Content uploaded by Richard Mayer
Author content
All content in this area was uploaded by Richard Mayer on Jul 03, 2016
Content may be subject to copyright.
Memory &
Cognition
1981,9
(1),101-109
Effects
of
shadowing on prose comprehension
and problem solving
RICHARD
E.
MAYER
and
LINDA
K. COOK
University a/California, Santa Barbara, California 93106
In two experiments,
subjects
listened to a
text
about
radar
while either shadowing or
not
shadowing
the
passage. Results of
Experiment
1produced a
pattern
in which
both
groups per-
formed
at
similar levels on retention of single facts,
the
nonshadowers excelled on
transfer
of
the
presented
information to novel problem solving,
and
shadowers excelled on
verbatim
recognition of words. In
Experiment
2, there was a
pattern
in which
both
groups performed
at
similar levels for recall of specific information,
the
nonshadowers excelled on recall of
conceptual principles,
and
shadowers excelled on recall of modifying details. Implications for
transfer-appropriate processing
and
the
cognitive capacity hypothesis were examined.
When a person is presented new prose information,
how does the person's orienting strategy influence what
is learned? Recent research using word lists as stimuli has
produced two theories: The "levels-of-processing"
hypothesis (Craik &Lockhart, 1972) states
that
deeper
levels of processing (e.g., semantic orienting tasks)
during encoding will lead to better performance on
memory tests; the "transfer-appropriate processing"
hypothesis (Bransford, 1979; Morris, Bransford, &
Franks, 1977; Stein, 1978) states that the type of orient-
ing strategy may influence the type of encoding rather
than simply having an overall effect on the strength of
memory traces.
This paper investigates the role
of
the learner's orient-
ing strategy when the to-be-learned materials consist of
atechnical prose passage rather than a word list. In
particular, the
text
is presented auditorily in chunks,
and subjects are asked to engage in one
of
two orienting
strategies: normal listening (nonshadowing group), in
which the learner is free to use available cognitive
capacity for all comprehension processes, and delayed
shadowing (shadowing group), in which the learner
repeats each phrase after it is presented. There is some
evidence
that
shadowing reduces the amount of available
cognitive capacity (Kerr, 1973). Thus, the goal
of
this
study is to determine the effects
of
shadowing on the
process
of
comprehension and the outcome of learning.
This paper encourages the development of a theory
of
prose processing that specifies the role of alternative
processing strategies. In particular, the to-be-developed
theory must deal with two issues, which can be called
This
research
was
supported
by
Grant
NIE-G-78-Q162
from
the
National
Institute of
Education
and Grant
SED77-19875
from
the
National
Science
Foundation.
Requests
for
reprints
should
be sent to eitherauthor at the
following
address:
Depart-
ment of
Psychology,
University
of
California,
Santa
Barbara,
California
93106.
We
gratefully
acknowledge
the
advice
and
assistance
of Jim
Voss,
Sherman
Tyler,
and
Greg
Vesonder in
developing
a
macrostructure
and in use of
analysis
programs.
the "cognitive capacity" hypothesis:
(I)
The cognitive
capacity of the comprehender is limited, and (2) there
are many information processing activities required in
order to comprehend a passage.
Case (1978) notes
that
the first point,
that
working
memory is limited, was shown by Jacobs (1887) early in
the history
of
experimental psychology. More recently,
Kahneman (1973) has shown how "cognitive capacity"
can be measured and how it affects comprehension.
Britton and his colleagues (Britton, Holdredge, Curry,
&Westbrook, 1979; Britton, Piha, Davis, &Wehausen,
1978) have also investigated the role
of
cognitive
capacity in prose processing. For example, Britton
et
a1.
(1978) found
that
when
text
passages contained
test-relevant adjunct questions, the usage
of
cognitive
capacity increased. This increase was measured by
recording the response latency to clicks presented during
reading, with longer latencies suggesting a higher degree
of
usage of cognitive capacity. Apparently, adjunct
questions encouraged readers to identify and manipulate
key concepts in certain portions
of
the
text,
and this
required additional capacity.
The second part
of
the cognitive capacity hypothesis,
that
verbal comprehension can be analyzed as several
information processing activities, has been presented by
Hunt, Lunneborg, and Lewis (1975,
p.192).
These
authors suggest
that
the main information processing
activities involved in comprehending a passage are:
(I)
decoding-the identification of overlearned verbal
units such as required in a Posner letter-matching task
(Posner, Boies, Eichelman, &Taylor, 1969), (2) main-
tenance-the
processing and storage of a sequence
of
verbal units in short-term memory (STM) such as required
in a Peterson and Peterson (1959) task, and (3) integra-
tion-manipulating and reorganizing information so that
it fits with existing knowledge from long-term memory
(LTM).
The remainder of this section presents three possible
theories
of
how a reduction in cognitive capacity might
Copyright 1981 Psychonomic Society, Inc. 101 0090-502X/81/010101-09$OI.15/0
102
MAYER
AND
COOK
influence the comprehension process and the outcome
of learning in this prose processing task. If shadowing
reduces the amount of cognitive capacity (Kerr, 1973),
then one or more of the comprehension processes-
decoding, maintenance, or integration-may be affected
when there is an overload. Since decoding and mainte-
nance are necessary events for all subsequent processes,
since all precede integration, and since these are likely
to be more automatic than integration, it is plausible
that integration would be the first process to be affected
if cognitive capacity were overloaded. Therefore, each of
the theories described below is based on the idea that
shadowing will mainly affect the integration process,
that is, the storage of information into LTM. Simon
(1976) has suggested that the time to store information
in LTM depends on two subprocesses: the time to add
the information to memory and the time to modify and
elaborate the newly acquired information.
Theory 1: Amount Added to
LTM
The most straightforward idea is that shadowing
affects how much information can be transformed
(or added) to LTM. This theory predicts that shadowing
will mainly hinder the process of adding new informa-
tion to LTM, and thus, the shadowing group should
perform poorer overall than the nonshadowing group on
all relevant tests such as recognition, application, recall
of
details, and recall of underlying principles. This
pattern would be reflected in an overall main effect for
treatment (shadowing vs. nonshadowing) on tests of
remembering the material.
Theory 2: Amount of Elaboration Processing
A second theory is that shadowing affects how much
effort may be given to modifying and elaborating
information that is added to LTM. This theory predicts
that shadowing and nonshadowing groups will be able to
add the same amount of information to LTM, but the
shadowing task reduces the amount of elaborative pro-
cessing on newly added information; thus, performance
will be poorer for shadowers than for nonshadowers on
tests that depend on a highly elaborated cognitive
structure, such as application problems on recall of
principles, but there should be no difference between
the groups on tasks that do not require highly elaborated
structures, such as verbatim recognition or recall of
details. This pattern would be reflected in an interaction
between treatment (shadowing and nonshadowing) and
type of test, in which the interaction is not disordinal
(i.e., the lines do
not
cross).
Theory 3: Type of Elaboration Process
The third idea is that shadowing affects the type of
elaborative processing that is performed on information
added to LTM. For example, the shadowers may use a
strategy that focuses on retention of information within
each presented chunk, whereas nonshadowers may focus
on integrating information between chunks. This theory
predicts that shadowers will perform worse than non-
shadowers on tests that require putting chunks of
information together, such as application problems
or abstracting out the conceptual theme, but non-
shadowers will perform better on tests that require
retention within chunks, such as verbatim recognition
or recall
of
marginal details. This pattern would be
reflected in an interaction between treatment (shadow-
ing vs. nonshadowing) and type of test, in which the
interaction is disordinal (i.e., the lines cross).
EXPERIMENT1
In Experiment 1, subjects either silently listened to or
shadowed a
passage
on radar. Learning outcomes were
measured by four posttests which varied from verbatim
recognition of presented sentences to application of
several text principles to solve novel problems.
Method
Subjects and Design. The subjects were 34 undergraduates
from the University of California who participated in order to
fulfill a course requirement. None of the subjects reported
having previous experience with or knowledge about radar.
Seventeen subjects served in the shadowing group, and 17 served
in the nonshadowing group. There were four within-subjects
measures corresponding to scores on four types of question sets.
Materials and Apparatus. A 640-word tape-recorded passage
describing the basic principles of radar was developed from a
lengthier treatment by Clarke (1977). The passage avoided
unnecessary physics principles and used examples drawn from
ordinary experience to explain the phenomena. Table 1 presents
a portion of the text. In recording the passage, short pauses were
placed between sequences of words corresponding to the natural
phrase boundaries within each sentence. This was to ensure that
subjects would have both time and an indication of when to
repeat what they had just heard. All sequences of words con-
tained no fewer than four and no more than seven words each.
The total recorded passage was read at a rate of 70 words/minute
in a female voice.
A three-page test booklet was constructed, consisting of four
tests. The first sheet contained the syntactic recognition test and
lexical recognition test. The syntactic recognition test consisted
of five pairs of sentences; one member of each pair was a verba-
tim sentence from the passage, and the other member of the
pair was a syntactically changed version of the same sentence.
Syntactically changed statements were formed by taking a
sentence from the passage and rearranging sentence parts to
form a new sentence that conveyed the same meaning as the
original. As an example, the following is a sentence from the
passage: "When you think about the returning pulses or echoes
you should remember any object in the path of the transmitted
beam reflects some of the energy back to the radio receiver." A
syntactically changed version of this sentence is as follows: "You
should remember, when you think about the reception of the
returning pulse or echoes, any object in the path of the trans-
mitted beam reflects some of the energy back to the radio
receiver." .
The lexical recognition test consisted of five pairs of sen-
tences; one member of each pair was a verbatim sentence from
the passage, and the other member of the pair was a lexically
changed version of the same sentence. Lexical changes involved
EFFECTS OF SHADOWING 103
Table I
Portion of RadarText
1) Radar means the detection and location of remote objects
2) by the reflection of radio waves.
3) The phenomenon of acoustic echoes is familiar.
4) Sound waves reflected from a building or cliff
5) are received back at the observer
6) after a lapse of a short interval.
7) The effect is similar to you shouting in a canyon
8) and seconds later
( 9) hearing a nearly exact replication of your voice.
(10) Radar uses exactly the same principle
(11) except that the waves involved are radio waves, not sound waves.
(12) These travel very much faster than sound waves
(13) 186,000 miles per second
(14) and can cover much longer distances.
(15) Thus radar involves simply measuring the time
(16) between transmission of the waves
(17) and their subsequent return or echo
(18) and then converting that to a distance measure.
the substitution of similar words for key words in the sentence.
For example, the original sentence "When an object is present
it leaves a bright spot on the face of the screen" became "When
an object is present, it leaves a lightened area on the face of the
screen." The five syntactic recognition and five lexical recogni-
tion questions were typed onto the same single 8.5 x 11 in.
sheet of paper, with type of item (syntactic versus lexical) and
position of correct answer (top or bottom) randomly ordered.
Directions indicated that the subjects were to choose the one of
the two paired statements that had come verbatim from the
original passage.
The second sheet presented the retention test. This test
consisted of 16 questions that assessed the learners' memory of
the basic facts presented in the passage. All questions in this set
required true-false responses. As an example of these, the text
explained that the first models of radar could tell only the speed
of an object and not its absolute location and that modern
systems now use a map-like screen to find both speed and
location. The question assessing knowledge of this information
was "Modern types of display systems are able to tell both
location and speed of an object while earlier models were only
able to tell location," to which the appropriate response was
false.
A last sheet contained five application questions. These
required conceptual understanding of several tex t principles
that would be applied to novel problem situations not addressed
in the text itself, for example:
"If
an object remained at a
constant distance with each measurement, what form would
the object's movement take?" or
"If
you were only able to send
out and receive one transmission, what is the most important
information you could learn about the object?"
Apparatus included a Panasonic cassette tape recorder and a
stereophonic headset. A practice tape was also constructed.
Procedure. Each subject was randomly assigned to either
a shadowing or a nonshadowing condition and was tested indi-
vidually. For the shadowing condition, subjects were told they
would be hearing, through stereophonic headphones, a short
description of the basic fundamentals of radar. While listening,
they were to repeat passage sequences at the pauses provided in
the tape. To facilitate familiarity with, the type of shadowing
required, they practiced on a sequence of six sentences taken
from a short story. They were always informed that they would
not be responsible for the contents of the practice tape. In the
nonshadowing condition, subjects were given the same descrip-
tion of the passage but were told to simply listen carefully
through the headphone set. They, however, did not hear the
practice tape. Prior to tape presentation,
both
groups of subjects
were told they would be given a test on the passage.
The testing phase occurred immediately after tape presenta-
tion. The three test sheets were administered in a counter-
balanced order across subjects. Directions to the subject on the
top of the syntactic/lexical recognition sheet asked them to pick
the one from the pair they thought had come directly from the
text. They were encouraged to guess if they were not sure. The
directions also explained the two types of possible changes in
the sentences. For the retention questions, subjects were
directed to think carefully before indicating whether or not the
statement was true or false. In the application question set, they
were asked to answer each question as thoroughly as possible.
Results and Discussion
Each answer was scored correct or incorrect, and the
percentage correct for each type of test was computed
for each subject. A two-way analysis of variance with
repeated measures was performed on the data. From the
perspective of Theory 1, it was expected the act of
shadowing would reduce the amount of information
encoded in LTM. Thus, the shadowing subjects' per-
formance across question sets, regardless of the type
of information assessed in each, should have been
reduced relative to the nonshadowing condition. The
significant main effect for shadowing vs. nonshadowing
[F(1 ,32) =14.02, P <.001] supported this contention.
Shadowing subjects correctly answered 68% of all the
question sets, compared with 75% for the nonshadow-
ing subjects.
There are, however, several pieces of information that
cast serious doubt on the adequacy of Theory 1 to
account for the obtained data. Table 2 shows the per-
formance of both groups by type of test. First,
if
shadowing reduced the amount of information encoded,
the overall effect noted above should be present for each
of the four tests. Yet, a significant interaction between
treatment group and type of question [F(3,96) =10.21,
P<.00I] clearly contradicted this prediction. Second,
it
should be noted that the retention test used in this
I 04 MAYER AND COOK
Nonshadow .64 .70 .82 .81
Shadow .71 .66 .79 .49
Note-Main effect
of
treatment: p <.001; interaction between
treatment and type
of
test: p <.001.
study is the test that is most commonly used in prose
memory studies. As can be seen, however, there is no
evidence to support the prediction that shadowing
would reduce the amount of information encoded;
the scores on the retention test were not significantly
different for the two groups, as measured by a t test
[t(32) =.25, p >.20]. In summary, Theory I predicts
an overall main effect but no interaction; however, an
interaction was obtained. Theory I predicts that the over-
all main effect should be present for the retention test;
however this is
not
the case. Thus, Theory I may be re-
jected as an adequate explanation of the data. Theories
2 and 3 both are consistent with both of the above
facts and thus will now be explored in more detail.
Theory 2 states that shadowing reduces the amount
of time available for elaborating on the information, and
thus it predicts a pattern of performance in which the
shadowers perform worse than the nonshadowers on
tests that require an integrated knowledge base (the
best example of such tests is the application test), but
there should be no differences on tests that do
not
require a highly elaborated cognitive structure (such as
the recognition tests and the retention test). Theory 3
predicts that since the two groups use qualitatively
different encoding strategies, the shadowers should
perform better on tests that require verbatim memory
within a chunk (the best candidate is the lexical recogni-
tion test, which requires verbatim memory of one word
within one chunk), whereas nonshadowers should
perform better on tests requiring integration across
chunks (the application test is the best example of
integrating across chunks). The general pattern of
Treatment by Type of Test interaction cited above and
shown in Table 2 is consistent with both predictions.
However, in order to more clearly discriminate the two
theories, t tests were performed on the differences
between the groups for each test. As predicted by both
theories, the shadowers performed significantly more
poorly than the nonshadowers on the application test
[t(32) =2.49, p <.05]. Performance for the two groups
was
not
significantly different on the retention test
[t(32) =.25, n>.20] or on the syntactic recognition
test [t(32) =.17, p >.20]. However, there was only
marginal support for Theory 3, since the shadowers
performed marginally better than the nonshadowers on
the lexical recognition test [t(32) =1.79, p <.10 (two-
tailed test)] .
Table 2
Proportion Correct Response by Type
of
Test
for Two Groups (Experiment 1)
EXPERIMENT2
Method
Subjects and Design. The subjects were 28 undergraduates
from the same population used in Experiment 1 (14 subjects in
the shadowing group and 14 subjects in the nonshadowing group).
Materials and Apparatus. The same taped passage, tape
recorder, and stereophonic headphones used in Experiment 1
were employed. A response booklet was constructed, which
included several sheets
of
8.5 x 11 in. paper. Directions at the
top of the first page asked subjects to write down all they could
remember from the tape recording and to organize their recall
in a manner that a reader unfamiliar with the phenomena would
understand.
The effects of shadowing can be summarized by
noting that (1) shadowing did
not
seem to reduce the
overall amount of information added to memory, since
performance on the retention test was not affected,
(2) shadowing did seem to reduce the amount of elabo-
rative processing for understanding and integrating
chunks together, since performance on the application
test was adversely affected, and (3) shadowing may have
actually enhanced elaborative processing for verbatim
recognition within chunks, since shadowing did posi-
tively (albeit only marginally) affect performance on the
lexical recognition test. These points are explored
further in Experiment 2.
Experiment 2 repeated the shadowing and non-
shadowing treatment conditions of Experiment I
but
employed a recall test rather than the four performance
tests used in Experiment 1. The results of Experiment I
indicated that both groups of subjects were equally
proficient with regard to learning of specific informa-
tion, as was assessed by the retention questions. The
differential performance on the application questions
in Experiment I raised the possibility that shadowing
precluded the opportunity for structuring an integrated
relationship between these individual chunks. Within
the passage itself, some of the units of information are
more crucial in promoting integration than others.
If
results parallel those of Experiment I (and support
Theory 3), the following is expected: The nonshadowing
group will excel in recall of this type of information,
which we call "conceptual idea units," but not in
recall of other specific information, which we call
"nonconceptual idea units." The differential perfor-
mance between the groups on lexical recognition
questions in Experiment 1 raised the possibility that
shadowing enhanced verbatim processing of words
within presented chunks. Within the
passage
itself, each
idea unit could be broken down into a central point
(such as an "action") and "modifiers" that presented
a detail that was not central to the main point of the
idea unit.
If
results parallel those
of
Experiment I (and
support Theory 3), the shadowing group is expected to
excel in recall of these modifying details, but not on
recall of the central information.
Appli-
cation
Reten-
tion
Recognition
Lexical Syntactic
Treatment
Group
EFFECTS OF
SHADOWING
105
Table 3
Proportion Recalled by Type of Idea Unit
for Two Groups (Experiment 2)
them, and how early models derived distance. The non-
conceptual category contained all other idea units,
including statements of fact, such as "Radar travels at
186,000 mps," descriptions of the parts
of
radar equip-
ment, and analogiessuch as tossing a pebble into a pond.
There were 20 conceptual idea units and 49 noncon-
ceptual idea units.
Table 3 summarizes the proportion recalled for
the groups by category of idea unit. Since protocols
contained less than 4% inferences or summarizations,
and since these were evenly distributed between groups,
these were not included in any of the analyses reported
in this paper. First, a t test was conducted to determine
whether the groups differed in overall number of idea
units recalled. Although there was a trend for the non-
shadowers to perform better than the shadowers, that
difference did not reach statistical significance [t(26) =
1.63, r>.10] . Thus, if we look only at overall amount
recalled in this study, or at overall content retained
(retention questions) in Experiment l , we would con-
clude that there is no difference in what the two groups
learned.
However, since Experiment 1 suggests a pattern of
interaction, a two-way analysisof variancewas conducted
on these data with treatment as a between-subjects
factor and type of category as a within-subjects factor.
This analysis revealed an interesting interaction between
treatment and type of category
[F(l
,26)=4.87,p<
.05],
in which the two treatment groups performed at about
the same level on nonconceptual idea units but the non-
shadowers recalled more than twice as many conceptual
idea units. Subsequent t tests indicated that the shadow
and nonshadow groups were not significantly different
in recall
of
nonconceptual idea units [t(26) = .99,
P < .20] but that the nonshadowers performed signifi-
cantly better than the shadowers in recall of conceptual
idea units
[t(26)=3.13,p<.01].
These results
nicely
parallel
the
results
of
Experi-
ment 1, in which nonshadowers and shadowers per-
formed at similar levels on retention questions (requiring
retention of specific segments of the text) but the
nonshadowers performed better on application questions
(requiring creative use of the learned materials). These
results suggest, but do not directly test, the idea that
creative use of information in problem solving depends
partly on recall of key conceptual ideas in the passage.
Propositional analysis. The foregoing analysis was
Idea Units
.26 .29
.11 .24
Nonconceptual
Conceptual
between treatment and
type
of
idea
unit:
Treatment
Group
Nonshadow
Shadow
Note-s Interaction
p<.05.
Results and Discussion
Protocol scoring. For purposes of protocol scoring,
the original
passage
was broken down into 69 idea units.
Each idea unit consisted of a sentence or clause that
conveyed one main idea.' A typical
passage
statement,
separated into idea units by slash marks, is the follow-
ing: "When no object impedes the traveling radar pulse/
it continues its travel/until lost from the screen on the
right." Additional examples are given in Table 1. In
addition to the 69 idea units, there were eight connec-
tives or introductory comments that were not counted
in the analysis.
Each subject's recall protocol was parsed into basic
idea units. Then, for each idea unit in the protocol,
an attempt was made to match it with one
of
the 69 idea
units in the original text. A match was made if the key
words in the protocol idea unit matched the key words
in the text idea unit. If a protocol idea unit did not
match any of the 69 idea units, it was assigned to one of
two alternative categories, inference or summarization.
An inference was a new idea unit that was not in the
text but was logically related to the text, for example,
"The dot's size can be used to determine the size of the
object." A summarization was a vague or brief statement
about information in the text, such as, "A transmitter
sweeps across a certain area." In all cases, the rater was
blind to the experimental condition of the subject.
Category analysis. A major question addressed in this
study is whether the recall protocols of subjects in the
shadowing group differed from those
of
the nonshadow-
ing group.
For
example, the results
of
Experiment I sug-
gest that the groups may be equivalent in recall of individ-
ual facts (as indicated in the retention tests in Experi-
ment
l)
but may differ in recall ot conceptual principles
needed for transfer (as indicated in the application test
in Experiment I). In order to investigate this idea, the
idea units were categorized into two separate sets:
conceptual idea units and nonconceptual idea units.
Conceptual idea units stated one of the five basic ideas:
transrnission-a pulse is sent out; reflection-it strikes a
remote object; reception-a pulse returns to the source;
measurement-the time and direction are recorded;
conversion-time may be converted to a measure of dis-
tance. The following are examples of conceptual idea
units: "Radar means the detection and location of
remote objects/by reflection of radio waves. Thus radar
involves simply measuring the time/between transmis-
sion of the waves/and their subsequent return or echo/
and then converting that to a measure of distance."
In addition, other conceptual units described how a
transmitter sends radar pulses, how an antenna receives
Procedure. Subjects were randomly assigned to either a
shadowing or a nonshadowing condition and given the same
instructions as in Experiment 1. At the end of the tape presenta-
tion, they were given the response booklet and asked to read
the printed directions. After the experimenter answered any
questions and reemphasized the directions, subjects began their
recall.
106 MAYER AND COOK
Note-Main effect
of
treatment, n.s.; interaction between treat-
ment and type, p <.001.
Table 4
Mean Number of Propositions Recalled
by
Type
for
Two
Groups
(Experiment
2)
based on idea units as the unit of analysis. In the present
section, a more "micro" level of analysis is reported,
based on Kintsch's "propositional analysis" (Kintsch,
1974; Turner &Greene, Note 1). Each
of
the first 32
idea units was further broken down into propositions,
with approximately three to six propositions found in
each idea unit. For example, the proposition "when
no object impedes the traveling radar pulse" would
be broken down into the following propositions:
(1)
IMPEDES
(0:
pulse; S: no object), (2) QUALITY
(when, 1), (3) MODIFY (pulse, radar), (4) MODIFY
(pulse, traveling). The recall of each proposition by each
of the 28 subjects was determined by rescoring the
original protocols.
Each proposition was classified as being a defining
part of the idea unit in which it was embedded (these
"major" propositions consisted of events such as
IMPEDES
and of some locations) or as being nonessen-
tial to the idea unit (these "minor" idea units consisted
of modifiers, qualifiers, conjunctions, and the like).
If
the nonshadowers attempt to listen for major meaning,
they might be less likely to retain minor propositions;
if the shadowers have no overall strategy for bringing
meaning to the passage, they would be likely to retain
equal amounts of major and minor propositions.
Table 4 shows the proportion recalled by type of
proposition for two groups. An analysis of variance
was conducted with treatment as a between-subjects
factor and type of proposition as a within-subjects
factor. As can be seen in Table 4, there is an interesting
pattern of interaction in which the nonshadowers
performed better on major propositions and shadowers
performed better on recall of minor propositions
[F(I,26) =13.86, P <.001].
The t tests revealed that the nonshadowers did not
perform significantly better than shadowers on recall
of major propositions [t(26) =1.23, p >.10] , but the
shadowers did recall significantly more minor details
than the nonshadowers did [t(26) =2.06, p <.05] .
This pattern of results is consistent with the pattern in
Experiment 1, in which shadowers outperformed non-
shadowers on lexical recognition but not on retention.
Apparently, the shadowers are more likely to retain
surface modifiers and details. For example, shadowers
overwhelmingly recall that a pebble is tossed into a
"calm" pond, whereas nonshadowers unanimously
ignore the adjective. A stepwise discriminant analysis
was performed, using score on minor and score on major
Treatment
Group
Nonshadow
Shadow
Major
Propositions
8.3
6.1
Minor
Propositions
3.4
6.9
propositions as independent variables. The resulting
function relied on only one factor: Minor propositions
loaded on the shadowing centroid with a weight of .44.
The function was able to correctly classify 23
of
the 28
subjects. Thus, relatively high performance on recall
of minor propositions seemed to distinguish most of
the shadowers from the nonshadowers.
The foregoing two analyses provided replicatory
evidence for the pattern of performance noted in Experi-
ment 1: namely, that shadowers and nonshadowers
performed at equivalent levels on tests of memory for
basic information, such as retention or recall of non-
conceptual idea units, nonshadowers excelled on tests
requiring integration of information, such as application
problems or recall
of
the conceptual theme, and shad-
owers outperformed nonshadowers on tests requiring
verbatim memory of details within each chunk of infor-
mation, such as lexical recognition or recall of minor
propositions. These findings are most consistent with
Theory 3, which is based on the idea that shadowers and
nonshadowers employ qualitatively different elaboration
processes and thus acquire qualitatively different learn-
ing outcomes.'
GENERAL DISCUSSION
Do subjects who are required to shadow a passage
acquire a learning outcome different from that of subjects
who just listen to the text?
If
we had used the traditional
measure of overall retention, such as that measured by
the retention test of Experiment 1, we would have
concluded that there were no major differences between
the groups. If we had used the traditional measure of
overall amount recalled, such as that measured by total
number of propositions recalled in Experiment 2, again
we would have found no treatment effect. However,
when we look deeper into the performance of our
subjects, important differences emerge.
Since the groups manifested different learning out-
comes, it is reasonable to suggest that the learning pro-
cesses were different for the two treatment groups. In
this paper, we have outlined three distinct theories of
the acquisition process and how it might be affected
when cognitive capacity is reduced.
Theory 1 has some similarities to the levels-of-
processing hypothesis (Craik &Lockhart, 1972). In
particular, Theory 1 states that shadowers have less
cognitive capacity available for adding new information
to memory; thus, decreasing the capacity available for
comprehension processes results in less information
being acquired. The levels-of-processing idea is that
deeper levels of processingwilllead to better performance
on memory tests. This proposal is based on research in
which recall for a list of words was better if subjects had
engaged in semantic orienting tasks during encoding
rather than orthographic orienting tasks (Craik &
Tulving, 1975; Hyde &Jenkins, 1969; Jenkins, 1974).
In the present experiments, using a scientific prose
passage, there was little evidence for the idea, stated in
Theory 1, that shadowing would result in overall decre-
ments in memory test performance: In Experiment I,
there was no significant difference between the groups
in retention of basic facts; in Experiment 2, there was
no significant difference between the groups in recall
of propositions; in both Experiments 1 and 2there
were significant Treatment by Type of Test interactions
in which shadowers performed better on some tasks
but nonshadowers performed better on others.
While
these results do not contradict all interpretations on the
levels-of-processing idea for prose material, they do
clearly allow one to reject a version of levels of process-
ing, as stated in Theory 1.
If
the results are highly inconsistent with the predic-
tions of Theory 1, they are only partially consistent
with Theory 2. Theory 2 suggests that shadowing reduces
the cognitive capacity available to perform integrative
elaboration among the chunks of acquired information.
As predicted, shadowers perform more poorly than
nonshadowers on tasks that require integration of
information (application problems in Experiment 1 and
recall of conceptual idea units in Experiment 2), but
not on remembering the basic information (retention
problems in Experiment 1 or recall of propositions in
Experiment 2). However, the main contradiction to
Theory 2 involves cases in which the shadowers actually
outperform nonshadowers on tasks requiring a verbatim
memory of the details. In Experiment 1, there is only a
marginally reliable superiority of shadowers over the
nonshadowers on lexical recognition (although the .05
level of significance would be achieved if a one-tailed
test were employed); in Experiment 2, the significant
superiority of the shadowers over the nonshadowers
on recall of minor propositions
gives
the strongest
evidence against Theory 2.
While
more research is
required, a tentative conclusion is that there is some
reason to suspect that Theory 2 may not be able to
account for all of the data.
The results of the present study tend to favor
Theory 3, the idea that shadowers and nonshadowers
used qualitatively different processes for elaborating on
the acquired information. This data is similar in many
respects to the concept of transfer-appropriate process-
ing (Bransford, 1979; Morris et
a1.,
1977). For example,
Morris et
a1.
(1977) auditorialy presented words and
asked some subjects to answer semantic questions for
each target word (e.g., does "dog" fit the description
"
__
has ears?") and other subjects to answer rhyming
questions for each target word (e.g., does "dog" fit the
description "rhymes with log"?). As predicted by the
levels-of-processing hypothesis, the semantic group
performed better on a test of recognizing that words
had been presented in the original list; however, a new
finding was that the rhyming group outperformed the
semantic group on a test
of
recognizing which words
rhymed with words from the original list. In another
study, Stein (1978) visually presented word lists con-
taining words such as "raDio." One group of subjects
EFFECTS OF
SHADOWING
107
was asked semantic questions (such as "Does this word
use electricity?") and the other group was asked ortho-
graphic questions (such as "Does this word have a
capital D?"). As in the previous research, the semantic
group performed better on recognizing which words had
been presented; however, the orthographic group out-
performed the semantic group on recognizing which
letter in a givenword had been capitalized in the original
list.
Morris et al. (1977) offer the transfer-appropriate
processing hypothesis to account for these results:
the idea that orienting strategy may influence the type
or quality of encoding rather than its simply having
overall effects on the strength of memory traces. In a
summary of this line of research, Bransford (1979,
p.67)
pointed out that
"it
is no longer profitable to
assume that certain items' memory traces are less dur-
able or adequate than others' simply because these
items were processed at a shallow level
...
and shallower
levels
of
processing can be superior to deeper levels if
one designs test situations that assess the kinds of
information that actually results from shallower levels
of analysis."
Much of the previous work on prose comprehension
has involved simple narrative texts or short factual
passages. One goal of such research has been to predict
probability of recall of text propositions. For example,
Kintsch (1974) and Meyer (1975) have developed a
case grammar procedure for producing a hierarchical
ordering of propositions and have shown that proba-
bility of recall of a proposition is related to its level
in the hierarchy. Similarly, Rumelhart (1975) and
Thorndyke (1977) have developed story grammars for
describing the hierarchical relations among events and
states in a simple narrative. More recently, Kintsch and
van Dijk (1978) have been able to specify a model for
learning from text that incorporates assumptions about
both case grammar and macrostructure.
The present study extends this work on narratives
in two ways. First, we used a scientific text that can be
used to generate problem solving.
like
Voss and his
colleagues (Chiesi, Spilich, &Voss, 1979; Spilich,
Vesonder, Chiesi, &Voss, 1979), we found that many
modifications have to be made to existing techniques
when new materials are used, including much larger
propositional units and an entirely different macro-
structure system.
Second, we used an experimental manipulation to
determine whether interfering with cognitive capacity
of the subject would influence ability to make use of
the macrostructure. Britton's work (Britton et al., 1978;
Britton et
a1.,
1979) has demonstrated that manipulations
in the prose processing task have measurable effects on
cognitive capacity. For example, presence of adjunct ques-
tions increases demands on cognitive capacity. Similarly,
in the present studies, the nonshadowers were allowed
more cognitive capacity to identify and integrate the
key conceptual information, whereas the shadowers
108 MAYER AND COOK
were more likely to attend to and store the surface
information as presented.
In the present studies, we manipulated the learner's
behavior during prose presentation. However, another
way to manipulate cognitive capacity is to identify
individual differences. For example, Hunt et al. (1975)
found that good readers (i.e., subjects scoring high on
tests of verbal ability) differ from poor readers on simple
information processing tasks that require holding and
manipulating information in memory.
Voss and his colleagues (Chiesi et al., 1979; Spilich
et al., 1979) have also investigated individual differences
in how subjects listen to an inning of a baseball game.
For example, subjects with high knowledge of baseball
are more likely than low-knowledge subjects to relate
the actions of the game to the goal structure and to
maintain important information in working memory.
In the present study, our nonshadowers were better
able than the shadowers to relate information to five
main macrostructure events (transmission, reflection,
reception, measurement, conversion).
In an earlier set of experiments, Mayer and Greeno
(1972) distinguished between two variables in learning
outcomes: Internal connections referred to how well
the presented elements were connected with each other
in memory, and external connectedness referred to how
well the presented elements were connected with an
organized body of existing knowledge. In the present
examples, we have provided a more specific and detailed
examination of the structural differences in what is
learned. There is certainly room for more detail and
precision,
but
the present example indicates one way to
describe differences in learning from technical prose.
One general implication of this work is if the teacher
says, "now repeat after me," he or she is insuring that
the students will not understand (Glaser, Note 2).
In summary, this work suggests that abetter under-
standing of the limits of cognitive capacity is an impor-
tant prerequisite to the development of theory of prose
comprehension. This work provides some new informa-
tion that indicates that draining some cognitive capacity
(e.g., through shadowing) affects the structure of what is
learned and the capacity for creative use of presented
information.
REFERENCE NOTES
J. Turner, A., &Greene, E. The construction and use
of
a
propositional text base (Tech. Rep. No. 63). Boulder, Colo:
Institute for the Study of Intellectual Behavior, 1977.
2. Glaser, R. Personal communication, September 1979.
3. Mayer, R.
E.,
&
Cook,
L. Effects
of
shadowing on prose
comprehension and problem solving (Tech. Rep. 80-3). Santa
Barbara,
Calif:
Department
of
Psychology Technical
Report
Series in Learning and Cognition, 1980.
REFERENCES
BRANSFORD,
.J.
D. Human cognition: Learning, understanding
and remembering. Belmont, Calif: Wadsworth, 1979.
BRITTON,
B. K.,
HOLDREDGE,
T. S.,
CURRY,
C., &
WESTBROOK,
R. D. Use of cognitive capacity in reading identical texts with
different amounts of discourse level meaning. Journal
of
Exper-
imental Psychology: Human Learning and Memory, 1979, 5,
262-270.
BRITTON,
B. K., PIHA, A., DAVIS,
J.,
&
WEHAUSEN,
E. Reading
and cognitive capacity usage: Adjunct Question effects. Memory
&Cognition, 1978, 6,266-273.
CASE, R. Implications of developmental psychology for the design
of effective instruction. In A. M. Lesgold, J. W. Pellegrino,
S. D. Fokkerna, &R. Glaser (Eds.), Cognitive psychology and
instruction. New York: Plenum, 1978.
CHIESI, H. L., SPILICH, G.
J.,
&
Voss,
J. F. Acquisition of
domain-related information in relation to high and low domain
knowledge. Journal
of
Verbal Learning and Verbal Behavior,
1979,18,257-273.
CLARKE,
D. The encyclopedia
of
how it works: From abacus to
zoom lens. New York: A &W, 1977.
CRAIK, F. 1. M., &
LOCKHART,
R. S. Levels of processing: A
framework for memory research. Journal
of
Verbal Learning
and Verbal Behavior, 1972,11,671-684.
CRAIK, F. 1. M., &
TULVING,
E. Depth of processing and the
retention of words in episodic memory. Journal
of
Experimental
Psychology: General, 1975,104,268-294.
HUNT,
E.,
LUNNEBORG,
C., &LEWIS,
J.
What
does it mean to be
high verbal? Cognitive Psychology, 1975, 7, 194-227.
HYDE, T. S., &JENKINS,
J.
J. Differential effects of incidental
tasks on the organization of recall of a list of highly associated
words. Journal
of
Experimental Psychology, 1969,82, 472-48J.
JACOBS,
J.
Experiments on prehension. Mind, 1887, 12, 75-79.
JENKINS,
J.
J. Can we have a theory of meaningful memory?
In R. L. Solso (Ed.), Theories in cognitive psychology: The
Loyola symposium. Hillsdale,
N.J:
Erlbaum, 1974.
KAHNEMAN,
D. Attention and effort. Englewood Cliffs,
N.J:
Prentice-Hall,1973.
KERR, B. Processing demands during mental operations. Memory
&Cognition, 1973, 1,401-412.
KINTSCH,
W. The representation
of
meaning in memory. Hillsdale,
N.J:
Erlbaum, 1974.
KINTSCH,
W., &
VAN
DIJK, T. A. Toward a model of text com-
prehension
and
production. Psychological Review, 1978, 85,
363-394.
MAYER,
R. E., &
GREENO,
J. G. Structural differences between
learning outcomes produced by different instructional methods.
Journal
of
Educational Psychology, 1972, 63, 165-173.
MEYER,
B.
J.
The organization
of
prose and its effects on memory.
Amsterdam: North-Holland, 1975.
MORRIS,
C. D.,
BRANSFORD,
J.
D., &
FRANKS,
J. J.
Levels of
processing versus transfer appropriate processing. Journal
of
Verbal Learning and Verbal Behavior, 1977,16,519-533.
PETERSON,
L. R., &
PETERSON,
M. J. Short-term retention of
individual verbal items. Journal
of
Experimental Psychology,
1959,58,193-198.
POSNER,
M. 1., BOIES, S.
J.,
EICHELMAN,
W. H., &
TAYLOR,
R. L. Retention
of
visual and name codes
of
single letters.
Journal
of
Experimental
Psychology,
1969, 79, 1-16.
(Monograph)
RUMELHART,
D. E. Notes on a schema for stories. In D. G.
Bobrow &A. M. Collins (Eds.), Representation and under-
standing: Studies in cognitive science. New York: Academic
Press, 1975.
SIMON,
H. A. The information storage system called
"human
memory."
In M. R. Rosenweig &E. L. Bennett (Eds.), Neural
mechanisms
of
learning and memory. Cambridge, Mass:
M.LT.
Press, 1976.
SPILICH, G.
J.,
VESONDER,
G. T., CHIESI, H. L., &
Voss,
J.
F.
Text processing of domain-related information for individuals
with high
and
low
domain
knowledge. Journal
of
Verbal
Learning and Verbal Behavior, 1979, 18, 275-290.
STEIN, B. S. Depth of processing re-examined: The effects of
precision encoding and test appropriateness. Journal
of
Verbal
Learning and Verbal Behavior, 1978,17,165-174.
THORNDYKE,
P. W. Cognitive structures in comprehension and
memory of narrative discourse. Cognitive Psychology, 1977,9,
77-110.
NOTES
I. We are using units
that
are much larger than those in
Kintsch's propositions (Kintsch, 1974; Turner &Greene, Note 1).
In this respect, we are consistent with Voss's (Chiesi et al.,
1979; Spilich et al., 1979) technique for analyzing real-world
prose.
EFFECTS
OF
SHADOWING
109
2. Supplemental analyses were carried
out
in the data from
Experiment 2. These include a detailed analysis of the structure
of the
text
and the development of macrostructure representa-
tions of the text. Asummary of these analyses is available
from the authors as
part
of a technical report (Mayer &Cook,
Note 3). These analyses generally provide replicatory support
for the findings and conclusions stated above.
(Received for publication March
II,
1980;
revision accepted April 15, 1980.)