Content uploaded by Yvette Kezilas
Author content
All content in this area was uploaded by Yvette Kezilas on Aug 08, 2016
Content may be subject to copyright.
Word and Pseudoword Superiority Effects on Letter Position Processing in
Developing and Skilled Readers
Yvette Kezilas1, Saskia Kohnen1, Meredith McKague2, Serje Robidoux1
& Anne Castles1
1Department of Cognitive Science, ARC Centre of Excellence in Cognition and its Disorders,
Macquarie University
2Melbourne School of Psychological Sciences, The University of Melbourne
This study was supported by a Macquarie University Research Excellence Scholarship
(MQRES) to the first author.
Correspondence concerning this article should be addressed to Yvette Kezilas, ARC Centre of
Excellence in Cognition and its Disorders, The Australian Hearing Hub, 16 University Avenue,
Macquarie University, NSW 2109, E-mail: kezilas.y@gmail.com
1
Abstract
Studies have shown that letter position processing changes as reading develops. Whether
these changes are driven by the development of the orthographic lexicon is currently unclear.
In this study, we administered a novel variant of the Reicher-Wheeler task to children aged 7-
12 years (Experiment 1) and adults (Experiment 2) to clarify the role of the developing
lexicon in letter position processing. The task required participants to report the identity of a
letter at a specified position within three orthographic contexts: anagram words (e.g. slime –
which has the anagram partner, smile), pseudowords (e.g., blire – brile) and illegal nonwords
(e.g. bfgsv – bsgfv). The influence of a reader’s whole-word orthographic representations was
investigated by comparing the performance of words to pseudowords (word superiority effect
or WSE), and the influence of their knowledge of orthotactic constraints was investigated by
comparing pseudowords to illegal nonwords (pseudoword superiority effect or PSE). Whilst
the PSE increased with developing orthographic skills (as indexed by irregular word reading)
in primary school children, the WSE emerged only in adult readers. Furthermore, the size of
the WSE increased with orthographic skill in adults. The findings are discussed in regards to
current models and theories of visual word recognition and reading development.
Key words: letter position processing, letter identity processing, development of orthographic
knowledge, word superiority effect, pseudoword superiority effect
2
Introduction
There has been much recent interest in how the mind processes the position of letters
within words (e.g., Andrews & Lo, 2012; Kinoshita, Castles, & Davis, 2009; Frost, 2012;
Kohnen & Castles, 2013; Lupker & Davis, 2009; Lupker, Perea, & Davis, 2008; Paterson,
Read, McGowan, & Jordan, 2014; Perea & Lupker, 2003; Perea & Lupker, 2004; Rayner,
White, Johnson, & Liversedge, 2006). Recognizing and comprehending a written word
requires the reader to not only identify each of its component letters but also to determine the
order of those letters, such that pat can be differentiated from apt and tap. Reports of
individuals with letter position dyslexia, who make excessive letter position errors such as
reading form as “from” or defining diary as “something that comes from a cow”, highlight
the importance of letter position processing for successful reading (Friedmann & Rahamim,
2007; Friedmann, Dotan, & Rahamim, 2010; Friedmann, Gvion, & Nisim, 2015; Friedmann
& Haddad-Hanna, 2012; Kohnen, Nickels, Castles, Friedmann, & McArthur, 2012; Kezilas,
Kohnen, McKague, & Castles, 2014). Therefore, determining how the letter position
processing system functions, and what the sources of individual variability are, is critical for
understanding typical as well as impaired reading.
A key unanswered question is whether the letter position processing system changes
as reading develops and, if so, in what way. Whilst there is evidence suggesting that letter
position coding changes across development (but see Paterson et al., 2015), the direction of
change is unclear. Some studies have found that letter position coding becomes more fine-
tuned with development: older children have been reported to process letter position more
accurately than younger children (Acha & Perea, 2008), and adults have been reported to
process letter position more accurately than children (Acha & Perea, 2008; Castles, Davis,
Cavalot, & Forster, 2007; Perea & Estevez, 2008). In contrast, other studies have found that
letter position is coded more coarsely with development: older children process letter position
3
less accurately than younger children (Ziegler, Bertrand, Lété, & Grainger, 2014; Lété &
Fayol, 2012), and adults code letter position less accurately than children (Tiffin-Richards &
Schroeder, 2015).
These mixed findings may reflect the fact that there is more than one possible source
of developmental change in the ability to resolve letter position during visual word
recognition. In models of reading, letter position is typically represented at the initial
encoding stage of visual word recognition, where a letter’s identity and position are analysed
prior to lexical access (see Grainger & van Heuven, 2003 for a review of these models). For
this reason, studies that have found a change in letter position effects across reading
development have suggested that this is most likely due to changes in the way letters are
encoded (Acha & Perea, 2008; Grainger et al., 2012; Grainger & Ziegler, 2011; Ziegler et al.,
2014). For example, based on Perea and Estevez’s (2008) finding that children make more
errors than adults when reading aloud nonwords that are anagrams of real words (e.g.,
cholocate “chocolate”) relative to control nonwords (e.g., choronate “chocolate”), it
has been proposed that the letter encoding system may become more precise (or less noisy) as
the visual system matures (Gomez, Ratcliff, & Perea, 2008; Perea & Estevez, 2008).
A second possibility that has yet to be systematically explored is that changes in letter
position effects across development are driven by changes occurring at a later stage in the
reading system than the initial encoding of letter position. It has been well-established, at
least for letter identification, that performance on basic detection tasks is influenced by
factors thought to reflect processes subsequent to letter encoding, such as orthographic
context. Indeed, letters tend to be identified more accurately when they are presented within
the context of a word than within a pseudoword (word superiority effect or WSE), and when
presented within the context of a pseudoword than within an illegal nonword (pseudoword
superiority effect or PSE; e.g., Chase & Tallal, 1990; Coch & Mitra, 2010; Grainger &
4
Jacobs, 1994; Johnston & McClelland, 1974; Juola, Schadler, Chabot, & McCaughey, 1978;
Reilhac, Jucla, Iannuzzi, Valdois, & Demonet, 2012). A common interpretation of these
findings is that the WSE and PSE reflect the influence of two somewhat independent
components of orthographic knowledge on letter processing: the PSE is thought to reflect the
influence of a reader’s knowledge of orthotactic constraints (e.g., knowing that mb does not
form a legal onset in English), and the WSE is thought to reflect the influence of a reader’s
knowledge of stored whole-word orthographic representations for words (over and above
their knowledge of orthotactic constraints; Grainger et al., 2003). Whilst there is some
contention regarding how separable the mechanisms underpinning the WSE and PSE really
are, with both effects potentially reflecting the activation of a reader’s stored whole-word
orthographic representations for words (e.g., both the word table and pseudoword toble may
activate the orthographic representation for table: Grainger & Jacobs, 1994; McClelland &
Rumelhart, 1981), the WSE and PSE undoubtedly demonstrate that orthographic knowledge
plays an important role in accurate letter identification.
There is also evidence that the two effects increase across reading development. For
example, skilled adult readers tend to show a larger word (Grainger et al., 2003; Lété &
Ducrot, 2008) and pseudoword (Chase & Tallal, 1990) superiority effect than children, older
children have been found to show a larger pseudoword superiority effect than younger
children (Juola et al., 1978), and children with dyslexia have been found to show no word or
pseudoword superiority effects at all (Chase & Tallal, 1990). If the ability to determine letter
position is similarly influenced by orthographic knowledge, then reported changes in letter
position effects across reading development may be driven, at least in part, by changes in the
influence of higher-level orthographic knowledge on letter processing.
The present study was designed to test this possibility. To do so, we developed a
novel adaptation of the classic Reicher-Wheeler task (Reicher, 1969; Wheeler, 1970). Since
5
its development, the Reicher-Wheeler task has been the most popular tool to investigate the
influence of higher-level orthographic knowledge on letter identification. In the standard
version of the task, participants are asked to identify which of two letters was displayed at a
cued position within a briefly presented string of letters. For example, the letter-string slime
would be presented, the fourth position cued, and the participant asked to decide whether they
saw m or d. Critically, when the target is a word, both letter options produce a word (e.g.,
substituting the fourth position in the word slime with the letter d produces the word slide),
but only one is consistent with the target word, obviating a lexical guessing strategy.
To investigate the influence of orthographic knowledge on letter position processing,
we modified the standard task in one key way: both the target letter and the foil letter are
present within the input string. For example, the letter-string slime is presented, the fourth
position cued, and the participant then asked to decide whether they saw m or l in that
position. This differs from the standard letter identity version of the task in that the
participant must have processed the exact position of the letters in the input string in order to
discriminate the two options. This simple modification provides a means to investigate the
influence of orthographic context on the fine discrimination of letter positions within words.
In Experiment 1, we administered our novel letter position Reicher-Wheeler task, as
well as the standard letter identification version of the task, to a group of 81 children (7-12
years old), and measured their level of orthographic reading skill using their scores on an
irregular word reading test. Based on previous research, we expected the WSE and PSE for
the standard letter identification version of the Riecher-Wheeler task to be negligible in
children with limited orthographic skills, but that the WSE and PSE should begin to emerge
as orthographic skills develop, with the children with the highest irregular word reading
scores displaying the largest WSE and PSE. Crucially, if orthographic knowledge influences
letter-position processing similarly to letter identity processing, then the same pattern of
6
results should be observed between the development of orthographic skill and the magnitudes
of the word and pseudoword superiority effects in our novel letter-position variant of the
Reicher-Wheeler task.
Experiment 1
Method
Participants. Eighty-one children (41 males) between the ages of 7 and 12 (M = 9
years, 6 months; SD = 1 year, 3 months) from a suburban school in Sydney, Australia, took
part in the experiment.
Orthographic skill measure. The irregular words from the Castles and Coltheart
Reading Test (CC2; Castles, Coltheart, Larsen, Jones, Saunders, & McArthur, 2009) were
used to index the development of orthographic skill in our sample of participants. Irregular
word reading is an ideal measure of orthographic skill because reading aloud an irregular
word correctly requires direct access to the orthographic representation or memory trace for
the word within the lexicon (using letter-sound rules will result in a mispronunciation of the
word).
Participants read aloud each word, one at a time, on separate flashcards. The CC2 also
includes the administration of 40 regular words and 40 pronounceable nonwords, which are
intermixed with the irregular words in the test. As per the standardised test instructions, a
discontinue rule of five consecutive errors for each subtest was applied. Participants’ CC2
scores are reported in Table 1.
Reicher-Wheeler materials. Targets were 32 words, 32 pseudowords and 32 illegal
nonwords that were 4 to 6 letters in length (see Appendix). According to N-Watch (Davis,
2005), word targets had a mean CELEX written word frequency of 15.27 (SD = 20.74) and a
token bigram frequency of 1071.65 (SD = 599.39). Each word was selected to have an
internal substitution neighbor and an internal migration neighbor (e.g., slime – slide, smile;
7
bread – broad, beard). Each word’s substitution neighbour and migration neighbour was
used to create the foil letters in the letter identity and letter position conditions respectively.
For example, if the target slime was presented and the fourth position cued, the letter m would
be the correct alternative, the letter d would be the foil in the letter identity condition, and the
letter l would be the foil in the letter position condition. Substitution and migration neighbors
were matched list-wise on CELEX written word frequency (substitution frequency M =
27.22, SD = 41.05, migration frequency M = 25.53, SD = 45.35, p = .869 two-tailed), and did
not differ significantly on the combined number of substitution, deletion and addition
neighbours (substitution M = 8.69, SD = 3.94; migration M = 7.72, SD = 5.26, t < 1), and
token bigram frequency (substitution M = 1314.89, SD = 1103.92; migration M = 1213.43,
SD = 601.11, p = .654). Of the 64 distractors, all but 6 (2 substitution, 4 migration), were
either in the children’s printed word database (Masterson, Stuart, Dixon, & Lovejoy, 2003),
or had a root word that was in the database (e.g., sighs is not in the database, but sigh is).
Analyses based on the neighbours (and root words) that were in the children’s printed word
database revealed that the substitution and migration conditions did not differ significantly on
children’s printed word frequency (substitution M = 113.07, SD = 328.65; migration M =
55.25, SD = 86.21, p = .371).
Pseudoword targets had a mean token bigram frequency of 979.39 (SD = 551.21),
which did not differ significantly from word targets (p = .521). Pseudoword targets were
selected to have an internal substitution neighbour (so that changing one letter could result in
another legal pseudoword) and an internal migration neighbor (e.g., blire – blide, brile; kirlp
– kirmp, klirp). As was the case for words, each pseudoword’s substitution neighbour and
migration neighbour was used to create the foil letters in the letter identity and letter position
condition respectively. For example, if the word blire was presented and the fourth position
cued, r would be the correct letter, d would be the foil in the letter identity condition, and the
8
letter l would be the foil in the letter position condition. CELEX bigram token frequency was
matched on average for the substitution and migration neighbours (substitution M = 997.12,
SD = 678.09; migration M = 1058.08, SD = 584.91, p = .657).
Illegal nonwords were created in the same way as the word and pseudoword items.
Only consonants were included in the strings to ensure a clear difference between
pseudoword and illegal nonword conditions in orthotactic legality. Care was also taken to
ensure that commonly contiguous letters (e.g., sh) were not present. Most of the illegal
nonwords had a token bigram frequency of zero, resulting in illegal nonwords having a
significantly lower token bigram frequency than pseudowords (illegal nonwords M = .03, SD
= .11, p < .0001).
Reicher-Wheeler procedure. The Reicher-Wheeler task was administered in groups
of up to four students at a time. Care was taken to ensure that each child understood the task
instructions before commencing the experimental trials. The task was administered to
participants using DMDX software (Forster & Forster, 2003). Stimuli were presented on a
laptop screen in black, lower-case courier-new font on a light grey background. On each trial,
participants saw a fixation cross for approximately 2000ms (119 ticks at 16.70ms per tick),
followed by the target for 284ms (17 ticks). Each letter of the target was then simultaneously
backward masked by a hashmark. To ensure that it was clear which position participants were
required to respond to, a single vertical bar ( | ) was presented as a probe both above and
below the hashmark of interest. The target and foil letters were presented in lower-case above
and below the probed hashmark. The probed position varied for each target in a fixed
manner, but was always a letter that was internal to the string (i.e., exterior letters such as s
and e in slime were never probed). For each level of orthographic context, the letter identity
and position conditions were matched as closely as possible on the position probed. While it
was not always possible to match items pair-wise on probe position, there was no systematic
9
difference between the letter identity and position conditions on probe position (words: χ2(2)
= .09, p =.957; pseudowords: χ2(2) = 2.46, p =.292; illegal nonwords: χ2(2) = .12, p =.942).
Words, pseudowords and illegal nonwords were intermixed and randomized for each
participant. Participants were told to respond as accurately and as quickly as they could, using
the arrow buttons on the keyboard, as to whether the top or the bottom letter had appeared in
the letter string at the position probed. Participants were informed that for some trials both the
target and foil letters would be presented in the letter string, and hence they needed to pay
close attention to the position they were asked to respond to. The two-alternative forced
choice task remained on the screen until a response was made.
The 284ms target duration was decided based on careful consideration of previous
research with young primary school children: Grainger, Bouttevin, Truc, Bastien, and Ziegler
(2003) presented targets for 200ms to 6-9 year old children, and Lété & Ducrot (2008)
presented targets for 250ms to 6-7 year old children. We opted for a slightly longer exposure
duration than these studies due to a critical difference in our task design – for half of our trials
participants were required to encode the exact position of letters within the target in order to
form an accurate response in the letter position condition. Pilot data confirmed that our
exposure duration was appropriate for children in grades 2 to 5.
Four versions of the task were created to ensure that participants only saw each item
once in a single condition1. Participants received a practice block of 12 items before
commencing the 48 experimental trials (24 identity trials, 24 position trials; 16 word trials, 16
pseudoword trials, 16 illegal nonword trials). Experimental trials were broken up over 4
blocks of 12 items, with an equal number of items per condition in each block.
Results and Discussion
1 This task formed part of a larger study. We created four rather than two versions in order to accommodate with
other tasks that were administered during the same testing session.
10
Nine participants who were at ceiling (100% accuracy) on both conditions of a
comparison that was considered critical to addressing our hypotheses were removed from
analyses. Critical comparisons included: letter position words vs pseduowords, letter position
pseudowords vs illegal nonwords, letter identity words vs pseudowords, letter identity
pseudowords vs illegal nonwords. This step was necessary as ceiling performance suggests
that the task was too easy for these participants, and hence they were not given adequate
opportunity to display the effects of interest2. No participants were identified as being at floor
(0% accuracy) on any condition. Responses with reaction times less than 200ms were also
removed from the analysis (< 1% of trials).
Three groups were created to investigate the influence of orthographic skill on task
performance (low-, intermediate- and high- orthographic skill) based on the total number of
irregular words read correctly on the CC2. Twenty-one participants were in the low group, 23
in the intermediate group, and 28 in the advanced group. Unequal numbers in the three
groups was necessary due to tied CC2 scores (e.g., eight participants had a score of 23/40).
Participants were grouped in order to assist in the interpretation of results and for ease of
comparison to previous studies.
Participants’ irregular word reading scores on the CC2 were correlated with age (rs
(N=72) = .45, p < .0001), regular word reading (rs (N=71) = .66, p < .0001) and nonword
reading (rs (N=71) = .64, p < .0001)3, resulting in the three groups differing on irregular word
reading (all ts > 6.70), as well as age (ts > 2.04), regular word reading (all ts > 3.31), and
nonword reading, (all ts > 2.14; see Table 1). No attempt was made to control for the
influence of age, regular word reading and nonword reading on task performance. This is
2 Including these nine children in the analyses, and regrouping participants accordingly, produced the same
pattern of results to those reported here. In fact, due to the increase in statistical power, the marginal three-way
interaction for the WSE reported below was significant with the inclusion of participants at ceiling (p = .057
changed to p = .013).
3 Note that due to missing regular word reading and nonword reading data, one participant was removed from all
analyses and descriptive statistics (reported in Table 1) pertaining to these variables.
11
because the systematic difference between the three orthographic skill groups in age and
reading ability is inherent (rather than due to chance or noise), and hence it makes little
theoretical sense to try to “equate” the groups on these measures (see Miller & Chapman,
2001). Therefore, whilst we consider irregular word reading to be our best proxy for the
development of orthographic skill, we acknowledge that this measure does not enable us to
make an exclusive link between orthographic skill and task performance (this point is
elaborated on in the General Discussion).
-Table 1 about here-
The data were analysed with logit mixed effects modelling (Jaeger, 2008), using the
lme4 package (Bates, Maecher, Bolker, & Walker, 2015) in R (R Core Team, 2015).
Two sets of analyses were created to investigate the word and pseudoword superiority
effects separately. For each effect, we considered the influence of orthographic context
(words vs. pseudowords for the WSE model; pseudowords vs. illegal nonwords for the PSE
model), orthographic skill group (low, intermediate, and high) and condition (letter identity
vs. letter position). We relied on competitive model testing to first settle on a general model,
before undertaking more detailed analysis. We tested five models that increased in
complexity as follows: (1) a model including only the three main effects, (2) a model adding
the interaction between orthographic skill and orthographic context, (3) a model adding the
interaction between orthographic skill and orthographic context, and the interaction between
orthographic context and condition, (4) a model adding all two-way interactions, and (5) a
model adding the three-way interaction. Intercepts were allowed to vary by subjects and
items for each model. The models were compared pair-wise in order of increasing
complexity.
12
Word Superiority Effect (WSE). Adding the two-way interactions (models 2, 3 and
4) did not improve the fit for the WSE, all ps > .31. Adding the three-way interaction (model
5) improved the fit marginally, χ2(2) = 5.75, p = .057. Though the three-way interaction did
not reach significance overall, when looking at the parameters in the model (using summary()
function in R) it is clear that the three way interaction is present when comparing those with
low orthographic skills to those with high orthographic skills (
^
β
= 1.19, SE = .50, Z = 2.38, p
=.017). Since these differences are at the heart of our research question, we opted to run a set
of exploratory analyses to further examine any theoretically relevant trends within the data.
The estimates (converted from the logit function to the original percentage accuracy
using predictSE from the AICcmodavg package (Mazarolle, 2015) in R) from the three-way
interaction model are presented in Figure 1. Contrasts were conducted using lsmeans package
(Lenth, 2010). Overall, performance was better for the letter identity condition than for the
letter position condition,
^
β
= .68, SE = .10, Z = 6.75, p < .0001, and better for words than for
pseudowords,
^
β
= .39, SE = .16, Z = 2.38, p = .017. The size of the WSE did not differ
between the two tasks,
^
β
= .23, SE = .20, Z = 1.13, p = .260.
For the letter identity condition, the WSE was significant collapsed across group,
^
β
=
.50, SE = .20, Z = 2.52, p = .012. This effect appears to be modulated by orthographic skill, as
only those in the high orthographic skill group displayed the effect (low
^
β
= .13, SE = .28, Z
= 0.46, p = .646; intermediate
^
β
= .33, SE = .29, Z = 1.13, p = .259; high
^
β
= 1.05, SE = .32,
Z = 3.31, p = .0009). Furthermore, the WSE was larger for those in the high orthographic
skill group than for those in the low and intermediate groups combined,
^
β
= .41, SE = .17, Z
= 2.39, p = .017.
For the letter position condition, the WSE was not significant collapsed across groups,
^
β
= .27, SE = .18, Z = 1.50, p = .135, nor was it significant for any of the three groups
13
separately (low:
^
β
= .43, SE = .26, Z = 1.61, p = .107; intermediate:
^
β
= .24, SE = .26, Z
= .92, p = .358; high:
^
β
= .16, SE = .26, Z = 0.59, p = .556.
-Figure 1 about here-
Pseudoword Superiority Effect (PSE). Including the interaction between
orthographic skill and orthographic context improved the fit (model 2: χ2(2) = 12.76, p
= .002), as did the interaction between orthographic context and condition (model 3: χ2(1) =
9.86, p = .002). Neither the interaction between orthographic skill and condition, nor the
three-way interaction significantly improved the fit (model 4: χ2(2) = .55, p = .760; model 5:
χ2(2) = 3.07, p = .216) 4. Figure 2 presents the model estimates for each condition from model
5 (three-way interaction model) so as to not obscure potential non-significant trends in the
data. Subsequent analysis considers only Model 3 with the two significant interactions
between orthographic skill and orthographic context and orthographic context and condition.
Performance was better for the letter identity condition than the letter position
condition,
^
β
= .26, SE = .09, Z = 2.91, p = .004, and better for pseudowords than for illegal
nonwords,
^
β
= .61, SE = .13, Z = 4.58, p < .0001. Follow-up of the interaction between
orthographic context and condition revealed that the PSE was significant for both the letter
identity,
^
β
= .89, SE = .16, Z = 5.46, p < .0001, and position condition,
^
β
= .32, SE = .16, Z =
2.05, p =.040, but was significantly larger for the letter identity condition,
^
β
=.57, SE = .18, Z
= 3.13, p = .002.
4 Whilst children were placed into groups, the pattern of results was the same when orthographic skill was
included in the analyses as a continuous variable. Specifically, for the WSE the three-way interaction between
orthographic skill, orthographic context and condition was marginal (p = .067), and for the PSE, the interaction
between orthographic skill and orthographic context as well as the interaction between condition and
orthographic context was significant (p = .0002 and p = .002 respectively).
14
The interaction between orthographic context and orthographic skill was followed up
by looking at the size of the PSE for each group. Because the three-way interaction between
orthographic context, orthographic skill and condition did not significantly improve the fit of
the model, we do not investigate the interaction between orthographic context and
orthographic skill for each condition separately. Planned comparisons revealed that the PSE
(collapsed across condition) was significant for the intermediate,
^
β
= .65, SE = .19, Z = 3.49,
p = .0005, and high orthographic skill groups,
^
β
= .98, SE = .18, Z = 5.48, p < .0001, but not
for the low group,
^
β
= .19, SE = .19, Z = 1.01, p = .312. Furthermore, the PSE was
significantly larger for the intermediate and high group combined relative to the low group,
^
β
= .62, SE = .19, Z = 3.20, p = .001.
-Figure 2 about here-
The findings regarding the PSE were straightforward. The effect was larger for the
letter identity condition than for the letter position condition, indicating that knowledge of
orthotactic constraints plays a more prominent role in accurate letter identity processing than
in letter position processing during the primary school years. Furthermore, the PSE (collapsed
across condition) increased with orthographic skill, such that the high and intermediate
groups displayed a larger PSE than the low orthographic skill group.
The pattern of results regarding the WSE was more complex. For the letter identity
condition, there was a significant WSE, which was driven by the high orthographic skill
group, who were the only group in the sample to display the effect significantly. This finding
suggests that there is an influence of whole-word orthographic representations on letter
identification, but that this influence only emerges as orthographic skill becomes more
advanced. This pattern was not observed for the letter position condition – none of the groups
15
displayed a significant WSE, indicating that whole-word orthographic representations have
little influence on a child’s ability to process letter position during the primary school years.
Two alternative interpretations of the data are possible. It may be that whole-word
orthographic representations simply do not influence letter position processing, regardless of
how advanced a reader’s orthographic reading skills are. Alternatively, a reader’s whole-
word orthographic representations may influence letter position processing, but this influence
is delayed relative to letter identity processing, such that the WSE on the letter position
condition only emerges with further reading experience.
Evidence for the latter interpretation comes from Gomez, Ratcliff and Perea (2008),
who used a forced-choice perceptual identification paradigm with skilled adult readers
including letter position and letter identity manipulated foils. The authors reported that letter-
string recognition was better when the target was a word (e.g., for the letter position condition
participants saw bird and were asked whether they saw bird or brid) than when it was a
pseudoword (e.g., participants saw brid and asked whether they saw bird or brid), suggesting
that higher-level orthographic knowledge plays an important role in both accurate letter
identity and position processing in skilled readers.
Experiment 2 was designed to determine whether adults would display a WSE on
both the letter identity and letter position conditions of our modified Reicher-Wheeler task.
Additionally, a measure of irregular word reading proficiency was taken to determine
whether individual differences in orthographic skill mediates the magnitude of the WSE in a
sample of adult readers. Indeed, if we find that the WSE for the letter position condition
emerges in skilled adult readers, then we might also find an increase in the size of the effect
with orthographic skill, similar to the findings reported in Experiment 1 for the letter identity
condition.
Experiment 2
16
Method
Participants. Participants were 60 adults (25 males; Age M = 22.14, SD = 4.43) who
participated in exchange for either course credit or for a small monetary reward. Those who
took part in exchange for course credit were undergraduate psychology students, and those
who took part in exchange for money were either university students or members of the wider
public recruited via the Macquarie University website, an advert posted on campus, or word
of mouth.
Orthographic skill measure. An extended version of the irregular words from the
Castles and Coltheart Reading Test (CC2; Castles et al., 2009) was used to index
orthographic reading skill in our sample of participants. The items were the same as those
presented to the children in Experiment 1, but included an additional 15 items that were more
challenging for adults to avoid ceiling effects. The task also included the same regular words
and pronounceable nonwords as Experiment 1, except with an additional 15 items for each
subtest. Test administration was the same as Experiment 1. On average, participants correctly
read 37.75 irregular words (SD = 5.81, Min = 25, Max =51), 50.77 regular words (SD = 3.26,
Min = 34, Max =55), and 41.52 nonwords (SD = 8.14, Min = 7, Max = 50). Irregular word
reading was correlated with regular word reading (rs = .54, p < .0001) and with nonword
reading (rs = .42, p = .001), but not with age (rs = .17, p = .193).
Reicher-Wheeler materials and procedure. The task items were the same as those
administered in Experiment 1. An additional 16 word targets (e.g., the target brunt, with the
distractor burnt and blunt), 16 pseudoword targets (e.g., trave, tarve, traze) and 16 illegal
nonword targets (e.g., tkmfj, tmkfj, tkmrj) were also included in the task. These items were not
administered in Experiment 1 as the words were unlikely to be known by many of the
children within the sample. To enable direct comparison to the results reported in Experiment
1, the additional items were not included in the analyses.
17
The Reicher-Wheeler task was administered to participants individually. Two
versions of the task in Experiment 2 were created to ensure that participants saw each letter-
string only once in a single condition. Following 12 practice trials, participants received 144
experimental trials, 96 of which were analysed following the removal of the additional items.
Experimental trials were broken up over 12 blocks (12 trials per block), with an equal number
of items per condition in each block. Note that the children in Experiment 1 received 48 trials
each (rather than the 96 trials in Experiment 2) because the task in Experiment 1 was divided
into 4 versions (see footnote 1).
Following the finding that some participants in Experiment 1 were performing at
ceiling on the task (see first paragraph of Results and Discussion for Experiment 1), in
Experiment 2 we decided to adjust the exposure duration for each participant based on a
preliminary test phase (see Grainger et al., 2003 for a similar approach). Based on their
performance on the preliminary test phase, item presentation time during the experimental
phase was 17ms (1 tick, 16.70ms) for 8 participants, 33ms (2 ticks, 16.70ms) for 41
participants, and 50ms (3 ticks, 16.70ms) for 11 participants. These presentation durations are
similar to previous Reicher-Wheeler studies with skilled adult readers (e.g., Chase & Tallal,
1990; Coch & Mitra, 2010; Grainger et al., 2003; Lété & Ducrot, 2008).
Results and Discussion
None of the adult participants were at ceiling (100% accuracy) or floor (0% accuracy)
in both conditions of a comparison that was considered critical to addressing our hypotheses
(critical comparisons were the same as those reported in the ‘Results and Discussion’ section
in Experiment 1). Responses with reaction times less than 200ms were removed from the
analysis (< 1% of trials). The model selection process was the same as reported in
Experiment 1, except orthographic skill was included as a continuous variable in the models
instead of as a categorical variable. Two sets of analyses were created to investigate the word
18
and pseudoword superiority effects separately. Five models increasing in complexity were
compared (1) a model including only the three main effects, (2) a model adding the
interaction between orthographic skill and orthographic context, (3) a model adding the
interaction between orthographic skill and orthographic context, and the interaction between
orthographic context and condition, (4) a model adding all two-way interactions, and (5) a
model adding the three-way interaction. Intercepts were allowed to vary by subjects and
items for each model. The models were compared pair-wise in order of increasing
complexity. Model estimates were interpreted using summary() function in R. Figure 3 and 4
show the estimates (converted into accuracy percentages) from the full model including the
three-way interaction for the WSE and PSE respectively.
Word Superiority Effect (WSE). Including the interaction between orthographic
skill and orthographic context, as well as the main effect of condition, improved the model
fit, χ2(1) = 4.19, p = .041. The more complex models were not significant (model 3: χ2(1) <
1, p = .705; model 4: χ2(1) < 1, p = .815; model 5: χ2(1) = 1.31, p = .253). To aid in the
interpretation of model 2 we ran the summary() function on two separate models including
(1) the main effects of condition, orthographic skill, and orthographic context, and (2) the
interaction between orthographic skill and orthographic context. Performance was better for
the letter identity condition than for the letter position condition,
^
β
= .51, SE = .08, Z = 6.65,
p < .0001, and better for words than for pseudowords (
^
β
= .52, SE = .14, Z = 3.61, p
= .0003). The WSE increased with orthographic skill,
^
β
= .15, SE = .08, Z = 2.04, p = .041.
-Figure 3 about here-
Pseudoword Superiority Effect (PSE). In the case of the PSE, the fit was not
improved by including the interaction between orthographic context and orthographic skill
19
(model 2: χ2(1) < 1, p = .968), but was significantly improved by including the interaction
between orthographic context and condition (model 3: χ2(1) = 7.25, p = .007). More complex
models did not improve the fit (model 4: χ2(1) < 1, p = .417; model 5: χ2(1) < 1, p = .638).
Since the interaction between orthographic skill and orthographic context proved to be
redundant (i.e., model 2 did not provide a better fit than model 1), this interaction was
removed from the model. Hence, the final model included the main effects of condition,
orthographic context and orthographic skill, as well as the interaction between orthographic
context and condition. As was done for the WSE, we ran the summary() function on two
separate models including (1) the main effect of condition, orthographic skill and
orthographic context, and (2) the interaction between condition and orthographic context.
Performance was better for the letter identity than for the letter position condition,
^
β
= .27,
SE = .07, Z = 3.87, p = .0001, and better for pseudowords than for illegal nonwords,
^
β
= .65,
SE = .13, Z = 5.00, p < .0001. The PSE was significantly larger for the letter identity
condition than for the letter position condition,
^
β
= .37, SE = .14, Z = 2.69, p = .007.
-Figure 4 about here-
As was found in Experiment 1, participants showed a significant PSE, and the effect
was significantly larger for the letter identity condition than for the letter position condition.
The critical finding from Experiment 2, however, was that skilled adult readers displayed a
significant WSE for both the letter identity and position conditions, and the WSE increased
with orthographic skill at a similar rate for the letter identity and position conditions.
General Discussion
The present study investigated the influence of a reader’s word-specific orthographic
knowledge, and their knowledge of orthotactic constraints, on the development of letter
20
identity and letter position processing in children (Experiment 1) and adults (Experiment 2).
Word-specific orthographic knowledge was measured using the WSE, and knowledge of
orthotactic constraints was measured using the PSE. The standard Reicher-Wheeler task
provided a measure of letter identity processing, whilst a novel variant of the task, in which
the foil letter yields an anagram of the target, provided a measure of the precision of letter
position coding. The CC2 measure of irregular word reading was used to index orthographic
skill.
There were four key findings: (1) The only children in Experiment 1 to show a WSE
were those in the high orthographic skill group, and this effect was present for the letter
identity condition only; (2) the adults in Experiment 2 showed a WSE for both the letter
identity and position versions of the Reicher-Wheeler task, and the effect increased with
orthographic skill; (3) both children (Experiment 1) and adults (Experiment 2) showed a
larger PSE for the letter identity condition than for the letter position condition; (4) the
magnitude of the PSE increased with orthographic skill in Experiment 1 with children, but
this relationship was not seen in Experiment 2 with adults.
The Word Superiority Effect (WSE)
Our hypothesis that the WSE for the letter identity condition would emerge with the
development of orthographic skills was supported. Specifically, we found that the WSE for
the letter identity condition was only significant for children in the high orthographic skill
group in Experiment 1, and for adult readers in Experiment 2. This finding is consistent with
Lété and Ducrot (2008), who found a significant WSE for adults but not for 6 and 7-year-old
readers, and with Grainger et al. (2003), who found a small WSE for older children (mean
age = 11.5 years) and adults. Together with previous studies, our findings suggest that a
21
reader’s whole-word orthographic representations influence letter identification, and this
influence increases as orthographic skills develop.
In contrast to the letter identity condition, the WSE for the letter position condition
was not significant for any children in Experiment 1, irrespective of level of orthographic
skill. Following this finding, two alternative hypotheses were formed for Experiment 2. One
hypothesis is that whole-word orthographic representations never influence letter position
processing, regardless of how advanced a reader’s orthographic skills are. Alternatively, the
influence of a reader’s whole-word orthographic representations on accurate letter-position
coding may be delayed relative to letter identity processing, such that the WSE emerges only
for skilled adult readers. The results from Experiment 2 supported the latter hypothesis, and
were consistent with the findings reported by Gomez et al. (2008) using the forced-choice
perceptual identification paradigm; skilled adult readers displayed a significant WSE for both
the letter identity and position conditions. Furthermore, the finding that the WSE increased
with orthographic skill at the same rate for the two conditions (see Figure 3), suggests that by
adulthood letter identity and letter position processing are influenced similarly by a reader’s
whole word orthographic representations.
These results have important implications for the interpretation of findings regarding
the developmental trajectory of letter identity and letter position processing. Castles et al.
(2007), found that children in grade 3 were faster to make a lexical decision to a target word
when the target was preceded by a prime that was either one letter different from the target
(e.g., rlay-PLAY) or transposed two letters in the target (e.g., lpay-PLAY), relative to a
control prime (e.g,. meit-PLAY). When the same children were re-tested in grade 5, the
priming effect was still present for the transposed letter condition. In contrast, the one-letter
different priming effect had diminished entirely, suggesting that target words were no longer
22
facilitated by a prime that was not composed of the exact same letter identities as the target.
Priming for both conditions was not apparent in a separate sample of skilled adult readers.
Following these findings, the authors concluded that the ability to resolve letter
position information is slower to develop than the ability to resolve letter identity
information, and that these developmental effects are driven by changes occurring within the
lexicon. An alternative interpretation, however, is that these changes occur at the prelexical
level. Specifically, the reduction in priming effects with age could be due to a general
maturation of the visual system, independently of changes occurring simultaneously with
lexical development. The use of the Reicher-Wheeler task in the present study enables us to
explicitly investigate the role of the lexicon in letter identity and position processing, and
hence directly test Castles et al.’s (2007) conclusions. Our finding that the WSE emerged
later in development for the letter position condition relative to the letter identity condition
(adulthood vs. primary school) suggests that the developmental delay reported by Castles et
al. (see also Kohnen & Castles., 2012) is likely due to a reader’s whole-word orthographic
representations having a differential influence on letter identity and letter position processing
early in development.
From an ecological stand-point, the developmental delay in accurate letter position
processing reported in the present study makes sense. English orthography consists of many
words that differ from one another by the substitution of a single letter (e.g., ball, hall, call,
tall, fall), making the accurate encoding of letter identity essential to ensuring words are
discriminated from neighbouring words. In contrast, there are very few words that have the
same letters in common but that differ in letter position (i.e., anagrams), and hence the
necessity to code exact letter position is less paramount for successful reading acquisition.
Following the idea that a reader’s orthographic environment shapes the development of their
visual word recognition system (Frost, 2012), it would be interesting to investigate whether
23
the developmental delay in the WSE for the letter position condition reported here is present
in orthographies that are anagrammatically dense, such as Hebrew.
The question still remains as to the specific mechanisms that develop to give rise to
the WSE findings reported in the present study. The WSE is typically explained using two
alternative theories. One theory uses feedback from the word to the letter level of
representation to explain the effect. Specifically, activation of a word’s orthographic
representation at the word level feeds back to the letter level, reinforcing the activation of the
word’s component letter representations (McClelland & Rumelhart, 1981). An alternative
theory uses cascaded activation to the word level to explain the effect. For example, Grainger
and Jacobs’ dual read-out model (1994) proposes that letters can be identified by either the
activation of the letter representations (letter readout) or can be inferred following word
identification (word readout). The WSE reflects the advantage of having an additional word
readout function that can be drawn upon when individual letter readout fails (for similar
accounts using multiple sources of information to explain the WSE see Paap, Newsome,
McDonald, & Schvaneveldt, 1982; Massaro, 1979; Richman & Simon, 1989).
Both of these accounts could easily be extended to capture the increase in the WSE
with development for the letter identity condition in Experiment 1 with children. Specifically,
it could be that readers with more advanced orthographic skills are simply more likely to
have an orthographic representation in their lexicon for the target word presented in the
Reicher-Wheeler task, and this additional information could either be fed back to the letter
level (following an interactive-activation account), or used as a word readout mechanism
when letter readout fails (following a dual read-out approach).
However, these accounts fall short in explaining (1) the developmental delay in the
emergence of the WSE for the letter position condition, and (2) the increase in the WSE with
orthographic skill for the adults in Experiment 2. In regards to the first point, if the presence
24
of the WSE is simply dependent on a participant’s knowledge of the target word, we should
have found the WSE to emerge at similar developmental time points for the letter identity and
letter position conditions, since the same target words were used in both conditions. Instead,
we found the WSE to emerge in primary school for the letter identity condition, and in
adulthood for the letter position condition. Regarding the second point, since the target words
in both Experiment 1 and 2 were selected to be suitable for children as young as 7 years old,
it can be assumed that the full set of target words were known by all adult participants – even
those with relatively limited orthographic skills.
One explanation for these findings is that it is not just the number of orthographic
representations that a reader has within their lexicon that determines the magnitude of the
WSE, but the precision with which these representations are encoded, where precise
representations are fully rather than partially specified in regards to letter identity and
position, enabling written input to directly activate its matching orthographic representation
with minimal coactivation of visually similar words (Andews & Lo, 2012, Perfetti, 1992;
Perfetti, 2007; Ehri, 2005). This theory can account for the increase in the WSE with
orthographic skill in adults by assuming that those with more advanced orthographic reading
skills have more precise orthographic representations than those with relatively weak
orthographic skills. In the Reicher-Wheeler task, the coactivation of neighbouring
orthographic representations can induce errors in the word condition, as the distractors are
designed to be orthographically similar to the target (e.g., the distractor for the target males is
makes for the letter identity condition, and meals for the letter position condition). It could
therefore be the case that readers with superior orthographic skills show a larger WSE
because the precision of their orthographic representations enables the direct and automatic
activation of the word males with minimal coactivation of visually similar words such as
makes and meals. The finding that the WSE for the letter identity condition emerged earlier in
25
development than the letter position condition suggests that the precision of a reader’s
orthographic representations develops first with regards to the letter identities that make up
words, and then later to the position of these letter identities within words.
The Pseudoword Superiority Effect (PSE)
Whilst the PSE was on average larger for the letter identity condition than for the
letter position condition, the developmental trajectory of the effect was similar for the two
conditions. In contrast to the WSE findings, the PSE increased with orthographic skill at the
same rate for the letter identity and position conditions for children in Experiment 1.
Furthermore, adults displayed a large PSE in Experiment 2 and the effect for both conditions
was not mediated by orthographic skill. These findings suggest that the influence of
orthotactic constraints on both letter identity and position processing increases with
orthographic skill, and this influence asymptotes by adulthood.
Like the WSE, the exact mechanisms underlying the PSE are currently unclear, which
makes interpreting the developmental findings in the present study difficult. One school of
thought is that the PSE is driven by the same processes as the WSE – that is, either by
feedback from the word to letter level, or cascaded activation to the word level (Grainger &
Jacobs, 1994; McClelland & Rumelhart, 1981). The pseudoword partially activates its real-
word neighbours (e.g., toble would activate table), resulting in a pseudoword advantage over
illegal nonwords. An alternative account is that the PSE reflects perceptual fluency that is
produced by stimulus familiarity independently of lexical status. This perceptual fluency may
arise due to the pronounceability of the pseudoword (Ziegler & Jacobs, 1995) or the uniquely
orthographic aspects of the pseudoword, such as frequency of letter combinations (Grainger
et al., 2003).
The finding that many of the children in the sample displayed a PSE in the absence of
a WSE may present challenges to models that propose a lexical locus for the PSE, such as the
26
interactive-activation model (McClelland & Rumelhart, 1981) and the dual read-out model
(Grainger & Jacobs, 1994). According to these models this dissociation between the WSE
and the PSE should not occur, as words always activate their matching orthographic
representations more so than pseudowords (Grainger et al., 2003). Indeed, based on a similar
dissociation with the WSE and PSE in a sample of developing readers, Grainger et al (2003)
concluded that the PSE must be driven by perceptual fluency rather than lexical influences.
However, it is important to remember that the interactive-activation model and the
dual read-out model are theories of skilled adult reading, and therefore assume that the
orthographic lexicon is fully developed. It could be that the same mechanisms underpin the
WSE and PSE, but that for novice readers, both words and pseudowords produce dispersed
activation within the lexicon, resulting in multiple orthographic representations being
activated simultaneously. This proposition can account for a significant advantage for
pseudowords over illegal nonwords (pseudowords activate visually similar words more so
than illegal nonwords), and no advantage for words over pseudowords (both words and
pseudowords activate many visually similar words). This interpretation is also supported by
the finding that the participants with the most advanced orthographic skills in the present
study showed both a WSE and PSE. These participants are likely to have well-developed and
precise orthographic representations, resulting in written words directly activating their
matching orthographic representations, and hence providing an advantage for words over
pseudowords.
The possibility that the PSE may reflect the same underlying mechanisms as the WSE
calls into question the theoretical distinction made between these two effects both in the
present study and elsewhere in the literature. If the PSE is simply an indirect way of
measuring the influence of a reader’s whole-word orthographic representations on letter
processing, then partitioning orthographic context effects into the PSE and WSE makes little
27
sense. If, however, knowledge of orthotactic constraints is a fundamental aspect of
orthographic knowledge, independently of a reader’s whole-word orthographic
representations, then the distinction between the WSE and the PSE has potential to be highly
informative.
A further possibility is that phonology plays a large role in the pseudoword advantage.
To ensure that pseudowords and nonwords differed on orthographic legality, our nonwords
consisted of consonants only (i.e., no vowels), and hence they were unpronounceable. Since
orthographic legality and phonological legality are tightly linked, it is difficult to determine
whether the PSE is driven by orthographic influences, phonological influences, or both
orthographic and phonological influences. Further research into the exact locus of the PSE is
therefore needed before strong claims can be made in regards to the PSE findings reported in
the present study.
A similar comment could also be made in regards to our use of irregular word reading
in the present study to measure the development of orthographic skill. Irregular word reading
was used as a proxy for orthographic skill because successful irregular word reading requires
direct access to the orthographic representation or memory trace for a written word (using
letter-sound rules will result in a mispronunciation). That being said, irregular word reading
is highly correlated with regular word and nonword reading, both of which could be
considered measures of phonological reading skill. It is therefore critical to reiterate that
irregular word reading was not selected as our measure of development on the basis that it
indexes orthographic knowledge exclusively, but rather it was selected to give us the closest
approximation of a reader’s orthographic reading skills. On this point, it is also important to
note that any attempts to ‘partial out’ the influence of phonological skill from our analyses
would have severely compromised the validity of our irregular word reading measure, since
28
the relationship between orthographic skill and phonological skill is inherent – as opposed to
the two constructs covarying due to chance or noise (see Miller & Chapman, 2001).
Conclusion
The findings from the present study suggest that a reader’s whole-word orthographic
representations and their knowledge of orthotactic constraints influence both letter identity
and letter position processing. How these higher-order influences play out as orthographic
skills develop, however, differs for letter identity and position processing. Specifically, the
influence of a reader’s whole-word orthographic representations on letter position processing
appears to be comparatively delayed, supporting the idea that the ongoing development of
precision in a reader’s orthographic representations influences accurate letter position
processing. These findings not only provide various challenges to models of visual word
recognition, but also suggest that future studies investigating the development of letter
identity and letter position processes must consider the role that an individual’s higher-level
orthographic knowledge may have on these lower-level processes.
29
Acknowledgements
We would like to thank Saskia Kohnen, Eva Marinus, Kathryn Preece, Xenia Schmalz, and
Nicholas Badcock for creating the extended version of the Castles and Coltheart Test 2, and
for providing us access to it.
30
References
Acha, J., & Perea, M. (2008). The effects of length and transposed-letter similarity in lexical
decision: Evidence with beginning, intermediate, and adult readers. British Journal of
Psychology, 99, 245-264. doi: 10.1348/000712607X224478
Andrews, S., & Lo, S. (2012). Not all skilled readers have cracked the code: individual
differences in masked form priming. Journal of Experimental Psychology: Learning,
Memory and Cognition, 38(1), 152-163. doi: 10.1037/a0024953
Castles, A., Coltheart, M., Larsen, L., Jones, P., Saunders, S., & McArthur, G. (2009).
Assessing the basic components of reading: A revision of the Castles and Coltheart
test with new norms. Australian Journal of Learning Difficulties, 14(1), 67-88. doi:
10.1080/19404150902783435
Castles, A., Davis, C., Cavalot, P., & Forster, K. (2007). Tracking the acquisition of
orthographic skills in developing readers: Masked priming effects. Journal of
Experimental Child Psychology, 97, 165-182. doi:10.1016/j.jecp.2007.01.006
Chase, C. H., Tallal, P. (1990). A developmental, interactive activation model of the word
superiority effect. Journal of Experimental Child Psychology, 49, 448-487.
Coch, D., Mitra, P. (2010). Word and pseudoword superiority effects reflected in the ERP
waveform. Brain Research, 1329, 159-174. doi:10.1016/j.brainres.2010.02.084
Douglas Bates, Martin Maechler, Ben Bolker, Steve Walker (2015). Fitting Linear Mixed-
Effects Models Using lme4. Journal of Statistical Software, 67(1), 1-
48. doi:10.18637/jss.v067.i01.
Ehri, L. C. (2005). Learning to read words: Theory, findings, and issues. Scientific Studies of
Reading, 9(2), 167-188.
Forster, K. I., & Forster, J. C. (2003). DMDX: A Windows display program with millisecond
accuracy. Behavior Research Methods, Instruments, & Computers, 35(1), 116-124.
31
Friedmann, N., Dotan, D., and Rahamim, E. (2010). Is the visual analyzer orthographic-
specific? Reading words and numbers in letter position dyslexia. Cortex, 46(8), 982-
1004. doi:10.1016/j.cortex.2009.08.007
Friedmann, N., Gvion, A., & Nisim, R. (2015). Insights from letter position dyslexia on
morphological decomposition in reading. Frontiers in Human Neuronscience, 9(143),
doi: 10.3389/fnhum.2015.00143
Friedmann, N., and Haddad-Hanna, M. (2012). Letter position dyslexia in Arabic: From form
to position. Behavioural Neuropsychology 25, 193-203, doi: 10.3233/BEN-2012-
119004. doi:10.3233/BEN-2012-119004
Friedmann, N., and Rahamim, E. (2007). Developmental letter position dyslexia. Journal of
Neuropsychology 1, 201-236. doi: 10.1348/174866407X204227
Frost, R. (2012). Towards a universal model of reading. Behavioural and Brain Sciences,
35(5), 263-279. doi: http://dx.doi.org/10.1017/S0140525X11001841
Gomez, P., Ratcliff, R., & Perea, M. (2008). The overlap model: a model of letter position
coding. Psychological Review, 115(3), 577. doi: 10.1037/a0012667
Grainger, J., Bouttevin, S., Truc, C., Bastien, M., & Ziegler, J. (2003). Word superiority,
pseudoword superiority, and learning to read: A comparison of dyslexic and normal
readers. Brain and Language. 87, 432-440. doi: 10.1016/S0093-934X(03)00145-7
Grainger, J., & Jacobs, A. M. (1994). A dual read-out model of word context effects in letter
perception: Further investigations of the word superiority effect. Journal of
Experimental Psychology: Human Perception and Performance, 20(6), 1158-1176
Grainger, J., Lété, B., Bertrand, D., Dufau, S., Ziegler, J. C. (2012). Evidence for multiple
routes in learning to read. Cognition, 123, 280-292. doi:
10.1016/j.cognition.2012.01.003
32
Grainger, J., & Ziegler., J. C. (2011). A dual-route approach to orthographic processing.
Frontiers in Psychology, 2(54), 1-13, doi: 10.3389/fpsyg.2011.00054
Jaeger, T. F. (2008). Categorical Data Analysis: Away from ANOVAs (transformation or
not) and towards Logit Mixed Models. Journal of Memory and Language, 59(4), 434-
446. doi: 10.1016/j.jml.2007.11.007
Johnston, J. C., & McClelland, J.L. (1974). Perception of letters in words: seek not and eye
shall find. Science, 184, 1192-1194.
Juola, J. F., Schadler, M., Chabot, R. J., & McCaughey, M. W. (1978). The development of
visual information processing skills related to reading. Journal of Experimental Child
Psychology, 25(3), 459-476. doi: 10.1016/0022-0965(78)90069-3
Kinoshita, S., Castles, A., & Davis, C. (2009). The role of neighbourhood density in
transposed-letter priming. Language and Cognitive Processes, 24(4), 506-526. doi:
10.1080/01690960802289213
Kezilas, Y., Kohnen, S., McKague, M., Castles, A. (2014). The locus of impairment in
developmental letter position dyslexia. Frontiers in Human Neuroscience, 8(356), 1-
14. doi: 10.3389/fnhum.2014.00356
Kohnen, S., & Castles, A. (2013). Pirates at parties: Letter position processing in developing
readers. Journal of Experimental Child Psychology, 50(14), 3681-3692. doi:
http://dx.doi.org/10.1016/j.jecp.2012.08.011
Kohnen, S., Nickels, L., Castles, A., Friedmann, N., McArthur, G. (2012). When ‘slime’
becomes ‘smile’: developmental letter position dyslexia in English.
Neuropsychologia, 50(14), 3681-3692. doi: 10.1016/j.neuropsychologia.2012.07.016
Lenth, R. (2015). lsmeans: Least-Squares Means. R package version 2.20-23.
http://CRAN.R-project.org/package=lsmeans
33
Lété, B., Ducrot, S. (2008). Visuo-attentional processing by dyslexic readers on the Reicher-
Wheeler task. Current Psychology Letters, 24(1), 25-40
Lupker, S. J., & Davis, C. J. (2009). Sandwich priming: a method for overcoming the
limitations of masked priming by reducing lexical competitor effects. Journal of
Experimental Psychology: Learning, Memory and Cognition, 35(3), 618-639. doi:
10.1037/a0015278
Lupker, S. J., Perea, M., & Davis, C. (2008). Transposed-letter effects: Consonants, vowels
and letter frequency. Language and Cognitive Processes, 23(1), 93-116. doi:
10.1080/01690960701579714
Massaro, D. W. (1979). Letter information and orthographic context in word perception.
Journal of Experimental Psychology: Human Perception and Performance, 5(4), 595-
609. doi:10.1037/0096-1523.5.4.595
Masterson, J., Stuart, M., Dixon, M., & Lovejoy, S. (2003). Children’s Printed Word
Database. Colchester, UK: University of Essex.
Mazerolle, M. J. (2015) AICcmodavg: Model selection and multimodel inference based on
(Q)AIC(c). R package version 2.0-3.
http://CRAN.R-project.org/package=AICcmodavg.
McClelland, J. L., Rumelhart, D. E. (1981). An interactive activation model of context effects
in letter perception: Part 1. An account of basic findings. Psychological Review, 88,
375-407
Miller, G. A., & Chapman, J. P. (2001). Misunderstanding analysis of covariance. Journal of
Abnormal Psychology, 110(1), 40-48, doi: http://dx.doi.org/10.1037/0021-
843X.110.1.40
34
Paap, K. R., Newsome, S. L., McDonald, J. E., & Schvaneveldt, R. W. (1982). An
activation–verification model for letter and word recognition: The word-superiority
effect. Psychological Review, 89(5), 573-594. doi:10.1037/0033-295X.89.5.573
Paterson, K. B., Read, J., McGowan, V. A., Jordan, T. R. (2014). Children and adults both
see ‘pirates’ in ‘parties’: letter-position effects for developing readers and skilled
adult readers. Developmental Science, 18(2), 1-9. doi: 10.1111/desc.12222
Perea, M., & Estevez, A. (2008). Transposed-letter similarity effects in naming pseudowords:
Evidence from children and adults. European Journal of Cognitive Psychology, 20,
33-46. doi: 10.1080/09541440701306941
Perea, M., & Lupker, S. J. (2003). Transposed-letter confusability effects in masked form
priming. In S. Kinoshita and S. J. Lupker (Eds.), Masked priming: State of the art.
Hove, UK: Psychology Press.
Perea, M., & Lupker, S. J. (2004). Can CANISO activate CASINO? Transposed-letter
similarity effects with nonadjacent letter positions. Journal of Memory and Language,
51, 231-246. doi:10.1016/j.jml.2004.05.005
Perfetti, C. A. (1992). The representation problem in reading acquisition. In P. B. Gough,
L.C. Ehri, & R. Treiman (Eds.), Reading acquisition. Hillsdale, NJ: Erlbaum.
Perfetti, C. A. (2007). Reading ability: Lexical quality to comprehension. Scientific Studies of
Reading, 11(4), 357-383. doi:10.1080/10888430701530730
Rayner, K., White, S. J., Johnson, R. L., & Liversedge, S. P. (2006). Raeding wrods with
jubmled lettres: There is a cost. Psychological Science, 17, 192-193. doi:
10.1111/j.1467-9280.2006.01684.x
R Core Team (2015). R: A language and environment for statistical computing. R Foundation
for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/.
35
Reicher, G. M. (1969). Perceptual recognition as a function of meaningfulness of stimulus
material. Journal of Experimental Psychology, 81, 275-280
Reilhac, C., Jucla, M., Iannuzzi, S., Valdois, S., & Démonet, J. (2012). Effect of orthographic
processes on letter identity and letter-position encoding in dyslexic children.
Frontiers in Psychology, 3, 1-11. doi: 10.3389/fpsyg.2012.00154
Richman, H. B., & Simon, H. A. (1989). Context effects in letter perception: Comparison of
two theories. Psychological Review, 96(3), 417-432. doi:10.1037/0033-295X.96.3.417
Tiffin-Richards, S. P., & Schroeder, S. (2015). Children’s and adults’ parafoveal processes in
German: Phonological and orthographic effects. Journal of Cognitive Psychology,
27(5), 531-548, doi:10.1080/20445911.2014.999076
Wheeler, D. D. (1970). Processes in word recognition. Cognitive Psychology, 1, 59-85. doi:
http://dx.doi.org/10.1016/0010-0285(70)90005-8
Ziegler, J. C., Bertrand, D., Lété, B., & Grainger, J. (2014). Orthographic and phonological
contributions to reading development: Tracking developmental trajectories using
masked priming. Developmental Psychology, 50(4), 1026-1036. doi:
10.1037/a0035187
Ziegler, J. C., & Jacobs, A. M. (1995). Phonological information provides early sources of
constraint in the processing of letter strings. Journal of Memory and Language, 34(5),
357-593. doi: 10.1006/jmla.1995.1026
36
Appendix. Word stimuli used in the Reicher-Wheeler task
LETTER IDENTITY LETTER POSITION
Target Distractor Target Distractor
beings brings beings begins
hares hires hares hears
polos poles polos pools
snug slug snug sung
bolt boat blot* bolt
filed fixed filed field
trap trip trap tarp
clams claps clams calms
sinks silks sinks skins
dairy daisy dairy diary
signs sighs signs sings
slime slide slime smile
feels feeds feels flees
stain slain stain satin
farmed formed farmed framed
seals seats seals sales
spine shine spine snipe
tired timed tired tried
snips slips snips spins
males makes males meals
wired wiped wired weird
bowls boils bowls blows
warps warns warps wraps
dared dated dared dread
angles ankles angles angels
fired fined fired fried
spelt spent spelt slept
spots shots spots stops
spans swans spans snaps
slate slave slate stale
bread broad bread beard
leaks leaps leaks lakes
*Note that due to an experimenter error, different targets for this item were presented for the
letter identity (bolt) and letter position (blot) conditions in both Experiment 1 and 2. Target
word descriptives reported in the method include both bolt and blot (i.e., N=33).
37
38
Appendix. Pseudoword stimuli used in the Reicher-Wheeler task
LETTER IDENTITY LETTER POSITION
Target Distractor Target Distractor
bergs berks bergs bregs
kirlp kirmp kirlp klirp
keabs keals keabs kabes
firth firch firth frith
blire blide blire brile
skole smole skole sloke
keings koings keings kegins
setag semag setag sateg
smule smupe smule slume
stalif shalif stalif slatif
smep snep smep semp
garth garsh garth grath
brog brug brog borg
parld porld parld plard
troms trogs troms torms
pilch polch pilch plich
garps garts garps graps
snech stech snech sench
firnch fornch firnch frinch
frish frich frish firsh
sputs snuts sputs stups
gorlt gormt gorlt glort
falds falms falds flads
glead gload glead galed
geals geavs geals gleas
krem klem krem kerm
kerlm korlm kerlm klerm
silofy simofy silofy sloify
neafs neals neafs nafes
splot splut splot spolt
falint famint falint fanilt
surks surds surks skurs
39
Appendix. Illegal nonword stimuli used in the Reicher-Wheeler task
LETTER IDENTITY LETTER POSITION
Target Distractor Target Distractor
xmvqw xnvqw mvqwz* mvwqz
zxqkn zxqcn jzxqk* jxzqk
wfgqz wfpqz wfgqz wqfgz
fxqwj fxqzj fxqwj fxwqj
bfkj bfdj bfkj bkfj
xqtvw xqnvw xqtvw xtqvw
kzxgt kjxgt kzxgt kgzxt
lxbzn lxvzn lxbzn lxzbn
qdsw qdzw qdsw qsdw
njxlv njtlv njxlv nxjlv
ncqvb ncqpb ncqvb nqcvb
tzdjp tzgjp tzdjp tjdzp
gcnjl gknjl gcnjl gjcnl
sxqjz sxqvz sxqjz sqxjz
cgjtd cgwtd cgjtd ctgjd
jzkqg jzfqg jzkqg jzqkg
zbpxv zbphv zbpxv zxpbv
rtwx rzwx rtwx rwtx
pxdzg pxwzg pxdzg pdxzg
rvzxk rgzxk rvzxk rxvzk
ptkvn ptksn ptkvn pvtkn
jzwpq jzfpq jzwpq jwpzq
dtzkj dqzkj dtzkj dzktj
xzlcg xzvcg xzlcg xlczg
czsgk cqsgk czsgk cgszk
kcgdf kcjdf kcgdf kgcdf
lnhxf lnhzf lnhxf lnxhf
mfjxz mfdxz mfjxz mxjfz
cdtzp cbtzp cdtzp cztdp
pfhmg pfhqg pfhmg pmhfg
bfgsv bzgsv bfgsv bsgfv
xrqkv xrwkv xrqkv xrqkv
*Due to an experimenter error, different targets for these two items were presented for the letter
identity and letter position conditions in Experiment 1. This error was corrected in Experiment 2,
where for the first listed item in the letter position condition participants saw xmvqw as the target
(with the distractor xvmqw), and for the second listed item in the letter position condition
participants saw zxqkn (with the distractor zxkqn). Descriptives for illegal nonword targets
reported in the method include xmvqw and zxqkn, as well as mvqwz and jzxqk (i.e.,
N=34).
Table and Figures
Table 1. Age and reading scores for the low, intermediate and high orthographic knowledge
groups
Low (N=21) Intermediate (N=23) High (N=28)
Mean SD Mean SD Mean SD
Age 8.74 1.12 9.44 1.16 10.22 1.10
Irregular word reading Ra
w
13.57 4.70 20.61 1.12 24.54 1.37
%il
e
25.52 22.48 44.96 26.37 55.89 23.05
Regular word reading Ra 26.00 9.04 34.00 3.06 36.50 2.12
40
w
%il
e
30.45 25.98 42.43 34.48 56.54 27.37
Nonword reading Ra
w
14.65 10.20 27.35 7.05 31.21 5.56
%il
e
26.15 27.06 44.35 33.31 48.43 30.38
41
Figure 1. Model estimates of the mean accuracy scores for words and pseudowords as a
function of orthographic skill (low, intermediate and high) and condition (letter identity and
letter position). Error bars denote the standard error of the model estimates.
42
Figure 2. Model estimates of the mean accuracy scores for pseudowords and illegal
nonwords as a function of orthographic skill (low, intermediate and high) and condition
(letter identity and letter position). Estimates are based on the model including the three-way
interaction between orthographic skill, orthographic context, and condition. Note that the
pseudoword condition reflected in this figure is the same as in Figure 1, however, the means
are slightly different as they are estimated from different models. Error bars denote the SE of
the model estimates.
43
Figure 3. Accuracy for words and pseudowords as a function of condition and orthographic
skill, as estimated by model 5 including the three-way interaction between orthographic
context, condition, and orthographic skill. A Z-score of 0 reflects the group mean accuracy.
44
Figure 4. Accuracy for pseudowords and illegal nonwords as a function of condition and
orthographic skill, as estimated by model 5 including the three-way interaction between
orthographic context, condition, and orthographic skill. A Z-score of 0 reflects the group
mean accuracy. Note that the pseudoword condition reflected in this figure is the same as in
Figure 3, however, the slopes are different as they are estimated from different models.
Pseudoword slopes in both Figure 3 and 4 are not statistically different from 0.
45
