Spelling of Children in Grade 4: Explicit Rules or Implicit Cues?

(1)

Research Master Child Development and Education Graduate School of Child Development and Education

University of Amsterdam

Research Internship

Student: Lotte Visser (11336706)

Supervisor and first reviewer: Madelon van den Boer Second reviewer: Elise de Bree

(2)

Abstract

Children have difficulty applying spelling rules in relation to long and short vowel spelling. A possible explanation is that children do not rely on explicit rules when spelling these long and short vowel words but on implicit cues. To gain more insight into children’s spelling, the present study examined to what extent Dutch typically developing children in Grade 4 use explicit rules and implicit cues in long and short vowel spelling. The investigated implicit cues were variations in morphology, orthography, and semantics. The spelling ability of 72 children in Grade 4 was measured through a dictation task. Multilevel logistic regression modeling was used to analyze the accuracy of the spelling of ten bisyllabic short vowel words and ten bisyllabic long vowel words. The words differed in morphological consistency, frequency of occurrence in text, and the age at which children generally become familiar with its meaning. Results showed that children did not make many spelling errors. Expectedly, morphological consistency affected children’s spelling. Unexpectedly, familiarity with a word’s orthography and meaning did not affect children’s spelling. Findings suggest that children in Grade 4 rely more on explicit rules when spelling long and short vowel bisyllabic words than on implicit cues. Teachers are advised to offer instruction, practice, and direct feedback to improve children’s use of explicit rules in spelling.

(3)

Spelling of Children in Grade 4: Explicit Rules or Implicit Cues? The ability to spell words is important for being communicative. In addition, if children are not able to spell words correctly this can have negative consequences for how their writing is evaluated by others (Graham, Harris, & Hebert, 2011). Because of its importance and because it is difficult to become a skilled speller (Bosman & van Orden, 1997), spelling has a big role in the curriculum of Dutch primary education (Bosman, 2011). Nevertheless, typically developing children still make many errors when spelling words (Landerl & Reitsma, 2005). Compared to reading, research about spelling and spelling errors is limited. To get more insight into the spelling development of typically developing children and to advise teachers about the focus of their spelling instruction, this study takes a closer look at the spelling of Dutch children in Grade 4. The current study focuses on two spelling patterns that are both known as difficult to learn: the spelling of bisyllabic words with long vowels and the spelling of bisyllabic words with short vowels. Both types of words are phonologically inconsistent: the words are written differently than they sound. Errors are made for both types of words, implying that the rules to spell these words are not applied (de Bree, Geelhoed, & van den Boer, 2018). Therefore, the present study investigates what other cues contribute to spelling words correctly. With the results of this study, more insight is gained into the spelling and spelling difficulties of children.

There is some knowledge about typically developing children’s spelling development and the errors they make when learning how to spell (Treiman, 2017). When children are around two years old, they already have some knowledge about words, such as that words look different from drawings (Treiman & Yin, 2011). They acquire this knowledge implicitly, meaning that they are exposed to words without getting explicit instruction about them. When children are around four years old, they have even implicitly learned how the letters in their writing system are ordered and arranged (Kessler, Pollo, Treiman, & Cardoso-Martins, 2013).

(4)

In Grade 1, at the age of five or six, children in the Netherlands start to receive spelling instruction (Bosman, 2011). The goal of spelling instruction is to teach children the written form of words, or orthography (Treiman & Cassar, 1997). This is done by focusing on several cues that are assumed to contribute to typically developing children’s spelling. The most basic and important cues for spelling are explicitly taught at school, that is phonology (i.e., knowledge about sounds; Seymour, 1997), orthography (i.e., knowledge about the written form of words), and most importantly the link between these two (Bosman, 2011). Children initially learn the link between phonology and orthography when teachers start teaching them sound-letter connections, or the representation of sounds with letters. As a result, children are able to spell phonologically consistent words, which are words in which the letters correspond exactly to the sounds (e.g., maan [moon]; Bosman, 2011). However, not every word can be written exactly as it sounds (e.g., zwanen [swans]). In Dutch, there are quite some phonologically inconsistent words. Of all phonemes in Dutch, 74.3% has two or more written possibilities (Bosman, de Graaff, & Gijsel, 2006).

From Grade 2 onward, children learn several rules to spell phonologically inconsistent words correctly. In the present study, two basic rules are central. The first rule, that is taught in Grade 2, is for the spelling of multisyllabic long vowel words (Bosman, 2011). The second rule, that is generally taught in Grade 3, is for the spelling of multisyllabic short vowel words (Landerl & Reitsma, 2005). In Dutch, long vowels can be spelled with two graphemes (vowel /a:/ in maan [moon]) or with one grapheme (vowel ‘a:’ in zwanen [swans]). The spelling with two graphemes is a consistent mapping from phoneme to grapheme: you hear a long /a:/ and you spell this with two graphemes. Hearing a long /a:/ but writing only one grapheme, is phonologically inconsistent. To spell these types of phonologically inconsistent long vowel words correctly, a child must know about the orthographic rule of multisyllabic long vowel spelling. This rule states that a vowel should be reduced to one if this vowel is heard at the

(5)

end of an open syllable (Bosman et al., 2006). Examples of multisyllabic long vowel words are bonen (beans), muren (walls), tenen (toes), and zwanen (swans). The rule does not only apply to plural nouns (bonen [beans]), but also to other words (lepel [spoon], lopen [to walk]). In Dutch, if the last sound of a syllable is a short vowel, the following consonant should be doubled. Thus, the short ‘a’ in tak (branch) results in a consonant doubling in the plural form takken (branches). Examples of multisyllabic short vowel words are botten (bones), takken

(branches), bruggen (bridges), and spinnen (spiders). The rule does not only apply to plural nouns (botten [bones]), but also to other words (dubbel [double], vallen [to fall]). For both multisyllabic long and multisyllabic short vowel words, there are no exceptions in spelling these types of phonologically inconsistent words: the spelling rule should always be applied to spell these types of words correctly.

Thus, for both phonologically inconsistent long vowel words and phonologically inconsistent short vowel words, a clear rule is taught in school (Bosman, 2011; Landerl & Reitsma, 2005). Therefore, it can be expected that children do not make spelling errors for these types of words, once they acquired the rules. However, Dutch typically developing children still make errors with vowel duration spelling (i.e., spelling of long and short vowels), even after being taught the multisyllabic long vowel rule (de Bree et al., 2018) and/or the multisyllabic short vowel rule (Landerl & Reitsma, 2005; Notenboom & Reitsma, 2007). For the multisyllabic long vowel rule (e.g., zwanen [swans]), children usually write two vowels instead of one (zwaanen). For the multisyllabic short vowel rule (e.g., bruggen [bridges]), children tend to write one consonant instead of two (brugen).

Notably, some of the phonologically inconsistent words that require the rules are spelled correctly and some are not (de Bree et al., 2018; Landerl & Reitsma, 2005). This indicates that children do not apply the explicit rules correctly and consistenly when spelling these words, because if they did, they would not make any spelling errors. Alternatively,

(6)

previous research suggests that children rely on certain implicit cues (i.e., consistencies in the language that are not explicitly taught) when spelling these types of words correctly (de Bree et al., 2018). From an early age, children acquire implicit knowledge, which contributes to their spelling. For example, children’s knowledge about the order and arrangement of letters in their writing system (Kessler et al., 2013). A previous study (de Bree et al., 2018) showed that for children in Grade 1 and Grade 2, semantics, orthography, and morphological

consistency are important implicit cues that affect children’s spelling. In the present study on Grade 4 children, I also focus on these three implicit cues.

The first implicit cue that is assumed to play a role in children’s spelling ability is semantics, which indicates a word’s meaning or its associations (Lupyan & Lewis, 2019). Children can acquire a word’s meaning explicitly at school or at home, but also implicitly through reading. According to the lexical quality hypothesis (Perfetti & Hart, 2002), a child should know the meaning, the orthography, and the phonology of a word to spell it correctly. Not only should these three word forms be known, they should also be linked to each other. For example, when children are asked to spell a certain word and hear how it is pronounced, they immediately think of its meaning and written form, which results in the correct spelling of the word. In Dutch spelling education, sound-letter connections and spelling rules are taught to strengthen the link between phonology and orthography. Although semantics is assumed to be important according to the lexical quality hypothesis (Perfetti & Hart, 2002), it is not part of Dutch spelling education. Therefore, semantics is considered an implicit cue because children can rely on it without receiving explicit instruction about it. In addition to theory, research shows that children who know the meaning of a word are more likely to recognize the correct spelling of a word (Ouellette & Fraser, 2009).

Another implicit cue that is assumed to play a role in children’s spelling ability is orthography. As mentioned before, orthography indicates the word’s written form (Treiman &

(7)

Cassar, 1997) and is the most important aspect of reading and spelling development. In Dutch spelling education, orthography is explicitly linked with phonology. However, orthography can also be acquired through exposure (i.e., reading) and can, therefore, also be considered an implicit cue. Similar to semantics, both theory (Perfetti & Hart, 2002) and research (Tamura, Castles, & Nation, 2017) show that children who are exposed to the written form of words are more likely to spell these words correctly.

Third, morphological consistency can contribute to word spelling. Morphology

concerns the representation of a word’s meaning in its spelling (Treiman & Cassar, 1997). For example, a morphologically consistent word in Dutch is taarten (English: cakes; singular: taart), because the singular form and meaning of taart is also present in the plural form

taarten. A morphological inconsistent word is ‘manen’ (English: moons; singular: maan),

because man does not represent the meaning of the singular form of manen. In Dutch, morphology has a weaker influence on spelling than phonology (Landerl & Reitsma, 2005). Therefore, morphology is not explicitly taught in Dutch spelling education. Nevertheless, some words in Dutch are not morphologically consistent, meaning that morphological cues can interfere with the correct spelling. This is also shown in a study by Landerl and Reitsma (2005) who found that children make many spelling errors in morphologically inconsistent words. Thus, it is also important to consider morphological consistency as implicit cue because it might influence the spelling of a word.

The joint role of semantics, orthography and morphology is already investigated in a Dutch sample of typically developing children (de Bree et al., 2018). In this study, 113 children in Grade 1 and 59 children in Grade 2, who were all fluent speakers of Dutch, participated. Their long vowel spelling of phonologically inconsistent words was measured through a dictation task. There were four categories of phonologically inconsistent words, which together represented the three implicit cues: monomorphemic words familiar in

(8)

semantics and orthography (water [water]), monomorphemic words familiar in semantics but not in orthography (bami [bami]), monomorphemic words unfamiliar in semantics and

orthography (krater [crater]), and inflected (i.e., morphologically inconsistent) words familiar in semantics and orthography (manen [moons]). Particularly the results of Grade 2 are

interesting because children in Grade 2 had already been taught the rule for spelling

multisyllabic long vowel words, but they still made many errors. These children spelled more words correctly when they were familiar in orthography and semantics than when they were only familiar in semantics or not familiar at all. Furthermore, they made more errors with morphologically inconsistent words than with monomorphemic words. These results imply that children do not consistently apply the rule for multisyllabic long vowel spelling, but that they rely on implicit cues instead. Especially when children have knowledge of all three implicit cues and if these cues all contribute to the correct spelling of a word, children in Grade 1 and Grade 2 are more likely to spell phonologically inconsistent long vowel words correctly.

It is also important to investigate the role of implicit cues for other spelling rules and in older children’s spelling. For example, it might be that the use of implicit cues and explicit rules differs for children of different ages (van der Ven & de Bree, 2019) and for different types of words. Therefore, in the current study it is examined whether semantics, orthography, and morphological consistency affect the spelling of typically developing children in Grade 4 for phonologically inconsistent long vowel words and short vowel words. Children’s spelling is measured through a dictation task. The focus was on children in Grade 4 because they have already been taught the rule for long vowel words and the rule for short vowel words.

Therefore, it was possible to investigate if they relied on explicit rules or implicit cues when spelling words.

(9)

Based on previous research, it is expected that children make spelling errors even after being taught the rule for multisyllabic long and short vowel words (de Bree et al., 2018; Landerl & Reitsma, 2005). Therefore, implicit cues might affect typically developing children’s spelling in Grade 4. In the current study, there are three expectations concerning the implicit cues. First, the earlier words are familiar in meaning, the better their vowels and consonants are spelled. In addition, the more exposure to a word’s orthography, the better its vowels and consonants are spelled. Lastly, vowels and consonants in morphologically consistent words are better spelled than vowels and consonants in morphologically inconsistent words.

Method Participants

The sample consisted of 72 children in Grade 4. There were approximately as many boys (54.9%) as girls (45.1%) in the sample. Children were between 8.4 years and 11 years old (M = 9.9 years; SD = 5.92 months). Most of the children spoke Dutch at home (95.8%). The remaining children spoke Polish (n = 1) and Surinamese (n = 2) at home. The children were recruited through four different schools. Of these schools, all children in Grade 4 participated, there were no inclusion or exclusion criteria applied.

Materials

Target words. A total of 20 words was analyzed in this study. The words were equally divided in two different categories: long vowel spelling (muren [walls]) and short vowel spelling (botten [bones]) (see Table 1). All bisyllabic long vowel words were morphologically inconsistent and all bisyllabic short vowel words were morphologically consistent. The 20 words differed in frequency based on SUBTLEX (Keuleers, Brysbaert, & New, 2010) and in Age of Acquisition (AoA; Brysbaert, Stevens, De Deyne, Voorspoels, & Storms, 2014). Frequency and AoA for each word are shown in Table 1. The higher the

(10)

frequency, the more often children are usually exposed to a word’s orthography. The AoA indicates the mean age children are familiar with a word’s meaning. The two sets of words did not differ from each other in frequency, t(18) = 1.661, p =.114, and AoA, t(18) = 0.444, p = .662.

Table 1

Frequencies and the Age of Acquisition (AoA) of the Words

Bisyllabic long vowel words Bisyllabic short vowel words

Word Translation Frequency1 AoA2 Word Translation Frequency1 AoA2

bonen beans 11.14 6.53 bonnen receipts 2.15 7.21

boten boats 12.14 4.50 botten bones 18.82 6.71

knopen buttons 11.89 6.23 knoppen knobs 3.22 5.12

maten sizes 11.21 7.38 matten mats 0.87 6.56

taken tasks 5.26 6.45 takken branches 3.66 5.53

blaren blisters 2.08 8.48 bessen berries 3.36 6.08

muren walls 26.00 4.98 blokken blocks 6.22 5.15

noten nuts 8.14 5.54 bruggen bridges 4.00 5.47

tenen toes 12.33 4.08 spinnen spiders 6.31 4.91

Zwanen swans 1.19 6.10 tassen bags 7.50 5.37

Note. 1Frequency per million; 2Age of Acquisition is based on the word’s singular form.

Dictation task. The target words were embedded in short sentences that were dictated to the children. The target word was pronounced, followed by a sentence in which its singular form was embedded and then a sentence in which its plural form was embedded. Next, the target word was read again two times. An example is “Zwanen. Hier zwemt één zwaan. Daar

(11)

zwemmen veertien zwanen. Schrijf op: zwanen … zwanen.” [“Swans. One swan is swimming here. Over there, fourteen swans are swimming. Write down: swans … swans.”]. Children’s spellings were scored as correct or incorrect in two different ways. First, a word was

distinguished as an incorrect or the correct spelling. Thus, zwaanen and swanen were both considered wrong and only zwanen was considered right. Second, a word was only scored as incorrect if the vowel/consonant was not spelled according to the rule. Thus, zwaanen was considered wrong, but not swanen. In addition, the errors were divided in four categories: (1) a vowel error (zwaanen, beessen), (2) a consonant error (zwannen, besen), (3) both a vowel error and a consonant error (zwaannen, beesen), or (4) another error. The dictation task was a reliable measure for both the phonologically inconsistent long vowel words (α = .851) and the phonologically inconsistent short vowel words (α = .835).

Procedure

This study was part of a larger study in which the spelling of different kinds of words was measured in children in Grade 2 to children in Grade 5. Parents gave permission via passive informed consent. The children were tested in the second semester of the school year (march-april). The dictation task consisted of 60 words and was therefore conducted in two class-based sessions of 30 words each. Each session took approximately 30 minutes. The ethics committee of the University of Amsterdam approved the conduction of this study (2018-CDE-8741).

Analysis

To test if orthography, semantics, and morphology affected children’s spelling of long and short vowel words, multilevel logistic regression models were used. Because the words (level-1) were nested in a child (level-2), multilevel modeling was used, as words written by the same child might be more similar than words written by different children. Independent variables were word frequency (i.e., orthography), the Age of Acquisition (i.e., semantics), and morphological consistency. Two different dependent variables were analyzed: a binary

(12)

variable in which a word was scored as correct (1) if it was spelled correctly and incorrect (0) if it was not, and a binary variable in which a word was scored as correct (1) if the spelling rule for spelling bisyllabic long vowel or short vowel words was applied and incorrect (0) if it was not. The dependent variables thus followed a binomial distribution, indicating that

multilevel logistic regression models were used.

To come to the final model, deviance tests were conducted and  = .05 was used as a criterion for significance (Agresti & Franklin, 2015). Effect sizes of deviance tests were interpreted based on Cohen’s (1988) criteria: below 2% was considered negligible, between 2 and 13% small, between 13 and 26% moderate and above 26% large. Odds ratios (OR) were interpreted based on Chen, Cohen, and Chen’s (2010) criteria: OR = 1.68 (small), 3.47 (medium), and 6.71 (large). Analyses were carried out in RStudio using the lme4 package with maximum likelihood estimation with Laplace Approximation (Finch, Holden, & Kelley, 2014).

Results

Dictation data of twenty target words was available for 72 children in Dutch Grade 4. However, three children were not present at one of the two dictation sessions. Therefore, information of 30 words was not available (2%). Inspection of the data showed that

orthography was not normally distributed, but skewed to the right. Thus, most words had a low frequency of occurrence in text (M = 7.875, SD = 6.213). Regarding semantics, the words were quite similar in age of acquisition of meaning, resulting in limited variation (M = 5.919, SD = 1.038). Mean scores for all words on both outcomes are shown in Table 2 and Table 3.

Both tables show that many words have a high proportion rule applied correctly score and a high proportion word spelled correctly score. Nevertheless, there is still some variation in the words, indicating that some children are more deviated from a word’s mean than other children. Several models were fitted and selected to test if orthography, semantics, and

(13)

morphology affected children’s spelling of long and short vowel words. These models are shown in Table 1 and Table 2 of the Appendix.

Table 2

Mean Score Rule Applied Correctly and Standard Deviation per Word

Word Translation M SD Word Translation M SD

taken tasks 0.957 0.205 takken branches 0.889 0.316 blaren blisters 0.903 0.298 bessen berries 0.971 0.168 muren walls 0.958 0.201 blokken blocks 0.917 0.278 noten nuts 0.986 0.120 bruggen bridges 0.847 0.362 tenen toes 0.928 0.261 spinnen spiders 0.957 0.205

(14)

Table 3

Mean Score Word Spelled Correctly and Standard Deviation per Word

Word Translation M SD Word Translation M SD

taken tasks 0.870 0.339 takken branches 0.889 0.316 blaren blisters 0.875 0.333 bessen berries 0.971 0.169 muren walls 0.903 0.298 blokken blocks 0.917 0.278 noten nuts 0.928 0.261 bruggen bridges 0.847 0.362 tenen toes 0.870 0.339 spinnen spiders 0.957 0.205

zwanen swans 0.861 0.348 tassen bags 0.928 0.261

To check if multilevel modelling was needed, empty models with a random intercept (Model 0) were fitted to calculate the intraclass correlation coefficient (ICC) for chances of applying spelling rules correctly and chances of spelling words correctly. For both outcomes, the ICC was 0.53, which indicates that words spelled by one child were much alike: variance at level-1 (within-group variance) was relatively small (2 = 3.289) and variance at level-2 (between-group variance) was relatively large (02 = 3.750 for applying rules correctly, 02 = 3.667 for spelling words correctly). Therefore, for both outcomes, the random intercept was kept in the models.

(15)

Effect of Orthography and Semantics Dependent on Morphological Consistency First, the fullest model was fitted to test if there was an effect of semantics and orthography on the chances of applying the spelling rules correctly and if these effects differed for morphologically inconsistent and consistent words. This was examined in a multilevel logistic regression model including the main effects of orthography, semantics, morphological consistency, and the interactions between orthography and morphological consistency and semantics and morphological consistency (Model 1). Results showed that Model 1 did not explain the data significantly better than a model with only main effects of orthography, semantics, and morphological consistency (Model 2), 2(2) = 0.308, p = .857, PRD = 0.052% (negligible effect).

For the other outcome, chances of spelling words correctly, models with the same predictors were fitted and compared. Results showed that the higher-order model (Model 1) did not explain the data significantly better than a model with only main effects of

orthography, semantics, and morphological consistency (Model 2), 2(2) = 0.003, p = .999, PRD = 0.000% (negligible effect). In sum, the effect of orthography and semantics on

chances of applying rules correctly and chances of spelling words correctly were the same for morphologically consistent and inconsistent words.

Combined Effects of Orthography, Semantics, and Morphological Consistency

Next, the expected positive effect of orthography and morphological consistency and negative (because lower scores are associated with an earlier acquisition of the meaning) effect of semantics on the chances of applying the spelling rules correctly was tested. Results showed that Model 2 did not explain the data significantly better than the empty model (Model 0), 2(3) = 7.35, p = .061, PRD = 1.209% (negligible effect). For chances of spelling words correctly, models with the same predictors were fitted and compared. Results showed that Model 2 did not explain the data significantly better than the empty model (Model 0),

(16)

2(3) = 7.466, p = .058, PRD = .927% (negligible effect). In sum, when combined together, orthography, semantics, and morphological consistency did not have an effect on chances of applying spelling rules correctly and chances of spelling words correctly.

Effect of Morphological Consistency

Subsequently, the expected positive effect of morphological consistency on chances of applying rules correctly was examined separately. Results showed that a random intercept model with only morphological consistency as predictor (Model 3) explained the data significantly better than Model 0, 2(1) = 6.051, p = .014, PRD = 0.995% (negligible effect). Unexpectedly, morphologically consistent words had a significantly lower log-odds of applying spelling rules correctly than morphologically inconsistent words, b = 0.581, z = -2.428, p = .015. The estimated odds of applying the spelling rules correctly for

morphologically consistent words were e-0.581 = 0.559 times as high as the odds for

morphological inconsistent words, 95% CI [0.347, 0.889]. In other words, the estimated odds of applying the spelling rules correctly for morphologically inconsistent words is 1.788 times as high as for morphologically consistent words. This is a small effect. In terms of

probabilities, expected chances for applying the spelling rule correctly is 97.20% for morphologically consistent words and 98.41% for morphologically inconsistent words.

For chances of spelling words correctly, a similar model was fitted (Model 3) and compared to Model 0 to test if morphological consistency positively influenced the chances of spelling words correctly. Results showed that Model 3 explained the data significantly better than Model 0, 2(1) = 7.015, p = .008, PRD = 0.871% (negligible effect). As expected, morphologically consistent words had a significantly higher log-odds of spelling words correctly than morphologically inconsistent words, b = 0.523, z = 2.624, p = .009. The estimated odds of applying the spelling rules correctly for morphological consistent words were e 0.523 = 1.687 times as high as the odds for morphological inconsistent words, 95% CI

(17)

[1.145, 2.503]. This is a small effect. In terms of probabilities, morphologically consistent words have an expected chance of 97.02% to be spelled correctly, whereas morphologically inconsistent words have an expected chance of 95.08% to be spelled correctly.

Effect of Orthography Above Morphological Consistency

To test the expected additional positive effect of orthography over and above morphological consistency on chances of applying the spelling rules correctly, orthography was added to Model 3, resulting in Model 4. Results showed that Model 4 did not explain the data significantly better than Model 3, 2(1) = 1.30, p = .254, PRD = 0.216% (negligible effect). For chances of spelling words correctly, a similar model was fitted (Model 4) and compared to Model 3 to test if orthography had an additional positive effect on chances of spelling words correctly over and above morphological consistency. Results showed that Model 4 did not explain the data significantly better than Model 3, 2(1) = 0.308, p = .579, PRD = 0.038% (negligible effect). In sum, orthography did not have an additional effect on chances of applying the spelling rules correctly and chances of spelling words correctly over and above morphological consistency.

Effect of Semantics Above Morphological Consistency

Subsequently, to test the expected additional negative effect of semantics over and above morphological consistency on chances of applying the spelling rules correctly,

semantics was added to Model 3, resulting in Model 5. Results showed that Model 5 did not explain the data significantly better than Model 3, 2(1) = 0.077, p = .781, PRD = 0.013% (negligible effect). For chances of spelling words correctly, a similar model was fitted (Model 5) and compared to Model 3 to test if semantics had an additional positive effect on chances of spelling words correctly over and above morphological consistency. Results showed that Model 5 did not explain the data significantly better than Model 3, 2(1) = 0.021, p = .885, PRD = 0.000% (negligible effect). In sum, semantics did not have an additional effect on

(18)

chances of applying the spelling rules correctly and chances of spelling words correctly over and above morphological consistency.

Types of Errors

To get more insight into the spelling of children in Grade 4, the proportion rule applied correctly, proportion words spelled correctly, and types of errors are analyzed for bisyllabic long vowel words and bisyllabic short vowel words. Proportion correct scores are shown in Table 4. The table shows that proportion rule applied correctly scores are higher (92.6%) than proportion words spelled correctly scores (89.2%). This is logical because all errors are considered in proportion words spelled correctly whereas in proportion rule applied correctly only errors related to the spelling rule are considered. The table also shows that, when comparing the two spelling rules, bisyllabic long vowel words have a higher proportion rule applied correctly score, whereas bisyllabic short vowel words have a higher proportion word spelled correctly score. Furthermore, the table shows that children in Grade 4 made some spelling errors. The types of possible errors were vowel errors, consonant errors, both vowel and consonant errors, or other errors. For bisyllabic long vowel words (e.g., boten [boats]), children mostly made consonant errors (botten), followed by vowel errors (booten). For bisyllabic short vowel words (e.g., botten [bones]) only consonant errors were made (boten). Thus, from all errors, most were consonant errors (71%). However, consonant errors occurred in only 8% of all words.

(19)

Table 4

Mean Score Rule Applied Correctly, Words Spelled Correctly and Types of Errors for

Bisyllabic Long Vowel Words and Bisyllabic Short Vowel Words

Type of word Rule applied correctly Word spelled correctly Vowel error Consonant error Vowel and consonant error Other error Bisyllabic long vowel words 0.940 0.871 0.057 0.068 0.003 0.001 Bisyllabic short vowel words 0.911 0.913 0.000 0.085 0.000 0.001 All words 0.926 0.892 0.028 0.077 0.001 0.001 Discussion

In this study, long and short vowel spelling of Dutch children in Grade 4 was assessed. Target words were all phonologically inconsistent but differed in morphology (consistent vs. inconsistent), frequency of occurrence in text (orthography), and the age children generally become familiar with its meaning (semantics). The goal was to evaluate whether these

implicit cues (i.e., consistencies in the language that are not explicitly taught) or explicit rules affected children’s spelling. In addition, the types of errors that were made were analyzed. The study yielded four main results. First, semantics and orthography did not affect children’s long and short vowel spelling. Second, morphology affected children’s long and short vowel spelling. Third, children in Grade 4 did not make many errors when spelling phonologically inconsistent long and short vowel words. Fourth, the errors that occurred the most in

(20)

The finding that semantics and orthography did not have an effect on children’s spelling in Grade 4 was unexpected. Although previous research found that words familiar in meaning and orthography positively affected Dutch children’s spelling in Grade 1 and 2 (de Bree et al., 2018), this is not found for children in Grade 4. An explanation for this finding could be that, in line with previous research (Landerl & Reitsma, 2005), most children in Grade 4 spelled the target words correctly. In other words, children did not make many spelling errors, resulting in limited variation in the spelling. Therefore, it is difficult to investigate what predicts children’s spelling. For future research, it is thus recommended to include more difficult words to increase the variation in spelling. For example, multisyllabic words consisting of three or more syllables or words with other complexity. The absence of an effect of semantics and orthography on spelling could also mean that children in Grade 4 rely on the explicit rules instead of on implicit cues, because they spell a lot of words correctly. It thus seems that these children rely on the explicit rules, but this is cannot be concluded from the spellings of a dictation task. Therefore, future research should explicitly ask children why they spell a word in a certain way when measuring spelling ability through a dictation task (de Bree et al., 2018). By doing this, more insight into if and when explicit rules overrule implicit cues is gained.

The finding that morphology affected children’s spelling in Grade 4 was expected. Unexpectedly, however, the results differed for the two outcome measures. First, and in line with previous research (Landerl & Reitsma, 2005), morphologically consistent words were spelled better than morphologically inconsistent words. In the other outcome measure, it was checked if the spelling rules were applied correctly or not. This resulted in another and unexpected finding: rules were applied correctly more often for morphologically inconsistent words than for morphologically consistent words. The different results can be explained by the (types of) errors the children made. When looking in-depth into the spelling errors it

(21)

becomes visible that most errors are made for morphologically inconsistent words (e.g, taken [tasks], which tended to be spelled with an extra vowel (taaken) or an extra consonant

(takken). Morphologically consistent words (e.g., takken [branches]) only tended to be spelled with one consonant (taken). Focusing on the outcome of applying the spelling rules correctly or not, the consonant errors of morphologically inconsistent words are not considered

anymore, suddenly resulting in less errors in morphologically inconsistent words than in morphologically consistent words. This can thus explain why morphologically inconsistent words had a higher proportion rule applied correctly score than morphologically consistent words.

Despite this finding, the outcome measure in which a word was spelled correctly or not seems to be a better representation of the results than the outcome measure in which the spelling rule was applied correctly or not. The reason for this is that consonant errors in morphologically inconsistent words, which are only visible in the outcome measure in which a word was spelled correctly or not, are also relevant for insight into spelling errors and spelling patterns so that teachers can give attention to these typically made errors. Because children in Grade 4 did not make other errors besides consonant and vowel errors, it is particularly relevant to analyze if they spelled a word correctly or not. However, younger children might also make other types of errors (Drijver, 2018). Therefore, for these children it might be valuable to also analyze the outcome measure in which the spelling rule is applied correctly or not. For future research it is recommended to look at both outcome measures to see which one represents the results best because errors might differ for children of various ages.

The finding that consonant errors occurred the most in children’s spelling can also be explained. Morphologically inconsistent words (e.g., taken [tasks]) tended to be spelled with an extra vowel (taaken), following the common phoneme-grapheme associations, or an extra

(22)

consonant (takken). The addition of an extra consonant might be explained because takken (branches) also is a word that exists in Dutch. Therefore, children might be confused when asked to spell taken because they are also familiar with its parallel form takken (Drijver, van den Boer, & de Bree, in revision). Morphologically consistent words (takken [branches]) only tended to be spelled with one consonant (taken), following the common phoneme-grapheme associations and because children are also familiar with the existing parallel form taken (tasks). Because the vowel in morphologically consistent words follows phoneme-grapheme associations, it is not surprising that children do not make vowel errors when spelling

morphologically consistent words. Thus, these results suggest that consonant errors are made for both types of words, whereas vowel errors are only made for morphologically inconsistent words. Future research should focus more on words with parallel forms and see if children make more errors for these words than words without parallel forms. In doing so, more insight into the types of errors is gained and can be valuable for teachers’ spelling instruction.

Despite that this study adds to the current knowledge about typically developing children’s spelling, it also has several limitations. First, the findings might not be completely reliable because only twenty target words were analyzed. To increase the chances of finding small true effects, future research should not necessarily include more children, but should include more target words to be sure of the current findings. For multilevel logistic regression modeling, ideally 50 level-1 units (words) and 40 level-2 units (children) are needed

(Schoeneberger, 2016). By including and analyzing more words in future research, the variation in frequency of occurrence in text and the age children are familiar with a word’s meaning will also increase, making it more likely to find small effects. This is desirable, because the words in the present study had limited variation in semantics and frequency of occurrence in text was rather low. Nevertheless, this study was a first step in investigating the

(23)

potential role of implicit cues on children’s spelling ability and multilevel logistic regression modeling was an appropriate method to analyze the data.

Second, the validity of the measurement of orthography and the criterion for morphological consistency might not be accurate. Concerning orthography, the frequency measure of Keuleers et al. (2010) is based on corpora of television subtitles of adult programs and is, therefore, more suitable for students and adults. It is plausible that children are not as much exposed to certain words as adults (Brysbaert, Mandera, & Keuleers, 2018). This could mean that the words’ frequency of occurrence in text based on Keulers et al. (2010) is

overestimated for children. Alternatively, the words’ frequency of occurrence in the text could be underestimated for children because the dictated words in the current study might not appear as much in adults’ television programs as in children’s. In addition, some children might generally watch more television than others, resulting in variation in exposure (Brysbaert et al., 2018). Thus, these comments suggest that future research should explore other possibilities of frequency measures that are also more suitable for children. Concerning morphological consistency, the analyzed target words might not be completely

distinguishable as morphologically inconsistent and morphologically consistent. The morphologically consistent words in the current study were plural short vowel nouns (e.g., takken [branches]). Although the meaning of the singular form of these nouns (e.g., tak) is

recognizable in its plural form, an extra consonant is added in the plural form (takken). Thus, the current results cannot be completely generalized to morphologically consistent and inconsistent words. Future research should incorporate words that really follow

morphological consistency to make better comparisons.

Third, the school children attended was not included as a level in the multilevel model, while it could have influenced children’s spelling ability. The type of instruction and method children receive for spelling differs per school (Inspectie van het Onderwijs, 1997). As a

(24)

result, it might be that some children rely more on explicit rules than other children because their teacher has focused more on explicit rules. Alternatively, in schools in which there is less instruction and practice of the explicit rules, children might rely more on implicit cues. To control for these possibilities, the school children attended should be included in a

multilevel model in future research. In this study, this was not possible because children only attended four different schools, resulting in too little variation between schools.

Fourth, and related to the third limitation, children’s language at home was not

controlled for although it could have influenced children’s reliance on implicit cues or explicit rules (Marinova-Todd & Hall, 2013). It is possible that children who do not speak Dutch at home might rely more on implicit cues than typically developing Dutch speaking children because they have more difficulty with acquiring the spelling rules of the Dutch language. If these children know the meaning of a word or if they are being exposed to a word more often, this might foster their spelling ability. Alternatively, these children have less word knowledge than native children, and as a result they have to rely more on the explicit rules than on implicit cues. Because only three children did not speak Dutch at home in the sample, both possibilities could not be investigated in the current study. Future research could focus on the use of implicit cues and explicit rules of children who do not speak Dutch at home because they might be more at risk for having spelling difficulty if their vocabulary size is low (Marinova-Todd & Hall, 2013), with consequences for how their writing is evaluated by others (Graham et al., 2011).

As the present study focused on the role of implicit cues in typically developing children, future research could also focus on children with spelling difficulties. For example, children with Developmental Language Disorder (DLD) might rely more on implicit cues than on explicit rules because they have difficulty with phonology (American Psychiatric Association, 2013), resulting in difficulty with sound-letter connections and explicit

(25)

phonological spelling rules. In addition, children with DLD might also have difficulty with the implicit cues because they have less semantic knowledge resulting in problems with comprehension of language. Future research should thus investigate to what extent children with spelling problems use implicit cues and explicit rules. Especially for these children it is important to know what cues they use and what cues they do not use when spelling words. If this is known, children’s spelling ability can be improved by giving support to the cues they already use and by giving extra instruction about the cues they do not yet use.

In sum, the current findings suggest that typically developing children in Grade 4 rely more on explicit rules when spelling long and short vowel words than on implicit cues. Although for younger children offering the meaning of words and increasing exposure to the orthographic form of words is recommended in spelling instruction (de Bree et al., 2018), this is not necessarily and not particularly recommended for children in Grade 4. Instead, because most errors were made for morphologically inconsistent words, it is recommended that children mostly practice the explicit rule for multisyllabic long vowel words and that teachers provide direct feedback while children are spelling. Teachers can give extra instruction about the spelling rules if children do not yet apply these correctly (Hilte & Reitsma, 2011).

Although the study’s findings contribute to the current knowledge about typically developing children’s spelling which is valuable for theory and educational practice, much is still

unknown. Future research should expand on the current knowledge and follow children over a longer period of time to see if the same children make less or different errors over time. Only then, the use of possible implicit cues becomes clear and insight is given into when these cues are overruled by explicit spelling rules.

(26)

References

Agresti, A., & Franklin, C. (2015). Statistics: The art and science of learning from data. Boston: Pearson Education.

American Psychiatric Association. (2013). Diagnostic and Statistical Manual of mental disorders (5th ed.). Washington, D.C.: Author.

Bosman, A. M. T. (2011). Spellen. In P. de Jong & H. Koomen (Eds.), Interventie bij onderwijsleerproblemen (pp. 41–52). Antwerpen: Garant.

Bosman, A. M. T., de Graaff, S., & Gijsel, M. A. R. (2006). Double Dutch: The Dutch spelling system and learning to spell in Dutch. In R. M. Joshi & P. G. Aron

(Eds.), Handbook of orthography and literacy (pp. 135–150). Retrieved from https://b ooks.google.nl/books?hl=nl&lr=&id=1xX-am4mqWoC&oi=fnd&pg=PA135&dq=dou ble+dutch+bosman&ots=F1NT2QU3_W&sig=znlVDedN9Dd5updt7u83HBhVwA& edir_esc=y#v=onepage&q=double%20dutch%20bosman&f=false

Bosman, A. M. T., & van Orden, G. C. (1997). Why spelling is more difficult than reading. In C. A. Perfetti, L. Rieben, & M. Fayol (Eds.), Learning to spell: Research, theory, and practice across languages (pp. 173–194). Retrieved from http://www.annabosma

n.eu/documents/BosmanVanOrden1997b.pdf

de Bree, E. H., Geelhoed, J., & van den Boer, M. (2018). Overruled!: Implicit cues rather than an orthographic rule determine Dutch children’s vowel spelling. Learning and Instruction, 56, 30–41. doi:10.1016/j.learninstruc.2018.03.006

Brysbaert, M., Mandera, P., & Keuleers, E. (2018). The word frequency effect in word processing: An updated review. Current Directions in Psychological Science, 27, 45-50. doi:10.1177/0963721417727521

(27)

Brysbaert, M., Stevens, M., de Deyne, S., Voorspoels, W., & Storms, G. (2014). Norms of age of acquisition and concreteness for 30,000 Dutch words. ACTA PSYCHOLOGICA, 150, 80-84. doi:10.1016/j.actpsy.2014.04.010

Chen, H., Cohen, P., & Chen, S. (2010). How big is a big odds ratio? Interpreting the magnitudes of odds ratios in epidemiological studies. Communications in Statistics -Simulation and Computation, 39, 860–864. doi:10.1080/03610911003650383

Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Retrieved from http://www.utstat.toronto.edu/~brunner/oldclass/378f16/readings/CohenPower.p df

Drijver, S. (2018). De verwerving van en de relatie tussen spellingpatronen: Lange klinkerspelling en dubbele grafeemspelling (master’s thesis). Retrieved from UvA Scripties. (https://scripties.uba.uva.nl/search?id=710590)

Drijver, S., van den Boer, M., & de Bree, E. H. (in revision). Hoe taken en takken elkaar beïnvloeden: De invloed van parallelvormen op spelling. Pedagogische Studiën. Graham, S., Harris, K., & Hebert, M. (2011). It is more than just the message: Presentation

effects in scoring writing. Focus on Exceptional Children, 44(4), 1–12. doi:10.17161 /foec.v44i4.6687

Hilte, M., & Reitsma, P. (2011). Effects of explicit rules in learning to spell open- and

closed-syllable words. Learning and Instruction, 21(1), 34-45. doi:10.1016/j.learninstr uc.2009.10.002

Finch, W. H., Holden, J. E., & Kelley, K. (2014). Multilevel modeling using R. Retrieved from https://juancarloscastillo.github.io/jc-castillo/documents/diplomado_multinivel/ bibliografia/finch%20et%20al%202014%20multilevel%20modeling%20using%20R.p df

(28)

Inspectie van het Onderwijs (1997). Spellenderwijs: Een evaluatie van de kwaliteit van het onderwijs in spelling en interpunctie op de basisschool (Inspectierapport 1997-11).

Den Haag: Sdu Uitgevers.

Kessler, B., Pollo, T. C., Treiman, R., & Cardoso-Martins, C. (2013). Frequency analyses of prephonological spellings as predictors of success in conventional spelling. Journal of Learning Disabilities, 46, 252–259. doi:10.1177/0022219412449440

Keuleers, E., Brysbaert, M. & New, B. (2010). SUBTLEX-NL: A new measure for Dutch word frequency based on film subtitles. Behavior Research Methods, 42, 643-650. doi: 10.3758/BRM.42.3.643

Landerl, K., & Reitsma, P. (2005). Phonological and morphological consistency in the acquisition of vowel duration spelling in Dutch and German. Journal of Experimental Child Psychology, 92, 322–344. doi:10.1016/j.jecp.2005.04.005

Lupyan, G., & Lewis, M. (2019). From words-as-mappings to words-as-cues: The role of language in semantic knowledge. Language, Cognition and Neuroscience, 34, 1319-1337. doi:10.1080/23273798.2017.1404114

Marinova-Todd, S., & Hall, E. (2013). Predictors of English spelling in bilingual and monolingual children in Grade 1: The case of Cantonese and Tagalog. Educational Psychology, 33, 734–754. doi:10.1080/01443410.2013.797337

Notenboom, A., & Reitsma, P. (2007). Spelling Dutch doublets: Children’s learning of a phonological and morphological spelling rule. Scientific Studies of Reading, 11, 133-150. doi:10.1080/10888430709336556

Ouellette, G., & Fraser, J. R. (2009). What exactly is a “yait” anyway: The role of semantics in orthographic learning. Journal of Experimental Child Psychology, 104, 239-251. doi:10.1016/j.jecp.2009.05.001

(29)

Perfetti, C., & Hart, L. (2002). The lexical quality hypothesis. In L. Verhoeven, C. Elbro, & P. Reitsma (Eds.), Precursors of functional literacy (pp. 189–213). Retrieved from http:// www.pitt.edu/~perfetti/PDF/Lexical%20quality%20hypothesis.pdf

Schoeneberger, J. A. (2016). The impact of sample size and other factors when estimating multilevel logistic models. The Journal of Experimental Education, 84, 373–397. doi: 10.1080/00220973.2015.1027805

Seymour, P. H. K. (1997). Foundations of orthographic development. In C. A. Perfetti, L. Rieben, & M. Fayol (Eds.), Learning to spell: Research, theory, and practice across languages (pp. 319-337). Mahwah, NJ: Lawrence Erlbaum.

Tamura, N., Castles, A., & Nation, K. (2017). Orthographic learning, fast and slow: Lexical competition effects reveal the time course of word learning in developing

readers. Cognition, 163, 93–102. doi:10.1016/j.cognition.2017.03.002

Treiman, R. (2017). Learning to spell words: Findings, theories, and issues. Scientific Studies of Reading, 21, 265-276, doi:10.1080/10888438.2017.1296449

Treiman, R., & Cassar, M. (1997). Spelling acquisition in English. In C. A. Perfetti, L. Rieben, & M. Fayol (Eds.), Learning to spell: Research, theory, and practice across languages (pp. 61-80). Mahwah, NJ: Lawrence Erlbaum.

Treiman, R., & Yin, L. (2011). Early differentiation between drawing and writing in Chinese children. Journal of Experimental Child Psychology, 108, 786–801. doi:10.1016/j.jecp .2010.08.013

van der Ven, S., & de Bree, E. (2019). Variation is the spice of spelling: The effect of implicit cues on Dutch past tense spelling is dependent on age and literacy, but not on task format. Scientific Studies of Reading, 23(5), 369–385. doi:10.1080/10888438.2019.15 79217

(30)

APPENDIX Table 1

Models obtained at the Model Selection Process for Applying a Rule Correctly or Not

Note. COEFF = Coefficient; SE = Standard Error; p = p value; PARM = Parameter; Morph = Morphological Consistency; AoA = Age of Acquisition of meaning (Semantics); Freq = Frequency of occurrence in text (Orthography). PRD is proportion of reduction in deviance compared to the Empty Model (Model 0). This table shows that the final model is model 3.

Model 0 Model 1 Model 2 Model 3 Model 4 Model 5

Fixed effects COEFF SE p COEFF SE p COEFF SE p COEFF SE P COEFF SE p COEFF SE p

Intercept 3.791 0.376 <.001 3.521 1.293 .006 3.843 0.946 <.001 4.130 0.412 <.001 3.896 0.458 <.001 4.330 0.831 <.001 Morph -0.135 1.739 .938 -0.473 0.264 .073 -0.581 0.239 .015 -0.477 0.256 .062 -.0589 0.241 .015 AoA 0.041 0.170 .811 0.008 0.125 .949 -0.033 0.119 .781 Freq 0.038 0.034 .267 0.025 0.023 .280 0.024 0.021 .263 AoA:Morph -0.023 0.269 .932 Freq:Morph -0.026 0.047 .579

Random part PARM PARM PARM PARM PARM PARM

Level-two variance: 02 = var(U0j) 3.750 3.857 3.853 3.833 3.853 3.833 PRD - 1.26% 1.21% 1.00% 1.21% 1.01% Deviance 608.02 600.36 600.67 601.97 600.67 601.89

(31)

Table 2

Models obtained at the Model Selection Process for Spelling a Word Correctly or Not

Note. COEFF = Coefficient; SE = Standard Error; p = p value; PARM = Parameter; Morph = Morphological Consistency; AoA = Age of Acquisition of meaning (Semantics); Freq = Frequency of occurrence in text (Orthography). PRD is proportion of reduction in deviance compared to the Empty Model (Model 1). This table shows that the final model is model 3.

Model 0 Model 1 Model 2 Model 3 Model 4 Model 5

Fixed effects COEFF SE p COEFF SE p COEFF SE p COEFF SE p COEFF SE p COEFF SE p

Intercept 3.188 0.326 <.001 2.626 0.939 .005 2.616 0.767 <.001 2.961 0.337 <.001 2.869 0.375 <.001 2.880 0.651 <.001 Morph 0.530 1.485 .721 0.585 0.220 .008 0.523 0.009 .015 0.564 0.213 .008 0.526 0.200 .009 AoA 0.036 0.122 .768 0.038 0.100 .706 0.013 0.093 .885 Freq 0.012 0.023 .607 0.012 0.018 .516 0.009 0.017 .581 AoA:Morph 0.010 0.238 .965 Freq:Morph -0.001 0.040 .980

Random part PARM PARM PARM PARM PARM PARM

Level-two variance: 02 = var(U0j) 3.677 3.765 3.756 3.748 3.753 3.749 PRD - 0.93% 0.93% 0.87% 0.91% 0.87% Deviance 805.98 798.51 798.51 798.96 798.66 798.94