Fluency transfer: Diﬀerential gains in reading speed and accuracy
following isolated word and context training
SANDRA LYN MARTIN-CHANG and BETTY ANN LEVY
McMaster University, Canada
Abstract. While ﬂuent reading is recognized as a primary goal of educational
instruction, the methods that best promote the development of ﬂuency remain
unclear. Two experiments are reported that examined increases in reading ﬂuency of
a novel passage following two types of training. In the context training condition,
children learned to read a set of target words in a story context, while in the isolated
word training condition, ﬂuency with a target word set was gained from a computer-
ized word naming game. Transfer of ﬂuency to reading these words in a new context
was then measured by gains in reading speed, accuracy, and comprehension of a
novel story. Results indicated that young readers showed speed beneﬁts on transfer
stories following both context and isolated word training, but the increases were
larger following context training.
Key words: Contextual facilitation, Reading ﬂuency, Rereading
Fluent reading involves the accurate and rapid rendering of a text and
is often accompanied by superior comprehension (Breznitz, 2001; Kuhn
& Stahl, 2003; Levy, Abello, & Lysynchuk, 1997). Each of these aspects
of ﬂuency develops with experience, but it is not yet clear what pro-
cesses underlie each measure. There has been considerable
debate about the relationship between reading accuracy, speed, and
comprehension (Breznitz, 2001; Bourassa, Levy, Dowin, & Casey, 1998;
Fleisher, Jenkins, & Pany, 1979). In the present study, each aspect of
ﬂuency was measured following two experimental training regimes.
Speciﬁcally, we were interested in identifying whether learning words in
context brought additional speed and accuracy beneﬁts over and above
learning words in isolation; and if so, if such beneﬁts would result in
corresponding gains in comprehension.
Thus far, empirical investigations of the role of context have derived
from two theoretical positions, each focused on a diﬀerent aspect of
skilled reading. The ﬁrst emphasizes the importance of context during
the on-line naming of words. It asks what children, if any, are able to
identify more items during a single exposure when words are presented
in a meaningful context compared to in isolation. The second position
Reading and Writing (2005) 18:343–376 !Springer 2005
focuses on understanding the role context plays in forming memory
representations. Here the question is whether learning words in context
changes the nature of memory representations, thereby resulting in
more ﬂuent reading of subsequent passages. Both positions are relevant
to the current discussion and the literature from each will be addressed
Goodman (1965, 1967) was one of the most prominent advocates of
using ‘meaning’ as both the goal and the method for reading. He pro-
posed that when a text was taken as a whole, it contained a rich source
of cues (e.g., syntax, intonation and referential meaning) that could be
used to facilitate skilled reading. Instead of decoding every word of a
passage, children were encouraged to ‘‘sample’’ from the text in order
to ‘‘guess’’ subsequent words. The idea of reading as a ‘‘psycholinguis-
tic guessing game’’ (Goodman, 1967) formed the theoretical backbone
for the ‘whole language’ movement, which has been an inﬂuential force
in forming educational policies for much of the western world (Alexan-
der, 1998). The whole language perspective posited that teaching chil-
dren how to sound out individual letters detracted from the meaning of
the text as a whole, and was therefore detrimental to reading advance-
ment (Goodman, 1965). It was this claim that ignited the longstanding
debate between those who argued that phonics should be explicitly
taught to children (e.g., Spector, 1995; Stanovich, 1980, 1986, 1994),
and those who thought that grapheme–phoneme regularities would be
acquired indirectly through the act of reading itself (Goodman, 1965,
1967; Oaken, Wiener, & Cromer, 1971).
One of the predictions made by the whole language movement was
that children would read more words correctly in context than in
isolation. In accordance with this hypothesis, Goodman (1965) reported
that children from grades 1–3 made 60% to 80% fewer errors with
words presented in a short story rather than these same words
presented randomly in a list. While this seemed like very persuasive
evidence in support of teaching in context, no precautions had been
taken to prevent order eﬀects between the story and list conditions.
Because the short stories were always read last, it was diﬃcult to
ascertain how much of the increased accuracy was due to contextual
support, and how much could be credited to repeated word exposure.
Indeed, Pearson (1978) found improvements after participants were
given the opportunity to simply read the list twice in succession. There-
fore, by Pearson’s (1978) estimation, context only accounted for 20%
or 30% of the gain in reading accuracy found by Goodman (1965).
Goodman (1967) argued that skilled reading stems from exploiting
the linguistic cues found in the text, as opposed to the precise decoding
344 SANDRA LYN MARTIN-CHANG AND BETTY ANN LEVY
of words. By this reasoning, a good reader was one who used the fewest
cues possible in order to reach the correct meaning of a word
(Goodman, 1965, 1967). Allington (1978) was the ﬁrst to directly
compare the eﬀect of context on good and poor fourth grade readers.
In doing so, he found support for the claim that context aided reading
accuracy although it was in the opposite direction of Goodman’s
predictions. Allington found that poor readers proﬁted substantially
from reading in context, whereas skilled readers performed equally well
in both list and story conditions.
In a more recent set of investigations, Nicholson and his colleagues
(Nicholson, 1991; Nicholson, Bailey & McArthur, 1991) extended the
ﬁndings of both Pearson (1978) and Allington (1978) by counterbalanc-
ing the order of presentation between context and list conditions, with
children of diﬀerent ages and abilities. They reported that only poor
readers (Nicholson et al., 1991) and young readers (6 and 7-year old
average readers; Nicholson, 1991) performed better in context
regardless of the order of presentation. On the other hand, good
readers (6 and 8 year olds; Nicholson, 1991; Nicholson et al., 1991) and
older average readers (8 year olds; Nicholson, 1991) showed consistent
improvements during whichever condition was presented last.
However, Archer and Bryant (2001) questioned whether counterbal-
ancing the presentation of the context and list conditions was stringent
enough to control for the order eﬀects in Goodman’s (1965) original
study. They maintained that using the same words for both list and
context conditions, even when counterbalanced, led to results that were
diﬃcult to interpret. Because each word was seen twice, it was diﬃcult
to know precisely how much of the improvement seen in trial 2 was
due to the current format of presentation (context or list) and how
much was due to the permanent learning that resulted from trial 1 (in
context or list). To avoid this diﬃculty, they had 6 and 7-year-old
average students read through a list of words until 8 items were missed.
The 8 unknown words for each participant were then divided into 2
groups of 4—half of which were taught within the context of
meaningful sentences, while the others were taught via ﬂash cards.
Archer and Bryant (2001) found that the children were able to read
more words in context vs. in isolation. They concluded that a
contextual facilitation eﬀect remained even when word exposure was
equated by using two diﬀerent material sets.
Thus, the results of Archer and Bryant (2001) add to the literature
suggesting that young and/or poor readers beneﬁt from the eﬀects of
contextual facilitation (see also Allington, 1978; Kim & Goetz, 1994;
Nation & Snowling, 1998; Nicholson, 1991; Stanovich, 1980, 1994).
Importantly, this is not akin to the ‘‘psycholinguistic guessing game’’
ﬁrst advanced by Goodman (1967). Instead, it is proposed that:
language prediction skill facilitates the development of word
recognition skill by enabling beginning readers to combine sentence
context cues with incomplete graphophonemic information to
identify unfamiliar words. Children should (...) use context to
supplement word-level information rather than substitute it
(Tunmer & Chapman, 1995, p. 98).
The combined use of phonological skills and context in order to
arrive at correct word pronunciations has been conceptualized as a
‘self-teaching mechanism’ that improves general reading ability (Nation
& Snowling, 1998, Share, 1995; Tunmer & Chapman, 1995). The
unspoken assertion behind this notion is that the contextual facilitation
observed on-line is responsible for a lasting change in the way words
are processed on future encounters. Viewing the eﬀects of context from
such a perspective poses a question central to reading development;
does the contextual facilitation observed during reading result in supe-
rior memorial representations thereby enhancing the later reading of
those words in a diﬀerent context? A number of experiments have doc-
umented the ways in which mastering one passage (often via the use of
repeated readings) generalizes to produce a processing advantage while
reading novel passages. These are referred to as ‘‘transfer’’ experiments
because the variable of interest is the manner in which learning trans-
fers from a training phase to later reading tasks.
For example, Dowhower (1987) trained beginning readers to read
the ﬁrst half of passages until they reached a speed criteria of 100
words per minute. She found that the accuracy, reading rate, and
comprehension of the practiced text improved with each rereading.
Moreover, she reported that the latter half of the text was read with
better accuracy, speed, and comprehension when compared to the ﬁrst
reading of the practiced text. Thus, Dowhower (1987) found that the
improvements accrued by practicing within one passage resulted in
overall reading gains on an ensuing task.
Similarly, Bourassa et al. (1998) reported evidence that training
words in context also led to faster and more accurate reading of target
words in a diﬀerent context (i.e., a new story). In addition, they found
comprehension beneﬁts on the new stories, indicating that training
words in context leads to better understanding when target words are
encountered at another time. Bourassa et al. also examined the inverse
relationship between context and lists. That is, they compared words
346 SANDRA LYN MARTIN-CHANG AND BETTY ANN LEVY
trained in context to words that received no training, and measured
how quickly and accurately both types of words could later be read in
lists. Using this paradigm, improvements in both reading rate and
accuracy were reported for words that were ﬁrst trained in context and
later presented in isolation. This led Bourassa et al. to conclude that
the beneﬁts gained from training words in context transfer to several
other levels of reading.
Rashotte and Torgesen (1985) examined the text characteristics that
were necessary for mediating ﬂuency transfer between passages. In one
condition, they had learning disabled students read a new passage each
day over a seven-day period. These 7 stories contained little content
overlap and they used very few of the same words. Within each session
the individual passages were read four times each. In the second condi-
tion, they had the same students read a new set of stories over a seven
day period. Once again these stories shared little content and were read
four times during each session. However, unlike the ﬁrst condition,
these passages contained 3 times as many shared words. Thus, the criti-
cal manipulation was the number of overlapping words found between
sessions. In condition 1, the between session stories shared very few
words, while in condition 2, many of the same word were used in all 7
Rashotte and Torgesen (1985) reported that the shared word
condition resulted in greater transfer of reading speed between reading
sessions. They concluded that a high degree of word overlap was
required for a transfer eﬀect to be observed. Repeatedly reading a text,
then, does not necessarily improve the ability to read novel passages
per se, rather it improves the ability to read novel texts that contain the
same words as the training texts.
If the amount of transfer from one text to another is mediated at
the word level for poor readers (see also Faulkner & Levy, 1994, 1999)
as suggested by Rashotte and Torgesen (1985), then training words out
of context should also result in ﬂuency transfer. Fleisher et al. (1979)
were among the ﬁrst to ask if training words in lists would transfer to
the more ﬂuent reading of texts. They had poor readers build up
reading speed for words presented on ﬂash cards. When each word
could be read in less than one second, they asked the children to read
two short passages. The transfer passage was made predominately of
trained words, while the control passage contained only unpracticed
words. In one of two studies, Fleisher et al. reported that stories were
read faster if the content words had been trained. However, this eﬀect
was not robust and was not found in a second similar study. Moreover,
comprehension beneﬁts for the stories containing trained words were
not found following either of the two experiments. This led Fleisher
et al. to conclude that training words out of context may not be
suﬃcient to promote ﬂuency among less skilled readers.
More than a decade later, Levy et al. (1997) revisited Fleisher et al.’s
(1979) classic study. They noted that the transfer stories in Fleicher’s
study were very short (less than 100 words) and that the level of
content diﬃculty was above the grade level for the children in the
study. This led Levy et al. to question whether the lack of ﬂuency
transfer was a testing artifact. Therefore, Levy et al. replicated
Fleisher’s study, using much longer transfer stories (288 words) that
were written to match the comprehension level of their participants.
Under these revised circumstances, faster reading times and improved
comprehension were reported for even the most severely deﬁcient
Thus, there is evidence indicating that training in both context and
lists leads to improved reading ability in transfer tasks. However, the
literature comparing the eﬃcacy of these methods is sparse. Dahl
(1979) was one of the ﬁrst investigators to contrast story reading
compared with single word identiﬁcation. She had children in grade 2
read a 100 word passage until it could be read in under 1 min. When
the children met the speed criterion, training on a new story began. A
second group of children were trained to read 800 individual words to
a criterion of one word a second. Finally, a third group of control chil-
dren received only classroom instruction over the eight month training
program. Dahl reported the greatest reading gains in accuracy, speed,
and comprehension, in the children who practiced reading in context.
In contrast, the children who received isolated word training were not
discernibly diﬀerent from the children in the control group. Dahl
concluded that ‘‘practice on isolated skills alone is not suﬃcient [for
ﬂuent reading]’’ (Dahl, 1979, p. 62).
In a more recent study comparing the eﬀects of learning words in
context versus in isolation, Levy (2001) reported somewhat diﬀerent
results. Levy taught groups of skilled and less skilled readers using both
context and isolated word training methods. In the isolated word condi-
tion, the grade four students practiced reading in a computer game set-
ting. In contrast, in the context condition, the same children practiced a
new set of words by re-reading an entire passage several times. Levy
reported that the two methods of training (context vs. isolated word)
both resulted in reading gains. Moreover, she found no diﬀerential
eﬀects favoring context or isolated word training in either skill group
when she measured the accuracy, reading rate or comprehension for
novel passages containing trained words.
348 SANDRA LYN MARTIN-CHANG AND BETTY ANN LEVY
Thus, Levy’s (2001) ﬁndings stand in stark contrast to those ﬁrst put
forth by Dahl (1979); however, they follow from the argument that the
reading improvements within transfer paradigms are the direct result of
the number of trained words found in the novel story. If one assumes
the beneﬁt of training is mediated at the word level (Rashotte &
Torgesen, 1985), at least for poor readers, it follows that many diﬀerent
methods of training individual words (be it in isolation or in context)
may be expected to produce similar reading outcomes on passages
containing an equal number of target items. Indeed, when Levy
compared isolated word and context training, the two methods
appeared to produce equivalent gains. Yet one limitation was that the
number of word exposures was not held constant between the word
and context conditions. Due to the fact that the children could only
read a passage once or twice during the same amount of time in which
they could read several lists, the items in the isolated word condition
received, on average, double the number of repetitions during training
compared with those in the context condition (Levy, 2001).
The conclusions regarding isolated word training, at this point in
time, are mixed. Whereas Dahl (1979) found no beneﬁts following
single word training, Levy (2001) reported substantial gains contingent
on improving word recognition. In the experiments to be reported here,
ﬂuency will be evaluated in terms of the improvements on the reading
speed, accuracy, and comprehension of novel transfer passages in good,
poor, and average readers.
It is widely accepted that good and poor readers use context as an
end to diﬀerent means. As outlined in the interactive-compensatory
model (Stanovich, 1980), good readers are better able to draw on
context under impoverished conditions (Perfetti & Roth, 1981), and for
comprehension based tasks (i.e. cloze tests). However, they seldom use
context to support on-going word recognition. In contrast, poor readers
tend to show larger relative eﬀects of contextual facilitation due to
faltering word identiﬁcation skills (Perfetti & Roth, 1981; Stanovich,
1980). Therefore, children with diﬀerent reading abilities were selected
in order to examine interactions between reader skill and method of
Experiments 1a and 1b were aimed at documenting what eﬀects, if
any, resulted from teaching good and poor readers using isolated word
training (1a) and context training (1b). The results of Experiments 1a
and 1b were also compared to determine whether one method of
training was more advantageous than its counterpart. In Experiment 2,
the ﬂuency beneﬁts following isolated word and context training were
examined once more in new population of younger readers.
Experiments 1a and 1b
Good and poor readers in grade 4 participated in isolated word and
context training programs. The children practiced new words in each
condition. Following training, the children read two novel passages: a
transfer passage containing trained words, and a control passage
containing untrained words. For the sake of clarity, the general
methods for Experiments 1a and 1b will ﬁrst be discussed which will in
turn be followed by the speciﬁc procedures, results, and discussions for
Participants. Forty-eight students in grade 4 between the ages of 8 and
9 (ranging from 106–119 months) participated in this study. These chil-
dren were chosen from 150 students who were screened in 18 elemen-
tary schools from a local public school board. The school board
adheres to a largely whole language instructional method therefore the
systematic teaching of phonics is not emphasized in regular classroom
instruction. These schools represent families from a broad socio-
economic base and a large number of racial backgrounds. All children
with parental consent were tested with the reading subtest of the Wide
Range Achievement Test—3rd Edition (WRAT3; Wilkinson, 1993).
The data from two students were excluded due to extended absenteeism
(more than 3 days missed between training and test). This resulted in a
total of 46 participants.
The good reader group consisted of 22 students (10 males and 12
females). The average age of the children in the good reading group
was 9 years and 3 months (range ¼104–118 months). The mean
standard score on the WRAT3 for good readers was 119.45
(SD ¼8.14, range ¼111–133), which corresponds to a grade 7 reading
level (Wilkinson, 1993). The poor reader group consisted of 24 students
(10 males and 14 females) whose standard scores on the WRAT3 were
lower than 90. The mean score for poor readers was 83.46, which
corresponds to a grade 2.0 reading level (SD ¼4.12, range ¼70–89).
The average age of the children in the poor reading group was 9 years
and 5 months (range ¼105–119 months).
In addition, all children selected for the study were asked to read a
list of 19 high frequency words that would be required in a later phase
of the experiment. Those achieving at least 85% accuracy on the high
frequency word list were included in the ﬁnal participant selection. In
general, the participants performed very well at this task (Good
350 SANDRA LYN MARTIN-CHANG AND BETTY ANN LEVY
M¼100%, Poor M¼93%). No children were excluded from the
study based on this criterion.
It should be noted that the schools participating in this experiment
are fully integrated, therefore, whenever possible, all children are
instructed within the regular classroom setting. Information was not
collected pertaining ‘learning disabled’ and/or ‘gifted’ designations
assigned to individual participants. Instead, the inclusion criteria for
the good and poor reading groups were based entirely on experimenter
collected data (the WRAT3 as outlined above). This sample did not
include any children who were not ﬂuent in spoken English (e.g., ESL
students) or any children with signiﬁcant developmental delays (e.g.,
Downs Syndrome or Autism).
Design and materials. Two yoked experiments (1a and 1b) were
conducted using the same participants and materials. Both experiments
employed a mixed design where a set of words was trained in phase 1,
and the transfer of learning was measured during a subsequent transfer
passage in phase 2. The critical diﬀerence between the experiments
occurred during the training phase. In Experiment 1a, the target words
were presented individually on a computer screen and were read aloud
by the children (isolated word training). In Experiment 1b the target
words appeared as part of an age appropriate story that was read with
the experimenter in a shared reading task (context training). That is, in
Experiment 1b the experimenter read the contextual words, while the
target words were read by the child. It is important to note that in both
training paradigms the children read only the designated target words.
In both experiments transfer in phase 2 was assessed by asking the
child to read a novel passage. The target words, then, were always read
in context during transfer. What varied was the method of training in
phase 1, so that the transfer conditions were deﬁned by the type of
training the children received leading up to the transfer task. In the
isolated word condition the target words were trained individually, but
then read as part of a transfer passage. In the context condition, the
target words were trained in context and transfer was measured using a
new story context. In each of the control conditions, the children read
control passages containing no trained words.
Every child participated in both training conditions separated by at
least 1 week. The order of the training conditions was counterbalanced
over all participants. The training conditions contained diﬀerent items,
so the child learned new words in each condition. At the end of each of
the two experiments (1a and 1b), the child was given a transfer story
(containing trained words) and a control story (containing new words).
Therefore, four sets of materials were required. Each material set
contained a training list, a training story, and a transfer passage
including ten relevant comprehension questions (see Appendix A).
To meet these material requirements, four mutually exclusive lists
containing 85 words each were created. Four training stories were then
written so that they corresponded to each of the lists. The training
stories contained two repetitions of each target word, resulting in 170
(2 ·85) target words per story. They also contained a number of
contextual words that were necessary to create plausible children’s tales.
Although the number of contextual words in the training stories varied
(ranging from 506–698 words), the children were only required to read
the 170 target words. Therefore, the number of critical responses was
equated over all training stories. The target words were printed in red
ink to make them clearly distinguishable from the contextual words.
All of the training stories were analyzed by the Flesch–Kincaid formula
to be at a grade 4.0 level of diﬃculty.
One day after the completion of each training session, the children
read the corresponding transfer passage and a control passage. The
shared reading task employed during training was not used in phase 2.
All four transfer stories were read entirely by the child. Thus, by the
end of the investigation each child had read four transfer passages: an
isolated word transfer, a context transfer and two control passages. The
transfer passages were analyzed by the Flesch–Kincaid formula to
range between 3.5 and 3.6 grade level of diﬃculty. They were 286 words
long and were written using primarily the target words from their
corresponding training lists/stories. The target words were repeated, on
average, 1.5 times each—resulting in 124 target words per story. The
remaining 162 items within the transfer stories were high frequency
words. These items were not included in the training materials because
it was anticipated that the children would be familiar with them (e.g.,
‘‘a, the, and, on,’’ etc). However, there were 19 high frequency words
that were over 3 letters in length and deemed to be moderately diﬃcult.
Therefore, these words were included in the screening battery to
guarantee the children could read them with a high degree of accuracy.
Thus, the transfer stories were comprised solely of: (a) trained words,
(b) words that were screened prior to training and, (c) words that
contained no more than three letters.
Ten comprehension questions tested each transfer story. The
questions were based on material that was taken directly from the texts.
They were intended to measure whether the child could remember
speciﬁc events from the passages. The questions were constructed to
avoid answers that relied on higher order skills, such as inferencing. They
352 SANDRA LYN MARTIN-CHANG AND BETTY ANN LEVY
were also written in a manner that discouraged yes/no responses. The
questions were asked and answered orally. The experimenter recorded
the child’s responses verbatim. Two independent, blind raters scored all
of the comprehension questions. Each question could be awarded a
maximum of two points: 0 for the incorrect or no response, 1 point for a
partially correct response, and 2 points for the correct response. A
Pearson correlation was used to examine inter-rater reliability. A
signiﬁcant correlation was found (r¼.972, p< .001) indicating that the
raters were in good agreement in their judgments. The mean score of the
two raters was taken to form the ﬁnal score for each transfer story,
resulting in a comprehension score that ranged from 0 to 20.
To summarize, the target words remained constant within the mate-
rial sets; however, between the sets there were no shared study items.
The materials were counterbalanced across participants, so that each of
the 4 material sets was used equally often for the isolated word, context
and the two control conditions. This counterbalance ensured that there
were no diﬀerences in materials on the critical transfer task.
Procedure: Experiment 1a. All children were tested individually in a
quiet room of their school. They received training for approximately 15–
20 minutes a day for four consecutive days (phase 1), with a testing
session occurring on the ﬁfth day (phase 2). During phase 1, the target
words were presented as a computer game. The goal was to make the
words ‘‘disappear’’ from the computer screen as quickly as possible by
reading each word aloud. Participants completed three list repetitions
each day, resulting in a total of 12 repetitions for each target word. The
85 words were randomized for each list presentation. The target words
appeared individually in the center of a computer screen. The experi-
menter controlled the onset of each item. A maximum of 1.5 seconds was
allotted for the words to be read out loud into a microphone. The word
left the screen either when the voice key was activated via the micro-
phone, or 1.5 seconds elapsed. If the child accurately read the word, no
feedback was given and the experimenter recorded the item as ‘correct’.
If, however, the child misread the word or failed to respond within the
1.5 second time criterion, the experimenter provided the correct pronun-
ciation and scored the item as ‘incorrect’. In those cases where the voice
key was activated by something other than the child’s voice, the trial was
recorded as ‘spoiled’. The computer program automatically recorded the
time between word onset and the activation of the voice key up to and
including 1.5 seconds. The number of correctly read items was summed
for the accuracy measure for each trial. The mean response times
reported below reﬂect only the words read correctly.
During phase 2, the children were tested individually on two new
passages. The order of the presentation of the transfer and control
passages was counterbalanced over all participants. The children were
asked to read both passages as fast as they could without making any
mistakes. They were also asked to read for meaning and alerted to the
fact that comprehension questions would follow upon the completion
of each story.
If a child misread a word or failed to read a word within 1.5 seconds
(as estimated by trained experimenters), the correct pronunciation was
provided in order to preserve the overall comprehension of the story.
Reading time was measured as the time it took each child to read the
story from beginning to end. The experimenter recorded reading time
on a millisecond stopwatch, as well as any errors the child made while
reading. The accuracy score was the total number of words read
correctly from the 286-word passage. Immediately after each passage
the child answered 10 verbal comprehension questions.
Procedure: Experiment 1b. Experiment 1b used a shared reading
paradigm where the target words were embedded in an age-appropriate
story. This equated the number of critical responses in Experiments 1a
and 1b. The training story was presented without illustrations. It was
presented in a binder and the child was encouraged to follow the text
with his or her ﬁnger. The experimenter read from a modiﬁed sheet on
which reading errors could be recorded. In Phase 1, the training stories
were read once a day for 6 days, resulting in a total of 12 repetitions
for each target word (two repetitions of each word per story). The
experimenter read the stories at a consistent pace, but the child was
encouraged to read the target words as fast as possible without making
any mistakes. If the child misread a word, or failed to make a correct
response within 1.5 seconds (as estimated by trained experimenters), the
correct pronunciation was provided and the word was recorded as an
error. Accuracy was recorded as the number of target words (out of
170) that the child read correctly. Response time was measured as the
time it took for the child and the experimenter to read the entire story
from beginning to end. The transfer phase of Experiment 1b was
identical to Experiment 1a.
Results and discussion: Experiment 1a
Training phase. The analyses reported below were conducted using
combined list repetitions. Pairs of list repetitions (e.g. 1 & 2, 3 & 4,
etc.) were summed so that one combined isolated word trial from
354 SANDRA LYN MARTIN-CHANG AND BETTY ANN LEVY
Experiment 1a contained the same number of target words as one
context trial in Experiment 1b (2 ·85 ¼170). The average reading time
for correct responses was recorded as the reading time measure.
The mean number of correct responses and the mean reading times,
for both good and poor readers, are presented in Table 1. Reading
times decreased by more than 65 milliseconds per word for poor
readers and 19 milliseconds for good readers. A 2 ·6 mixed ANOVA,
where the between-subject factor was group (good and poor) and the
within-subject factor was repetition (1 through 6), conﬁrmed this obser-
vation. There were signiﬁcant main eﬀects of group, F(1,44) ¼143.96,
MSE ¼5.184, p<.001, and repetition, F(5,220) ¼7.48, MSE ¼.01,
p< .001, indicating that good readers were faster than poor readers
and that both groups were getting faster across trials. However, a sig-
niﬁcant Group ·Repetition interaction, F(5,220) ¼2.68, MSE ¼.003,
p< .05, suggests that this gain in speed was disproportionate. The
poor reading group showed greater speed beneﬁts due to training
compared to the good readers.
As indicated by the accuracy data from Table 1, the data from the
good readers were at ceiling throughout training. Consequently, a
one-way ANOVA was carried out on the accuracy scores from the poor
Table 1. Mean accuracy scores (out of 170) and reading times (in seconds) in the
training phase of Experiments 1a and 1b (standard deviations in parenthesis).
Good readers Poor readers
Accuracy Reading time Accuracy Reading time
Experiment 1a (isolated word)
1 165.5 (4.16) .515 (058) 121.87 (26.34) .804 (.094)
2 168.45 (2.42) .494 (.053) 135.88 (25.7) .769 (.087)
3 169.04 (1.68) .477 (.053) 142.92 (24.86) .772 (.104)
4 169.14 (1.39) .488 (.056) 149.29 (18.64) .766 (.094)
5 168.86 (1.93) .487 (.056) 156.13 (13.92) .753 (.106)
6 169.36 (1.18) .496 (.066) 157.33 (13) .739 (.131)
Experiment 1b (context)
1 168.73 (1.6) 374.09 (31.39) 135.95 (19.86) 530.75 (86.49)
2 169.45 (.74) 340.14 (21.20) 151.63 (13.07) 460.08 (60.89)
3 169.82 (.66) 332.18 (19.46) 159.29 (10.21) 423.04 (50.98)
4 169.68 (.65) 332.04 (26.83) 163.66 (7.16) 402.62 (44.35)
5 169.86 (.35) 335.95 (59.54) 165.5 (5.32) 383.38 (40.11)
6 170 (0) 304.14 (41.88) 167.7 (3.39) 369.37 (27.77)
reading group alone. The poor readers, on average, learned over 35
words throughout training. An ANOVA revealed a signiﬁcant eﬀect of
repetition, F(5,115) ¼42.36, MSE ¼4365.328, p< .001, supporting the
claim that learning occurred during isolated word training. Taken
together, the results from the training phase of Experiment 1a show
that isolated word training results in faster reading times for good and
poor readers, and more accurate word recognition for poor readers.
Transfer phase. The reading times, accuracy scores and comprehension
scores for the Transfer phase of Experiment 1a are presented in
Table 2. Table 2 shows that the good readers were able to complete the
transfer stories faster than the poor readers, and the transfer passages
were read faster than the control passages regardless of skill group. A
2·2 mixed ANOVA of the reading time data, with group (good and
poor) as the between-subject factor and training (isolated word and
control) as the within-subject factor, conﬁrmed these observations with
signiﬁcant main eﬀects of group, F(1,44) ¼78.56, MSE ¼543585.25,
p< .001, and training, F(1,44) ¼35.21, MSE ¼14551.02, p< .001.
However, the Group ·Training interaction was also signiﬁcant,
F(1,44) ¼25.84, MSE ¼10678.65, p< .001, indicating that the poor
readers showed a greater decrease in reading time for the transfer story
over the control story compared with the good readers.
The accuracy data from Table 2 indicate that the good readers made
very few errors on either the transfer passage or the control passage.
Therefore, the poor readers were analyzed separately. A t-test
conducted on the data from the poor reading group alone, conﬁrmed
that the transfer passages were read more accurately than the control
passages, t(1,23) ¼6.77, p< .001. Table 2 shows that the comprehen-
sion scores for the transfer and control passages were similar for both
good and poor readers. A 2 ·2 mixed ANOVA of the comprehension
scores, with group (good and poor) as the between-subject factor and
training (isolated word and control) as the within-subject factor,
conﬁrmed that no comparisons approached signiﬁcance.
The results from the transfer phase of Experiment 1a show that
isolated word training contributes to faster reading times on novel
passages containing trained words for good and poor readers, and more
accurate reading of novel passages containing trained words for poor
readers. However, the gains in accuracy and reading speed did not
appear to inﬂuence the degree of reading comprehension of the transfer
story compared to the control story. This was true of both the skilled
and less skilled readers. Thus, word level training results in both
accuracy and speed aspects of subsequent reading ﬂuency.
356 SANDRA LYN MARTIN-CHANG AND BETTY ANN LEVY
Results and discussion: Experiment 1b
Training phase. The mean number of accurate responses and the mean
reading times, for both good and poor readers, are presented in
Table 1. The pattern of data suggest that both groups decreased their
reading times over training and good readers completed reading the
training stories faster than poor readers. A 2 ·6 mixed ANOVA with
group (good and poor) as the between-subject factor and repetition (1
through 6) as the within-subject factor conﬁrmed this observation with
signiﬁcant main eﬀects of repetition, F(5,220) ¼80.57, MSE ¼
74072.71, p< .001, and group, F(1,44) ¼67.30, MSE ¼580179.21,
p< .001. The poor readers were able to read the stories 162 seconds
faster after the six repetitions of training, while the good readers
decreased their reading speed over 70 seconds from the rereading para-
digm. The disproportional improvement in reading rate between poor
and good readers was conﬁrmed by the signiﬁcant Repetition ·Group
interaction, F(5,220) ¼20.29, MSE ¼18652.62, p< .001. The less
skilled readers showed greater advances in reading speed following
training compared to the skilled readers.
As indicated by the accuracy data from Table 1, the good readers
made very few errors, thus precluding them from any further analyses.
In contrast, the poor readers learned, on average, over 31 words
throughout training. A one-way ANOVA on the accuracy scores from
the poor reading group alone revealed a signiﬁcant main eﬀect of repe-
tition, F(5,115) ¼67.50, MSE ¼3397.18, p< .001. Consistent with
iso1ated word training in Experiment 1a, the results from the training
phase of Experiment 1b showed that training words in context results
in faster reading times for good readers and poor readers, as well as
more accurate word recognition for poor readers.
Transfer phase. The reading rate, accuracy scores, and comprehension
scores for the Transfer phase of Experiment 1b are presented in
Table 2. As illustrated in Table 2 good readers were able to complete
the passages faster than poor readers, and transfer passages containing
trained words were read faster than control passages, regardless of skill
group. A 2 ·2 mixed ANOVA, where the between-subject factor was
group (good and poor) and the within-subject factor was training
(context and control), conﬁrmed these observations with signiﬁcant
main eﬀects of training, F(l,44) ¼38.18, MSE ¼22445.44, p< .001
and of group, F(l,44) ¼70.23, MSE ¼534194.55, p< .001. The
Training ·Group interaction was also signiﬁcant, F(l,44) ¼19.25,
Table 2. Mean Accuracy Scores (out of 286) and Reading Times (in seconds) and Comprehension scores (out of 20) in the Transfer Phase of
Experiments 1a and 1b.
Good Readers Poor Readers
Accuracy Reading Time Comprehension Accuracy Reading Time Comprehension
Word Transfer 285.68 120.85 14.43 278.75 253.16 13.85
(Words Trained in Lists) (.48) (23.26) (3.44) (7.24) (67.97) (2.9)
Word Control 285 124.46 15.27 263.88 299.9 13.25
(Untrained Words) (1.57) (22.93) (2.52) (14.46) (91.83) (3.3)
Context Transfer 285.45 115.76 15.27 279.08 246.1 14.15
(Words Trained in Stories) (.8) (18.1) (2.59) (4.35) (67.89) (3.65)
Context Control 284.94 124.83 15.18 262.83 299.57 13.62
(Untrained Words) (1.76) (21.17) (3.47) (12.07) (101.77) (3.32)
358 SANDRA LYN MARTIN-CHANG AND BETTY ANN LEVY
MSE ¼11316.83, p< .001, indicating that the training gains were
greater for poor than for good readers.
The accuracy data from Table 2 show that the good readers, again,
made very few errors on both the transfer passage and the control
passage. Therefore, the data from the poor reading group was analyzed
alone. A t-test conﬁrmed that the transfer passages were read
more accurately than the control passages by the poor readers,
t(1,23) ¼8.24, p < .001. Once again, the comprehension scores for the
transfer and control passages were similar for both good and poor
readers. A 2 ·2 mixed ANOVA of the comprehension scores, with
group (good and poor) as the between-subject factor and training
(context and control) as the within-subject factor, conﬁrmed that no
comparisons approached signiﬁcance.
The results from the transfer phase of Experiment 1b show that
training words in context leads to faster reading times on new passages
containing trained words for good and poor readers. In addition, poor
readers read transfer passages more accurately than control passages.
However, as discussed in Experiment 1a, the ﬂuency gains resulting
from improvements in accuracy and speed did not translate to
comparable comprehension gains on the transfer passages.
Results and discussion: Cross training comparisons
The results of Experiments 1a and 1b showed that both isolated word
and context training improved the reading of trained words in new
story contexts. This ﬁnding is consistent with the corpus of data
reported by Levy and her colleagues (Bourassa et al., 1998; Faulkner &
Levy, 1994, 1999; Levy et al., 1997), which found that training at one
level of reading transfers to several other levels. Indeed, the current
experiments reported similar patterns of results, whereby skilled and
less skilled children read passages containing trained words faster than
passages containing new words, and poor readers made fewer errors
while reading passages with trained words. However it is not clear from
the above analyses if one method of training is superior to the other.
Therefore, to determine whether the degree of transfer that resulted
from context and isolated word training was equal, cross training
comparisons were conducted.
The mean number of accurate responses made by poor readers in each
training condition (good readers were at ceiling) are presented in
Table 1. During the ﬁrst presentation in each training condition, poor
readers were able to read more words in context than in isolation. This
observation was conﬁrmed with a t-test, t(23) ¼4.23, p< .001. It has
been well documented that poor readers are more successful when
decoding words in context rather than in isolation (Alexander, 1998;
Kim & Goetz, 1994; Nicholson, 1991). The current ﬁndings are consis-
tent with the notion that poor readers use contextual cues to compen-
sate for faltering phonological skills (Stanovich, 1980). The contextual
beneﬁt was maintained throughout training, and the children were able
to read approximately 10 more words correctly at the completion of
story training compared to isolated word training, t(23) ¼4.42,
p< .001. A 2 ·6 repeated measure ANOVA, where the factors were
training (isolated word and context) and repetitions (1 through 6),
revealed main eﬀects of training, F(1,23) ¼28.66, MSE ¼12906.89,
p< .001, and repetition, F(5,115) ¼70.64, MSE ¼7663.77, p< .001.
Taken alone, these ﬁndings suggest that isolated word training is a less
valuable method for teaching poor readers new words. However, it
should be noted that throughout the duration of training, the children
learned over 35 words in the isolated word condition compared to 31
words in the context condition—a diﬀerence that approached
signiﬁcance, F(5,115) ¼2.2, MSE ¼98.74, p¼.059. Therefore, the
claim could also be made that although poor readers are able to read
more words in context than in isolation, they acquire more words
through isolated word training compared to context training.
Transfer phase. The reading times, accuracy scores, and comprehension
scores for the Transfer Phases of Experiments 1a and 1b are presented
in Table 2. As Table 2 indicates, the control conditions from Experi-
ments 1a and 1b yielded comparable levels of baseline performance.
T-tests conﬁrmed this observation, revealing no signiﬁcant diﬀerences
in accuracy, t(45) ¼.618, p> .5, reading rate, t(45) ¼.00, p> 1 or
comprehension, t(45) ¼.302, p> .764 between the two control
Table 2 also shows that the good readers were able to complete the
transfer passages faster than the poor readers. Of more interest here,
the context transfer passages were read faster than the isolated word
transfer passages regardless of skill group (approximately 5 seconds
faster by the good readers and 7 seconds faster by the poor readers). A
2·2 mixed ANOVA, where the between-subject factor was group
(good and poor) and the within-subject factor was type of training
(isolated word and context), conﬁrmed these observations with
signiﬁcant main eﬀects of training type, F(l,44) ¼4.12, MSE ¼846.98,
360 SANDRA LYN MARTIN-CHANG AND BETTY ANN LEVY
p< .05 and of group, F(l,44) ¼78.65, MSE ¼395924.32, p< .001.
Thus, learning words in context leads to the faster reading of a novel
passage containing trained words, compared with learning words in
isolation. The Training Type ·Group interaction did not reach
signiﬁcance, F(l,44) ¼.108, MSE ¼22.28, p> .74, which indicates
that the training methods did not have diﬀerential transfer beneﬁts for
the good and poor readers.
The accuracy data from Table 2 show that while both methods of
training succeeded in teaching the children most of the unknown target
words, the good readers continued to read more accurately than the
poor readers. A 2 ·2 mixed ANOVA of the accuracy scores, with
group (good and poor) as the between-subject factor and training type
(isolated word and context) as the within-subject factor, conﬁrmed
signiﬁcant main eﬀects of group, F(1,44) ¼34.19, MSE ¼1015.65,
p< .001. No other comparisons approached signiﬁcance. Finally, a
2·2 mixed ANOVA of the comprehension scores, with group (good
and poor) as the between-subject factor and training type (isolated
word and context) as the within-subject factor, revealed that there were
no signiﬁcant eﬀects. Overall, then, there was a transfer beneﬁt from
learning words in context, compared to learning words in isolation, for
reading time, but not for accuracy or comprehension.
The results reported by Levy (2001) found no additional beneﬁts
attributed to contextual facilitation. However, the unequal number of
word presentations throughout training, favoring the isolated word
condition, left open the possibility that any beneﬁts due to context
training may have been overlooked. The present experiment used a
shared reading paradigm that substantially reduced the task demands
for the poor readers in the context condition. Moreover, this change
in procedure from Levy’s original (2001) experiments also enabled the
two training conditions to be equated. Indeed, the current ﬁndings
clarify the question raised by Levy (2001). The combined results from
Experiments 1a and 1b show that training words in context, as
compared to isolation, leads to the faster reading of those words
when they are later encountered in a new context. Importantly, the
increased reading speed gained from story training was just as
prominent for good readers as it was for poor readers. This ﬁnding
was unexpected based on the literature, which suggests that, under
normal circumstances, contextual eﬀects are more readily observed in
poor readers (e.g., Nicholson, 1991; Nicholson et al., 1991; Perfetti &
Roth, 1981). However, at present, the ﬁndings indicate that training
words in context may bring additional advantages in reading speed
for all children, over and above training in isolation.
As in Experiments 1a and 1b, Experiment 2 explored the merits of
training words in context and in isolation. However, the participants in
Experiment 2 were average readers in grade 2. The purpose of the
experiment was to replicate, with a younger population, the greater
improvement in reading time observed following context training
compared to training in isolation.
In Experiments 1a and 1b, an interesting pattern of acquisition was
noted, whereby poor readers were able to read more words in total dur-
ing context training (over 10, on average), and yet they acquired more
words during isolated word training (approximately 4, on average). To
further examine the learning curve of new words in Experiment 2,
average readers were presented with material that was 1 year above
grade level to ensure that the target lists would contain unfamiliar
Participants. Children in Experiment 2 were selected from the same
local school board as Experiments 1a and 1b (see above for a descrip-
tion of the school board). The objective of Experiment 2 was to study
younger average readers. Thus, while only 30% of the children screened
for Experiments 1a and 1b participated in the study, 82% of the
children screened were included in Experiment 2. In total, 28 children
obtained parental consent and were given the reading subtest of the
WRAT3. Of those, 25 fell within the normal reading range. The data
from three students were excluded from analyses due to extended
absenteeism (n¼2) and non-compliance (n¼1). This resulted in a ﬁnal
sample of 23 participants (17 males and 6 females) who were between 7
and 8 years old (ranging from 86 to 98 months, M¼91). They had a
mean standard score on the WRAT3 of 102.8 (SD ¼3.6, range=
89–113), which corresponded to a grade 2 reading level. Similar to
Experiments 1a and 1b, children selected for the study were asked to
read a list of 19 high frequency words that would be required in a later
phase of the experiment. Those achieving at least 85% accuracy on the
high frequency word list were included in the ﬁnal participant selection.
Once again, the participants performed very well at this task
(M¼90%) and no children were excluded from the study based on this
criterion. Information was not collected pertaining ‘learning disabled’
and/or ‘gifted’ designations assigned to individual participants. Instead,
the inclusion criteria were based entirely on experimenter collected data
362 SANDRA LYN MARTIN-CHANG AND BETTY ANN LEVY
(the WRAT3, as outlined above). This sample did not include any chil-
dren who were not ﬂuent in spoken English (e.g., ESL students) or any
children with signiﬁcant developmental delays (e.g., Downs Syndrome
Design and materials. A within-subject design was employed where each
participant received both isolated word training and story training. The
order of training was counter balanced across all participants. Three
sets of materials were used, one for each of the training conditions and
one for a control condition. Three of the material sets from
Experiments 1a and 1b were used in the current investigation.
The children were trained and tested individually in a quiet room of the
school. They received training for approximately 15–20 minutes a day
for three consecutive days (phase 1), followed by transfer testing on the
fourth day (phase 2). The four-day sessions (one for isolated word and
one for story) were conducted in successive weeks. The presentation of
the control passage was counterbalanced to ensure that it appeared
equally often after story and isolated word training, as well as after the
ﬁrst and second week of training.
The procedure for isolated word training was identical to Experiment
1a, with the exception that the children completed four complete list
repetitions each day over a three-day training period, resulting in 12
total repetitions of each target word. Likewise, the context training
condition was the same as Experiment 1b, with the exception that the
children completed two complete story readings each day, over a
three-day training period, resulting in 12 total repetitions of each target
word (two repetitions of each target word per story). The transfer phase
was similar to Experiments 1a and 1b, with the exception that each child
read only one control passage, and corrective feedback was not supplied
to the participants during test. At the completion of Experiment 2 each
child had read three transfer passages (isolated word transfer, context
transfer, and a control passage) as opposed to four in Experiments 1a
Results and discussion
Training phase. The mean reading times for story and isolated word
training are presented in Table 3. The mean reading times in both
training conditions decreased as training progressed. Two separate
one-way ANOVAs supported this observation with signiﬁcant eﬀects of
Repetition, F(5,22) ¼62.97, MSE ¼114184.44, p< .001, and
F(5,110) ¼7.76, MSE ¼.01, p< .001, for context and iso1ated word
training, respectively. Thus, consistent with the results of Experiment
1a and 1b, the average readers were getting faster with both types of
The mean accuracy scores achieved during isolated word and context
training are also presented in Table 3. On the ﬁrst trial of training, the
children were able to read more than 20 additional words when the
words were presented in context compared to in isolation, t(22) ¼30.13,
p< .001. This beneﬁt continued throughout training, as conﬁrmed by a
two-way within- subject ANOVA where the factors were training
(context and isolated word) and repetitions (1 through 6). A signiﬁcant
eﬀect of repetition, F(5,110) ¼43.33, MSE ¼6886.119, p< .001 and of
training, F(1,22) ¼28.38, MSE ¼18669.93, p< .001, were noted. This
indicates that although children acquired more words on each trial in
both conditions, children continued to read more words accurately in
context training compared to isolated word training.
Consistent with the pattern of results found in Experiment 1,
children were able to read more words in context, yet they acquired
more words in isolation. During isolated word training children
learned, on average, over 37 words in total while only 29 words were
learned in context. This diﬀerence was real as indicated by a signiﬁcant
Training ·Trial interaction, F(1,110) ¼3.32, MSE ¼103.36, p< .05.
Thus, two factors appear to inﬂuence word reading: (a) the contextual
beneﬁt observed on trial 1 that is carried through each repetition, and
(b) the better word acquisition observed in isolation compared with
Table 3. Mean accuracy scores (out of 170) and reading times (in seconds) in the
training phase of experiment 2 (standard deviations in parenthesis).
Repetition Isolated word condition Context condition
Accuracy Reading time Accuracy Reading time
1 118.09 (37.22) .847 (.172) 138.87 (27.06) 556.03 (141)
2 134.7 (29.14) .804 (.163) 151.43 (19.43) 477.82 (107)
3 140.35 (27.87) .802 (.162) 159.13 (12.32) 432.31 (76)
4 147.57 (23.91) .787 (.172) 163.22 (8.42) 402.87 (58)
5 151.7 (22.18) .784 (.182) 165.78 (5.75) 383.77 (47)
6 155.22 (17.07) .773 (.169) 167.87 (3.58) 366.51 (39)
364 SANDRA LYN MARTIN-CHANG AND BETTY ANN LEVY
Transfer phase. The reading times, accuracy scores, and comprehension
scores, for Experiment 2 are presented in Table 4. The measure of
reading time was the total number of seconds required for each child to
read the transfer stories from beginning to end. As the mean reading
times in Table 4 indicate, the transfer passages were read the most
quickly in the context condition, followed by the isolated word
condition, with the control condition resulting in the slowest reading
times. A one-way repeated measure ANOVA conﬁrmed a signiﬁcant
eﬀect of training type, F(2,44) ¼24.24, MSE ¼59802.63, p< .001. A
Bonferroni post hoc comparison showed that all possible pair wise
comparisons were signiﬁcant. This patterns suggests that while training
in context promotes greater reading speed on a new passage, training in
isolation is more beneﬁcial than receiving no training. These ﬁndings
replicate those found in the ﬁrst set of experiments with good and poor
readers in grade 4.
The accuracy data show a similar tendency to favor the context
condition. Table 4 illustrates that the highest accuracy scores were
obtained after context training, followed by isolated word training,
while the lowest scores occurred in the control condition. Once again, a
one-way repeated measure ANOVA conﬁrmed a signiﬁcant eﬀect of
training type, F(2,44) ¼21.92, MSE ¼1674.45, p< .001. A Bonferroni
post hoc comparison showed that all possible pair-wise comparisons
were signiﬁcantly diﬀerent. The comprehension scores presented in
Table 4 show that there were no diﬀerences among the experimental
and control conditions. A one-way repeated measure ANOVA
conﬁrmed that no comparisons approached signiﬁcance.
In summary, Experiment 2 replicated the same reading speed beneﬁts
after context training, for average readers, which was observed for good
and poor readers in Experiments 1a and 1b. The results of Experiment
2 also showed a corresponding beneﬁt in accuracy following training in
context over training in isolation. Finally, Experiment 2 replicated the
same pattern of accuracy acquisition that was noted in Experiments 1a
Table 4. Mean reading time (in seconds) and accuracy score (N= 286) in the isolated
word, context, and control conditions in the transfer phase of Experiment 2 (standard
deviations in parenthesis).
Accuracy Reading time Comprehension
Context 279.00 (8.50) 221.49 (87.09) 11.91 (4.33)
Word 276.26 (9.27) 242.96 (105.87) 11.91 (4.24)
Control 263.04 (20.75) 293.60 (132.37) 11.35 (4.84)
and 1b, whereby children could read more words during context
training, and yet learned more words during isolated word training.
Dahl (1979) compared the eﬃcacy of learning to read words in and out
of context. In doing so, she reported reading gains following context
training and concluded that reading practice on individual words
oﬀered no beneﬁts beyond those that would be expected for regular
classroom instruction. The results reported here do not support this
claim. Experiments 1a, 1b and 2 show that both isolated word and
context training result in faster, more accurate reading of transfer
passages compared to control passages. This is in keeping with the bulk
of empirical evidence, which suggests that both isolated word and
context training can improve several aspects of reading ﬂuency (e.g.
Bourassa et al., 1998; Dowhower, 1987; Fleisher et al., 1979; Levy,
2001; Levy et al., 1997).
The automatic information processing model (LaBerge & Samuels,
1974) suggested that the attentional resources needed for comprehension
could be made available by automatizing reading at the word level.
Indeed, Rashotte and Torgesen (1985) found that the beneﬁts observed
from repeated reading was the result of the number of individual words
shared among the training and transfer passages. If ﬂuent reading results
from the rapid identiﬁcation of words, then there is no reason to predict
diﬀerent reading outcomes on a passage made up of words trained in
context, vs. a passage made up of words trained in isolation. Yet our
data did not wholly support this view either. The cross-training
comparisons of Experiment 1 show that while training words in isolation
did transfer to the faster reading of a novel passage, the increase in
reading rate was more pronounced when children were trained in context.
This pattern was further substantiated in Experiment 2, where context
training was found to exceed isolated word training for both reading
speed and accuracy in a younger group of average readers. In
Experiment 2, average readers were presented with words above their
reading level providing more scope to observe accuracy gains. Therefore,
in both experiments reading words in context oﬀered beneﬁts, over and
above, those observed after practicing words in isolation, implicating
processes beyond automatization at the word level alone.
Importantly, this beneﬁt does not appear to be a result of simply
reading passages per se. Gains in reading speed were only noted in
those cases where the novel passages contained trained words. Word
366 SANDRA LYN MARTIN-CHANG AND BETTY ANN LEVY
overlap is necessary for the transfer of ﬂuency to occur. Therefore, the
current investigations are consistent with the view that speciﬁc
processing advantages at the word level are the key to the transfer of
ﬂuency gains (Rashotte & Torgesen, 1985). However, our data extend
this view by suggesting that the context in which the words are trained
can also aﬀect later reading performance.
As outlined above, Goodman’s (1965, 1967) claim that skilled
readers make better use of contextual cues than unskilled readers, has
not been backed by empirical evidence. In fact, the opposite pattern
has been reported; namely, the performance of less skilled readers has
been shown to improve from reading in context, whereas skilled readers
perform equally well reading words in context and in isolation
(Allington, 1978; Archer & Bryant, 2001; Nicholson, 1991; Nicholson
et al., 1991; Stanovich, 1980, 1994). According to the interactive-
compensatory model (Stanovich, 1980), this improvement in poor
readers results from the enlistment of higher order processes to
supplement deﬁcient word analysis skills. Good readers, on the other
hand, have fast identiﬁcation skills which limit the inﬂuence context
exerts on word recognition (Perfetti & Roth, 1981; Stanovich, 1980).
Consequently, the second result from the present studies that
merits attention is that good and poor readers showed equivalent
gains in reading speed after context training compared to isolated
word training. This suggests that the processes involved in on-line
word identiﬁcation (i.e., word naming during a single exposure) are
diﬀerent than those involved during training words across repeated
exposures. The pattern of performance from good and poor readers
in the current investigations indicate that when word representations
are being formed over training, higher-order processes are employed
in conjunction with word identiﬁcation processes by all readers. If
higher-order processes were only enlisted when word identiﬁcation
processes were overwhelmed, such as the case with on-line word
recognition, then poor readers would have once again shown a
greater degree of contextual facilitation. Rather, the current data
suggest that information from several levels of the processing
hierarchy contribute to forming word representations. This is not to
say that the contributions made by all levels are equal—on the
contrary, we would assume a position similar to the one forwarded
by Perfetti and Roth (1981) where the:
processes are at once interactive and asymmetrical (...). An
important consequence of this asymmetry is that so called bottom-
up processes, can carry on reasonably well without top–down
processes but not visa versa. No matter how helpful top down
processes are, they are neither deﬁnitive, nor essential (p. 270).
Yet, the important point to be made is that when higher order
processes are engaged, they have a part to play in forming word
representations in memory.
What is it, then, that accounts for the additional boost in reading
speed (and accuracy) following context training? One possible explana-
tion is that building up sight recognition of words in context might
engage more semantic processes than reading words in isolation.
Accordingly, words learned in context might have ‘deeper’ or more
meaningful representations in memory. This position is reminiscent of
the levels of processing account of memory (Craik &Lockhart, 1972).
It would follow that words with more sophisticated representations
would be accessed with greater speed and precision when they are
encountered during subsequent reading tasks (Perfetti, 1992).
An alternative explanation of the increased ﬂuency beneﬁts observed
after context training vs. isolated word training, highlights the use of
similar processes during phase 1 and phase 2 of the current paradigm.
In the present experiments, the transfer tasks always involved reading
in context at phase 2. A byproduct of this procedure was that there
was a high degree of congruity between the training and testing phases
in the context condition, compared with the list condition. Following
the logic of transfer appropriate processing (TAP), the greater contex-
tual facilitation would result from the fact that ‘‘memory performance
is assumed to beneﬁt, to the extent that processes used during study are
captured by the task used at test’’ (Rajaram, Srinivas & Roediger,
1998: 1993). If higher order processes contribute to forming word
representations in phase 1, those representations might be accessed
more easily if higher order processing is reinstated at phase 2.
Investigations are currently underway in order to delineate between
these two alternate hypotheses.
Comprehension was the one aspect of ﬂuency that failed to show
training induced improvements. A cursory review of the transfer litera-
ture reveals several discrepant ﬁndings with respect to comprehension.
For instance, comprehension gains have been reported following
isolated word training (Levy et al., 1997) and context training
(Bourassa et al., 1998; Dahl, 1979; Dowhower, 1987). In contrast, other
authors have failed to move comprehension after isolated word training
(Dahl, 1979; Fleisher et al., 1979). Nevertheless, the absence of a
comprehension gain was surprising in light of the fact that Breznitz and
her colleagues (Breznitz, 1997, 2001; Breznitz & Berman, 2003; Breznitz
368 SANDRA LYN MARTIN-CHANG AND BETTY ANN LEVY
& Share, 1992) have provided a wealth of evidence showing that
accelerating individualized reading rates improves comprehension. For
example, when Breznitz (1997) presented texts to learning disabled
children at rates 25% faster than usual, she observed increases in both
accuracy and comprehension. In the current set of investigations, both
the poor readers in grade 4 and the average readers in grade 2 showed
markedly increased reading rates in the trained transfer passages
compared to control (mean increases of 15.6% and 17.2%, respectively,
after isolated word training and 17.8% and 24.6%, respectively, after
context training), and yet there was no hint of improved comprehen-
sion. Perhaps the gains in reading accuracy were not large enough to
induce comprehension improvements. Indeed, Markell and Deno (1997)
reported that an improvement in accuracy equivalent to roughly 20
words had to be observed in order to ﬁnd comprehension beneﬁts.
In summary, the current ﬁndings appear to suggest that more
eﬃcient text processing (observed via improved reading speed and
accuracy) does not necessarily result in corresponding improvements in
text interpretation. However, these data must be interpreted with
caution because the expected diﬀerences in comprehension between
good and poor readers in Experiments 1a and 1b were also absent.
Therefore, the comprehension measures used here may not have been
sensitive enough to reﬂect subtle gains in understanding.
With regards to the training data, the results of Experiments 1a, 1b
and 2 showed that poor readers in grade 4 and average readers in grade
2 were more successful in reading words on the ﬁrst presentation of
context training compared with isolated word training. This replicated
the well-documented contextual facilitation eﬀect shown for poor and
young readers (e.g., Alexander, 1998; Archer & Bryant, 2001; Kim &
Goetz, 1994; Nation & Snowling, 1998; Nicholson, 1991; Nicholson,
et al., 1991; Stanovich 1980, 1994). In addition, our experiments
substantiated previous ﬁndings showing that the advantages incurred
by context were observed over the entire course of training. However,
interpretation of these results must be tempered by the ﬁnding that the
actual number of new words learned over the course of training was
reliably larger for isolated word, compared to context, training. This
ﬁnding is in keeping with the notion that young readers are more
successful in learning words when those words are stripped from the
contextual support provided by predictable text (Johnston, 2000).
Further investigation is currently underway in order to fully
understand: (a) the diﬀerences between intercept and slope eﬀects, and
(b) whether the accuracy gains observed over training are permanent or
transitory in nature.
In summary, context training resulted in faster reading of transfer
passages for all readers, as well as the more accurate reading of
transfer passages for younger readers, compared with isolated word
training. Employing a shared reading paradigm resulted in the
beneﬁts associated with contextual reading while avoiding many of
the cravats normally associated with reading passages. For example,
the frustration level of both the young and poor readers was much
reduced by having the experimenter read the majority of the training
passage. This procedure also equated the contextual cues available to
good and poor readers, giving the struggling readers access to
ﬂuently rendered text that would otherwise escape them. Thus, if the
goal of instruction is to develop meaning based reading uncluttered
by problems in word recognition, the present procedure oﬀers a way
to promote enjoyable reading while children develop word recognition
This paper is part of the ﬁrst author’s doctoral thesis. We thank
Carolyn Breukelman, Dr. Nicole Conrad, and Joann Lim for their
valuable input into this project, as well as Johanna Weststar and
Rupa Parakh for their eﬀorts in data collection. We also oﬀer our
sincerest thanks to the principals, teachers, and children for their
continued support and participation. This research was supported by
a graduate fellowship to the ﬁrst author and an operating grant to
the second author, both from the Social Sciences and Humanities
Research Council of Canada.
370 SANDRA LYN MARTIN-CHANG AND BETTY ANN LEVY
The Halloween Story
It was Halloween and the kids in grade four were creating a haunted
house. Their room was made into a small forest, with paper trees every-
where. There was even an old owl’s nest halfway up one tree. The whole
class went to the park and took some leaves and dead sticks. When they
returned they put them underfoot, trying to make the woods look more
real. One child had brought a teddy bear; it was given black wings to
look like a crow. Miss Mayer, the teacher, decided that because it was
such a good haunted house, when everything was ready the children
could invite the older boys and girls to check it out. The grade fours
‘‘I think we’re just about ready ,’’ the teacher told the children while
she helped some kids climb into costume boots, ‘‘but we better hurry.
We all need to help, we mustn’t waste time!’’
When it was time, the grade ﬁve children ﬁled in the back of the
room. A few were giggling and looked like they were going to crack up.
They whispered to each other quietly.Asmall sign was at the beginning
of the forest, it read ‘‘WARNING–KEEP OUT!’’ The older kids just
grinned at the sign. Not much would scare these kids!
‘‘We mustn’t make a noise, be sure to tiptoe as quietly as mice, ’’
Miss Mayer told the children, ‘‘I hope you took notice how softly the
older kids tiptoe.’’
The only sound the kids heard was a branch in the breeze tapping
on the window. The children’s feet crunched on the dead sticks from
the park that were underfoot.
‘‘Look!’’ whispered one student looking at the owl’s nest, ‘‘I think I
Without warning the sound of the branch in the breeze was replaced
by a loud CRACK. The children gasped as the teddy bear was set free.
He fell, diving down towards the heads of the older students from a
place high above. The crow’s foot hit the edge of a desk and ﬂuttered
against a grade ﬁve. He did not realize that it was the bear and he
yelled. The next moment, the grade fours returned from the place where
they had been sitting and ran by ﬂapping long costume wings that
ﬂuttered up and down.
The older children had stopped giggling now. Everyone was getting
a little nervous. The older kids made the climb one foot at a time. They
looked around amoment later as an awful sound broke the silence. It
sounded like cannon shots or a thousand ﬁrecrackers going oﬀat once.
As it sounded a light ﬂashed. It was as bright as a camera ! Some of the
372 SANDRA LYN MARTIN-CHANG AND BETTY ANN LEVY
children yelled and called out as an awful crow’s screeching was heard.
It made such a racket! Then the screeching stopped and the children
again moved forward, trying to keep going, not turn back. The older
kids didn’t realize but some of the younger kids had been sitting under
some desks. These kids now slid as softly as mice around the edge of
the desks without being seen.Next, the silent children caught up to the
older boys and girls and jumped out at them. The bigger kids had seen
enough. They were getting really scared and couldn’t take anymore. It
was time to escape the cannon shots,camera,ﬁrecrackers, and the
ﬂapping and diving of the bird. The kids fell over each other to try and
‘‘HELP!!’’ some of the students called. The kids all tried to hurry
towards the door. Everyone stopped. Something big caught their eye...
feet covered in boots... abelt halfway up... a hat high above covering a
bald head?!? Was it a ghost or a ghoul? The lights went up. It was only
the school janitor! Everyone laughed. The old bald janitor tipped the
cap on his head with one thumb looped under his belt. He walked away
as he said, ‘‘Take care, children.’’
‘‘We sure gave you a scare!’’ the grade fours said and broke into
laughter,’’ It was a good haunted house wasn’t it?’’
The kids all grinned and decided it was a good haunted house.
Everyone was glad they had come to check it out because they had
enjoyed them selves very much.
The Owl’s Nest
One day Jim and Teddy told the teacher, Mr. Mayer, they had seen
an owl sitting high on a nest. The teacher decided to take his camera
and go check it out. The next day, the boys took him to the place they
had seen the nest. At the edge of the park, Mr. Mayer took out his
camera and got it all ready. When he was ready he whispered, ‘‘Tiptoe
as softly as mice, we mustn’t scare the owl!’’
‘‘You cannot tiptoe quietly because of these dead sticks underfoot,’’
whispered Jim to Teddy.
‘‘They crack like ﬁrecrackers!’’ Teddy grinned. Under Mr. Mayer’s
boots they sounded like cannon shots.
At the tree there was no sign of the owl. Mr. Mayer told the boys,
‘‘I think that it is an old crow’s nest.’’
But the teacher decided to climb up the tree to make sure. He had
on a big hat to make sure his bald head was not getting too much sun.
Jim and Teddy couldn’t help giggling while it ﬂuttered in the breeze.
Without warning, an owl came diving down towards Mr. Mayer when
he was only halfway up the tree. Mr. Mayer yelled. He was in such a
hurry to climb down, he did not realize a branch he put his foot on
was too small. It broke and he fell down ﬁve feet when his belt got
caught on a small branch. While he was trying to get away the owl
returned. There was an awful ﬂapping and screeching. Then the owl
took oﬀwith Mr. Mayer’s hat.
‘‘Was it an owl’s nest?’’ called Teddy.
Mr. Mayer looked down for a moment, ‘‘Yes, Teddy, I think it
The Owls Nest
1. What did Mr. Mayer decide to take with him?
2. What did Mr. Mayer say they shouldn’t scare?
3. Why couldn’t Jim and Teddy tiptoe softly?
4. What did the dead sticks sound like under Mr. Mayer’s boots?
5. Who decided to climb the tree?
6. What made Jim and Teddy start to giggle?
7. How did the owl react to Mr. Mayer?
8. What item of Mr. Mayer’s got caught on a small branch?
9. What did the owl take oﬀwith?
10. What did Teddy ask if Mr. Mayer saw?
Alexander, J.C. (1998). Reading skill and context facilitation: A classic study
revisited. The Journal of Educational Research, 9, 314–318.
Allington, R.L. (1978). Eﬀects of contextual constraints upon rate and accuracy.
Perceptual and Motor Skills, 48, 1318.
Archer, N. & Bryant, P. (2001). Investigating the role of content in learning to read:
A direct test of Goodman’s model. British Journal of Psychology, 92, 579–591.
Bourassa, D., Levy, B.A., Dowin, S. & Casey A. (1998). Transfer eﬀects across
contextual and linguistic boundaries: Evidence from poor readers. Journal of
Experimental Child Psychology, 71, 45–61.
Breznitz, Z. (1997). Eﬀects of accelerated reading rate on memory for text among
dyslexic readers. Journal of Educational Psychology, 89, 289–297.
Breznitz, Z. (2001). The determinants of reading ﬂuency: A comparison of dyslexic
and average readers. In M. Wolf (Ed.), Dyslexia, ﬂuency, and the brain (pp. 245–
276). Timonium, MD: York Press.
Breznitz, Z. & Berman, L. (2003). The underlying factors of word reading rate.
Educational Psychology Review, 15, 247–265.
374 SANDRA LYN MARTIN-CHANG AND BETTY ANN LEVY
Breznitz, Z. & Share, D. (1992). Eﬀects of accelerated reading rate on memory for
text. Journal of Educational Psychology, 84, 193–199.
Craik, F.I.M. & Lockhart, R.S. (1972). Levels of processing: A framework for
memory research. Journal of Verbal Learning and Verbal Behavior, 11, 671–684.
Dahl, P. R. (1979). An experimental program for teaching high speed word
recognition and comprehension skills. In J.E. Button, T.C. Lovitt, & T.D.
Rowland (Eds.) Communications research in learning disabilities and mental
retardation (pp.33–65). Baltimore, MD: University Park Press.
Dowhower, S.L. (1987). Eﬀects of repeated reading on second-grade transitional
reader’s ﬂuency and comprehension. Reading Research Quarterly, 22, 389–406.
Faulkner, H. J. & Levy, B.A. (1994). How text diﬃculty and reader skill interact to
produce diﬀerential reliance on word and content overlap in reading transfer.
Journal of Experimental Child Psychology, 58, 1–24.
Faulkner, H.J. & Levy, B.A. (1999). Fluent and nonﬂuent forms of transfer in
reading: Words and their message. Psychonomic Bulletin and Review, 6, 111–
Fleisher, L. S., Jenkins, J.R. & Pany, D. (1979). Eﬀects on poor readers’ comprehension
of training in rapid decoding. Reading Research Quarterly, 15, 30–48.
Goodman, K.S. (1965). A linguistic study of cues and miscues in reading. Elementary
English, 42 , 639– 643.
Goodman, K.S. (1967). Reading: A psychological guessing game. Journal of the
Reading Specialist, 6, 126–135.
Johnston, F.R. (2000). Word learning in predictable text. Journal of Educational
Psychology, 92 , 248–255.
Kim, Y.H. & Goetz, E.T. (1994). Context eﬀects on word recognition and reading
comprehension of poor and good readers: A test of the interaction-compensatory
hypothesis. Reading Research Quarterly, 29, 178–188.
Kuhn, M.R. & Stahl, S.A. (2003). Fluency: A review of developmental and remedial
practices. Journal of Educational Psychology, 95, 3–21.
LaBerge, D. & Samuels, S.J. (1974). Toward a theory of automatic information
processing in reading. Cognitive Psychology, 6, 293–323.
Levy, B.A. (2001). Moving the bottom. In M. Wolf (Ed.), Dyslexia, ﬂuency, and the
brain (pp. 357–379). Timonium, MD: York Press.
Levy, B.A., Abello, B. & Lysynchuk, L. (1997). Transfer from word training to
reading in context: Gains in reading ﬂuency and comprehension. Learning
Disability Quarterly, 20 , 173– 188.
Markell, M.A. & Deno, S.L. (1997). Eﬀects of increasing oral reading: Generalization
across reading tasks. The Journal of Special Education, 31, 233–250.
Nation, K. & Snowling, M.J. (1998). Individual diﬀerences in contextual facilitation:
Evidence from dyslexia and poor reading comprehension. Child Development, 69,
Nicholson, T. (1991). Do children read words better in context or lists? A classic
study revisited. Journal of Educational Psychology, 83, 444–450.
Nicholson, T., Bailey, J. & McArthur, J. (1991). Context cues in reading: The gap
between research and popular opinion. Reading, Writing, and Learning
Disabilities, 7 , 33– 41.
Oaken, R., Wiener, M. & Cromer, W. (1971). Identiﬁcation, organization, and
reading comprehension for good and poor readers. Journal of Educational
Psychology, 62 , 71– 78.
Pearson, D.P. (1978). On bridging gaps and spinning chasms. Curriculum Inquiry, 8,
Perfetti, C.A. (1992). The representation problem in reading acquisition. In P. Gough,
L. Ehri & R. Trieman (Eds.), Reading acquisition (pp. 147–174). Hillsdale, NJ:
lawrence Erlbaum Associates.
Perfetti, C.A. & Roth, S.F. (1981). Some of the interactive processes in reading and
their role in reading skill. In A.M. Lesgold & C.A. Perfetti (Eds.), Interactive
processing in reading (pp. 269–297). Hillsdale, NJ: lawrence Erlbaum Associates.
Rajaram, S., Srinivas, K. & Roediger, H.L. (1998). A transfer-appropriate processing
account of context eﬀects in word-fragment completion, Journal of Experimental
Psychology. Learning, Memory, and Cognition, 24, 993–1004.
Rashotte, C.A. & Torgesen, J.K. (1985). Repeated reading and reading ﬂuency in
learning disabled children. Reading Research Quarterly, 20, 180–202.
Share, D. (1995). Phonological recoding and self-teaching: sine qua non of reading
acquisition. Cognition, 55, 151–218.
Spector, J.E., (1995). Phonemic awareness training: Application of principles of
direct instruction. Reading and Writing Quarterly: Overcoming Learning
Diﬃculties, 11, 37– 51.
Stanovich, K. E. (1980). Toward an interactive-compensatory model of individual
diﬀerences in the development of reading ﬂuency. Reading Research Quarterly, 16,
Stanovich, K. E. (1986). Matthew eﬀects in reading: Some consequences of indivi-
dual diﬀerences in the acquisition of literacy. Reading Research Quarterly, 21,
Stanovich, K. E. (1994). Romance and reality. The Reading Teacher, 47, 280–291.
Tunmer, W.E. & Chapman, J.W. (1995). Context use in early reading development:
Premature exclusion of a source of individual diﬀerences? Issues in Education,
Wilkinson, G.S. (1993). The wide range achievement test. Wilmington, DE: Wide
Address for correspondence: Sandra Martin-Chang, Department of Psychology, Mount
Allison University, Crabtree Building, 49A York Street, Sackville NB, E4L 1G7,
Canada; Phone: +1-506-364-2455; Fax: +1-506-364-2467; E-mail: firstname.lastname@example.org
376 SANDRA LYN MARTIN-CHANG AND BETTY ANN LEVY