect during a single reading. This is consistent with the ERP data from Experiment 2, which also showed no levels effect. Since the reaction time data do not show any evidence of a levels effect, it is not likely that the lack of ERP differences in Experiment 2 were due to the

PAGE 107

Figure 3-2 ERP waveform for Experiment 2B separated byreading span and presentation for site CZ For each panel, the solid line represents the ERP during the reading of high-importance information and the dashed line represents the ERP during the reading of low-importance information. Onset times of the target word (T) probe (P) and the word following the target (T+l) are indicated on the abscissa. Panel A represents the high reading span group, first reading. Panel B represents the low reading span group, first reading. Panel C represents the high reading span group, second reading. Panel D represents the low reading span group, second reading.

PAGE 108

102 -10 Site = CZ 10 U) ~o > -10 o L u 3 10 -10 Site = CZ 10 T P -10 T+l j L 100 300 500 700 Time (msec) 900

PAGE 109

103 ERP measure not being sensitive to changes in processing load. In sum, there is simply no evidence for a levels effect when the subjects were given only a single reading of a passage and were instructed to recall everything they could remember. This is reflected in the free recall data and ERP data from Experiment 2, and the reaction time data presented here. Table 3-4 Mean Response Times (msec) and Standard Deviations (in parentheses) to Detect Probes Embedded in Highand LowImportance Information for High and Low Reading Span Groups Importance Reading Span High Low Mean High

PAGE 110

104 not reflect a levels effect when only a single reading of the passages was allowed, which is consistent with the ERP data of Experiment 2 Overall, there is limited evidence based on the N1-P2 data for a levels effect. It appears, however, that certain conditions must exist for the levels effect to occur. To obtain a levels effect in the ERP data, and potentially the recall data, the task needs to be structured in a way that allows subjects to selectively focus their attention on highor low-importance information. This agrees with prior research which shows that the levels effect can by altered as a function of subject strategies and task demands (Meyer, 1985) Additional support for the ability of readers to selectively allocate attentional resources comes from research involving the insertion of guestions into a text during reading. When information relevant to a previously presented question is encountered, reading speed slows and response to secondary probes also slows, indicating increased cognitive load or attentional focus on the material at that point (Britton, Piha, Davis, & Wehausen, 1978; Reynolds & Anderson, 1982). This is known as the adjunct-questions effect, and demonstrates that subjects can selectively vary the amount of attentional resources allocated to processing the text. The fact that a levels effect in the ERP data can be obtained by the proper manipulation of subject strategies

PAGE 111

105 provides support for selection hypotheses. When given the opportunity, subjects appear to be able to change how they approach a task. In Experiment 2, subjects did not report focusing their attention on high-importance information. In Experiment 2B, subjects specifically reported focusing their attention on important information during the first reading. Regardless of what information attention is focused upon, subjects do appear to be able to selectively choose how their attention is distributed. Under the proper circumstances, increased resources may indeed be focused on the processing of high-importance information.

PAGE 112

CHAPTER 4 SUMMARY AND CONCLUSIONS One goal of these studies was to determine if ERPs may be used as a measure of cognitive load during reading. The evidence reviewed below supports the conclusion that N1-P2 responses to probes embedded in a primary task can be used as a measure of certain aspects of cognitive load. Importantly, the ERP can be used without requiring any response to the probes. This makes the ERP nonintrusive. A secondary goal was to compare the ERP with a behavioral (reaction time) measure of cognitive load. It was concluded that reaction time to detect a probe and the N1-P2 response to a probe both reflect changes in cognitive load. However, each measure appears to index different aspects of cognitive load. Two factors which may effect cognitive load when reading were examined. The first was the difficulty of the text. The effect of difficulty on cognitive load was discussed primarily in terms of two theories, the prior knowledge hypothesis and the cognitive contents hypothesis. The second factor examined was the levels effect which was described as reflecting the structure of a text. Two contrasting theories which explain the levels effect were discussed; selection hypotheses and retrieval hypotheses. 106

PAGE 113

107 Text Difficulty The studies presented in Chapter 2 examined the effect of rereading a text on cognitive load. It was hypothesized that processing difficulty should be greater during the first than during the second reading of a text. Four main findings emerged from these studies. First, the N1-P2 response to secondary auditory probes was larger during the second reading of a text. In Experiment 1, this result only occurred for high reading span subjects. In Experiment IB, this result occurred for both reading span groups. The increase in N1-P2 amplitude during the second reading was taken as indicating reduced cognitive load during the second reading. Second, the ERP measure demonstrated sensitivity to individual differences in memory capacity. In Experiment 1, subjects read long passages which contained a large number of details. In this case, only the high reading span group showed a reduction in load during a second reading as indexed by N1-P2 amplitude. It was suggested that both the first and second reading were difficult for the low reading span group, so they did not show a reduction in load. This was confirmed in Experiment IB. When shorter, less complex passages were used as stimuli-which place a smaller demand on memory capacity — both reading span groups showed a reduction in load during the second reading.

PAGE 114

108 Third, the pattern of ERP responses was the same regardless of whether a subject was required to respond to the probes. This shows that the ERP can be used as a measure of cognitive load in the absence of a response, which makes the ERP a very nonintrusive measure. In addition, this would allow ERP measures to be used in situations in which an overt response is not compatible with the primary task. Fourth, the probe detection (reaction time) data from Experiment 1 indicated the seemingly opposite result of the ERP data; probe detection times were faster during the first reading. This suggests increased cognitive load during the second reading. Experiment 1C indicated that this result was not due to a lack of response competition between the primary and secondary tasks. It initially appears that neither the cognitive contents nor the prior knowledge hypotheses fully explain the pattern of results. The ERP data are consistent with the cognitive contents hypothesis, and the reaction time data are consistent with the prior knowledge hypothesis. To explain the conflicting pattern of ERP and reaction time data, it was suggested that N1-P2 amplitude and reaction time to secondary probes measure different aspects of cognitive load. Specifically, it was suggested that the N1-P2 response, which precedes probe detection times by about 2 50 msec, reflects early-occurring, lowerlevel processes. These may include perceptual processes,

PAGE 115

109 such as individual letter and word recognition, as well as lexical access and semantic analysis. The reaction time measure was proposed to reflect later-occurring, higherlevel processes. These may include functions such as text-level integration of meaning, and comparisons between information in memory with what is being read from the text. If N1-P2 amplitude and reaction time to probes do reflect different aspects of load, then the results are explainable. Past research has shown that processing efficiency increases during a second reading, and that this allows additional resources to be allocated to the task as needed during the second reading (Levy & Begin, 1984). In terms of the present studies, the lower-level processes become more efficient, and therefore less demanding of resources during the second reading. This is reflected by an increased amplitude N1-P2 during the second reading. Increased efficiency of lower-level processes allows more resources to be devoted to the higher-level processes during the second reading. That is, more resources are given to remembering the text, which is the important task demand. This is reflected by increased reaction time during the second reading. For either the cognitive contents or prior knowledge hypotheses to explain the present data, they need to be modified to allow distinct components of a task to make independent contributions to overall cognitive load.

PAGE 116

110 One question that arises is, why has the opposite pattern of N1-P2 and reaction time data not been previously reported? The answer is simple; these two measures have never been contrasted. Previous research has not applied the N1-P2 measure to the analysis of complex tasks such as reading. Others have reported congruence between ERP and reaction time measures of load during the performance of complex tasks (Isreal, Wickens, Chesney, & Donchin, 1980; Kramer, Sirevaag, & Braune, 1987; Wickens, Kramer, Vanasse, & Donchin, 1983). However, these studies used a temporally later component of the ERP; the P3 00. The P3 00 typically begins around 250 to 350 msec after the Nl, and has been described as representing processes such as memory updating as well as resource allocation (Gopher & Donchin, 1986) Thus, the P3 00 may be an index of later-occurring, high-level processes. Therefore, this measure should, indeed, be consistent with reaction time measures. This leads to the conclusion that different components of the ERP may be used to measure different aspects of cognitive load. The results of these studies have an important implication for theories of cognitive load and resource allocation. The evidence reviewed above supports the conclusion that different elements of a task may independently contribute to the overall cognitive load, or resource demands of a given task. This is consistent with multiple resource models of capacity. To review, multiple

PAGE 117

Ill resource models posit the existence of separate pools of capacity for different types of cognitive processes. However, there still appears to be an overall limit on the total amount of resources available. Text Structure The studies presented in Chapter 3 examined a phenomena known as the levels effect The levels effects refers to the fact that when a passage is recalled, more high-importance information is usually recalled than lowimportance information. The levels effect was described as reflecting the structure of a text. Two classes of theories which explain the levels effect on recall were examined. The first, designated selection hypotheses, suggest that important information is "selected" during reading for extra or unique processing. This extra processing makes the storage of the information more likely, and, consequently, increases the probability of its recall. The second type of theory described was retrieval hypotheses. Retrieval hypotheses imply that both highand lowimportance information are equally likely to be stored, but the information is organized in memory in a manner that makes recall of the highimportance material more probable. The studies in Chapter 2 were designed to test for encoding differences during processing. Four major findings emerged from the studies in Chapter 2. First, when subjects read short texts under

PAGE 118

112 conditions that stressed complete recall of the material, no levels effect occurred in the free recall data. Under these conditions, the amplitude of the N1-P2 response to probes embedded in highand lowimportance information did not differ. That is, there was no evidence for a levels effect in the ERP data. This suggests no differences in encoding for the highand low-importance information. Therefore, no support for selection hypotheses were found in Experiment 1 in either the free recall or ERP data. The second finding is that when subjects are allowed to engage in strategies which led them to focus on highimportance information, there was evidence for a levels effect in the ERP data. In Experiment 2B, each text was read twice. Subjects reported focussing on the main ideas during the first reading, and on details during the second reading. In this case, there was evidence of a levels effect in the ERP data. Specifically, N1-P2 amplitude was greater for probes embedded in lowimportance information, but only for high reading span subjects. Third, sensitivity to individual differences was again demonstrated. In Experiment 2B, the levels effect was demonstrated in the ERP data only for high reading span subjects. Recall that reading span is correlated with reading ability (Daneman, 1987) and that good readers are more sensitive to the structure of a text (Meyer, 1985) Therefore, high reading span individuals

PAGE 119

113 should be better at identifying the high-importance information as it is being encountered. The ERP data is consistent with this conclusion. Fourth, when the conditions of Experiment 2 were repeated using probe detection latencies as a measure of load (Experiment 2C) there was no levels effect in the reaction time data. This suggests that the results of Experiment 1 were not due to a lack of sensitivity in the ERP measure. In summary, Experiment 2B provides some evidence supporting selection hypotheses. Based on the studies reported in Chapter 3, it was concluded that subjects are able to selectively distribute their attentional resources as needed. If the task is structured in a manner that allows subjects to freely allocate their processing resources, attentional resources may be increased when reading high-importance information. Future Research One important element in determining cognitive load in the present studies was individual differences in cognitive capacity. In each study, reading span was used as an index of capacity. It seems somewhat obvious that one individual may be better or worse at performing a given task than another individual, and that ability may be an important aspect of cognitive load. Future research may benefit by including measures of ability or cognitive capacity as a factor in the experimental design. However,

PAGE 120

114 measures of performance need to be chosen carefully. To be informative, the individual difference measure should involve the primary elements of the task being studied. For example, one reason that reading span was a particularly appropriate measure in the present studies is that it mimicked two important processes involved in reading; simultaneously processing and storing information. Future research is needed to test the conclusion that the N1-P2 and probe detection times reflect different aspects of cognitive load. If this conclusion is true, then it should be possible to independently manipulate the difficulty of lowand high-level processes. In the studies described in Chapter 2 the resources required for lower-level processes decreased after the first reading, allowing more resources to be devoted to higher-level processes during the second reading. Therefore, if the lower-level processes were not allowed to become more efficient, additional resources could not be allocated to higher-level functions. For example, if the rate of presentation was increased during the second reading, or the font was changed to a very unfamiliar style, then this may not allow a decrease in the lower-level demands. If this manipulation was successful, then additional resources could not be devoted to higher-level processes during the second reading. In terms of N1-P2 amplitude and probe detection times, this should lead to no change

PAGE 121

115 in N1-P2 amplitude across readings, since the lower-level demands would not be reduced; and this should lead to no change in reaction times, since additional resources could not be released to perform higher-level processing. An alternative finding for the above manipulation could be that changing the lower-level demands of a task would not effect higher-level demands, and vice versa. This finding would indicate independence between the various components contributing to overall cognitive load. Future research also needs to establish parameters which effect the N1-P2 measure. For example, one reason for concluding that N1-P2 amplitude and reaction time are sensitive to different aspects of load was their time of occurance after a target word's onset. One question is, what would happen if probe onset was delayed until 3 00 msec after a target word's onset? Would the N1-P2 be more sensitive to later occurring, higher-level processes; or would it still primarily reflect early occurring, lowerlevel processes? There are other factors which may effect the N1-P2 response to a probe which also need to be studied. These include the difficulty of detecting a probe; the frequency of the probes; and the effect of probe modality, such as visual versus auditory versus tactile, to name a few. Lastly, it was previously mentioned that different components of the ERP may index different aspects of cognitive load. The relation between

PAGE 122

116 various ERP components and cognitive load is certainly an important area for future research.

PAGE 123

APPENDIX A PASSAGES FOR EXPERIMENTS 1 AND 1C Sample Passage Key to Passage Markings = Location of probes during first reading. ** = Location of probes during second reading. Hyphenated (-) words were presented in two parts. Note: Titles were not presented with passages. Anthropology (Text #1, Group A) The study of human beings and human culture is the subject matter of anthro-pology Culture can be defined as the way of life of a specific race* or social group. Culture includes all behavior** that is learned and passed on from gener-ation to gener-ation. This aspect of social group existence is unique to humans. What is meant by culture** varies widely from place to place and from one period in time to another. The two basic divisions of anthro-pology are physical* anthro-pology and cultural anthro-pology. Physical anthro-pology is the study of the human** race as a biological organism. It includes* the study of differences in human size and form for both current races of people and for fossils of earlier humans and primates. The focus of physical anthro-pology is on biological evolution. This includes how humans are related** to other mammals and to other primates. Physical anthro-pology now employs the techniques of modern sciences such as genetics and bio-chemistry. This makes it possible to study* subtle differences in blood type as well as the effects of disease on a group of humans. Cultural anthro-pology is concerned with the study of human* cultures, both past and present. It includes three** divisions: archae-ology eth-nology, and lin-guistics. (The passage continued with a description of each division of archeology. Due to copyright restictions, the entire passage can not be reprintered here. ) 117

PAGE 124

118 Questions for the Anthropology text: 1. Describe how culture was defined and what it includes. 2 Describe what a physical anthropologist studies and what technigues are used. 3. Describe the goal and focus of archaeology. 4 Describe what an ethnologist studies and what technigues are used. 5. Describe what a linguist studies and how is this applied to the study of culture. Passage Titles and Sources for Experiment 1 and 1C : Anthropology (Text #1, Group A) Source: Martinson, Fazzone, Haynes, and Haynes (1988) p. 433. Food Supply (Text #2, Group A) Source: Gruber (1988) p. 371. Special Interest Groups (Text #3, Group A) Source: Martinson, Fazzone, Haynes, and Haynes (1988) p. 418. Japanese-Americans (Text #4, Group A) Source: The College Board (1988), p. 290. Hostile Takeovers (Text #5, Group A) Source: Martinson, Fazzone, Haynes, and Haynes (1988) p. 422. War (Text #6, Group A) Source: Martinson and Crocetti (1987) p. 596. Davis, CA (Text #7, Group B) Source: Martinson and Crocetti (1987), p. 368. Meteors (Text #8, Group B) Source: The College Board (1988), p. 40. High School (Text #9, Group B) Source: Gruber (1988), p. 672. Smoking (Text #10, Group B) Source: Gruber (1988), p. 673.

PAGE 125

119 Speech and Writing (Text #11, Group B) Source: Gruber (1988), p. 610. Trees (Text #12, Group B) Source: Gruber (1988) p. 611. The passages and guizzes may be obtained by writing the author at: Department of Psychology University of Florida Gainesville, FL 32611-2065 These may be obtained on computer disk (IBM DOS, 3 60 k, 720 k, or 1.44 meg) by sending the author a formatted disk with your reguest.

PAGE 126

APPENDIX B SENTENCES USED IN THE READING SPAN TEST 1 His arms and legs were sore from lifting the heavy boxes. 2 John believed that I would help him in his time of need. 3 My mother said that my dog was outside playing in the dirt. 4 The black house has been vacant for well over 100 years. 5 The student said that tennis has always been was her favorite sport. 6 Last week Susan decided that she wanted to become an artist. 7 Ninety percent of birds begin flying when they are three weeks old. 8 The bread had so much butter on it I could not taste the jelly. 9 I tried to pull the bent nail with a small wrench. 10 He looked on the shelf for something to keep himself amused. 11 I put salt, pepper, and butter all over the mashed potatoes. 12 The only weapon she had was a pistol she kept in her closet. 13 The land was divided into acres when plots were being sold. 14 John stood on a chair but he could not reach the shelf. 15 The first song we heard was loud, but it got our attention. 120

PAGE 127

121 16 Although the question was important, the mayor did not answer it. 17 The maid and the queen spent all day talking about the weather. 18 The animal was chased by the man all through the house. 19 The nurse went to get the blood pressure cuff from the cabinet. 20 Just when the disease was under control, more malaria was reported. 21 Since I broke my left foot I need to have special shoes made. 22 I was in there for fifty minutes before I began to relax. 23 The light was so intense that I covered my eyes with a towel 24 The flowers were pretty until they wilted in the sunlight. 2 5 The snow was hard to shovel, but it made our yard look clean. 2 6 The animal in the museum was the largest species ever found in Utah. 27 The woman liked vegetables, so she put some carrots in her soup. 28 The bird which was building a nest was a sparrow. 29 The missing toys were with the ball which the children had forgotten. 30 The man sitting three rows behind us left the building first. 31 He told his friends that he hated biology, but chemistry class was fun. 32 The woman asked all the members of the jury many questions. 33 All my friends wanted to read the new book again. 34 James did not realize the drink was not his soda.

PAGE 128

122 35 The boy was thinking about the girl he saw at school. 36 I gave my cat to a friend because I had to move. 37 Twelve of the new windows in the store broke during the storm. 3 8 The farmer said that the truck was used for carrying hay. 3 9 Tony thought the movie was so funny he watched it three times. 40 After the rain last week, the grass in the back yard turned green. 41 The piece of carpet was almost large enough to cover the floor. 42 Bob said that the ship in the harbor was a shrimp boat. 4 3 I told him to open a window because the house smelled like glue. 44 After sitting for two hours I wanted to stand for a while. 45 The fireman said that the burning fuel was gas and oil. 4 6 We knew he was drinking because the rum was almost gone. 47 When he returned, he discovered that all of his things had been stolen. 48 The store sold almost every type of classical music he liked. 49 She likes to sleep with pillows that are very soft and large. 50 I wanted to remove the tree and put a flower bed in its place. 51 Last year when he caught the flu he stayed sick for three months. 52 The officer stated that he did not know if robbery was the charge. 53 I looked through the tools for a chisel to give to my boss.

PAGE 129

123 54 The leg of the chair was broken so I glued it back together. 55 The painting I liked was now on display in the art gallery. 56 The food was not fresh, but it tasted better than as I expected. 57 In last years football game they defeated Ohio by seven points. 58 We decided to buy the wooden table and the matching chairs. 59 My car quit running after the oil pressure dropped below normal. 60 How he got past the guards and into the locker was never discovered. 61 During the convention the two candidates for governor never stopped arguing. 62 When he saw how much his home town had changed he was astonished. 63 The medicine was supposed to last three hours, but it lasted only one. 64 Many executives reach the top by crushing anyone that gets in their way. 65 The officers had been warned to watch soldiers who always complain. 66 The belief that we are descended from apes in now quite widespread. 67 The identity of the mass murderer was kept a complete mystery. 68 The chef tossed the Caesar salad with great skill and artistry. 69 An abundance of food was provided for the guests at the feast. 70 Sex, politics, and religion are topics which are bound to create controversy. 71 When the kitchen door swung open we could smell the food cooking.

PAGE 130

124 72 The design of electronic components requires great patience and skill. 73 All the survivors were hungry and each demanded an equal portion. 74 The loss of her parents was much more than she could endure. 75 Early theories in physics were based on principles that were not proven. 76 It may be years before the symptoms of the disease become apparent. 77 His recent performance at the Olympics is best described as apathetic. 78 As he strutted about the room his opponents knew he was vain. 7 9 The use of computer analysis showed that the transformation was exact. 10 I was not sure if his comments were always so absurd.

PAGE 131

APPENDIX C PASSAGES FOR EXPERIMENTS IB, 2, 2B, AND 2C Sample Passage Key to Passage Markings = Location of probes during first reading. ** = Location of probes during second reading (for Experiments IB and 2B only) Hyphenated (-) words were presented in two parts. Underlined target words represent HIGH importance information in the content structure. Non-underlined target words represent LOW importance information in the content structure. Note: Titles were not presented with passages. Use of Chemical Pesticides (Text #14) Several aspects of the use of chemical pesticides will be discussed. First, the use of chemical pesticides is frequently required by the American commercial growers of corn, potatoes, tomatoes, lettuce** and cotton* so that their crops will reach certain standards ** of quality These standards are much higher than those found in less developed countries like Nigeria, Ethiopia, and other** third* world countries. Second, very large amounts of chemical pesticide are normally used by commercial growers each year. Third, the use of chemical pesticides can harm ** the environment. More specific, the use of small amounts of chemical pesticides can harm a variety of small creatures like fish* and birds.** The use of medium amounts can result in harm to bigger animals such** as squirrels* and gophers. Large amounts of chemical pesticides can kill deer and bear. The excessive use of many chemical pesticides results in the development ** of chemical resistant insects and water pollution. Lastly, if pesticides are not carefully controlled ** permanent damage could happen. 125

PAGE 132

126 Content Structure for the Chemical Pesticides Passage: LI L2 L3 L4 L5 1

PAGE 133

127 51

PAGE 134

128 Key to Semantic Content Structure LI L5 = Levels 1 through 5 in the content structure CAPITALIZED WORDS = Content words from the text Lower case words = Role relations UNDERSCORED CAPITALIZED WORDS = Lexical Predicates Underlined lower case words = Rhetorical Predicates Passage Titles and Sources for Experiments IB. 2. 2B. and 2C: Loss of Body Water (Text #13) Source: Dunn (1985) Chemical Pesticides (Text #14) Source: Dunn (1985). Nuclear Breeder Reactors (Text #15) Source: Dunn (1985) Original source is Meyer (1975). Light Waves and Plant Growth (Text #16) Source: Unknown. Scientific Theory (Text #17) Source: Unkown. Tankers and Oil Spills (Text #18) Source: Meyer (1975). The passages and their content structures may be obtained by writing the author at: Department of Psychology University of Florida Gainesville, FL 32611-2065 These may be obtained on computer disk (IBM DOS, 360 k, 720 k, or 1.44 meg) by sending the author a formatted disk with your request.

PAGE 135

REFERENCES Britton, B. K. Glynn, S. M. Meyer, B. J. F. & Penland, M. J. (1982) Effects of text structure on use of cognitive capacity during reading. Journal of Educational Psychology 74 51-56. Britton, B. K. Glynn, S. M. & Smith, J. W. (1985). Cognitive demands of processing expository text: A cognitive workbench model. In B. Britton & J. Black (Eds.), Understanding expository text (pp. 227-248). Hillsdale: Lawrence Erlbaum. Britton, B. K. Holdredge, C. C, & Westbrook, R. D. (1979) Use of cognitive capacity in reading identical texts with different amounts of discourse level meaning. Journal of Experimental Psychology: Human Learning and Memory 5, 262-270. Britton, B. K. Meyer, B. J. F. Hodge, M. H. & Glynn, S. M. (1980) Effects of organization of text on memory: Tests of retrieval and response criterion hypotheses. Journal of Experimental Psychology: Human Learning & Memory 5, 620-629. Britton, B. K. Meyer, B. J. F. Simpson, R. Holdredge, T. S., & Curry, C. (1979). Effects of the organization of text on memory: Tests of two implications of a selective attention hypothesis. Journal of Experimental Psychology: Human Learning & Memory 5, 496-506. Britton, B. K. Muth, K. D. & Glynn, S. M. (1986). Effects of text organization on memory: Test of a cognitive effort hypothesis with limited exposure time. Discourse Processes 9, 475-487. Britton, B. K. Piha, A., Davis, J., & Wehausen, E. (1978) Reading and cognitive capacity usage: Adjunct question effects. Memory & Cognition 6, 266-273. Britton, B. K. & Price, K. (1981). Use of cognitive capacity in reading: A performance operating characteristic. Perceptual and Motor Skills 52 291-298. 129

PAGE 136

130 Britton, B. K. & Tesser, A. (1982). Effects of prior knowledge on use of cognitive capacity in three complex cognitive tasks. Journal of Verbal Learning and Verbal Behavior 21 421-436. Britton, B. K. Westbrook, R. D. & Holdredge, T. S. (1978) Reading and cognitive capacity usage: Effects of text difficulty. Journal of Experimental Psychology: Human Learning & Memory 4, 582-591. Britton, B. K. Ziegler, R. & Westbrook, R. (1980). Use of cognitive capacity in reading easy and difficult text: Two tests of an allocation of attention hypothesis. Journal of Reading Behavior 12, 23-30. Brown, I. D. (1978). Dual task methods of assessing work-load. Ergonomics 21 221-224. Cirilo, R. K. (1981) Referential coherence and text structure in story comprehension. Journal of Verbal Learning and Verbal Behavior 20 358-367. Cirilo, R. K. & Foss, D. J. (1980). Text structure and reading time for sentences. Journal of Verbal Learning and Verbal Behavior 19 96-109. The College Board (1988). 10 SATs (3rd ed.). New York: Author. Daneman, M. (1987). Reading and working memory. In R. Beech, & A. M. Colley, (Eds.), Cognitive approaches to reading (pp. 57-86) Chishester: John Wiley & Sons. Daneman, M. & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior 19 450-466. Daneman, M. & Carpenter, P. A. (1983). Individual differences in integrating information between and within sentences. Journal of Experimental Psychology: Learning, Memory, & Cognition 9, 561-584. Daneman, M. & Green, I. (1986). Individual differences in comprehending and producing words in context. Journal of Memory and Language 25 1-18. Dunn, B. R. (1985) Bimodal processing and memory from text. In V. M. Rentel, S. A. Corson, & B. R. Dunn, (Eds.), Psychophysiological aspects of reading and learning New York: Gordon and Breach.

PAGE 137

131 Fischler, I., & Raney, G. E. (in press). Reading by eye: Behavioral and psychophysiological approaches to reading. In J. R. Jennings, & M. G. H. Coles (Eds.), Handbook of cognitive psychology: Central and autonomic nervous system approaches London: John Wiley. Geschwind, N. (1979) Specializations of the human brain. Scientific American 241 158-168. Gopher, D. & Donchin, E. (1986). Workload: An examination of the concept. In K. Boff, L. Kaufman, & J. Thomas (Eds.), Handbook of perception and human performance: Vol. 2. Cognitive processes and performance (pp. 41.1-41.49). New York: Wiley. Gopher, D. & Navon, D. (1980). How is performance limited: Testing the notion of central capacity. Acta Psychol igica 46 161-180. Gratton, G. Coles, M. G. H. & Donchin, E. (1983). A new method for off-line removal of ocular artifact. Electroencephalagraphv and Clinical Neurophysiology 55, 468-484. Gruber, G. R. (1988) Gruber's complete preparation for the SAT (3rd ed.). New York: Barnes & Noble. Hillyard, S. A. (1985). Electrophysiology of human selective attention. Trends in Neurosciences 8, 400-405. Hillyard, S. A., & Woods, D. L. (1979). Electrophysiological analysis of human brain function. In M. Gazzaniga (Ed.), Handbook of behavioral neurobiology (pp. 345-378) New York: Plenum Press. Hink, R. F., Hillyard, S. A., & Benson, P. J. (1978). Event-related brain potentials and selective attention to acoustic and phonetic cues. Biological Psychology 6, 1-16. Hink, R. F., Van Voorhis, S. T. Hillyard, S. A., & Smith, T. S. (1977) The division of attention and the human auditory evoked potential. Neuropsychologia 15 597-605. Inhoff, A. W. (1983) Attentioanl strategies during the reading of short stories. In K. Rayner (Ed.), Eye movements in reading (pp. 181-192) New York: Academic Press.

PAGE 138

132 Inhoff, A. W. & Flemming, K. (1989). Probe-detection times during the reading of easy and difficult text. Journal of Experimental Psychology; Learning, Memory, & Cognition 15 339-351. Isreal, J. B. Wickens, C. D. Chesney, G. L. & Donchin, E. (1980) The event-related brain potential as an index of dislpay-monitoring workload. Human Factors 22 211-224 Jasper, H. H. (1958) Report to the committee on methods of clinical examination in electroencephalography. Appendix: The ten-twenty system of the International Federation. Electroencephalagraphy and Clinical Neurophysiology 10 371-375. Just, M. A., & Carpenter, P. A. (1987). The psychology of reading and language comprehension Boston: Allyn and Bacon. Kahneman, D. (1973). Attention and effort Englewood Cliffs, NJ: Prentice-Hall. Kerr, B. (1973) Processing demands during mental operations. Memory & Cognition 4., 401-412. Kintsch, W. & van Dijk, T. A. (1978). Toward a model of text comprehension and production. Psychological Review 85 363-394. Kramer, A. F. Sirevaag, E. J. & Braune, R. (1987). A psychophysiological assessment of operator workload during simulated flight missions. Human Factors 29, 145-160. Levy, B. A. (1983) Proofreading familiar text: Constraints on visual processing. Memory & Cognition 11, 1-12. Levy, B. A., & Begin, J. (1984). Proofreading familiar text: Allocating resources to perceptual and conceptual processes. Memory & Cognition 12 621-632. Levy, B. A., Newell, S. Snyder, J., & Timmins K. (1986). Processing changes across reading encounters. Journal of Experimental Psychology: Learning. Memory, and Cognition 12 467-478. Levy, B. A., & Kirsner, K. (1989). Reprocessing text: Indirect measures of word and message level processes. Journal of Experimental Psychology: Learning. Memory, and Cognition 15 407-417.

PAGE 139

133 Martinson, T. H. St Crocetti, G. (1987). Graduate record examination general test (2nd ed.). New York: Simon & Schuster. Martinson, T. H. Fazzone, J., Haynes, R. S., & Haynes, R. A. (1988) Supercourse for the ACT New York: Simon & Schuster. Meyer, B. J. F. (1985) Prose analysis: Purposes, procedures, and problems. In B. Britton & J. Black (Eds.), Understanding expository text (pp. 11-64). Hillsdale: Lawrence Erlbaum. Meyer, B. J. F. (1975) The organization of prose and its effects on memory Amsterdam: North-Holland. Miller, J. R. (1985) A knowledge-based model of prose comprehension: Applications to expository texts. In B. Britton & J. Black (Eds.), Understanding expository text (pp. 199-226) Hillsdale: Lawrence Erlbaum. Naatanen, R. (1988). Implications of ERP data for psychological theories of attention. Biological Psychology 26 117-163. Naatanen, R. & Michie, P. T. (1979). Early selectiveattention effects on the evoked potential: A critical review and reinterpretation. Biological Psychology 8, 81-136. Nakahara, D. & Ikeda, T. (1987). Effects of attention and background verbal stimuli on event-related potentials to tone pips in humans. The Japanese Journal of Psychiatry and Neurology 41 699-707. Navon, D. & Gopher, D. (1979). On the economy of the human-processing system. Psychological Review 86, 214-255. Parasuraman, R. (1978) Auditory evoked potentials and divided attention. Psychophysiology 15 460-465. Parasuraman, R. (1980) Effects of information processing demands on slow negative shift latencies and N100 amplitude in selective and divided attention. Biological Psychology 11 217-233. Picton, T. W. & Hillyard, S. A. (1974). Human auditory evoked potentials II: Effects of attention. Electroencephalography and Clinical Neurophysiology 36, 191-200.

PAGE 140

134 Picton, T. W. Hillyard, S. A., Krausz, H. I., & Galambos, R. (1974) Human auditory evoked potentials I: Evaluation of components. Electroencephalography and Clinical Neurophysiology 36 179-190. Potter, M. C. (1984) Rapid serial visual presentation (RSVP) : A method for studying language processing. In D. E. Kieras & M. A. Just (Eds.), New methods in reading comprehension research (pp. 91-118) Hillsdale: Lawrence Erlbaum. Raney, G. E., Dunn, B. R. & Rust, D. T. (1989, March). EEG correlates of cognitive style and text recall in 11 year old children Presented at the meeting of the Southeastern Psychological Association, Washington, DC. Raney, G. E. & Shuman, D. (1989). A delay timer for presenting auditory and visual probes. Behavior Research Methods, Instruments, & Computers 21 608-610. Reynolds, R. E., & Anderson, R. C. (1982). Influence of guestions on the allocation of attention during reading. Journal of Educational Psychology 74 623-632. Schwent, V. L. & Hillyard, S. A. (1975). Evoked potential correlates of selective attention with mul it-channel auditory inputs. Electroencephalograp h y and Clinical Neurophysiology 38 131-138. Tyler, S. W. Hertel P. T. McCallum, M. C. & Ellis, H. C. (1979) Cognitive effort and memory. Journal of Experimental Psychology: Human Learning and Memory 5, 607-617. Walker, N. Jones, J. P., & Mar, H. H. (1983). Encoding processes and the recall of text. Memory & Cognition 11, 275-282. Wickens, C. D. (1976) The effects of divided attention in information processing in tracking. Journal of Experimental Psychology: Human Perception and Performance 2, 1-13. Wickens, C. D. (1983) Engineering psychology and human performance Columbus: Merrill Publishing Co. Wickens, C. D. Kramer, A. Vanasse, L. & Donchin, E. (1983). Performance of concurrent tasks: A psychophysiological analysis of the reciprocity of information-processing resources. Science 221 1080-1082.

PAGE 141

BIOGRAPHICAL SKETCH After living in many parts of the United States, Gary Ernest Raney began his college years at Pensacola Junior College. He received an Associate of Arts degree in 1981 and decided to pursue a bachelor's degree in psychology. While working on his bachelor's degree, Gary developed an interest in research and set his academic goal for a Ph.D. in psychology. Gary received his bachelor's and master's degrees from the University of West Florida in 1983 and 1987, respectively. Gary was married in the spring of 1986 to Jill Johnson, also a graduate student in the psychology program at UWF. Both having goals of doctoral degrees in psychology, they entered the psychology doctoral program at the University of Florida. At the time of this writing, Jill has received her degree and Gary anticipates graduating in August, 1990. Following graduation, they will both begin academic careers. 135

PAGE 142

I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. :ra Fischler I] Professor of Psychology I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. M. Je/f/fey7Farrar Assistant Trofessor of Psychology I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. jckM&t Richard A. Griggs Professor of Psychology ^r I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. K^di 4) • CoLtg Keith D. White Associate Professor of Psychology

PAGE 143

I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the deqree of Doctor of Philosophy. Donald G. Childers Professor of Electrical Engineerinq This dissertation was submitted to the Graduate Faculty of the Department of Psychology in the Colleqe of Liberal Arts and Sciences and to the Graduate School and was accepted as partial fulfillment of the requirements for the deqree of Doctor of Philosophy. August, 1990 Dean, Graduate School

PAGE 144

UNIVERSITY OF FLORIDA 3 1262 08553 4856


xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID EP6SOCQNP_665ZVL INGEST_TIME 2017-07-11T21:22:08Z PACKAGE UF00102746_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES