Koninklijke Bibliotheek, National Library of the Netherlands
IP1584439464192
Children’s knowledge of single- and multiple-letter grapheme-phoneme correspondences: An exploratory study
Larsen, Linda
Schauber, Stefan Kilian
Kohnen, Saskia
Nickels, Lyndsey
McArthur, Genevieve
text
article
monographic
Early childhood research quarterly
continuing
08852006
0000000010834
51
C
text
Born digital tijdschriften
KB collectiekavel
text
Elektronische Wetenschappelijke Tijdschriften
EWTIJ
10.1016/j.ecresq.2019.12.001
urn:nbn:nl:kb-1584439463936
Automatisch gegenereerd op basis van de EWTIJ XML in release 1.5 van het digitaal magazijn.
EARCHI
1230
S0885-2006(19)30161-9
10.1016/j.ecresq.2019.12.001
The Authors
Fig. 1
Percentage of younger and older children producing the target response (phoneme) for each grapheme (the solid line represents the total sample of children). Items presented in order of decreasing accuracy.
Fig. 1
Table 1
Example entropy calculations as indices of grapheme-phoneme correspondence consistency.
Table 1
Grapheme
Phoneme
Example
Number of occurrences (token frequency)
Probability (token consistency)
Entropy components
B
→
/b/
BIN
509843
1
0.00
Entropy value: H(B) = 0.00
EA
→
/eɪ/
STEAK
13340
0.1027
−0.337249396
→
/ɪə/
HEAR
5085
0.0392
−0.183035783
→
/eə/
PEAR
366
0.0028
−0.023873174
→
/ε/
BREAKFAST
35564
0.2738
−0.511698234
→
/i:/
PEACH
75514
0.5815
−0.454844119
Entropy value: H(EA) = 1.511
Table 2
Study sample demographics by grade (class), group and age.
Table 2
Grade level
N
Group
Age (years)
Boys
Girls
Range
Mean (SD)
Kindergarten
82
39
43
5.45–6.72
6.06 (0.33)
Grade 1
87
34
53
6.45–7. 79
7.13 (0.30)
Grade 2
75
34
41
7.36–8.77
8.14 (0.32)
Grade 3
93
53
40
8.43–9.85
9.18 (0.33)
Table 3
Single- and multiple-letter graphemes used in the LeST (ordered from highest to lowest written frequency).
Table 3
Index card
1
t
n
s
i
l
r
a
d
c
p
e
m
o
b
2
er
g
f
u
v
k
h
j
w
y
ar
z
th
sh
3
ng
ch
ee
x
qu
oo
ph
oi
ai
kn
ay
oa
oy
au
4
wr
ea
gn
aw
ir
wh
ou
ur
igh
Note. LeST = Letter-sound Test.
Table 4
Modeling random effects for students, items (GPCs), classes, and schools. The null models were not significantly different and thus, the random effect for schools was excluded when modeling fixed effects.
Table 4
Null model 1
Null model 2
OR
CI
p
OR
CI
p
Fixed effects
(Intercept)
7.39
3.99–13.70
<0.001
7.39
3.98–13.70
<0.001
Variance
ICC
Variance
ICC
Random effects
Students
0.858
0.110
0.858
0.110
Items
3.075
0.395
3.075
0.395
Classes
0.557
0.072
0.557
0.072
Schools
0.000
0.000
Note. OR = Odds Ratio; CI = Confidence Interval; ICC = Intra Class Correlation.
Table 5
Results from linear mixed effects models analyses.
Table 5
Null model
Model 1: item level predictors
Model 2: item level predictors
Model 3: item level predictors
Model 4: person level predictors
Model 5: person level predictors
OR
p
OR
p
OR
p
OR
p
OR
p
OR
p
Fixed effects
(Intercept)
7.39
<0.001
30.98
<0.001
29.75
<0.001
29.32
<0.001
29.25
<0.001
29.52
<0.001
Grapheme complexity
0.17
<0.001
0.16
<0.001
0.17
<0.001
0.17
<0.001
0.17
<0.001
Phoneme status
0.45
0.013
0.56
0.079
0.52
0.090
0.52
0.091
0.52
0.091
GPC entropy
a
0.56
0.045
0.54
0.034
0.70
0.433
0.70
0.432
0.70
0.431
GPC frequency (token)b
1.81
<0.001
1.85
<0.001
1.84
<0.001
1.84
<0.001
GPC frequency (type)
b
1.78
0.001
Phoneme status × GPC entropy
0.69
0.515
0.69
0.510
0.69
0.511
Own-name
1.54
0.134
1.26
0.402
Age (in months)
b
1.46
0.042
1.47
0.038
Age × Own name
0.39
<0.001
Random effects
Students
0.914
0.914
0.914
0.914
0.915
0.918
Items
3.150
0.789
0.820
0.783
0.780
0.785
Classes
0.526
0.496
0.497
0.496
0.198
0.199
Explained variance (percentage) in relation to the null model
Students
0.00
0.00
0.00
−0.11
−0.44
Items
74.95
73.97
75.14
75.24
75.08
Classes
5.70
5.51
5.70
62.36
62.17
Note. GPC = Grapheme-phoneme correspondence; OR = Odds Ratio. Bold values indicate significant effects.
a
Index of consistency using token count.
b
Log-scaled.
Children’s knowledge of single- and multiple-letter grapheme-phoneme correspondences: An exploratory study
Linda
Larsen
a
b
c
⁎
linda.larsen@fhi.no
Stefan Kilian
Schauber
d
Saskia
Kohnen
b
Lyndsey
Nickels
b
Genevieve
McArthur
b
a
The Department of Special Needs Education, Faculty of Educational Science, University of Oslo, Oslo, Norway
The Department of Special Needs Education
Faculty of Educational Science
University of Oslo
Oslo
Norway
b
ARC Centre of Excellence in Cognition and its Disorders, Department of Cognitive Science, Macquarie University, Sydney, Australia
ARC Centre of Excellence in Cognition and its Disorders
Department of Cognitive Science
Macquarie University
Sydney
Australia
c
Norwegian Institute of Public Health, Oslo, Norway
Norwegian Institute of Public Health
Oslo
Norway
d
Centre for Health Sciences Education, Faculty of Medicine, University of Oslo, Oslo, Norway
Centre for Health Sciences Education
Faculty of Medicine
University of Oslo
Oslo
Norway
⁎
Corresponding author at: Department of Child Health and Development, Norwegian Institute of Public Health, 0213 Oslo, Norway.
Department of Child Health and Development
Norwegian Institute of Public Health
Oslo
0213
Norway
Highlights
•
We investigated children’s knowledge of single- and multiple-letter grapheme-phoneme correspondences (GPCs).
•
There was variability in GPC knowledge across children and across GPCs.
•
GPCs that are vowels, consist of multiple letters or occur less frequently in text were more difficult for children.
•
Type and token GPC frequency did not differentially predict GPC knowledge.
•
Children did not show an advantage for the initial grapheme of their first name when asked to provide its associated phoneme.
Abstract
In this study, we examined Australian children’s knowledge of single- and multiple-letter grapheme-phoneme correspondences (GPCs), and the influence of five different factors – GPC complexity, phoneme status, the child’s name, GPC entropy, and GPC frequency – on GPC knowledge. Data from 337 Australian children enrolled in Kindergarten to Grade 3 were included in the study and analyses were performed using mixed effects models. Results indicate that GPC knowledge varied across children and GPCs, children were almost twice as likely to accurately pronounce single-letter graphemes compared to multiple-letter graphemes, and performance was better for GPCs which occur more frequently in text. GPCs with higher entropy values (less consistent) had close to 40% lower odds of being known by children. The study has practical implications by providing an evidence-based guide for the order in which GPCs should be introduced to children in schools.
Keywords
Grapheme-phoneme correspondence knowledge
Grapheme-phoneme correspondence complexity
Phoneme status
Grapheme-phoneme correspondence entropy
Grapheme-phoneme correspondence frequency
Multilevel modelling
1
Introduction
Learning the names of letters (letter-name knowledge) and the sounds that letters represent in written text (letter-sound knowledge) are two important skills for learning to read an alphabetic script. Letter-name knowledge in kindergarten is a strong predictor of early reading skills (Bond & Dykstra, 1997; Leppänen, Aunola, Niemi, & Nurmi, 2008; Scarborought, 1998; Schatschneider, Fletcher, Francis, Carlson, & Foorman, 2004; Torppa, Lyytinen, Erskine, Eklund, & Lyytinen, 2010), because it appears to assist children’s development of letter-sound knowledge (for a review, see Foulin, 2005). Letter-sound knowledge, in turn, enables children to phonologically decode written words that they have not yet learned (Hulme, Bowyer-Crane, Carroll, Duff, & Snowling, 2012; Share, 1999, 2008).
At least two factors have been suggested to play a role in children’s development of letter-name knowledge: a child’s name and the phonological structure of letter names. Research has shown that children learn the names of letters (and how to write letters) in their own name earlier than other letters, because they have a particular interest in learning to write their own name and are exposed to the written form of their name more often than other written words (containing other letters; Bloodgood, 1999; Justice, Pence, Bowles, & Wiggins, 2006; Treiman & Broderick, 1998). Children also appear to learn the names of letters more easily when those names have a consonant + vowel (CV) structure (e.g., /b/ + /i:/ for the letter B; Treiman & Broderick, 1998; Treiman, Tincoff, & Richmond-Welty, 1997), however, at least one study has not found this result (McBride-Chang, 1999).
While it is not yet clear if the phonological structure of letter names plays a critical role in letter-name knowledge, the research overwhelmingly supports that the phonological structure of letter names has a role in letter-sound knowledge (e.g., Evans, Bell, Shaw, Moretti, & Page, 2006; Justice et al., 2006; McBride-Chang, 1999; Treiman, Tincoff, Rodriguez, Mouzaki, & Francis, 1998; Treiman, Weatherston, & Berch, 1994). This has been explained in terms of the iconicity of letters; that is, most letter names contain the sound that the letter represents (Treiman & Kessler, 2003). For example, children find it easier to learn the sounds for letters with names that contain the corresponding sound (e.g., T is pronounced /ti:/ which contains the sound /t/) than for letters that do not (e.g., Y is pronounced /waɪ/ like the word WHY, which does not contain the sound /j/; Treiman et al., 1998). Further, children find it easier to learn the sounds for letters that have a CV name structure than a VC name structure (Treiman et al., 1998). That is, letter sounds are easier to learn for letters like B and K, where the letter sound is at the beginning of the letter name (i.e., /bi:/ and /keɪ/, respectively) than letters like F and M where the sounds is at the end of the letter name (e.g., /εf/ and /εm/, respectively). Several other factors have been shown to affect the development of letter-sound knowledge including the method of literacy instruction (Piasta & Wagner, 2010), the visual similarity of letters (e.g., b and d) and similarity of sounds (e.g., /m/ and /n/; Carnine, 1976, 1980), and the consistency (or ambiguity) of letter sounds (e.g., D is pronounced /d/ versus C pronounced /k/ or /s/; Huang, Tortorelli, & Invernizzi, 2014; Treiman, 1993; Treiman et al., 1998).
Of these various factors, it is the phonological structure of letter names that has received the most attention in the literature. However, the English writing system is complex and children need to learn many other spelling units than the 26 letters, or graphemes, of the alphabet and their associated sounds, or phonemes. In particular, children must learn many grapheme-phoneme correspondences (GPCs) where more than one letter corresponds to a single sound such as EA-/i:/ as in HEAT and SH-/ʃ/ as in SHIP. However, the phonological structure of letter names cannot account for how children learn these multiple-letter GPCs, since these graphemes (i.e., EA, SH) are compounds of single letters (or graphs) and therefore do not have names per se. Single-letter GPCs may be useful for children’s early reading development and allow children to read simple one-syllable words, however, many words contain multiple-letter GPCs that children must learn in order to become skilled readers who master complex words and texts. It is therefore important to understand what factors are associated with children’s knowledge of single-letter and multiple-letter GPCs. However, surprisingly few studies have focused on multiple-letter GPCs, and even fewer have investigated factors associated with children’s knowledge of single- and multiple-letter GPCs in the same study.
To address this gap in the literature, we explored the association of several factors (see below) with children’s knowledge of both single- and multiple-letter GPCs. Based on previous research, we selected five factors that could be applied to both single- and multiple-letter GPCs: (1) GPC complexity (i.e., single- or multiple-letter grapheme), (2) phoneme status (i.e., consonant or vowel phoneme), (3) the child’s own name (i.e., own-name advantage), (4) GPC entropy (i.e., measure of consistency), and (5) GPC frequency. Below we provide a review of the literature for each of the five factors in order to explain our predictions for the current study.
1.1
GPC complexity
In English, there is a mismatch between the number of speech sounds (phonemes) and the number of letters in the alphabet. There are many more phonemes (approximately 44 depending on the accent) than there are letters, but the concatenation of letters (e.g., C and H to CH) compensates for this (Venezky, 1999). A distinction between GPCs can thus be made in terms of the complexity of the grapheme, that is, whether a grapheme consists of a single letter (e.g., P, E) or multiple letters (e.g., PH, EA). Single- and multiple-letter GPCs differ in (at least) one important way. As the name implies, a multiple-letter GPC contains at least two letters that each map onto a phoneme (or phonemes). This may interfere with learning the phoneme associated with the grapheme. For example, learning the OI-/ɔɪ/ association, as in COIN, may be hampered by knowledge of the O-/ɒ/ and I-/ɪ/ associations as in HOT and HIT, respectively. Indeed, research with adults has shown that when presented with a multiple-letter grapheme, the phonemes associated with the individual letters of the multiple-letter grapheme are activated (Peereman, Brand, & Rey, 2006). Moreover, we have observed that children in the Macquarie Univeristy Reading Clinic often respond with two successive phonemes for multiple-letter graphemes, for example, saying /ɒ/ and then /œ/ for OA, and this pattern has also been reported in acquired dyslexia (e.g., /ʌ/ and /ɪ/ for UI when reading the word SUIT; Newcombe & Marshall, 1985). Research with children does indeed suggest that multiple-letter GPCs are more difficult than single-letter GPCs irrespective of whether the GPC is a consonant or vowel (Frederiksen & Kroll, 1976; Marinus & de Jong, 2008; Olson, Forsberg, Wise, & Rack, 1994). While this may be the case, it is important to note that literacy instruction programs often (if not always) teach single-letter GPCs before multiple-letter GPCs, which means that there may be an order-of-teaching confound. Nonetheless, we predict based on previous research that children in the present study will find multiple-letter GPCs more difficult than single-letter GPCs and thus provide fewer correct target responses to multiple-letter than single-letter GPCs.
1.2
Phoneme status
A distinction can be made in terms of whether a GPC is a consonant or vowel. Several studies have demonstrated that a GPC’s phoneme status (as consonant or vowel) influences children’s GPC knowledge. For example, Stuart and Coltheart (1988) found that young children were less likely to know vowels than consonants when asked to provide the sound associated with each of the letters of the alphabet. A further analysis of children’s errors showed that incorrect responses tended to be letter names, which for vowels are often referred to as the “long vowel sound” (e.g., A pronounced /eɪ/ as in BABY). It was argued that the relatively poorer performance for vowels was not due to the (low) consistency of vowels relative to consonants (i.e., generally, vowel letters have more possible pronunciations than consonant letters) as children also made letter name responses for consonants. However, responding with /eɪ/ to the letter A and /bi:/ to the letter B is different, because A does make the /eɪ/-sound in words (e.g., BABY, TABLE), while B does not make the /bi:/-sound in words (it makes the /b/-sound as in BAG). Graham (1980) assessed primary-school students’ knowledge of vowel and consonant single- and multiple-letter GPCs using an extensive nonword reading task and also found that children struggled more with vowels than consonants. Finally, a more recent study shows that children have more difficulty with multiple-letter GPCs for vowels than consonants in nonword reading (Gilbert, Compton, & Kearns, 2011). Based on this research, we predict that the children in the present study will have more difficulty with vowels relative to consonants.
1.3
Own-name advantage
Children have been found to be better able to learn those letters of the alphabet that occur in their own name (Justice et al., 2006; Treiman & Broderick, 1998; Villaume & Wilson, 1989). Whether the own-name advantage extends to children’s GPC knowledge is less clear. For example, Treiman and Broderick (1998) found that 4- and 5-year-olds were no more likely to know the phoneme associated with a single-letter grapheme if the grapheme was in the child’s name, even if the grapheme was the initial of their name. However, a more recent study by Huang et al. (2014) did find evidence in support of the own-name advantage as children in their study were more likely to know the phoneme associated with individually presented graphemes, if the grapheme occurred as the initial of the child’s first name. Why did Huang et al. (2014) find own-name advantage for GPC knowledge, when Treiman and Broderick (1998) did not? One possibility relates to how the initial grapheme of children’s names was categorised. Huang et al. (2014) “made distinctions between child names in which the letter represents its most common sound [i.e., the target sound] and when it does not”. When asked to provide the phoneme for the grapheme C, children named Cathy or Carl would be expected to show an own-name advantage, whereas Celine or Charlie would not, and this is what was found. Treiman and Broderick (1998) did include this important distinction (in Study 2 only), but they did not find a significant own-name advantage for GPC knowledge when this was assessed using either a free choice task (similar to Huang et al.; i.e., what sound does this letter make?) or a two-choice task (e.g., is this a /də/ or a /pə/?). This could be in part due to Treiman and Broderick’s small sample size (n = 47; Huang et al.: n = 1197), restricted number of (single-letter) graphemes (n = 6) or as the authors suggest, because of a confound in the design. Namely, half of the consonants included have names with a VC structure (i.e., M, R, and S), whereas the other half have names with a CV structure (i.e., D, J, and K) and the children in their study performed significantly better on graphemes with a CV than VC name structure. This effect may have distorted the own-name advantage as it was not factored into the own-name analyses.
But what about multiple-letter GPCs and the own-name advantage – is Charlie more likely to know the CH-/tʃ/ association than Chris, Cathy, or Danielle? The only study that we are aware of reporting on this (Huang et al., 2014) did not find evidence to support an own-name advantage on knowledge of multiple-letter GPCs. In the present study, we wish to further explore this issue. Specifically, we include (1) single- and multiple-letter GPCs, (2) a more extensive list of both vowel and consonant multiple-letter GPCs (e.g., AI for Aidan, PH for Phillip), and (3) we use the name distinction similar to Huang et al. (2014). Given the mixed research findings and relative lack of research on the own-name advantage on knowledge of GPCs and in particular multiple-letter GPCs, we do not have any predictions regarding own-name advantage.
1.4
GPC entropy
English is often described as a deep orthography, which means that the relationship between orthographic and corresponding phonological units is not one-to-one. Some graphemes can represent more than one phoneme (orthography to phonology in reading; e.g., TH as in THIS versus THIN, OO as in NOOK versus NOON) and some phonemes can be represented by different graphemes (phonology to orthography in spelling; e.g., /i:/: SEA and SEE; /εə/: STARE and STAIR). Of interest in this study is the translation from orthography to phonology; consequently, we refer specifically to GPC consistency (referred to as ambiguity in Huang et al., 2014). The consistency of a GPC is defined as the relative frequency of that GPC (e.g., EA-/i:/) as a proportion of all occurrences of that grapheme (i.e., EA). Previous research has shown that GPC consistency affects children’s GPC knowledge with inconsistent GPCs being more difficult for children to learn. This is the case both when GPC knowledge is assessed using a standard GPC task (i.e., the child sees a grapheme and is asked to provide the associated phoneme) and a nonword reading task (Huang et al., 2014; Siegel & Faux, 1989).
While previous studies have used a dichotomous measure of GPC consistency (i.e., consistent versus inconsistent), we suggest that this is a rather crude measure that does not adequately reflect the nuances of GPC consistency. Instead, we use a more sophisticated and sensitive measure, GPC entropy, which takes into account both the number of different pronunciations associated with a grapheme and the relative proportions of these pronunciations (see Section 2 and Table 1
for details of how entropy values are calculated and example calculations). While entropy values for GPCs have been calculated for several different orthographies (Borgwaldt, Hellwig, & de Groot, 2004; Protopapas & Vlahou, 2009), we are not aware of any other studies that have investigated the effect of GPC entropy (as an index of GPC consistency/ambiguity) on children’s GPC knowledge. We predict that children in the present study will show more difficulty with more inconsistent GPCs, that is, those with higher entropy values.
1.5
GPC frequency
The frequency with which children encounter their own name has been suggested as a possible explanation for the own-name advantage. However, there is another frequency index which might be important for children’s development of GPC knowledge; namely, the frequency with which GPCs occur in text. Previous research has shown that children may implicitly learn about GPCs in spelling (e.g., D is pronounced /d/ versus C pronounced /k/ or /s/ Gingras & Sénéchal, 2019; Treiman & Kessler, 2011) and the graphotactic and morphological patterns in spelling (Deacon, Conrad, & Pacton, 2008). Such implicit statistical learning has also been found for developing readers in terms of GPC acquisition, where vowel GPC frequency and the consonantal context of vowels is predictive of GPC knowledge (Steacy et al., 2019; Treiman, Kessler, & Bick, 2003; Treiman, Kessler, Zevin, Bick, & Davis, 2006). While these studies have focused specifically on the pronunciation of a limited set of vowel GPCs during nonword reading, if the observed frequency effect generalizes for vowel and consonant GPCs more broadly, then children should find it easier to learn and therefore know a greater proportion of more frequent GPCs compared to the proportion of less frequent GPCs. If children do not make use of the written frequency of GPCs, then we would not expect to find a difference in knowledge across GPCs with different frequencies.
Several studies have investigated the effect of frequency on GPC knowledge for vowels and consonants; however, critically, these studies have used a measure of grapheme frequency (and not GPC frequency). That is, how often a particular grapheme occurs in written text. We believe it is important to differentiate between the two frequency indices as they measure something quite different, especially when a grapheme has multiple possible pronunciations (e.g., A has at least three different pronunciations; Coltheart, Rastle, Perry, Langdon, & Ziegler, 2001). While the results of previous studies are mixed in terms of the effect of grapheme frequency on GPC knowledge (Ecalle, 2004; Evans et al., 2006; Huang et al., 2014; Treiman, Levin, & Kessler, 2012), we suggest that GPC frequency might be a more appropriate index for predicting children’s GPC knowledge and hence, we include this measure in the present study. We predict that children will have better knowledge of GPCs that occur more frequently in text than GPCs that occur less frequently.
In summary, the current study investigated children’s knowledge of the mappings between letters and sounds, GPCs. GPC knowledge is crucial for children’s reading development and it is therefore important to understand what factors influence this knowledge. As far as we are aware, this is the first study to investigate children’s GPC knowledge using a comprehensive list of both single- and multiple-letter GPCs. More specifically, we explore how five factors – GPC complexity (single- or multiple-letter grapheme), phoneme status (consonant or vowel), a child’s own name (own-name advantage), GPC entropy (an index of consistency), and GPC frequency (frequency in written text) – are associated with children’s knowledge of single- and multiple-letter GPCs.
2
Method
The Macquarie University Human Ethics Committee approved the methods of this research. Children participated in the study with parental consent. Verbal assent was also obtained from each child at the beginning of the testing session.
2.1
Participants
We analysed data from a total of 337 children (177 girls, 160 boys) enrolled in Kindergarten, Grade 1, Grade 2, and Grade 3 who participated in the study. Children ranged in age from 5 years 5 months to 9 years 10 months. See Table 2
for sample demographics.
Children were recruited from two independent co-educational schools in the Sydney metropolitan area. Both schools were selected on the basis that they performed within the average range for Australia on the National Assessment Program – Literacy and Numeracy (NAPLAN), and were in areas that were in the average range on two socio-economic status (SES) measures (Australian Bureau of Statistics, 2006). The schools were typical of Australian schools in that they taught reading using a combination of different methods. One school used a mix of phonics and sight words. Phonics instruction was focused on teaching grapheme-phoneme associations where the pronunciation is the most common (e.g., A pronounced /æ/ as in HAT and CH is pronounced /tʃ/ as in CHECK) and sight words were to be memorised via frequent exposure (i.e., word recognition). This school did not use a particular reading instruction program, but rather used a mix of several programs. The other school predominately used the THRASS method (Davies & Ritchie, 2003). THRASS is a comprehensive teaching method which covers handwriting, reading and spelling. The THRASS method has a strong focus on phonemic awareness (i.e., segmenting words into sounds), and teaches children the 44 phonemes in English and how these may be represented in writing (e.g., the phoneme /f/ can be spelled F as in FIONA, FF as in COFFEE, or PH as in PHONE). Children also learn a word example for each of the sound spelling associations (e.g., BIRD for B, RABBIT for BB), in addition to learning frequently occurring other words (e.g., THE, SAID, YOU). Teachers use a THRASS chart with the 44 phonemes, divided into a consonant and a vowel section, with the most common spellings (graphemes) for each phoneme shown with a word and a picture example. If a phoneme has other less common spellings (e.g., PB for /b/ as in CUPBOARD) these are represented with an asterisk on the chart. Less common spellings are called Grapheme Catch-Alls (GCAs) and are used to teach frequently occurring words that do not have a common spelling pattern (e.g., SAID where the /ε/-sound is spelled with AI). The THRASS method does not have a prescribed teaching sequence for phoneme grapheme associations and teachers can teach these associations in the order they wish (e.g., to match their other resources), and words are introduced as they become useful for children (i.e., depends entirely on the individual child and class room). While the THRASS method teaches sound-to-spelling associations (phoneme-to-grapheme) and not spelling-to-sound associations (grapheme-to-phoneme), research suggests that neither approach is superior, at least during beginning literacy instruction (Callinan & van der Zee, 2010). In the present study, we explored whether there were any difference across schools given that the approaches to literacy instruction varied between the schools, but we did not find this to be the case (see Analytic Approach below).
2.2
Procedure
Children were tested individually on the Letter-Sound Test (LeST; Larsen, Kohnen, Nickels, & McArthur, 2015) in a quiet room of their school in the final term of the academic year (in Australia the academic year has four terms and the final term runs from October to December). The LeST, which took 5−10 min to administer, was administered as part of a battery of reading and spelling tests which formed a large norming study. The test battery was administered in a fixed order to all students (Marinus, Kohnen, & McArthur, 2013; McArthur et al., 2013).
2.3
Measures
2.3.1
GPC knowledge
Children’s knowledge of single- and multiple-letter GPCs was assessed using a single task (no parallel forms), namely, the LeST (Larsen et al., 2015). The LeST comprises 51 items presented on A4 index cards (9–14 items per card) in Arial 24-point font. See Table 3
for a full list of the GPCs included in the LeST. Children were shown the first of the index cards and asked to provide the sound that each of the single- or multiple-letter GPCs makes. The scoring procedure for the LeST is similar to previous studies in that only the most common phoneme associated with a grapheme is accepted – for single-letter vowel GPCs the short vowel sound was accepted, the hard sound for C and G (i.e., /k/ and /g/, respectively) and /ks/ for X.
1
1
There was one exception to the scoring procedure. For TH both the voiced (e.g., THIS) and unvoiced (e.g., THANK) pronunciations were accepted. For this reason, TH was excluded from all analyses.
We refer to this as the target responses or target phonemes. The LeST has high test-retest reliability (ICC = 0.88) and moderate to strong criterion validity (Pearson’s r between 0.49 and 0.70) with the nonword list from the Castles and Coltheart 2 (CC2; Castles et al., 2009) and a 39-item nonwords list (experimental nonwords varying in length and complexity). Internal consistency of the LeST in the present study was good (rKR-20
= 0.88). For a full description of the LeST including how the test was constructed (e.g., item selection), normative data, and further information regarding reliability and validity, see Larsen et al. (2015).
For the present study, GPCs included in the LeST were coded in terms of complexity, phoneme status, entropy and frequency. Below we outline the classifications used and an overview of the coding of each of the GPCs is presented in Appendix A.
2.3.2
GPC complexity
GPCs were classified according to whether the grapheme contained a single letter and coded 0 or contained multiple letters and coded 1. There was an equal number (25) of single-letter and multiple-letter GPCs.
2.3.3
Phoneme status
GPCs were coded 0 if they were consonants and 1 if they were vowels. There were 29 consonant GPCs and 21 vowel GPCs.
2.3.4
Own-name advantage
To code the data for the own-name advantage analysis, we checked if the pronunciation of the initial grapheme of a child’s first name corresponded to a target response on the LeST. If this was the case it was coded 1, otherwise it was coded 0. For example, Charlie was coded 1 as CH is pronounced /tʃ/ and Cathy was coded 1 as C is pronounced /k/ (i.e., both are target responses on the LeST), but Charlotte and Celine were coded 0 as CH and C is pronounced /ʃ/ and /s/, respectively (i.e., neither are target responses on the LeST). First names were scored independently by the first author and a second scorer (a native English speaker) and agreement was high (94%). Any discrepancies were resolved through discussion. A total of 86.9% of children had a first name that began with a target phoneme.
2.3.5
GPC entropy
It is possible to calculate entropy using either type frequency (i.e., number of different word types in which a GPC occurs) or token frequency (i.e., the total number of times all of the words, in which a GPC appears, occur in a written corpus) counts, but following the recommendations of Balota, Yap, and Cortese (2006) and Protopapas and Vlahou (2009), we used token frequency measures in this study. More specifically, the entropy (H) value for each GPC on the LeST was calculated as the negative sum, over the alternative mappings, of the products of each probability times its logarithm:
H
=
-
∑
i
=
1
n
p
i
log
2
p
i
GPCs with an entropy value of zero have a perfect one-to-one mapping (i.e., consistent correspondence) between grapheme and phoneme (e.g., B-/b/). However, as entropy increases, the consistency of the GPC decreases. For example, the grapheme EA has five possible phoneme mappings of which /i:/ (as in TEA) is the most frequent. Overall, the grapheme EA occurs 129,869 times and on 75,514 occurrences it is pronounced /i:/, amounting to a probability (p) of 0.1027 for the mapping EA to /i:/. The probabilities for the other mappings are 0.1027 for /eɪ/, 0.092 for /ɪə/, 0.0028 for /eə/, and 0.2738 for /ε/, respectively (see Table 1). By inserting these probabilities into the entropy formula, we get the entropy value H(EA) = 1.1511, describing the (in)consistency of the pronunciation of the grapheme EA.
2.3.6
GPC frequency
The frequency measure used in the present study refers to the frequency of GPCs rather than graphemes as discussed above. That is, how often a particular GPC occurs in written text. GPC frequency counts are generated from the empirical investigations of large word corpora and there have been several such investigations (e.g., Coltheart et al., 2001; Fry, 2004; Hanna, Hanna, Hodges, & Rudorf, 1966). In the present study, we used the frequency counts by Coltheart et al. (2001), which are based on the monosyllabic words (over 7500 different words) from the widely used CELEX database
2
2
CELEX is based on dictionary data from the Oxford Advanced Learner’s Dictionary (1974, 41,000 entries) and the Longman Dictionary of Contemporary English (1978, 53,000 entries), and text corpus data from the COBUILD/Birmingham corpus (17.9 million words), which has been specifically used to develop CELEX frequency measures. While the CELEX frequencies are based on text samples from adult texts, evidence suggests that the distribution of GPCs in adult and children’s books is similar (Solity & Vousden, 2009).
(Baayen, Piepenbrock, & Gulikers, 1995). In their look-up table, Coltheart and colleagues provide both type and token frequency counts. We chose to use Coltheart and colleagues’ frequency counts as it allowed us to explore if GPC type and token frequencies are differently associated with children’s knowledge of GPCs.
2.4
Analytic approach
All analyses were performed using the R software environment for statistical computing and graphics (R Core Team, 2017). The lme4 package (Bates, Maechler, & Bolker, 2012) was used to perform generalized linear mixed models (GLMM) analyses. We employed a two-step approach. In the first step, we estimated a number of decreasingly complex models incorporating only the intercept as a fixed effect and, successively, random effects for students, items, classes, and schools. The purpose was to determine the appropriate random effects structure and differences between models were assessed using likelihood ratio tests. In the second step, we estimated the random effects structure determined in Step One and included fixed effects for GPC complexity, phoneme status, GPC entropy, GPC frequency and the interaction term between phoneme status and GPC entropy. We then added fixed effects for own-name and age,
3
3
Thank you to two anonymous reviewers for suggesting the inclusion of Age as a predictor.
before estimating a model adding the interaction term between own-name and age. Finally, we explored a complex interaction model in which we included fixed effects for GPC complexity, phoneme status, GPC entropy, and GPC frequency, and the interaction terms between GPC complexity, phoneme status, GPC entropy, GPC frequency and age, respectively.
3
Results
Descriptive data (see Fig. 1
) shows that accuracy – the ability to provide the target phoneme for single or multiple-letter graphemes – varied across items and children. More than 80% of children correctly produced the target phoneme for 23 out of 50 items. Dividing the sample into younger and older children by a median split, we found that more than 80% of younger children correctly produced the target phoneme for 25 out of 50 items. This was 30 out of 50 items for older children (see Appendix B).
As expected, children produced single-letter GPCs (e.g., P-/p/, 98%) more accurately, but two multiple-letter GPCs were also fairly accurate (i.e., SH-/ʃ/, 92%; CH-/tʃ/, 89%). As predicted, children in general were less accurate on vowels than consonants, and in particular multiple-letter vowel GPCs. Only two single-letter vowel GPCs were produced accurately by more than 80% of children (i.e., I-/ɪ/ and O-/ɒ/) and only one multiple-letter vowel GPC (i.e., EE-/i:/) was known by more than 75% of children. The items that children had the most difficulty with were GN-/n/ as in GNOME (39% correct), OU-/aʊ/ as in SOUND (35%), and AU-/ɔ:/ as in AUTOMATIC (25%).
Results from the mixed effects model analyses are presented in Tables 4 and 5
. In the first step, we performed a preliminary multilevel analysis of null models with four levels (i.e., students, items, classes, and schools) and three levels (i.e., students, items, and classes), respectively (see Table 4). In the first model, the random effect of school explained zero variance. Excluding the random effect for school in the second model did not significantly reduce the fit of the model and we therefore collapsed the data across schools for subsequent analyses.
In the second step, we explored the effect of GPC complexity, phoneme status, GPC entropy, GPC frequency (token and type), own-name, and age on children’s GPC knowledge (see Table 5). Results are reported as odds ratios (OR), where a significant OR greater (or less) than 1 indicates a positive (or negative) association with the predictor and an OR of 1 indicates no association with the predictor. For example, an OR of 1.35 indicates a 35% increase in the odds of correctly producing the target phoneme associated with a one unit increase in the predictor while controlling for all other variables. An OR of 0.65 indicates a 35% decrease in the odds. In Model 1 we included GPC complexity, phoneme status, GPC entropy and GPC frequency (token). The results indicated that all predictors were statistically significant in the model, which was in accord with our predictions. Children were less likely to produce the target phoneme for multiple-letter graphemes (OR = 0.17), vowels (OR = 0.45), and GPCs that had higher entropy values (i.e., less consistent; OR = 0.56). Further, more frequent GPCs were almost twice as likely to be accurately pronounced compared to less frequent GPCs (OR = 1.81).
In Model 2 we ran the same analysis, but token frequency was replaced with type frequency to further explore if these two indices of GPC frequency differentially predicted children’s GPC knowledge. This was not the case: more frequent GPCs (using type frequency) still had 78% higher odds of being accurately pronounced compared to less frequent GPCs (OR = 1.78). GPC complexity and GPC entropy were still significant predictors with only very minor changes to the odds ratios (i.e., complexity: OR = 0.16 and entropy: OR = 0.54). However, while the odds ratio for phoneme status remained almost the same, the associated p-value indicated that it was no longer a significant predictor in the model (OR = 0.56). This result may be due to a lack of statistical power based on the (relatively small) sample size (as discussed in the Section 4.1, below).
In Model 3 we included the phoneme status by GPC entropy interaction term to (tentatively) explore whether GPC entropy acts differently across vowels and consonants, and help understand why children seem to be less accurate on vowels compared to consonants. However, neither the interaction nor phoneme status nor GPC entropy were significant in the model.
In Models 4 and 5 we added fixed effects for own-name and age, and the own-name by age interaction term, respectively. In Model 4 we found that age, but not own-name, was a significant predictor in the model (age: OR = 1.46 and own-name: OR = 1.54). GPC complexity and token frequency remained significant in the model, while phoneme status, GPC entropy, and the phoneme type by GPC entropy interaction remained non-significant. In Model 5 we found a significant own-name by age interaction effect (OR = 0.39), indicating that older children benefitted less from the own-name advantage. Finally, in Model 6 (identical to Model 1, but with the inclusion of higher order interaction terms between GPC complexity, phoneme status, GPC entropy, and GPC frequency, and age, respectively), we found that GPC complexity, phoneme status, GPC entropy, and GPC frequency were significant predictors in the model with only minor changes to odds ratios compared to the other models. We also found significant interaction effects, except for age by phoneme status. For example, the largest odds ratio was for the age by GPC complexity interaction (OR = 2.51, p < 0.001), indicating that older children were two and a half times more likely to know complex GPCs. However, we interpret these results with caution as Model 6 failed to converge (see Appendix C for model results).
When comparing the overall fit of the models (fit indices Model 1: AIC = 12,846 and BIC = 12,908; Model 2: AIC = 12,848 and BIC = 12,909; Model 3: AIC = 12,848 and BIC = 12,917; Model 4: AIC = 12,846 and BIC = 12,931; Model 5: AIC = 12,835 and BIC = 12,928), we did not find a significant difference between models and, as can be seen, fit indices are very similar. About 75% of the variance at the item (GPC) level was accounted for across the five fitted models.
4
Discussion
Learning the sounds that letters represent is a crucial skill for children’s reading development in an alphabetic language. It is therefore important to understand how this skill develops and what factors are associated with it. The purpose of this research was to investigate children’s knowledge of single- and multiple-letter GPCs, and the association between GPC knowledge and five different factors, namely: (1) GPC complexity (i.e., single- or multiple-letter grapheme), (2) phoneme status (i.e., consonant or vowel phoneme), (3) the child’s own name (initial GPC in first name), (4) GPC entropy (i.e., measure of consistency), and (5) GPC frequency.
From our descriptive analysis, it was evident that there was variation in children’s knowledge of GPCs, both across children and across GPCs. The majority (more than 80%) of younger children correctly provided the target phoneme for half of the graphemes, while older children unsurprisingly were better and provided the target phoneme for 30 out of 50 graphemes. Further and as predicted, children performed better on single-letter GPCs, with the exception of two multiple-letter consonant GPCs (i.e., SH-/ʃ/ and CH-/tʃ/). A possible explanation as to why children performed so well on these two items is that both are highly consistent (low entropy values) and occur frequently in written text (see Appendix A for entropy and frequency values). This may also account for why PH-/f/ (i.e., PH as in PHONE) was relatively more difficult, as, while consistent, it is less frequent.
From the descriptive statistics children also appeared to exhibit greater difficulty with vowels and in particular multiple-letter vowel GPCs compared to consonants, although this finding may not be as robust as phoneme status (consonant or vowel GPC) was not a significant predictor across the five models where the other predictors are controlled for. Only two single-letter vowel GPCs were known by more than 80% of children and only one multiple-letter vowel GPC was known by more than 75% of children.
It is interesting to note that A-/æ/ (as in CAP) was not among the easiest vowels, even though it is generally recommended that children be introduced to this vowel very early in teaching (e.g., Jolly Phonics; Lloyd, 1992, Letters and Sounds; Primary National Strategy, 2007). Instead, I-/ɪ/ as in PIT and O-/ɒ/ as in POT were the vowels on which accuracy was the highest. If easier letter sounds should be taught before harder letter sounds, our data suggests that I-/ɪ/ and O-/ɒ/ should be taught first, followed by E-/ε/, A-/æ/, and finally U-/ʌ/. However, our findings contrast with Huang et al. (2014), who found A-/æ/ to be the easiest vowel, followed by O-/ɒ/, E-/ε/, I-/ɪ/, and U-/ʌ/. While we do not have a clear explanation for the contrasting results, it is possible that the easily distinguishable letter-form of O may have affected the acquisition of the O-/ɒ/ association and hence, making this item easier for children in our study. Another possibility is the difference in English accent, with the present study using Australian-English and Huang et al. (2014) used American-English. Thus, orthographic features of graphemes and different accents affecting vowels may play a role in children’s letter sound acquisition.
For multiple-letter vowel GPCs, two were particularly difficult for children, namely, OU-/aʊ/ as in SOUND and AU-/ɔ:/ as in AUTOMATIC. Only 35% and 25% of children provided the target phoneme for these, respectively. Many teachers use the rule “when two vowels go walking the first does the talking” to teach multiple-letter vowel GPCs. What this means is that in the word BOAT the O says its name /əʊ/ and the A is silent. While this rule is useful for some multiple-letter vowel GPCs (e.g., OA as in BOAT, EA as in TEA, and IE as in PIE), this is not helpful for learning OU-/aʊ/ or AU-/ɔ:/. It may be confusing for children to learn this rule that applies sometimes and may hamper learning of some multiple-letter GPCs.
We now turn to the results from the mixed effects models and the factors associated with children’s GPC knowledge. First, we looked at GPC complexity, which is concerned with whether the grapheme consists of a single letter (e.g., P, E) or multiple letters (e.g., PH, EA). GPC complexity was a significant predictor in all our models. Single-letter graphemes had around an 83% higher odds of being accurately pronounced compared to multiple-letter graphemes (e.g., Model 1: OR = 0.17). This is in line with our prediction and adds further support to previous research (Frederiksen & Kroll, 1976; Marinus & de Jong, 2008; Olson et al., 1994). It is not surprising that multiple-letter GPCs should be more difficult for children and a possible explanation is that multiple-letter GPCs contain two (or more) single letters that each map onto their own associated phoneme or phonemes. This may cause interference or inhibit access to the phoneme with which the multiple-letter grapheme is associated. Another possible explanation is related to the age at which children are taught different GPCs. When children first begin receiving literacy instruction they are most often (if not always) taught single-letter GPCs before being taught multiple-letter GPCs that allow them to read more complex text. While children in the present study were sampled from schools that use a mixed approach to literacy instruction and hence, minimised the possible confound of type of literacy instruction, it is possible that our results could be confounded by age of acquisition effects. That is, that children had been introduced to and taught single-letter GPCs at an earlier age and only later on had been taught multiple-letter GPCs. Tentative evidence from Model 6 lends indirect support for this as the GPC complexity by age interaction was found to be significant.
Second, we found evidence to suggest that children may find vowels more difficult than consonants. In fact, for every one vowel a child knew, they knew two consonants. However, the effect of phoneme status (i.e., consonant or vowel) on GPC knowledge was only statistically supported in Model 1, despite the odds ratio (and associated confidence interval) being stable across models. This indicates that the effect of phoneme status may lack robustness. Nonetheless, considering the results of the descriptive data and results from the mixed effects models, we cautiously suggest that our results are in accord with our prediction and previous research (Gilbert et al., 2011; Graham, 1980; Stuart & Coltheart, 1988). Also, to our knowledge, this is the first time that the degree of difficulty of vowels relative to consonants has been quantified using both single- and multiple-letter GPCs of which 29 were consonants and 21 were vowels.
We might then ask, what it is that make vowels GPCs more difficult than consonants GPCs for children? Undoubtedly, the acquisition of vowel GPCs is particularly interesting and complex as the name of vowel letters (e.g., name of E is /i:/) is also the sound which these letters and other multiple-letter units are used to represent (i.e., E-/i:/ as in ME, EE-/i:/ as in TREE, and EA-/i:/ as in TEA). Further, there are phonetic similarities between vowel sounds (e.g., /ε/-/ɪ/, /æ/-/ε/) and there are many more vowel sounds than there are letters in the alphabet to represent these. It seems unlikely that the difficulty with vowels should be related to speech sound acquisition, as vowels are acquired earlier than consonants in language development (Priester, Post, & Goorhuis-Brouwer, 2011) and vowels present less difficulty for speech interpretation and intelligibility than do consonants (e.g., Cole, Yan, Mak, Fanty, & Bailey, 1996; Fogerty & Humes, 2012). Further, we note that the acquisition order of vowel GPCs in the present study does not fit clearly with the acquisition order of vowel phonemes in speech (McLeod, 2009). We are not aware of any studies that have used speech sound acquisition order to predict GPC knowledge. However, one study has focused on consonant speech sound acquisition and letter naming of consonants (Justice et al., 2006). This study found that children were better able to name letters that corresponded to consonant speech sounds acquired earlier rather than later in development.
4
4
In a supplementary analysis, we explored whether the consonant-order advantage extended to letter sounding, but we did not find evidence of this.
Returning to vowels, evidence from neuropsychological single-case studies of dysgraphia, aphasia and dyslexia does suggest that (phonologically and orthographically) consonants and vowels are distinct categorical representations (Khentov-Kraus & Friedmann, 2018; Miceli, Capasso, Benvegnù, & Caramazza, 2004). In particular, the study by Khentov-Kraus and Friedmann (2018) is relevant for our study as it reports 23 single cases of vowel letter dyslexia (i.e., selective impairment in reading vowels in words and nonwords). While this may lend support to the possibility that there is something special about vowels that make these more difficult for children there could be a more straightforward explanation. Namely, the spoken similarity of vowels (i.e., there is much greater spoken similarity between vowels than there is among consonants). The more limited discriminability between vowels may result in less distinct memories for vowels, which in turn could leave more room for error during vowel GPC acquisition. Taken together, there may be a complex interaction of several factors when learning vowel GPCs, and more direct studies are required to better understand this.
Third, we investigated the own-name advantage, which states that children should be more likely to know those letters that occur in their name. In this study, we were specifically interested in whether children showed an advantage for the initial grapheme (single- or multiple-letter) of their name when asked to provide its associated phoneme. To our knowledge, only one study has explored the own-name advantage on (a limited set of) knowledge of multiple-letter GPCs in a preliminary analysis. In the present study, we included a comprehensive list of both single- and multiple-letter GPCs. In line with Huang et al. (2014) and Treiman and Broderick (1998), our results do not support the own-name advantage. For example, a child named Charlie was no more likely to know the CH-/tʃ/ association than a child whose name was Chris, Cathy or Danielle. Despite the lack of a significant effect of own-name on GPC knowledge, we did find a significant interaction with age, with older children benefitting less from the own-name advantage than younger children. It is possible that the own-name advantage could be similar to a (specific) frequency effect (i.e., frequency of own name), which washes out over time, possibly with increased exposure to text. That is, older children are exposed to many more words than just their name, whereas younger children are exposed to fewer words overall, thus increasing the impact of seeing their name.
Fourth, in the present study, we used a relatively sophisticated measure to index consistency, namely entropy, which takes into account all the phoneme mappings for a given grapheme and their relative proportion (e.g., Protopapas & Vlahou, 2009). We found that GPC entropy was a significant predictor of children’s GPC knowledge. This is in line with our prediction and provides stronger (more conclusive) evidence for the effect of consistency on children’s GPC knowledge compared to previous studies that have used coarser measures of GPC consistency (Huang et al., 2014; Siegel & Faux, 1989). This result has practical implications for classroom teachers when planning literacy instruction – specifically GPC instruction – as it suggests that GPCs with lower entropy values (i.e., that are less ambiguous) should be introduced before GPCs with higher entropy values (more ambiguous). To do this, teachers may find the GPC entropy values listed in Appendix A a useful resource.
Finally, this study found, as hypothesized, that GPC frequency was a significant predictor of children’s GPC knowledge. Children had around an 80% higher odds of knowing more frequent GPCs relative to less frequent GPCs. GPC knowledge was not differentially predicted by the number of word types a particular GPC occurs in (type frequency) compared to the sum of the frequency of those word types (token frequency). As far as we are aware, the present study is the first to use GPC frequency (rather than letter (or grapheme) frequency) and investigate if type and token frequency have differential effects on GPC knowledge.
4.1
Limitations
The present study has a number of limitations which future studies may wish to address. One is that the sample was collapsed across four different grades in order to increase the sample size (and hence, statistical power) when performing the mixed effects analyses. We acknowledge that our sample size of more than 330 children may seem relatively small compared to some other studies investigating GPC (or letter-name) knowledge (Huang et al., 2014; Justice et al., 2006; Phillips, Piasta, Anthony, Lonigan, & Francis, 2012). Nonetheless, given that this is the first study to investigate children’s knowledge of a comprehensive list of single- and multiple-letter GPCs and factors associated with children’s GPC knowledge, it represents a solid starting point.
A second limitation relates to the fact that children who participated in the present study were drawn from only two schools. It could be argued that, for example, the literacy instruction or method used at each school may have influenced the results. However, both schools were typical of Australian schools and used a combination of phonics and sight words (whole words) to teach reading. One of the schools used the THRASS method predominately. Critically, however, the main findings of the study were robust even when effects of school on accuracy and interactions between school and grapheme complexity, phoneme status, GPC entropy, GPC frequency, and own-name, respectively, were explored in supplementary analyses. Nevertheless, we suggest that future studies would benefit from using a larger sample of children drawn form a wider variety of schools.
We encourage future studies to use GPC (rather than letter or grapheme) frequency (type and token) and GPC entropy as an index of consistency, and to investigate speech sound acquisition order or the orthographic structure of letters/graphemes and how this relates to children’s GPC acquisition. Other child-related factors may also be included such as language background and specific measures of children’s speech or oral language. Finally, we strongly encourage future studies to explore literacy instruction practices within schools and their relationship with children’s GPC acquisition.
4.2
Implications for instruction
Letter-sound (GPC) knowledge directly influences children’s reading development by allowing children to phonologically decode words they have not yet learned (Share, 1999). While children need to learn both single- and multiple-letter GPCs to become proficient readers, many studies to date have only focused on children’s knowledge of single-letter GPCs. The present study is the first to explore children’s knowledge of a comprehensive list of single- and multiple-letter GPCs and several factors associated with children’s GPC knowledge. It therefore has direct implications for literacy instruction. First, the study provides new information regarding the sequence of acquisition of a comprehensive list of GPCs that includes both single- and multiple-letter GPCs, and consonants and vowels. As it seems natural to introduce children to easier GPCs before more difficult GPCs, Fig. 1 (earlier) can be used to guide the order in which teachers and literacy instructors introduce GPCs to beginning readers. As suggested earlier, it seems that the order in which many literacy instruction programs introduce single-letter GPCs does not correspond with what the current research shows about the acquisition order of GPCs. Second, the acquisition order of GPCs is also helpful for teachers as a guide for how much time should be devoted to teaching different GPCs and the amount of repetition for different GPCs. That is, the introduction of earlier acquired GPCs (i.e., easier GPCs) may progress quicker than the introduction of later acquired GPCs (i.e., harder GPCs), and easier GPCs may also require less time dedicated to repetition than harder GPCs.
Third, the present study provides new information regarding why some GPCs are easy for children to acquire and why some are harder to acquire. For example, we found that children struggled the most with GN-/n/ (as in GNOME), OU-/aʊ/ (as in SOUND), and AU-/ɔ:/ (as in AUTOMATIC). The two last items are both multiple-letter vowel GPCs that have relatively low consistency and frequency. In other words, these two items have multiple characteristics that make GPCs harder for children to learn. It is noteworthy that GN-/n/ – also a multiple-letter consonant low in frequency – was surprisingly difficult for children. Why could this be? One possibility is the low frequency alone. However, another possibly is that it has to do with the word position specifics of the item. That is, GN most often appears in word final position (e.g., SIGN, ALIGN) and only rarely in word initial or medial position (e.g., GNOME, POIGNANT). While we did not investigate word position in the present study, it is possible that this could somehow have negatively influenced performance on the LeST where test items were presented in isolation (and in a sense in word initial position).
In conclusion, the present study extends previous research on children’s GPC knowledge by investigating a comprehensive set of both single- and multiple-letter GPCs in the same study. Our findings suggest that children’s GPC knowledge is associated with GPC complexity (i.e., single- versus multiple-letter grapheme), phoneme status (i.e., consonant or vowel), GPC entropy, and GPC frequency. Further, the effect of a child’s own name (initial grapheme) on GPC knowledge is stronger for younger than older children. The present study provides new insight into the order in which single- and multiple-letter GPCs should be introduced to children during beginning literacy instruction where the aim is generally to introduce easier GPCs before more difficult GPCs.
5
Declaration of interest
None.
Acknowledgments
We would like to thank the teachers and children for their time and effort donated to this research. We thank Eva Marinus and Thushara Anandakumar for coordinating the testing and the research assistants for testing the many participants. We also thank Athanasios Protopapas and two anonymous reviewers for very useful comments on the manuscript. This paper was prepared while Linda Larsen was funded by a Postdoctoral Fellowship from the Department of Special Needs Education at University of Olso, Saskia Kohnen was funded by a Macquarie University Research Fellowship (MQRF), Genevieve McArthur was funded by an ARC Australian Research Fellowship (0879556), and Lyndsey Nickels was funded by an ARC Future Fellowship (120100102).
Appendix A
GPC
GPC complexity
Phoneme status
GPC entropy
Type frequency
Token frequency
t-/t/
0
0
0.000
1641
2081414
n-/n/
0
0
0.000
1169
1836402
s-/s/
0
0
1.285
2411
918372
i-/ɪ/
0
1
0.578
818
1359101
l-/l/
0
0
0.000
1639
562756
r-/r/
0
0
0.000
1364
443701
a-/æ/
0
1
1.636
729
1329249
d-/d/
0
0
0.340
1334
1232701
c-/k/
0
0
0.735
724
213457
p-/p/
0
0
0.000
1277
315754
e-/ε/
0
1
1.068
506
396199
m-/m/
0
0
0.000
833
746321
o-/ɒ/
0
1
1.605
483
901467
b-/b/
0
0
0.000
794
508525
er-/ʌr/
1
1
0.061
58
80649
g-/g/
0
0
0.986
513
165363
f-/f/
0
0
0.999
629
523968
u-/ʌ/
0
1
0.801
620
283584
v-/v/
0
0
0.000
286
155171
k-/k/
0
0
0.000
637
257224
h-/h/
0
0
0.000
307
740592
j-/ʤ/
0
0
0.003
122
42149
w-/w/
0
0
0.000
396
641173
y-/j/
0
0
0.969
67
77988
ar-/a:/
1
1
0.622
134
63256
z-/z/
0
0
0.128
83
6507
sh-/ʃ/
1
0
0.000
293
137319
ng-/ŋ/
1
0
0.786
59
49583
ch-/tʃ/
1
0
0.327
296
169611
ee-/i:/
1
1
0.002
194
169656
x-/ks/
0
0
0.000
60
21244
qu-/kw/
1
0
0.000
111
14532
oo-/u:/
1
1
1.142
137
56366
ph-/f/
1
0
0.000
29
4307
oi-/ɔɪ/
1
1
0.000
44
14108
ai-/eɪ/
1
1
0.995
119
33900
kn-/n/
1
0
0.000
40
37724
ay-/eɪ/
1
1
0.000
57
95589
oa-/əʊ/
1
1
0.288
91
15205
oy-/ɔɪ/
1
1
0.000
14
8053
au-/ɔ:/
1
1
0.535
56
7295
wr-/r/
1
0
0.000
43
10073
ea-/ɪə/
1
1
1.511
189
76804
gn-/n/
1
0
0.000
26
3568
aw-/ɔ:/
1
1
0.000
102
17051
ir-/ʌr/
1
1
0.004
75
40128
wh-/w/
1
0
0.280
76
190827
au-/aʊ/
1
1
0.952
101
82304
ur-/ʌr/
1
1
0.000
73
11988
igh-/aɪ/
1
1
0.000
40
55873
Note: GPC = grapheme-phoneme correspondence. GPC complexity: single-letter grapheme = 0 and multiple-letter grapheme = 1. Phoneme status: consonant = 0 and vowel = 1.
Appendix B
GPC accuracy (%) ordered by decreasing accuracy
Item
Total sample
Younger children
Older children
T
99
99
99;1
S
99
99
99
K
99
98
99
N
99
99
98
M
98
97
99
H
98
97
98
P
98
98
98
V
98
97
98
L
97
96
98
W
96
95
97
F
96
97
96
J
96
96
96
R
95
94
96
Z
95
95
95
O
95
95
94
I
94
93
95
D
93
89
96
SH
92
88
96
CH
90
86
93
G
89
88
89
B
87
82
92
X
82
83
80
Y
81
81
81
WH
80
70
90
E
80
80
79
AR
79
68
91
A
79
77
82
EE
78
72
84
AY
76
64
88
C
75
84
66
OY
74
65
83
U
72
81
63
PH
71
53
89
ER
70
51
90
NG
66
61
70
QU
61
49
73
OI
60
52
69
OO
59
55
63
AI
57
47
66
OA
57
49
64
WR
56
37
76
IR
56
44
69
KN
54
36
73
AW
51
34
68
UR
50
34
67
EA
46
38
54
IGH
45
38
52
GN
39
27
52
OU
35
36
34
AU
25
12
39
Appendix C
Results from supplementary age interaction analysis (model failed to converge)
Model 6
OR
p
Fixed effects
(Intercept)
28.72
<0.001
Age (in months)b
0.96
0.0843
Grapheme complexity
0.21
<0.001
Phoneme status
0.43
0.011
GPC entropya
0.55
0.046
GPC frequency (token)b
1.91
<0.001
Age × Grapheme complexity
2.51
<0.001
Age × Phoneme status
0.92
0.120
Age × GPC entropy
0.77
<0.001
Age × GPC frequency (token)
1.18
<0.001
Random effects
Students
0.99
Items
0.83
Classes
0.22
Note. GPC = Grapheme-phoneme correspondence; OR = Odds Ratio. Bold values indicate significant effects.
aIndex of consistency using token count.
bLog-scaled.
References
Australian Bureau of Statistics, 2006
Australian Bureau of Statistics (2006). Census of population and housing: Socio-economic indexes for areas (SEIFA), Australia [Online]. Available: http://abs.gov.au/AUSSTATS/abs@.nsf/DetailsPage/2033.0.55.0012006?OpenDocument
Baayen et al., 1995
R.H.
Baayen
R.
Piepenbrock
L.
Gulikers
The CELEX lexical database (CDROM)
1995
Linguistic Data Consortium, University of Pennsylvania
Philadelphia, PA
Balota et al., 2006
D.A.
Balota
M.J.
Yap
M.J.
Cortese
Visual word recognition: The journey from features to meaning (a travel update)
M.A.
Gernsbacher
Handbook of psycholinguistics (second edition)
2006
Academic Press
London
285
375
Bates et al., 2012
D.
Bates
M.
Maechler
B.
Bolker
lme4. 0: Linear mixed-effects models using S4 classes (2012). R package version 0.9999-1/r1692
2012
Bloodgood, 1999
J.W.
Bloodgood
What’s in a name? Children’s name writing and literacy acquisition
Reading Research Quarterly
34
3
1999
267
342
10.1598/RRQ.34.3.5
Bond and Dykstra, 1997
G.L.
Bond
R.
Dykstra
The cooperative research program in first‐grade reading instruction
Reading Research Quarterly
32
4
1997
348
427
10.1598/RRQ.32.4.4
Borgwaldt et al., 2004
S.R.
Borgwaldt
F.M.
Hellwig
A.M.
de Groot
Word-initial entropy in five languages: Letter to sound, and sound to letter
Written Language and Literacy
7
2
2004
165
184
Callinan and van der Zee, 2010
C.
Callinan
E.
van der Zee
A comparative study of two methods of synthetic phonics instruction for learning how to read: Jolly Phonics and THRASS
The Psychology of Education Review
34
1
2010
21
31
Carnine, 1976
D.W.
Carnine
Effects of two teacher-presentation rates on off-task behavior, answering correctly, and participation
Journal of Applied Behavior Analysis
9
2
1976
199
206
10.1901/jaba.1976.9-199
16795521
Carnine, 1980
D.W.
Carnine
Two letter discrimination sequences: High-confusion-alternatives first versus low-confusion-alternatives first
Journal of Reading Behavior
12
1
1980
41
47
Castles et al., 2009
A.
Castles
M.
Coltheart
L.
Larsen
P.
Jones
S.
Saunders
G.
McArthur
Assessing the basic components of reading: A revision of the Castles and Coltheart test with new norms
Australian Journal of Learning Difficulties
14
1
2009
67
88
10.1080/19404150902783435
Cole et al., 1996
R.A.
Cole
Y.
Yan
B.
Mak
M.
Fanty
T.
Bailey
The contribution of consonants versus vowels to word recognition in fluent speech
1996 IEEE International Conference on Acoustics, Speech, and Signal Processing Conference Proceedings, Vol. 2
1996
853
856
IEEE
Coltheart et al., 2001
M.
Coltheart
K.
Rastle
C.
Perry
R.
Langdon
J.
Ziegler
DRC: A dual route cascaded model of visual word recognition and reading aloud
Psychological Review
108
1
2001
204
256
10.1037/0033-295X.108.1.204
11212628
Davies and Ritchie, 2003
A.
Davies
D.
Ritchie
Teaching THRASS whole-picture keyword phonics: The essential guide to progression and assessment for all teachers of English
2003
THRASS
Deacon et al., 2008
S.H.
Deacon
N.
Conrad
S.
Pacton
A statistical learning perspective on children’s learning about graphotactic and morphological regularities in spelling
Canadian Psychology/Psychologie Canadienne
49
2
2008
118
124
10.1037/0708-5591.49.2.118
Ecalle, 2004
J.
Ecalle
Les connaissances des lettres et l’écriture du prénom chez 1’enfant français avant l’enseignement formel de la lecture-écriture. [Knowing the letters and writing the first name in a French child before formally teaching reading and writing]
Canadian Psychology/Psychologie Canadienne
45
1
2004
111
119
10.1037/h0086975
Evans et al., 2006
M.A.
Evans
M.
Bell
D.
Shaw
S.
Moretti
J.
Page
Letter names, letter sounds and phonological awareness: An examination of kindergarten children across letters and of letters across children
Reading and Writing
19
9
2006
959
989
10.1007/s11145-006-9026-x
Fogerty and Humes, 2012
D.
Fogerty
L.E.
Humes
The role of vowel and consonant fundamental frequency, envelope, and temporal fine structure cues to the intelligibility of words and sentences
The Journal of the Acoustical Society of America
131
2
2012
1490
1501
Foulin, 2005
J.N.
Foulin
Why is letter-name knowledge such a good predictor of learning to read?
Reading and Writing
18
2
2005
125
155
10.1007/s11145-004-5892-2
Frederiksen and Kroll, 1976
J.R.
Frederiksen
J.F.
Kroll
Spelling and sound: Approaches to the internal lexicon
Journal of Experimental Psychology Human Perception and Performance
2
3
1976
361
379
10.1037/0096-1523.2.3.361
Fry, 2004
E.
Fry
Phonics: A large phoneme-grapheme frequency count revised
Journal of Literacy Research
36
1
2004
85
98
10.1207/s15548430jlr3601_5
Gilbert et al., 2011
J.K.
Gilbert
D.L.
Compton
D.M.
Kearns
Word and person effects on decoding accuracy: A new look at an old question
Journal of Educational Psychology
103
2
2011
489
507
10.1037/a0023001
Gingras and Sénéchal, 2019
M.
Gingras
M.
Sénéchal
Evidence of statistical learning of orthographic representations in grades 1–5: The case of silent letters and double consonants in French
Scientific Studies of Reading
23
1
2019
37
48
10.1080/10888438.2018.1482303
Graham, 1980
S.
Graham
Word recognition skills of learning disabled children and average students
Reading Psychology
2
1
1980
23
33
10.1080/0270271800020106
Hanna et al., 1966
P.R.
Hanna
J.S.
Hanna
R.E.
Hodges
E.H.
Rudorf
Phoneme-grapheme correspondences as cues to spelling improvement
1966
US Government Printing Office
Washington DC
Huang et al., 2014
F.L.
Huang
L.S.
Tortorelli
M.A.
Invernizzi
An investigation of factors associated with letter-sound knowledge at kindergarten entry
Early Childhood Research Quarterly
29
2
2014
182
192
10.1016/j.ecresq.2014.02.001
Hulme et al., 2012
C.
Hulme
C.
Bowyer-Crane
J.M.
Carroll
F.J.
Duff
M.J.
Snowling
The causal role of phoneme awareness and letter-sound knowledge in learning to read: Combining intervention studies with mediation analyses
Psychological Science
23
6
2012
572
577
Justice et al., 2006
L.M.
Justice
K.
Pence
R.B.
Bowles
A.
Wiggins
An investigation of four hypotheses concerning the order by which 4-year-old children learn the alphabet letters
Early Childhood Research Quarterly
21
3
2006
374
389
10.1016/j.ecresq.2006.07.010
Khentov-Kraus and Friedmann, 2018
L.
Khentov-Kraus
N.
Friedmann
Vowel letter dyslexia
Cognitive Neuropsychology
2018
1
48
10.1080/02643294.2018.1457517
Larsen et al., 2015
L.
Larsen
S.
Kohnen
L.
Nickels
G.
McArthur
The Letter-Sound Test (LeST): A reliable and valid comprehensive measure of grapheme-phoneme knowledge
Australian Journal of Learning Difficulties
20
2
2015
129
142
10.1080/19404158.2015.1037323
Leppänen et al., 2008
U.
Leppänen
K.
Aunola
P.
Niemi
J.E.
Nurmi
Letter knowledge predicts Grade 4 reading fluency and reading comprehension
Learning and Instruction
18
6
2008
548
564
Lloyd, 1992
Lloyd
Phonics handbook (UK). A handbook for teaching reading, writing and spelling
1992
Jolly Learning
Essex, UK
Marinus and de Jong, 2008
E.
Marinus
P.F.
de Jong
The use of sublexical clusters in normal and dyslexic readers
Scientific Studies of Reading
12
3
2008
253
280
10.1080/10888430802132246
Marinus et al., 2013
E.
Marinus
S.
Kohnen
G.
McArthur
Australian comparison data for the Test of Word reading Efficiency (TOWRE)
Australian Journal of Learning Difficulties
18
2
2013
199
212
10.1080/19404158.2013.852981
McArthur et al., 2013
G.
McArthur
K.
Jones
T.
Anandakumar
L.
Larsen
A.
Castles
M.
Coltheart
A test of everyday reading comprehension (TERC)
Australian Journal of Learning Difficulties
18
1
2013
35
85
10.1080/19404158.2013.779588
McBride-Chang, 1999
C.
McBride-Chang
The ABCs of the ABCs: The development of letter-name and letter-sound knowledge
Merrill-Palmer Quarterly-Journal of Developmental Psychology
45
2
1999
285
308
McLeod, 2009
S.
McLeod
Speech sound acquisition
J.E.
Bernthal
N.W.
Bankson
P.
Flipsen
Jr
Articulation and phonological disorders: Speech sound disorders in children
6th ed.
2009
pp. 63–120 and 385–405
Miceli et al., 2004
G.
Miceli
R.
Capasso
B.
Benvegnù
A.
Caramazza
The categorical distinction of vowel and consonant representations: Evidence from dysgraphia
Neurocase
10
2
2004
109
121
10.1080/13554790409609942
Newcombe and Marshall, 1985
F.
Newcombe
J.C.
Marshall
Sound-by-sound reading and writing
K.E.
Patterson
J.C.
Marshall
M.
Coltheart
Surface dyslexia: Neuropsychological and cognitive studies of phonological reading
1985
Lawrence Erlbaum Associates, Ltd.
London
Olson et al., 1994
R.
Olson
H.
Forsberg
B.
Wise
J.
Rack
Measurement of word recognition, orthographic, and phonological skills
Frames of reference for the assessment of learning disabilities: New views on measurement issues
1994
Brookes Publishing
Baltimore
243
277
Peereman et al., 2006
R.
Peereman
M.
Brand
A.
Rey
Letter-by-letter processing in the phonological conversion of multiletter graphemes: Searching for sounds in printed pseudowords
Psychonomic Bulletin & Review
13
1
2006
38
44
Phillips et al., 2012
B.M.
Phillips
S.B.
Piasta
J.L.
Anthony
C.J.
Lonigan
D.J.
Francis
IRTs of the ABCs: Children’s letter name acquisition
Journal of School Psychology
50
4
2012
461
481
10.1016/j.jsp.2012.05.002
Piasta and Wagner, 2010
S.B.
Piasta
R.K.
Wagner
Learning letter names and sounds: Effects of instruction, letter type, and phonological processing skill
Journal of Experimental Child Psychology
105
4
2010
8
38
10.1016/j.jecp.2009.12.008
20097352
Priester et al., 2011
G.H.
Priester
W.J.
Post
S.M.
Goorhuis-Brouwer
Phonetic and phonemic acquisition: Normative data in English and Dutch speech sound development
International Journal of Pediatric Otorhinolaryngology
75
4
2011
592
596
10.1016/j.ijporl.2011.01.027
Primary National Strategy, 2007
Primary National Strategy
Letters and sounds
Available online at: http://www.standards.dfes.gov.uk/clld/las.html (accessed 12 January 2019)
2007
Protopapas and Vlahou, 2009
A.
Protopapas
E.L.
Vlahou
A comparative quantitative analysis of Greek orthographic transparency
Behavior Research Methods Instruments & Computers
41
4
2009
991
1008
10.3758/BRM.41.4.991
R Core Team, 2017
R Core Team (2017). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/.
Scarborought, 1998
H.E.
Scarborought
Early identification of children at risk for reading disabilityes: Phonological awareness and some other promising predictores
B.K.
Shapiro
P.J.
Accardo
A.J.
Capute
Specific reading disability: A view of the spectrum
1998
York Press
Tmonium, MD
75
119
Schatschneider et al., 2004
C.
Schatschneider
J.M.
Fletcher
D.J.
Francis
C.D.
Carlson
B.R.
Foorman
Kindergarten prediction of reading skills: A longitudinal comparative analysis
Journal of Educational Psychology
96
2
2004
265
282
Share, 1999
D.L.
Share
Phonological recoding and orthographic learning: A direct test of the self-teaching hypothesis
Journal of Experimental Child Psychology
72
2
1999
95
129
10.1006/jecp.1998.2481
9927525
Share, 2008
D.L.
Share
Orthographic learning, phonological recoding, and self-teaching
Advances in Child Development and Behavior
36
2008
31
82
Siegel and Faux, 1989
L.S.
Siegel
D.
Faux
Acquisition of certain grapheme-phoneme correspondences in normally achieving and disabled readers
Reading and Writing
1
1
1989
37
52
10.1007/bf00178836
Solity and Vousden, 2009
J.
Solity
J.
Vousden
Real books vs reading schemes: A new perspective from instructional psychology
Educational Psychology
29
4
2009
469
511
10.1080/01443410903103657
Steacy et al., 2019
L.M.
Steacy
D.L.
Compton
Y.
Petscher
J.D.
Elliott
K.
Smith
J.G.
Rueckl
K.R.
Pugh
Development and prediction of context-dependent vowel pronunciation in elementary readers
Scientific Studies of Reading
23
1
2019
49
63
10.1080/10888438.2018.1466303
Stuart and Coltheart, 1988
M.
Stuart
M.
Coltheart
Does reading develop in a sequence of stages?
Cognition
30
2
1988
139
181
Torppa et al., 2010
M.
Torppa
P.
Lyytinen
J.
Erskine
K.
Eklund
H.
Lyytinen
Language development, literacy skills, and predictive connections to reading in Finnish children with and without familial risk for dyslexia
Journal of Learning Disabilities
43
4
2010
308
321
Treiman, 1993
R.
Treiman
Beginning to spell: A study of first-grade children
1993
Oxford University Press
New York
Treiman and Broderick, 1998
R.
Treiman
V.
Broderick
What’s in a name: Children’s knowledge about the letters in their own names
Journal of Experimental Child Psychology
70
2
1998
97
116
10.1006/jecp.1998.2448
Treiman and Kessler, 2003
R.
Treiman
B.
Kessler
The role of letter names in the acquisition of literacy
Advances in Child Development and Behavior
31
2003
105
135
Treiman and Kessler, 2011
R.
Treiman
B.
Kessler
Similarities among the shapes of writing and their effects on learning
Written Language and Literacy
14
1
2011
39
57
10.1075/wll.14.1.03tre
Treiman et al., 2003
R.
Treiman
B.
Kessler
S.
Bick
Influence of consonantal context on the pronunciation of vowels: A comparison of human readers and computational models
Cognition
88
1
2003
49
78
10.1016/S0010-0277(03)00003-9
Treiman et al., 2006
R.
Treiman
B.
Kessler
J.D.
Zevin
S.
Bick
M.
Davis
Influence of consonantal context on the reading of vowels: Evidence from children
Journal of Experimental Child Psychology
93
1
2006
1
24
10.1016/j.jecp.2005.06.008
Treiman et al., 2012
R.
Treiman
I.
Levin
B.
Kessler
Linking the shapes of alphabet letters to their sounds: The case of Hebrew
Reading and Writing
25
2
2012
569
585
10.1007/s11145-010-9286-3
Treiman et al., 1997
R.
Treiman
R.
Tincoff
E.D.
Richmond-Welty
Beyond zebra: Preschoolers’ knowledge about letters
Applied Psycholinguistics
18
4
1997
391
409
10.1017/S0142716400010900
Treiman et al., 1998
R.
Treiman
R.
Tincoff
K.
Rodriguez
A.
Mouzaki
D.J.
Francis
The foundations of literacy: Learning the sounds of letters
Child Development
69
6
1998
1524
1540
10.1111/j.1467-8624.1998.tb06175.x
Treiman et al., 1994
R.
Treiman
S.
Weatherston
D.
Berch
The role of letter names in childrens learning of phoneme grapheme relations
Applied Psycholinguistics
15
1
1994
97
122
10.1017/S0142716400006998
Venezky, 1999
R.L.
Venezky
The American way of spelling: The structure and origins of American English orthography
1999
Guilford Press
New York
Villaume and Wilson, 1989
S.K.
Villaume
L.C.
Wilson
Preschool children’s explorations of letters in their own names
Applied Psycholinguistics
10
3
1989
283
300
10.1017/S0142716400008638
EARCHI
1230
S0885-2006(19)30161-9
10.1016/j.ecresq.2019.12.001
The Authors
Children’s knowledge of single- and multiple-letter grapheme-phoneme correspondences: An exploratory study
Linda
Larsen
a
b
c
⁎
linda.larsen@fhi.no
Stefan Kilian
Schauber
d
Saskia
Kohnen
b
Lyndsey
Nickels
b
Genevieve
McArthur
b
a
The Department of Special Needs Education, Faculty of Educational Science, University of Oslo, Oslo, Norway
The Department of Special Needs Education
Faculty of Educational Science
University of Oslo
Oslo
Norway
b
ARC Centre of Excellence in Cognition and its Disorders, Department of Cognitive Science, Macquarie University, Sydney, Australia
ARC Centre of Excellence in Cognition and its Disorders
Department of Cognitive Science
Macquarie University
Sydney
Australia
c
Norwegian Institute of Public Health, Oslo, Norway
Norwegian Institute of Public Health
Oslo
Norway
d
Centre for Health Sciences Education, Faculty of Medicine, University of Oslo, Oslo, Norway
Centre for Health Sciences Education
Faculty of Medicine
University of Oslo
Oslo
Norway
⁎
Corresponding author at: Department of Child Health and Development, Norwegian Institute of Public Health, 0213 Oslo, Norway.
Department of Child Health and Development
Norwegian Institute of Public Health
Oslo
0213
Norway
Highlights
•
We investigated children’s knowledge of single- and multiple-letter grapheme-phoneme correspondences (GPCs).
•
There was variability in GPC knowledge across children and across GPCs.
•
GPCs that are vowels, consist of multiple letters or occur less frequently in text were more difficult for children.
•
Type and token GPC frequency did not differentially predict GPC knowledge.
•
Children did not show an advantage for the initial grapheme of their first name when asked to provide its associated phoneme.
Abstract
In this study, we examined Australian children’s knowledge of single- and multiple-letter grapheme-phoneme correspondences (GPCs), and the influence of five different factors – GPC complexity, phoneme status, the child’s name, GPC entropy, and GPC frequency – on GPC knowledge. Data from 337 Australian children enrolled in Kindergarten to Grade 3 were included in the study and analyses were performed using mixed effects models. Results indicate that GPC knowledge varied across children and GPCs, children were almost twice as likely to accurately pronounce single-letter graphemes compared to multiple-letter graphemes, and performance was better for GPCs which occur more frequently in text. GPCs with higher entropy values (less consistent) had close to 40% lower odds of being known by children. The study has practical implications by providing an evidence-based guide for the order in which GPCs should be introduced to children in schools.
Keywords
Grapheme-phoneme correspondence knowledge
Grapheme-phoneme correspondence complexity
Phoneme status
Grapheme-phoneme correspondence entropy
Grapheme-phoneme correspondence frequency
Multilevel modelling
KBJ00000000012480
2020-03-13T04:46:25
S300.1
S300
S0885-2006(19)30161-9
10.1016/j.ecresq.2019.12.001
EARCHI
0885-2006
1230
FLA
NON-CRC
UNLIMITED
UNIT
2020-01-21T10:26:34Z
08852006/v51sC/S0885200619301619/main.xml
181509
MAIN
JA 5.5.0 ARTICLE
FULL-TEXT
08852006/v51sC/S0885200619301619/main.assets/gr1.jpg
33286
IMAGE-DOWNSAMPLED
08852006/v51sC/S0885200619301619/main.assets/gr1.sml
3852
IMAGE-THUMBNAIL
08852006/v51sC/S0885200619301619/main.pdf
913753
MAIN
1.7 6.5
DISTILLED OPTIMIZED BOOKMARKED
08852006/v51sC/S0885200619301619/main.raw
78155
S0885-2006(19)X0004-6
EARCHI
0885-2006
51
C
202032
1
490
S0885-2006(19)30161-9
10.1016/j.ecresq.2019.12.001
379
391
main.pdf
PDF
1.7
local
1584439463936
collectiebehoudsniveau 1
2020-03-17T10:16:21.793+01:00
local
1584439464448
0
SHA-512
fda83fe5f60591ec83d592fd637ad7f0c23e093a8c130a28fded23a8712caa50a7383047d1e9ed13e5b2687b45cef87648297d457ea40f9a062e66b3dcba63af
java.security.MessageDigest
913753
Adobe Acrobat Document
1.7
DIAS
62
DIAS tentative identification
main.pdf
local
1584439464449
0
SHA-512
826bfe21a59ddd9a959f56ac2e671a87c5a8ebcdbebc70d7e363f83c2662e26a2742f3333867d3e042192b137dd7544b810117b2645e35ddf505f98f6becc16b
java.security.MessageDigest
78155
not checked
main.raw
local
1584439464450
0
SHA-512
7239857e1479308f226546e689ed3f722b4ba3df284a50469d5c242ae34dfdd827de77f2381f33846ef6a2709f9e47865fd50d8dd69d17d8a46b6b84cc614e28
java.security.MessageDigest
181509
not checked
main.xml
local
1584439464451
0
SHA-512
c9afe7d3c1eb38efe9774b9d9680abae66406a53fcd9ae3d0cb4e9e38ab71d5c5820a9d3100eb24ef873bc695e8c9ca5dbfa7ab4cc763d5257baeeab12372ba2
java.security.MessageDigest
570
not checked
si1.gif
local
1584439464452
0
SHA-512
23d69c03a2f9294931a0af7954ffb7eb295dd55723332db1ca904494e273b5aa49d913869a49878db79bfe7d8ed7593969c1df683d80799087562f0f0d24b155
java.security.MessageDigest
33286
not checked
gr1.jpg
local
1584439464453
0
SHA-512
83893cfeeec2b8a7615caba96bec368606df58a0114a776efaeb765eb5e424fed2e0c269f8124322ded4085208551414e2e7422f66acfaf2ece9182ee6818341
java.security.MessageDigest
3852
not checked
gr1.sml
local
1584439464454
0
SHA-512
402b910f9c5f15bac4d7ea1e6bcad13eb04e4f86ad788a0ec69af416c39092fd5d965bd66fdbb37904abaabbc68eb58b81f8a88785d2e1b1180859371c549b2a
java.security.MessageDigest
14381
not checked
metadata.xml
free
00001
The Authors
KB-agent-id
1
supplier
KB-owner-id
00001
KB-agent-id
1
Elsevier
organization
The Authors
ingestion
2020-03-17T10:16:21.793+01:00
Connector
software
Digitaal Magazijn release 1.5
ejournals_esp_1
streamprofile
ingestion2020-03-17T18:31:24.241+01:00Generic IngestsoftwareDigitaal Magazijn release 1.5