Feature Article Theory and Practice: Uses of the Computer in Reading

Feature Article
Editor's N o t e : This paper was invited as part of the special
issue, "The Challenge of Reading with Understanding in the Intermediate Grades" (RASE 9:1).
Theory and Practice: Uses of the
Computer in Reading
Beth Warren and Ann S. Rosebery
Our goal in this article is to examine two ways in which recent
psychological theory has influenced the use of computers in reading
instruction. In particular, we look at two competing theories, the
componential and the constructivist, and the implications that each
carries for computer-based practice. To exemplify these perspectives,
we examine two computer environments, RACER and the Reader's
Assistant. In conclusion, we put forward the notion of "contextualization" as a basis for integrating the componential and constructivist approaches into a sensible model of computer use in the
special education classroom.
I
N A R E C E N T article, Brown and Campione (1986) trace
the influence o f psychological theories o f intelligence and
learning on conceptions o f developmental delay and special
education. They argue that as psychological theory has
evolved to focus increasingly on understanding the kinds o f
thinking and learning that people actually do in specific
situations (e.g., when they read and write or solve algebra
word problems), it has helped to bring about an important
shift in emphasis in the diagnosis and remediation o f learning difficulties. Specifically, the shift is away from implicating a general deficit (e.g., memory) as the principal cause o f
learning difficulities to a focus on the specific types o f
knowledge, skills, and strategies that underlie competence
in a particular domain or academic task.
Working from this perspective, our purpose is to examine two ways in which recent psychological theory has influenced the use o f the computer as a tool for instruction in
reading. T h e change in focus to domain-specific skills and
knowledge has given rise to two distinct approaches to instruction, in general, and to computer-mediated instruction,
in particular. These two approaches differ principally in (1)
the focus o f instruction (basic skill components vs. com-
prehension strategies) and (2) the context o f instruction
(isolated practice o f individual skills vs. integrated practice
o f specific strategies). T h e first approach emphasizes a
principled decomposition o f the cognitive requirements o f
reading, leading to focused instruction in particular skill
components (e.g., isolated practice in decoding, use o f context, etc.). This we will refer to as the componential approach (Frederiksen & Warren, 1987; Frederiksen, Warren,
& Rosebery, 1985a, 1985b). T h e second approach also emphasizes direct teaching o f specific components—most
commonly comprehension strategies (e.g., summarization,
question formulation)—but situates teaching and learning
in the social and functional context in which the strategies
are actually used (e.g., that o f a group cooperatively trying
to understand the ideas in a science text). This we will refer
to as the constructivist approach, which is exemplified in the
reciprocal teaching method o f Palincsar and Brown (1984)
and, more generally, in cognitive apprenticeship models o f
learning as described in Collins, Brown, and Newman (in
press).
Frequently these differences in focus and context in the
two approaches have caused them to be viewed as compet-
Downloaded from rse.sagepub.com at PENNSYLVANIA STATE UNIV on May 18, 2016
ing models for instruction. In fact, there are important differences between the two. Our position, however, is that
maintaining these differences in classroom practice may not
maximally benefit student learning. Rather, presenting
students with a mix o f alternative approaches may be best,
particularly when there is a unifying framework for integrating the approaches into a coherent model o f classroom
instruction. Our goal in this article, therefore, is to explore
both the componential and constructivist approaches, including the theories that underlie them and the computerbased practices that they promote, with an eye toward
understanding their potential complementarity for computer-based reading instruction in the special education
classroom. In each case, a representative computer-based
system ( R A C E R and the Reader's Assistant, respectively) is
featured as part o f a scenario in which low achieving, middle school students are shown using the computer as a tool
for learning to read. (It is important to note here that both
R A C E R and Reader's Assistant are at this point intended as
research tools, not commercial products. Both have been—
and, in the case o f Reader's Assistant, continue to be—
developed with two goals in mind: (a) to test and refine the
theories on which their design is based, and (b) to evaluate
their effectiveness as instructional tools.) W e conclude by
suggesting "contextualization" as a basis for integrating the
componential and constructivist approaches into a sensible
model o f computer use in the special education classroom.
A Componential Approach
Theory. Like the more familiar subskill approach, a
componential approach to reading instruction focuses on
developing individual skill components. WTiat distinguishes
a componential approach, however, is its grounding in a
cognitive theory o f expertise in reading (Frederiksen &
Warren, 1987; Frederiksen et al., 1985a, 1985b). T h e theory
specifies not only the skill components and knowledge that
contribute to expertise in reading but also the ways in
which the components interact. For example, the correlation between decoding processes and comprehension is well
known (Curtis, 1980). But precisely how is decoding skill
related to comprehension? This is the type o f question that
a componential theory tries to answer. And, through the
answers to such questions, the theory also offers suggestions for practice, in particular, for sequencing instructional
activities based on patterns o f component interactions.
How does componential theory explain the decodingcomprehension connection? In a componential view, the effect o f decoding skill on comprehension is not direct. It is
not simply that efficient decoding processes let the reader
concentrate on thinking about a text's meaning rather than
on sounding out a particular word. Instead, efficient decoding facilitates the operation o f other process components
such as word identification and lexical retrieval. These processes, in turn, facilitate others that are thought to directly
support comprehension, such as semantic integration and
referent identification. From this view, then, comprehension is the product o f a complex interaction among components operating at various levels o f the reading process,
ranging from multiletter unit identification to complex
processes o f inference (Frederiksen & Warren, 1987;
Rosebery, 1986a; Warren, 1986a).
Practice. Consistent with this view o f component interaction, one instructional focus o f a componential approach is on developing critical components to automatic
levels (that is, criteria o f speed as well as accuracy). Wlien a
process like decoding is automatic, it operates in concert
with other processes and facilitates the operation o f those
processes to which it is directly linked. Inefficient decoding
processes, on the other hand, disrupt comprehension
through their effect on those processes to which they are
directly linked. T h e goal o f a componentially oriented instructional program is therefore to develop these critical
components to automatic levels. A second instructional
focus is on the skill sequence. In a componential approach,
instruction in specific components follows a particular sequence based on patterns o f component interaction. Thus,
decoding instruction takes place prior to instruction in use
o f context, but after (or simultaneously with) instruction in
multiletter unit identification.
In the next section we take a look at a student with learning difficulties as he works at the computer to improve his
decoding skill. T h e scenario is based on a set o f studies conducted to evaluate the effectiveness o f the featured instructional system, R A C E R (see Frederiksen et al., 1985b, for
details). W e then briefly look at R A C E R ' s effectiveness in
promoting skill improvement and at its usefulness for special education students.
RACER
Imagine Ron, an eighth grader, who reads at a third-grade
level. He takes basic academic subjects (English, mathematics, science, etc.). In addition, he is assigned to a resource
center where he receives individualized instruction.
At the resource center, Ron concentrates on improving
basic reading skills. Three times a week for 20 minutes, he
practices decoding at the computer using R A C E R . W h e n he
started, Ron had a limited sight vocabulary and experienced
difficulty decoding one-syllable words. For example, he read
"sweet" for the word sweat. T w o months later, Ron can read
five-syllable words like anthropology and crystallization with
ease. W h a t changes have taken place in Ron's decoding skill
and how have these changes been fostered?
Downloaded from rse.sagepub.com at PENNSYLVANIA STATE UNIV on May 18, 2016
The computer is an ideal medium for developing
automatic decoding skill. It can provide immediate and
direct feedback, adapt to a student's progress, and offer a
motivating learning experience. R A C E R , the system Ron is
using, is designed to help him develop both efficient and accurate decoding skill. This dual focus is the key to developing
automatic decoding skill.
Set in a videogame-like context, R A C E R challenges the
student to read aloud 20 words as quickly and accurately as
possible. T h e game adopts the metaphor o f running a race;
the student's goal is to get his runner to cross the finish line
ahead o f the computer's runner. T h e race track is divided
into 20 steps represented by 20 flags, each o f which corresponds to a word the student must pronounce (see Figure
1). When the "starting gun" goes off, both runners begin
the race. W h e n they come to the first flag, a word pops up.
The student's task is to read the word into a microphone as
quickly as possible.
Speed. T h e dynamics o f the contest encourage the student to read the target words as quickly as possible. At the
moment the target word appears on the screen, the student's
runner temporarily stops and does not advance until the
word is read aloud. Meanwhile, however, the computer's
runner continues to advance at a steady pace (which reflects
the student's current ability, as we discuss below under
Adaptiveness). I f the student pronounces the word faster
than the computer's runner moves, his runner will outpace
the computer's. I f the student is unsure o f a word and
spends time decoding it, his runner will begin to fall behind
the computer's. In short, the faster the student pronounces
the word, the faster his runner will go and the sooner he will
overtake his opponent.
I f the student is unable to decode a word, he can press a
Help key and have the word pronounced for him via synthesized speech. Requests for Help are not penalized; the
computer's runner simply stops at the time the request is
made. T h e contest resumes as soon as the synthesized pronunciation ends. Interestingly, our research has shown that
students rarely use this option, perhaps because it tends to
disrupt the pace o f the contest.
Accuracy.
Speed o f performance is balanced in the contest with a check on accuracy. At random times (up to a total
o f five times) throughout the race, the runners are momentarily frozen and a test word is presented via a speech synthesizer. The student must judge whether the test word he
hears is the race word he just decoded or a sound-alike distractor (e.g., bark vs. barn). Our experience with this method
is that i f the student has accurately decoded the race word,
he can easily judge the test word (see N o t e ) . Correct
judgments advance his runner. If, however, the student has
decoded the race word incorrectly, he is more likely to make
a mistake on the test word. Errors in judgment push the
computer's runner ahead one step. Errors also result in
repeated presentation o f the appropriate race words at the
RON
C-MAN
treat
Figure 1. Screen mock-up of the Reader's Assistant showing the Annotator.
Downloaded from rse.sagepub.com at PENNSYLVANIA STATE UNIV on May 18, 2016
end o f the race, providing the student with additional practice. While this method may not be as effective as the potential suggested by speech recognition, it has the advantage o f
using currently available and relatively inexpensive technology.
Feedback.
Like all successful videogames, R A C E R responds to a student's performance with dynamic representations o f action and immediate feedback reflecting his performance. In R A C E R both graphics and sound effects are
freely used. T h e pace o f a student's decoding, for example, is
reflected in the relative positions o f the runners. I f the pace
o f decoding begins to slow, the student's runner begins to
fall behind the computer's runner; i f his pace increases,the
student's runner advances on the computer's runner. Thus, a
quick look at the relative positions o f the runners lets the
student gauge the efficiency o f his decoding performance .
In the same way, graphics and sound effects are used to inform the student about his decoding accuracy as checked
through the identification task. These forms o f immediate
feedback make it possible for students to monitor and adjust
their decoding performance for both efficiency and accuracy.
Adaptiveness.
In R A C E R , the student is always challenging his own level o f skill rather than an arbitrarily established standard, as in most drill-and-practice programs.
T h e system automatically adapts itself to the student's progress in two ways.
First, it adjusts the pace o f the race (i.e., the amount o f
time in which the student is expected to decode the set o f 20
words) to represent the mean performance o f the student's
previous three races. This means that i f a student's pace has
slowed in the last three races (say, in order to increase accuracy or because o f errors in the identification task), the
amount o f time he has to complete a race before the computer's runner crosses the finish line will be increased. Conversely, i f his pace has quickened and he has maintained his
accuracy, R A C E R responds by gradually reducing the
amount o f time he has to decode the set o f 20 words. Thus,
there is a constant interplay between a student's overall progress and his current decoding performance. T o win a race,
the student must respond more quickly than his average race
time over the last three races.
Second, as a student demonstrates mastery o f words at a
given level o f difficulty (e.g., one-syllable words with short
vowels), R A C E R gradually begins to substitute more difficult words (e.g., replacing the one-syllable, short vowel
words with one-syllable, long vowel words). T h e substitution continues provided that the student can maintain accuracy. A total o f 8,000 words constitutes the R A C E R corpus, ranging from those representing simple, consistent
decoding rules to those representing complex, five-syllable
words. W o r d difficulty is defined by frequency (high, middle, and low frequency) and length (one, two, three, four,
and five syllables). I f necessary, students can begin R A C E R
instruction with a set o f words chosen to represent basic
decoding principles, for example, simple short and long
vowels, digraphs, r-controlled vowels, initial and final consonant blends. Thus, it is possible to customize materials ac-
cording to the individual student's needs, either by structuring the existing corpus or by adding to it. In this regard, it is
important to note that explicit decoding rules are not taught
to students as part o f R A C E R training. Instead, decoding
skill is developed progressively through a deliberate structuring o f task materials and conditions.
Motivation.
Finally, R A C E R instruction is highly motivating. Elements o f challenge and fantasy are deliberately
built into the game to make students eager to play (cf.
Malone, 1981). This is an important consideration i f
students are to engage in the several thousand practice trials
(where one trial equals one pronunciation attempt) that are
required to achieve the goal o f automaticity. More powerful
than these external stimuli, however, is the internal motivation that comes from achievement in an area in which the
student has a prior history o f failure. Students like Ron,
whose school experience is one o f frustration and failure,
see dramatic, demonstrable improvements in their skill on a
day-to-day basis. In fact, in research studies investigating the
effectiveness o f R A C E R , students have requested additional
practice, arriving at unscheduled times and asking for longer
practice sessions.
Effectiveness
W^hat have evaluation studies revealed
about the effectiveness o f R A C E R instruction? W e briefly
review results from a training study that evaluated the effectiveness o f an earlier version o f R A C E R with low achieving
9th- through 12-grade students. Six students were intensively trained. O n the Nelson-Denny Reading Test (Brown,
Nelson, & Denny, 1973), these students' raw total scores
(obtained by combining scores from the Vocabulary and
Comprehension subtests) ranged from 18 to 3 9 ; percentile
ranks ranged from the 4th to the 29th. R A C E R training extended over 24 sessions, each lasting about 30 minutes. T w o
types o f results are presented here, those relating to skill acquisition itself and those relating to transfer o f decoding to
reading fluency in the context o f a demanding comprehension measure (for details, see Frederiksen et al., 1985b).
Training resulted in marked improvements in both speed
and accuracy o f isolated word and pseudoword decoding.
Students showed overall gains in pronunciation speed, the
greatest gains being for the most difficult test items. After
R A C E R instruction, students decoded two-syllable words
(e.g., habit) as efficiently as one-syllable words (e.g., come)
and two-syllable pseudowords (e.g., kbit) as efficiently as
one-syllable pseudowords (e.g., brend). Importantly, this improvement in efficiency did not come at the expense o f accuracy. Students' overall accuracy for words and pseudowords also improved, from 73% on the pretest to 87% on
the posttest. These results, especially the finding that improvements in pseudoword pronunciation were equal to or
greater than those for words, suggest that R A C E R instruction promoted development o f automatic phonological
decoding skill, over and above any improvements attributable to development o f automatic word recognition
skill.
T h e transfer results o f R A C E R training to a hard test o f
inferential comprehension that was constructed as part o f a
Downloaded from rse.sagepub.com at PENNSYLVANIA STATE UNIV on May 18, 2016
transfer battery were more complicated. T h e general purpose o f the task was to test students' skill in inferring a
semantic relation (represented by a connective such as but or
as a result) linking the ideas in a three-sentence paragraph.
Dependent measures included (a) reading time per word for
the third and final sentence (the one requiring the inference) and (b) accuracy in choosing the appropriate conjunctive expression from a pair o f expressions. T h e results
supported the notion o f interactions among linked components. By itself, R A C E R training did not have an impact
on students' performance on the inference task. However,
R A C E R training in conjunction with training in use o f
context did have an effect on reading time for the third sentence in the inference task. Students who were trained in
both decoding and use o f context to to retrieve and integrate word meanings showed reductions in reading time
for the third sentence. Students trained in use o f context
without the benefit o f R A C E R training showed no reductions in reading time. T h e reductions in reading time in a
task involving a specific and demanding comprehension
problem were not due to a loss o f accuracy in selecting the
appropriate connective; accuracy rates remained above 70%.
In addition, following training in use o f context, students
who had prior R A C E R training continued to show improvements in decoding speed. These results suggest some
support for a componential model in that reductions in the
effort required to make sense o f demanding texts such as
those used in the inference task came about through improvements in the performance o f multiple skill components; namely, decoding and use o f context.
Usefulness.
W h a t does a system like R A C E R offer special education students? Many special education students
have trouble learning to read. T h e acquisition o f basic
decoding skill is often the first obstacle they meet. For some
students, this obstacle proves to be insurmountable, negatively affecting their schooling and learning over the long
term. It is not unusual for these students to have had several
years o f remedial reading instruction by the time they reach
middle or junior high school. Without concrete successes in
acquiring basic skills, they begin to lose confidence in their
abilities and, less obviously, in the school's ability to help
them.
R A C E R offers students a challenging and efficient environment for developing decoding skill. Features like game
design, individualization, and graphic and auditory feedback
make decoding practice not only a self-motivating experience but also a constructive one (although we are careful to
note that some o f these features may not be suitable for
every student). Because R A C E R is highly structured,
students quickly come to understand what the goals o f practice are; they know how to win the game. Moreover, the
rewards are tangible and progress is relatively quick, especially when compared to more conventional paper-andpencil activities. Because R A C E R monitors and continually
adapts to the student's current level o f performance, even
the least skilled students can experience success. As students
encounter difficulty, R A C E R slows the pace o f the game to
provide them with more time for decoding; as they
demonstrate mastery, R A C E R provides additional challenge, either by increasing the pace o f decoding or by introducing more difficult words. In these ways, students
become aware o f the learning process ( o f the tradeoff between speed and accuracy, for example). They also become
aware o f the progress they are making, how fast they are able
to decode words, and, specifically, how fast they can decode
words at different levels o f difficulty. Finally, because
students receive immediate feedback, they are able to monitor their decoding performance for accuracy and speed.
They see that their progress is tied directly to what they do
with each word; in short, they can exercise control over
their learning in a way not always possible with more conventional decoding tasks.
Although R A C E R can be an effective environment for
acquiring decoding skill, it provides students with only one
o f the prerequisite skills they need for comprehension. In
particular, R A C E R does not address the issue o f transfer. It
does not speak to the problem o f whether students will use
their newly acquired skills in the context o f meaningful
reading tasks. While this is a critical issue with regard to all
students, it is especially critical for special education
students who have tremendous difficulty generalizing skills
learned in one context to other contexts. An important
question, then, is: Once they have acquired basic decoding
skill in a highly structured, practice oriented environment,
what is needed to ensure that they will use those skills as
they read to learn? W i t h this question in mind, we turn to an
examination o f the constructivist approach to reading,
which emphasizes contextualization o f skill development
from the start.
A Constructivist Approach
Theory. A constructivist approach to reading instruction emphasizes reading as a problem-solving process
shaped by communicative purpose (Collins, Brown, &
Larkin, 1980; Rosebery et al., in press). From this perspective, readers solve problems o f meaning by actively building
understandings that draw on multiple knowledge sources
and comprehension strategies (Bock & Brewer, 1985;
Collins et al., 1980; Spiro, 1980). But the problem-solving
process is not simply an interaction between a reader and a
text; it involves interaction among a reader, a text, and an
author (Bruce, 1981; Rosebery et al., in press). T h e term
communicative purpose that we used above refers to both the
reader's purpose and his or her efforts to understand the
author's purpose (i.e., to solve the problem o f intended
meaning).
W h a t does it mean to say that readers solve problems o f
meaning? First, it means that what a reader understands is
not simply the literal content o f a text, but an interpreted
meaning. By interpreted meaning, we refer to a meaning
that is elaborated inferentially on the basis o f the reader's
knowledge and beliefs (e.g., knowledge o f the world, beliefs
about the author's intentions, etc.), purposes, skill, understanding o f the task situation, and the like (Bransford,
Barclay, & Franks, 1972; Collins et al,. 1980). Like the com-
Downloaded from rse.sagepub.com at PENNSYLVANIA STATE UNIV on May 18, 2016
ponential view, the constructivist view assumes that the
reading process is interactive, although the interactions are
focused on the communicative process as a whole rather
than on individual skill components.
Second, when readers solve problems o f meaning, they
engage in an active thinking process involving the progressive refinement o f their understanding (Collins et al.,
1980). According to this view, the reader may begin with a
general model o f what a discourse is about (because, for example, she knows something about the author, or the topic,
or because the title is suggestive). But, as she reads on, her
understanding evolves into a more elaborated, perhaps
radically restructured, model o f meaning. This evolution in
her understanding is mediated by a number o f problemsolving strategies. She may, for example, question the
assumptions that are implicit in the understanding she has
been building, as might occur in reading satire, for example.
She may reformulate initial hypotheses about the author's
purpose and then test the new hypotheses with predictions
(Collins & Smith, 1982). O r she may skim previous pages in
search o f specific kinds o f evidence to support a revised understanding or to clarify a confusion.
hender goes about making sense o f a text by modeling use o f
the four strategies. As the students' strategic skill expands,
the teacher fades increasingly from the process, acting as
monitor and critic and allowing the students to assume more
and more control over their learning (see Collins et al., in
press, for an analysis o f reciprocal teaching as a form o f cognitive apprenticeship).
Finally, a contructivist model o f reading implies that
there is typically not a single solution to a given problem o f
meaning; rather, a given text is open to more than one interpretation or understanding. T h e model also implies that the
process by which a reader charts a path through the "space
o f possible solutions" is as critical as the solution he or she
actually produces. In fact, the product cannot be separated
from the process, a view that stands in contrast to traditional
reading comprehension practices. Helping students acquire
control over the process o f understanding and the process
o f monitoring their understanding is, therefore, a key element in a constructivist approach to instruction.
Reader's Assistant
Practice. T h e reciprocal teaching approach o f Palincsar
& Brown (1984, in press) is one method that derives
from a constructivist view. In this approach to reading comprehension instruction, learning takes place cooperatively
within a group that includes students and a teacher. T h e
method focuses on providing members o f the group with
guided practice in using four simple strategic—or comprehension-fostering—skills: question formulation, summarization, prediction, and clarification. T h e strategies are
practiced in the context o f discussions about portions o f a
text that the group is cooperatively trying to understand.
The method is reciprocal in that the teacher and students
take on different roles in the discussion process, at times acting as discussion leader (the one who produces questions,
summaries, etc.) and at other times as critic (the one who
evaluates what the leader has produced).
Within the reciprocal teaching context, individual students' efforts are supported by the larger group as well as by
the teacher. T h e teacher acts as a coach, providing students
with guidance in how to pose good questions and construct
good summaries, to the extent that such guidance is needed.
For example, the teacher offers prompts to students or critiques the questions they produce. T h e teacher also acts as a
model or expert, demonstrating how a skilled compre-
W i t h this discussion o f a constructivist approach and
reciprocal teaching in mind, we now turn to a second
scenario in which we imagine two students who are working
with the Reader's Assistant, an experimental microcomputer environment that provides students with a variety o f
supports for solving problems as they read (Rosebery,
1986b; Warren, 1986b). T h e Reader's Assistant, unlike
R A C E R , is currently under development and, therefore, has
not been evaluated. Like R A C E R , its design is driven by
theory, in this case, constructivist theory. In the scenario we
present a vision o f one o f the ways in which the Reader's
Assistant, or tools like it, may be used to foster reading
comprehension.
Rebecca and Denise are two middle school students who,
like Ron, are assigned to a resource center. Also like Ron,
they cannot handle on their own the demands o f eighthgrade texts and assignments, although not necessarily for
the same reasons.
At the center, they have been working on improving their
comprehension. In particular, they have been working in a
reciprocal teaching situation with their teacher, Ms. Stone.
Recently, Ms. Stone has suggested that Rebecca and Denise
work together at the computer using the Reader's Assistant,
a tool-based environment for reading. T h e Reader's Assistant provides students with a set o f tools designed to support comprehension as a problem-solving activity. In this
environment, Rebecca and Denise can use their emerging
comprehension strategies independently o f the teacher and
at the same time receive support from the computer as well
as from each other.
Rebecca and Denise are collaborating on " W e s t Side
Story," an assignment for English. Their English class is putting on an informal production o f the play and Rebecca and
Denise are to play the two female leads, Maria and Anita.
Learning at the center is typically situated within contexts
that are meaningful to the students. In this case, Rebecca and
Denise have asked Ms. Stone for help in understanding the
play and, in particular, their characters. Together, they have
decided that the girls might use a question asking strategy
they have learned at the Center to explore a dialogue rhat
reveals the feelings and attitudes o f each character after Bernardo has been killed.
Because computers are used extensively throughout the
school curriculum, most student texts are in the computer
library. Rebecca and Denise call up the play and begin to
read the dialogue. Together, they read sections o f it, discuss
meaning and generate questions that they think highlight
each character's point o f view. Wlien they formulate a ques-
Downloaded from rse.sagepub.com at PENNSYLVANIA STATE UNIV on May 18, 2016
tion, they use the Annotator, one o f the process tools available in the Reader's Assistant, to attach the question directly
to a text segment (see Figure 2 ) .
Process Tools. T h e Annotator is a tool designed to
facilitate and make explicit the kinds o f comprehensionfostering and monitoring processes that form the basis o f
constructivist approaches to instruction, such as those
found in reciprocal teaching. T h e fundamental notion
behind the Annotator is to engage students in activities that
will help them understand reading as an active problemsolving process, involving both constructive and reflective
processes (cf. Collins et al., in press). In general, tools within
the Reader's Assistant are designed to help bring to the surface the cognitive and metacognitive processes that normally remain tacit in instructional situations so that students
can observe, practice, and refine them across different contexts (e.g., as independent reader, as reader-writer, as
collaborative reader, as critic, etc.).
In the current context, the process o f directly connecting
a question to a text has several benefits. It emphasizes question formulation as an active process, one that is integral to
understanding. This stands in contrast to the traditional
model o f comprehension presented to students in which
questions are asked o f a student primarily after they have
finished reading. In the same way, the process emphasizes
question asking as a process that the reader uses to monitor
and direct his or her ongoing comprehension rather than as a
File
Say
ECF
Define
The Most Dangi
test o f a static product o f comprehension, like main idea or
detail questions. T h e Annotator reinforces the notion o f
question asking as self-evaluation because, as students use it,
they have opportunities to reflect on the meaning or purpose o f the questions they generate. In the process they can,
for example, revise a question t o make it more precise or
relevant, flag it for later attention, or ask for help from a
peer or teacher. In these ways, the process o f question formulation leads the student to reflect on the meaning he or
she has constructed.
How, specifically can the Annotator, as a computerbased tool, support the student's problem solving? First, a
student's annotations might be recorded in a "process history." A process history records all the annotations and any
revisions a student makes to a text in the sequence they are
made. T h e process history can itself then become an object
that can be studied for various purposes, for self-evaluation,
as study notes to aid comprehension, as ideas that can evolve
into a plan for writing, or by the teacher as a basis for understanding what is being learned. Second, the Annotator
can be structured in such a way that it allows students (and
teachers) to operate on the contents o f the process history in
order to focus their problem solving on different aspects o f
the process. T h e student might, for example, filter his or her
process history according to topic (e.g., all annotations relating to a particular character, theme, or event) or type o f annotation (e.g., all summaries, questions, or predictions). T h e
teacher might, for example, want to filter a student's process
Annotate
Diagram
Summary
(Page t )
Rrowse
Question
- is a large island,
"Off there to the right said Whitney.
"It's rather Commcnl
"What island is it?" Rainsford asked.
"The old charts call it 'Ship-Trap Island'," Whitney replied.
"Sailors have a dread of the place, i don't know why. Some
superstition-"
"Can't see it," remarked
(he dank tropical night t h a i — I
I
thick warm blackness in up|
Figure 2. Screen mock-up of the RACER instructional system.
Remedial and Special Education
Downloaded from rse.sagepub.com at PENNSYLVANIA STATE UNIV on May 18, 2016
history by type o f annotation, extracting all the summaries
the student has produced over a given period o f time in
order to examine qualitative changes or to compare the
quality o f independent and collaborative work. Third, in the
hands o f a teacher, the Annotator can become an authoring
tool. The teacher can comment on a text, attaching information on content (e.g., background information, summaries)
or on process (e.g., marking a difficult passage and articulating why it is difficult or modeling some o f the kinds o f useful questions students might ask while reading). One o f our
main research goals in the Reader's Assistant project is to investigate what effects, i f any, the articulation o f processbased thinking, both on the part o f the student and teacher,
may contribute to students' comprehension.
Enabling Tools. T h e Reader's Assistant also includes
tools that help students solve some o f the "bottleneck"
problems o f reading like decoding and understanding word
meaning. A speech tool, for example, lets readers request
the pronunciation o f single words, phrases, sentences,
paragraphs, or entire screens o f text. In this way, students
can get support for weak or newly acquired decoding skills
when necessary. Another tool, an on-line dictionary, lets the
reader look up the meanings o f unfamiliar words; definitions include both graphic and textual information. In addition to typical dictionary information (e.g., pronunciation
guide, part o f speech, definitional information), the dictionary displays the sentence in which the unfamiliar word appeared. A third tool functions as a concordance, letting the
reader see in a single window the multiple contexts in which
a given word occurs in a text. In this way, the reader can examine the meaning and use o f a word, noting consistency in
meaning or subtle connotative or denotative differences.
When Denise and Rebecca have finished with the
dialogue, they will print out a copy o f their process history
and meet with Ms. Stone. In the conference, Ms. Stone will
ask them about the thinking that went into their question
generation process as well as about the content and quality
o f the questions themselves. Together, they will use the process history as a reference for discussing the girls' understanding o f their characters. In this context, Ms. Stone
can model for Denise and Rebecca the kinds o f reflective
and self-evaluative strategies she wants them to develop; she
can suggest ways that they can improve their question asking
strategies; and she can provide support for extending their
skills to increasingly difficult problems.
Usefulness.
T o date, development continues on a prototype version o f the Reader's Assistant. Although we have
not evaluated its effectiveness in promoting comprehension, we can describe some o f the ways we think it will facilitate reading instruction for special education students. T w o
features, in particular, are designed to support the development o f comprehension in poorly skilled readers: enabling
tools and process tools.
The enabling tools (e.g., speech synthesizer, on-line dictionary, etc.) provide support for some o f the bottleneck
problems o f reading like word recognition and word meaning that can disrupt comprehension. Bottleneck problems
can be especially troublesome for special education students,
who frequently lack the basic skills to solve such problems.
As a result, their comprehension efforts are side-tracked or
bog down. How often, for example, do students lose sentence or paragraph meaning as they struggle to decode an
unfamiliar word or as they skip over unrecognized word
meanings? By providing easily accessible help we believe
that the Reader's Assistant will support the comprehension
efforts o f poorly skilled readers by letting them concentrate
principally on solving problems o f meaning.
One possible negative side effect o f the enabling tools is
that students might become dependent upon them, deferring to the speech synthesizer for help rather than exercising
their own decoding skills. Although we cannot yet address
this question directly with respect to the Reader's Assistant,
we have found in our studies o f R A C E R that students prefer
to exercise their own skills whenever possible and request a
pronunciation only when they are completely baffled.
T h e process tools provide students with a different kind
o f support for comprehension, operating more at the level
o f reasoning about text. These tools focus students' attention on activities like question asking and prediction, aspects
o f comprehension that can otherwise be difficult for special
education students to conceptualize, let alone improve. By
turning these internal, mental processes into explicit activities that one does while reading, like looking up words in
a dictionary, the process tools support poorly skilled
readers' meaning-making efforts. Moreover, teachers can
use the very same tools to annotate texts with models of, or
other support for, the kinds o f problem-solving skills they
want students to acquire. In this function, the tools make
available a very carefully tailored source o f support for
comprehension.
Finally, tools are, by their very nature, flexible. Their uses
can be continually redefined, depending on the characteristics and purposes o f the user and the contexts in which they
are used. Thus, as a student's skill improves, his or her use o f
the tools is likely to change. T h e Annotator, for example,
may come to serve less as a framework for learning to read
(e.g., what kinds o f questions to ask) and more as a tool for
learning from text or for bridging from reading to writing
(e.g., using it to record notes about argument development
in an essay one is reading).
W e put forward these speculations cautiously, however.
Many o f the questions regarding the usefulness o f a learning environment like the Reader's Assistant, especially for
academically delayed students, remain open. For example,
with the Reader's Assistant, the student (rather than the
teacher) becomes the decision maker, deciding when and
what kind o f help is needed. T h e reader is responsible for
identifying the kinds o f problems that need to be solved,
locating the appropriate forms o f help, and structuring a
problem-solving path. Productive use o f the tools is
therefore predicated on the reader's being an active problem
solver. But developing readers might not be able effectively
to take on the kind o f active problem-solving role that use
o f the Reader's Assistant assumes. Student interaction with
the tools will, therefore, need to be structured to promote
active participation (e.g., by supporting collaborative learn-
Downloaded from rse.sagepub.com at PENNSYLVANIA STATE UNIV on May 18, 2016
ing or by scaffolding the particular process so that the student does not bear the full burden o f production). Other
questions that warrant consideration include the following:
What role, i f any, can expert modeling (e.g., teacher modeling o f strategies as in reciprocal teaching) play in the process? What should expert models look like to promote
learning? How is the appropriateness o f a model for a given
reader to be determined? Questions such as these, which are
really questions about the tradeoff between complexity and
clarity in a constructivist approach, are inevitable because
the approach is committed to teaching reading as a rich,
problem-solving process. They are also among the most
challenging issues we face as we continue to use psychological theory to drive the design o f practice, both on the computer and off.
An Integrative Approach
In the introduction to this article, we described the componential and constructivist approaches as distinct in terms
o f both the focus and context o f instruction. W e then went
on to show how each has led to different computer-based instructional practices. In conclusion, we would like to emphasize instead the potential complementarity o f the two
approaches, a complementarity that is meant to be responsive to the practical realities and goals o f reading instruction
in the special education classroom.
dents a sense o f reading as a purposeful, meaning-making
process, not as a passive information storage and retrieval
process. Collins et al. (in press), in an analysis o f reciprocal
teaching, suggest that part o f its effectiveness is in conveying to students a new conceptual model o f the reading process through a focus on active strategies like question formulation, prediction, summarization, and clarification.
How, then, does the notion o f contextualization apply to
computer use in the special education classroom and, in particular, to componentially and constructivist oriented tools
like R A C E R and Reader's Assistant? W e believe that an integrated approach is not only possible but also that it can
greatly facilitate the special education student's acquisition
o f reading skill. In conclusion, we want to suggest how contextualization relates to computer use in reading instruction.
A great deal is now known about the cognitive character
o f the reading process (Perfetti, 1985; Spiro, Bruce, &
Brewer, 1980). Theories have been developed to extend and
test our understanding o f the reading process, with certain
major points o f agreement (e.g., that reading is interactive)
but with clear differences in emphasis as well. These
theories, two o f which we have explored in this article, have
in turn given rise to instructional models that are typically
viewed as competing rather than complementary. But, in
fact, neither the componential nor the constructivist approach deals comprehensively with the reading process and,
in particular, with the range o f difficulties that special
education students actually meet when they try to learn to
read or to learn from reading (cf. Anderson, Hiebert, Scott,
& Wilkinson, 1985). This lack o f comprehensiveness argues,
we believe, for a sensible union o f the two approaches, the
goal o f which is to enable students to become good readers,
that is, readers who can independently learn from text.
At first glance, the notion o f contextualization seems
seriously at odds with componentially oriented instruction.
W e believe, however, that contextualization may greatly
enhance and be essential to sustaining the known benefits
o f such instruction. T h e issue here is not whether to teach
decoding at all, for we know that bottleneck processes like
decoding do interfere with reading progress and that a system like R A C E R is highly effective in developing skilled
decoding, especially for academically delayed students.
Rather, the issue is one o f understanding the purpose—or,
in our terms, the c o n t e x t — o f decoding instruction; the
focus on individual skills, where it is judged to be important, needs to be complemented by a constructivist perspective, i f students are to understand the purpose o f the reading activities and procedures they practice. Skills practice,
therefore, needs to be situated within an overall learning
context that reflects a constructivist reading model and
demonstrates to students in every possible way (e.g.,
through their activities, the teacher's behavior, and the class
activities) the constructive and communicative character
o f reading.
A crucial issue, then, for integrating the componential
and constructivist approaches into a sensible model o f computer use for the special education classroom is to determine
a basis for integration. W e propose "contextualization" as a
basis for bringing together the two approaches. By contextualization we mean a process o f tying reading instruction
explicitly to an understanding o f purpose both in the context o f the immediate instructional activity and in the larger
context o f learning. T h e concept o f purpose we have in
mind derives from a constructivist perspective that conceptualizes reading as an active, goal-driven process focused on
the construction and communication o f meaning. Contextualization, therefore, focuses on communicating to stu-
T h e notion o f contexualization directly applies to the
character o f the skills practice environment as well. T h e
practice environment itself should be transparent to the
students, allowing them to develop a sense o f the skill and
even, where possible, a sense o f its purpose in the larger
process. In the case o f R A C E R , for example, students
seemed to monitor their performance against changing task
conditions such as changes in the complexity o f the
materials, increases in speed, and decreases in accuracy.
They seemed to develop a model, however tacit, o f effective performance and to make use o f it during practice. As
we discussed earlier, many features o f the R A C E R environment may contribute to this observed effect, chief among
Downloaded from rse.sagepub.com at PENNSYLVANIA STATE UNIV on May 18, 2016
them being continuous feedback focused on critical dimensions o f skill performance and the explicit representation o f
goals.
Finally, there is evidence suggesting that contextualization itself can promote individual skill development. Recall
that students who had R A C E R training showed additional
improvement in decoding skill following training in use o f
context. This suggests that it is both important and beneficial to have students use their newly acquired skills in the
service o f comprehension.
The Reader's Assistant, in contrast to R A C E R and other
skills oriented software, has been designed with a constructivist model in mind. It can, in fact, support multiple kinds
o f contextualization. First, the Reader's Assistant can support contextualization o f newly acquired skills like decoding. Indeed, its enabling tools were built expressly for this
purpose, to enable students to read texts that are o f interest
to them but that, unsupported, would be beyond their
reach. Our feeling is that the speech tool, to take one example, will prove most effective as a resource for readers by
allowing them to gain a measure o f control over a critical
aspect o f the reading process. It will not, in our opinion, be
effective as the principal instructional tool for building effective decoding skill. In general, the transition from practice o f individual skills to use o f those skills in reading can
be supported at many levels within a framework like the
Reader's Assistant, through the enabling and process tools
as well as the problem-solving context.
The process tools within the Reader's Assistant represent
another form o f contextualization, focusing students' efforts on reflective and constructive activities as they read.
But the tools in and o f themselves are not sufficient to promote learning; they, too, need to be contextualized. This
means, as Brown and Palincsar (in press) have argued, creating learning situations that set into motion particular processes, like negotiation o f conflicting views and elaboration
o f new points o f view. Thus, while systems like the Reader's
Assistant provide support for particular aspects o f problem
solving, it is extremely important that their use be not only
scaffolded for students but also contextualized. Only in
these ways can such tools help facilitate the acquisition o f
expertise in reading.
In sum, we believe that the essence o f contextualization
lies in creating learning situations that help students understand the purposes and processes o f reading. T h e computer
can play an important role in contextualizing reading instruction, provided it is not viewed as an autonomous agent
o f change. Its potential as a tool for reading instruction
depends on the quality o f its design (e.g, the psychological
and pedagogical underpinnings) and on the nature o f the
contexts in which it is used. £ ±
B e t h W a r r e n is a scientist in the Education Department at
BBN Laboratories Inc. and a senior researcher with the
Reading Research and Education Center. Her current research
interests are in innovative uses of technology in reading, writing,
and second language learning, and in students' and teachers'
mental models of history. She has a doctorate in human
development and reading from Harvard University, Graduate
School of Education. Ann S. R o s e b e r y is a scientist in the
Education Department at BBN Laboratories Inc. and a senior
researcher with the Reading Research and Education Center.
Her current work focuses on developing educational technologies
in a wide range of disciplines (reading, writing, statistics) and
in developing teacher education programs. She has a doctorate in
human development and reading from Harvard University,
Graduate School of Education, and has an extensive teaching
background that includes middle and high school,
undergraduate, and special education.
Note
In RACER, the computer does not have speech recognition
capabilities; that is, it does not recognize and subsequently judge
the accuracy of the student's pronunciation.
References
Anderson, R., Hiebert, E., Scott, J . , & Wilkinson, I. (1985). Becoming a nation of readers: The report of the Commission on Reading.
Washington DC: National Institute of Education.
Bock, J . , & Brewer, W. (1985). Discourse structure and mental models
(Tech. Rep. No. 343). Champaign: University of Illinois,
Center for the Study of Reading.
Bransford, J . , Barclay, J . , & Franks, J . (1972). Sentence memory: A
constructive versus interpretive approach. Cognitive Psychology, 3,
193-209.
Brown, A., & Campione, J . (1986). Psychological theory and the
study of learning disabilities. American Psychologist, 14, 10591068.
Brown, A., & Palincsar, A.S. (in press). Guided, cooperative learning and individual knowledge acquisition. In L.B. Resnick (Ed.),
Cognition and instruction: Issues and agendas. Hillsdale, NJ:
Erlbaum.
Brown, J.I., Nelson, M.J., & Denny, E.C. (1973). The Nelson-Denny
reading test. Boston: Houghton Mifflin.
Bruce, B. (1981). A social interaction model of reading. Discourse
Processes, 4, 273-311.
Collins, A., Brown, J.S., & Larkin, K. (1980). Inference in text understanding. In R J . Spiro, B.C. Bruce, & W.F. Brewer (Eds.),
Theoretical issues in reading comprehension (pp. 385-410).
Hillsdale, NJ: Erlbaum.
Collins, A., & Brown, J.S. (in press). The computer as a tool for
learning through reflection. In H. Mandl & A.M. Lesgold
(Eds.), Learning issues for intelligent tutoring systems. New York:
Springer.
Collins, A., Brown, J.S., & Newman, S. (in press). Cognitive apprenticeship: Teaching the crafts of reading, writing and
mathematics. In L.B. Resnick (ed.), Cognition and instruction:
Issues and agendas. Hillsdale, NJ: Erlbaum.
Collins, A., & Smith, E.E. (1982). Teaching the process of reading
comprehension. In D.K. Detterman & R J . Sternberg (Eds.),
How much and how can intelligence be increased? (pp. 173-185).
Norwood, NJ: Ablex.
Curtis, M.E. (1980). Development of components of reading skill.
Journal of Educational Psychology, 12, 656-669.
(continued on p . 61)
Downloaded from rse.sagepub.com at PENNSYLVANIA STATE UNIV on May 18, 2016