Full Terms & Conditions of access and use can be found at
Scientific Studies of Reading
ISSN: 1088-8438 (Print) 1532-799X (Online) Journal homepage: http://www.tandfonline.com/loi/hssr20
Learning From Our Mistakes: Improvements in
Spelling Lead to Gains in Reading Speed
Gene Ouellette, Sandra Martin-Chang & Maya Rossi
To cite this article: Gene Ouellette, Sandra Martin-Chang & Maya Rossi (2017) Learning From
Our Mistakes: Improvements in Spelling Lead to Gains in Reading Speed, Scientific Studies of
Reading, 21:4, 350-357, DOI: 10.1080/10888438.2017.1306064
To link to this article: https://doi.org/10.1080/10888438.2017.1306064
Published online: 27 Apr 2017.
Submit your article to this journal
Article views: 488
View Crossmark data
Citing articles: 7 View citing articles
Learning From Our Mistakes: Improvements in Spelling Lead to
Gains in Reading Speed
, Sandra Martin-Chang
, and Maya Rossi
Mount Allison University;
The present study tested the hypothesis that underlying orthographic
representations vary in completeness within the individual, which is man-
ifested in both spelling accuracy and reading speed. Undergraduate stu-
dents were trained to improve their spelling of difficult words. Word
reading speed was then measured for these same words, allowing for a
direct evaluation of whether improvements in spelling would bring about
faster reading speeds. Results were clear: Spelling accuracy and reading
speed were strongly related across and within participants. Most important,
words that improved in spelling accuracy were read more rapidly at post-
test than words that did not show improvement in spelling. These results
provide direct evidence showing that the quality of orthographic represen-
tations, as indexed by spelling accuracy, directly relates to reading speed.
This is consistent with the lexical quality hypothesis and highlights the
relevance of spelling in literacy acquisition.
Sadly, proficient reading is not always associated with equally proficient spelling (Martin-Chang,
Ouellette, & Madden, 2014). This fits anecdotal experience; many strong readers struggle with
spelling, even into adulthood (Holmes & Castles, 2001; Moll & Landerl, 2009). The dissociation
between reading and spelling has been explained with reference to how words are stored in memory;
when orthographic representations are incomplete or unstable, their degree of specification may
suffice for reading (where the visual word is present) but not spelling (where the word must be
generated; Conrad, 2008). An intriguing extension of this argument is that even within good spellers,
some words will have orthographic representations that are fully and accurately stored in memory
(high quality), whereas other words will have representations that are incomplete or inaccurate
(poor quality; Perfetti, 2007). Viewed through the lens of lexical access, incomplete representations
should take longer to activate because of the mismatch between the internal representation and the
visual stimuli (Martin-Chang et al., 2014). In the present study we conducted a novel, stringent test
of this assertion by employing an experimental design in which we taught university students to spell
words they found difficult and subsequently measured their reading times for those same words.
Developmental theories (e.g., Ehri, 2005; Frith, 1985), cognitive explanations (e.g., Perfetti &
Hart, 2002), and connectionist models (e.g., Plaut, McClelland, Seidenberg, & Patterson, 1996) alike
have attributed the dissociation between reading and spelling to variations in stored orthographic
representations. For example, as illustrated by Conrad (2008) and Martin-Chang et al. (2014), the
word occasion may still be read when encountered in print, even if some information is missing from
the stored orthographic representation (e.g., oc?as?on). In contrast, this lack of orthographic
completeness will render accurate spelling difficult. Similarly, a word that is stored completely but
inaccurately (e.g., ocassion) would also prevent correct spelling but allow for a partial match to the
CONTACT Gene Ouellette email@example.com Department of Psychology, Mount Allison University, 49A York Street,
Sackville, New Brunswick, E4L 1C7, Canada.
© 2017 Society for the Scientific Study of Reading
SCIENTIFIC STUDIES OF READING
2017, VOL. 21, NO. 4, 350–357
word when encountered in print. Thus, according to such theory, having complete representations is
not mandatory for reading, yet it should still prove advantageous. Indeed, Perfetti and colleagues
proposed that words with higher quality representations would both be less dependent upon top-
down cues (Perfetti, 2007; Perfetti, Liu, & Tan, 2005) and read faster than words with lower lexical
Holmes and Carruthers (1998) were among the first to examine whether words with higher
quality representations (as reflected by spelling accuracy) were activated more rapidly than words
with lower quality representations. They asked adult readers to complete two tasks (silent reading
and spelling to dictation) using the same words. After individualizing the data, Holmes and
Carruthers reported that participants read words equally quickly regardless of whether they could
spell those words accurately. In contrast, Burt and Tate (2002) employed a similar paradigm but used
a lexical decision task instead of a silent reading task. When their data were individualized,
participants could recognize words that they could spell accurately more quickly than words they
had difficulty spelling.
Moll and Landerl (2009) recently studied children with discrepant reading and spelling abilities.
They reported that good readers/poor spellers read words they could spell faster than poor readers/
good spellers. Perhaps even more surprising, they also found that the words that good readers/poor
spellers could not spell accurately were read as quickly as words that good readers/good spellers could
spell accurately. Although these data imply that it is reading ability rather than spelling ability that is
driving reading speed, there were no direct comparisons between how quickly the same children
could read words they could or could not spell.
Martin-Chang et al. (2014) addressed this question by directly measuring the reading speed of
correctly and incorrectly spelled words within the same participants. They noted that words spelled
accurately over multiple trials were read faster than words that were spelled incorrectly over the
same number of attempts. In short, each participant read the words he or she could spell faster than
those he or she could not. These results, together with those of Burt and Tate (2002), provide
correlational support for two hypotheses: first, that orthographic representations vary in terms of
accuracy within each individual person, and second, that higher quality representations allow for
more rapid recognition/retrieval of words (Perfetti, 2007; Share, 2008). However, the data just
reported are limited by their correlational nature. Here, we directly test the hypothesis that
improvements in orthographic quality cause reductions in reading times by employing a tightly
controlled experimental design.
In the present study we trained undergraduate students to accurately spell words that were
initially difficult for them and then assessed the impact of improved spelling accuracy on the reading
speed for these same words. Difficult words were determined for each participant based on their
performance on an initial spelling assessment. These words were then assigned to one of two
conditions: a training condition designed to improve spelling accuracy, and a control condition
designed to equate the number of auditory and visual word exposures over training. The control
condition was included to alleviate concerns over priming/practice effects given that the words in the
training condition were heard and read multiple times. This design allowed us to directly compare
reading access speed for words that participants learned how to spell relative to words that they were
still unable to spell after training.
Fifty-six students (41 female), with a mean chronological age of 18.68 years (SD =1
participated in the first session; of these, 45 (34 female) completed the second session. The mean
age of this group was 18.64 years (SD = 1.21 years). All students were native English speakers
(five reported being bilingual with multilanguage exposure from birth), recruited from a
SCIENTIFIC STUDIES OF READING 351
Canadian undergraduate university. They received course credit in an introductory psychology
course for participating.
The Test of Word Reading Efficiency (TOWRE; Wagner, Torgesen, & Rashotte, 1999)determined
participants’general word reading and phonemic decoding ability. The TOWRE has excellent
reported reliability, above .90. Participants were asked to read as many words as possible out loud
(from a list of 104 real words) in 45 s. Subsequently, they were asked to read as many nonwords as
possible (from a list of 63 nonwords) in 45 s.
The Spelling subtest of the Woodcock–Johnson Test of Achievement–Third Edition (WJ-III;
Woodcock, McGrew, & Mathers, 2001) determined participants’general spelling abilities. The WJ-
III has high split-half reliability (Mdn r = .90; Schrank, McGrew, & Woodcock, 2001). Participants
were asked to write the words dictated by the researcher in an untimed test. Participants continued
until they had made six errors, or until the task was completed. The test contains 59 items.
Four additional items were added to 20 words taken from Martin-Chang et al. (2014). The resulting
24 words (presented in the appendix) were chosen to control for frequency, length, and number of
syllables. The items varied with regards to how difficult they were to spell.
Participants completed six tasks in total, three of which contained the target words. Reading speed of
the target words was first measured at the beginning of the session, followed by a visual maze task
(filler) and two standardized screening measures (TOWRE, WJ-III). The spelling task and the
definitions task both included the target words and occurred at the end of the session.
Reading response time was measured with SuperLab Pro 5.0 (Cedrus Corporation, 2014). The
words were presented individually, in random order, on a standard 24-in. computer screen. A
brief fixation point preceded each word. The reader’s voice triggered a voice key that prompted
the word to disappear from view. The experimenter manually scored accuracy of the reading
attempt on the keyboard, which triggered the presentation of the next fixation point. Each word
was read three times (once during each randomized 24-item list). The recorded reading speed
was averaged across all accurate readings of the word. Following the recommendations of Ratcliff
(1993), the response times were not transformed. Data were trimmed by setting the maximum
valid speed at 2,500 ms and discarding data from any trial in which the reading was not accurate
or in which the voice key was incorrectly triggered. In all, these procedures involved less than 2%
of the computed data points.
Immediately after the filler task and the two standardized screening measures, the participants
completed a spelling-to-dictation test. The experimenter read the target words in a fixed random
order. The participant wrote them in a list format using pen and paper. Each word appeared only
once per list. After the full list of target words had been written down, the participant received a new
sheet of paper and the list of words was dictated again in a different random order. This was done
three times. The score per word was computed as the number of correct spellings out of three
Finally, to confirm that participants had working knowledge of the target words, they wrote a
brief definition to dictation (without including the target word). The definitions were scored on a
maximum of 3 points. A complete definition with examples was awarded 3 points. Two points were
awarded if a key part of the definition was provided. One point was awarded if the word was
352 G. OUELLETTE ET AL.
appropriately used in context. Two assessors scored the definitions; in cases where there was a
discrepancy, consensus was reached via discussion.
One week later, participants returned for the training and posttest component of the research design.
After the training practice (see next), the participants completed a visual filler-task (maze comple-
tion) and then repeated the timed reading task from Session 1; a second visual filler-task was then
completed, followed by the spelling dictation test of the target words.
This was a within-subjects design; each participant took part in both the spelling condition and
the control condition (devised to control for priming/practice effects). Half of each participant’s
misspelled words from Session 1 were randomly assigned to the spelling condition; the remaining
misspelled words were assigned to the control condition. To limit the variability in the number of
words practiced across participants, and to facilitate the opportunities for learning, a maximum of 10
words were assigned for each participant (i.e., five per condition). The order was counterbalanced so
that half of the participants received the spelling condition before the control condition.
During the spelling training, participants were shown the correct spelling of the target word on an
index card while hearing the experimenter read the word aloud. The card was removed from view,
and the participant was given another card and pencil and asked to spell the word. The original
index card was then shown again and the two spellings compared. This process was repeated two
more times (for a total of three spelling attempts with feedback regardless of spelling accuracy).
No spelling took place in the control condition; the experimenter read the words within an
elaborate definition. The target word was present four times within each definition. Each time the
word occurred, the researcher would show the index card of that word (for equal visual and auditory
exposure to words as in the spelling condition, to control for any priming effects of repeated stimuli).
Descriptives and intercorrelations from the first session are presented in Table 1. Standardized tests
indicated that participants fell within the expected range of proficient reading and spelling, and the
scored definitions indicated working knowledge for the experimental word set (i.e., average score
above 1). The positive correlations between the standardized reading and spelling tests confirm the
predicted association between reading and spelling skills. Although the standardized spelling test was
not significantly correlated with our reading speed measure, the experimental spelling test
(which contained the same words as the reading speed task) was moderately correlated with the
reading task. This both supports the importance of considering word-specific measures and indicates
that higher spelling accuracy was associated with faster reading times within participants.
Table 1. Descriptives and correlation coefficients, Session 1.
Variable 1 2 3 4 5 6
1. TOWRE words —
2. TOWRE nonwords .59** —
3. WJ-III spelling .26* .50** —
.32** .38** .22 —
5. Reading time
−.19 −.38** −.19 −.33** —
6. Spelling accuracy
.23* .44** .75** .26* −.37** —
M95.11 57.80 51.50 1.83 831.46 1.66
SD 7.74 5.64 3.54 .39 242.71 .45
Note. N = 56. TOWRE = Test of Word Reading Efficiency; WJ-III = Woodcock–Johnson Test of Achievement–Third Edition.
Mean definition score per word.
Reading time averaged across all words.
Number of correct spelling attempts out of 3, averaged
across all words.
*p< .05. **p< .01.
SCIENTIFIC STUDIES OF READING 353
To follow up on the positive correlation between spelling accuracy and reading speed within
participants, we performed a within-item analysis to control for lexical properties of each word
(e.g., frequency of use, regularity). This entailed reorganizing the data so that separate reading
means were calculated for each word when it was spelled correctly (i.e., three of three) versus when
it was spelled incorrectly (i.e., none of three). In line with Martin-Chang et al. (2014), we found
that when a word was spelled consistently correctly, it was read on average, 20.2% faster compared
to when it was spelled consistently incorrectly. This represented a significant difference with a
moderate to large effect size, F(1, 23) = 7.90, p<.01,η
Session 2: Training data
To directly test the hypothesis that the quality oforthographic representations manifests in both
spelling accuracy and reading access speed, we turn to the results from the second session. In all,
18 of the words were included in the training set because they were consistently misspelled
(zero out of three accuracy) in Session 1 (see appendix). Of these 18 words, participants mis-
spelled, on average, approximately eight words each (M=8.07errors;range=4–12). The number
of words assigned to each condition, for each participant, was dependent upon the number of
spelling errors made by that participant in Session 1 (see the Methods section). Most participants
(n= 40) had four to five words assigned to each practice condition, only two participants had just
two words assigned to each condition, and 14 participants had three words assigned to each
condition. Overall, the spelling training was found to be very effective. When comparing the
number of correct spellings (out of three attempts) at posttest, spellings for words assigned to the
experimental condition were significantly higher (M= 2.19/3) than those assigned to the control
condition (M= 1.32/3), F(1, 44) = 25.03, p<.001,η
As evident in the means just reported, not all participants learned how to spell all of the words
they practiced. In addition, some incidental learning took place in the control condition; spelling
improvement was evident for 78% of the words assigned to the spelling training, whereas 33.6% of
the words in the control condition also improved. To test the hypothesis under study, it is necessary
to consider the effect on reading speed, not of assigned condition but of improved spelling
(regardless of assignment). Further, it is important to do so while also considering possible
participant and word-level effects. To accomplish this, linear mixed-effects modeling was employed.
For the linear mixed-effects model, words were first coded separately for each participant as to
whether they improved in spelling; this was dummy coded for each word with 0 indicating no
improvement and 1 indicating improvement in spelling between Time 1 and Time 2. In the analysis,
we entered this improvement variable as a fixed-effect predictor and reading speed change across
sessions as the dependent variable (to account for any time and general practice effects). We then
added word and participant as random effects with random intercepts and slopes. The intercept of
the completed model corresponds to the mean reading speed change of the no-improvement
condition, which was –147.83 ms (SE = 20.73), indicating faster reading across time, not associated
with spelling improvement. There was a significant further improvement in reading speed, repre-
sented by the estimated fixed effect of –55.93 ms (SE = 17.42) for the improved-spelling condition,
F(1, 647) = 10.31, p= .001. Participant was not statistically significant as a source of variance for the
intercept (p= .19) or slope (p= .35). Word was also not a significant source of variance (p= .25).
This model confirms faster reading speed consequential from improved spelling, further to any
general practice effects that may emerge over time, and regardless of any impact of participant or
Note that there were too few instances of inconsistent spelling (10.2%) to permit a separate analysis of this category of
performance (see also Martin-Chang et al., 2014).
354 G. OUELLETTE ET AL.
Here, we provide direct experimental evidence showing that the quality of orthographic representa-
tions, as indexed by spelling accuracy, directly relates to reading speed. Important to note, this
relation holds across and within participants. Initial reading and spelling assessments revealed
moderate correlations between both general reading and spelling skill, as well as word-specific
reading speed and spelling accuracy. Further, a within-word analysis indicated that when a given
word was spelled accurately, it tended to be read more rapidly. Even more convincing, words that
were spelled accurately after training were read more rapidly at posttest than words that showed no
such improvement, even when participant and word effects were included in the analysis. The
training data reported here provide the most direct support to date for the contention that improved
spelling brings about increased reading speed.
Together these results extend prior investigations (Burt & Tate, 2002; Martin-Chang et al., 2014;
Moll & Landerl, 2009) while supporting the contention by Perfetti and colleagues (Perfetti, 2007;
Perfetti & Hart, 2002) that underlying orthographic representations vary in completeness across and
within individuals and that this variation is manifested in both spelling accuracy and reading speed.
The training results showed that improvements in spelling brought about immediate improvements
in reading speed; impressively, this robust finding was in addition to any general improvement in
reading speed that occurred due to repeated oral and visual exposure to the words. Further, neither
participant nor lexical characteristics at the word level appear to account for the present findings, as
confirmed in the mixed-effects modeling.
In considering the training data, it is unclear why the spelling of only some of the words
improved over time. Certain items seemed easier to learn. For example, “paraffin,”despite
being very low frequency, has only one consonant doubling (20 instances of improved spelling,
seven of no improvement), compared to a higher frequency word like “hemorrhage”that has
one doubling, one vowel team, more letters, and a problematic “h”(eight instances of
improved spelling; 14 of no improvement). Future research should tackle this question by
focussing on lexical characteristics that underlie word complexity, as well as depth of semantic
knowledge held by participants. In a similar vein, more diverse populations learning to spell
larger word sets would also contribute to the understanding of how orthographic representa-
tions, spelling accuracy, and reading intersect.
In closing, recent research has proposed mechanisms that may help explain the gradual learning
of orthographic representations in literacy acquisition. For example, Ouellette and Sénéchal (2017)
described a developmental pathway to literacy in which the analytical process of generating spellings
contributes directly to word reading; Sénéchal, Gingras, and Heureux (2016) presented a “fuzzy-
representation model”that explicitly describes incremental, partial learning of orthographic repre-
sentations as manifested in spelling. Together with the present study, and in accord with a recent
meta-analysis (Graham & Hebert, 2011), these efforts highlight the connections between ortho-
graphic representations, spelling accuracy, and reading and point to spelling instruction as a
potential valuable means of improving underlying representations, and subsequent word reading.
The authors would like to extend gratitude to Dr. Lisa Dawn Hamilton for her assistance with the mixed-effects
This research was funded by grants to both authors from the Natural Sciences and Engineering Research Council of
SCIENTIFIC STUDIES OF READING 355
Brysbaert, M., & New, B. (2009). Moving beyond Kucera and Francis: A critical evaluation of current word frequency
norms and the introduction of a new and improved word frequency measure for American English. Behavior
Research Methods,41, 977–990. doi:10.3758/BRM.41.4.977
Burt, J. S., & Tate, H. (2002). Does a reading lexicon provide orthographic representations for spelling? Journal of
Memory and Language,46, 518–543. doi:10.1006/jmla.2001.2818
Cedrus Corporation. (2014). SuperLab Pro (5.0). San Pedro, CA. Retrieved from www.superlab.com
Conrad, N. (2008). From reading to spelling and spelling to reading: Transfer goes both ways. Journal of Educational
Psychology,100, 869–878. doi:10.1037/a0012544
Ehri, L. C. (2005). Learning to read words: Theory, findings, and issues. Scientific Studies of Reading,9, 167–188.
Frith, U. (1985). Beneath the surface of developmental dyslexia. In K. Patterson, J. Marshall, & M. Coltheart (Eds.),
Surface dyslexia: Neuropsychological and cognitive studies of phonological reading (pp. 83–103). The Hague, The
Netherlands: Martinus Nijhoff.
Graham, S., & Herbert, M. (2011). Writing to read: A meta-analysis of the impact of writing and writing instruction on
reading. Harvard Educational Review,81, 710–744. doi:10.17763/haer.81.4.t2k0m13756113566
Holmes, V. M., & Carruthers, J. J. (1998). The relation between reading and spelling in skilled adult readers. Journal of
Memory and Language,39, 264–289. doi:10.1006/jmla.1998.2583
Holmes, V. M., & Castles, A. (2001). Unexpectedly poor spelling in university students. Scientific Studies of Reading,5,319–350.
Martin-Chang, S., Ouellette, G., & Madden, M. (2014). Does poor spelling equate toslow reading? The relationship between
reading, spelling, and orthographic quality. Reading and Writing,27,1485–1505. doi:10.1007/s11145-014-9502-7
Moll, K., & Landerl, K. (2009). Double dissociation between reading and spelling deficits. Scientific Studies of Reading,
13(5), 359–382. doi:10.1080/10888430903162878
Ouellette, G., & Sénéchal, M. (2017). Invented spelling in Kindergarten as a predictor of reading and spelling in Grade
1: A new pathway to literacy, or just the same road, less known? Developmental Psychology,53,77–88. doi:10.1037/
Perfetti, C. A. (2007). Reading ability: Lexical quality to comprehension. Scientific Studies of Reading,11, 357–383.
Perfetti, C. A., & Hart, L. (2002). The lexical quality hypothesis. In L. Vehoeven, C. Elbro, & P. Reitsma (Eds.),
Precursors of functional literacy (pp. 189–213). Amsterdam, The Netherlands: John Benjamins.
Perfetti, C. A., Liu, Y., & Tan, L. H. (2005). The Lexical Constituency Model: Some implications of research on
Chinese for general theories of reading. Psychological Review,112,43–59. doi:10.1037/0033-295X.112.1.43
Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K. (1996). Understanding normal and impaired word
reading: Computational principles in quasi-regular domains. Psychological Review,103,56–115. doi:10.1037/0033-
Ratcliff, R. (1993). Methods for dealing with reaction time outliers. Psychological Bulletin,114, 510–532. doi:10.1037/
Schrank, F. A., McGrew, K. S., & Woodcock, R. W. (2001). Assessment service bulletin number 2: WJ-III technical
abstract. Itasca, IL: Riverside.
Sénéchal, M., Gingras, M., & L’Heureux, L. (2016). Modeling spelling acquisition: The effect of orthographic
regularities on silent-letter representations. Scientific Studies of Reading,20, 155–162. doi:10.1080/
Share, D. L. (2008). Orthographic learning, phonological recoding, and self-teaching. In R. V. Kail (Ed.), Advances in
child development and behavior (pp. 31–82). San Diego, CA: Elsevier Academic Press.
Wagner, R. K., Torgesen, J. K., & Rashotte, C. A. (1999). Comprehensive test of phonological processing (CTOPP).
Austin, TX: PRO-ED.
Woodcock, R. W., McGrew, K. S., & Mather, N. (2001). Woodcock-Johnson III test. Itasca, IL: Riverside.
356 G. OUELLETTE ET AL.
Table A1. Words used in experimental reading and spelling tasks.
No. of Letters No. of Phonemes No. of Syllables Frequency per Million
11 8 4 2.14
11 9 4 0.51
11 8 4 0.18
11 6 3 0.61
convergent 10 9 3 0.04
9 6 3 1.22
9 7 3 2.06
dividend 8 8 3 0.37
filtration 10 9 3 0.45
11 8 3 0.59
gradient 8 8 3 0.18
10 6 3 1.71
8 7 3 0.73
10 9 4 0.78
lollipop 8 7 3 1.78
8 5 2 2.10
8 7 3 0.43
8 7 3 0.43
10 8 3 0.35
9 7 3 1.53
salutation 10 10 4 0.12
10 6 3 0.57
8 7 3 0.08
8 6 3 0.96
Brysbarert and New (2009).
Words assigned in training study.
SCIENTIFIC STUDIES OF READING 357