Optimal reading and spelling development depend on phonological awareness, which is the ability to identify and manipulate phonological units (syllables, onset-rimes, phonemes) in spoken language (Caravolas, Hulme, & Snowling, Reference Caravolas, Hulme and Snowling2001; Ehri, Reference Ehri, Perfetti, Rieben and Fayol1997; Goswami & Bryant, Reference Goswami and Bryant1990; Liberman, Shankweiler, Fischer, & Carter, Reference Liberman, Shankweiler, Fischer and Carter1974; McBride-Chang, Reference McBride-Chang1995; Rayner, Foorman, Perfetti, Pesetsky, & Seidenberg, Reference Rayner, Foorman, Perfetti, Pesetsky and Seidenberg2001; Wagner & Torgesen, Reference Wagner and Torgesen1987). However, the development of phonological awareness, which mediates children's transcoding between speech and writing, depends, in turn, on aural linguistic experience (Caravolas & Bruck, Reference Caravolas and Bruck1993; Chaney, Reference Chaney1998; Maclean, Bryant, & Bradley, Reference Maclean, Bryant and Bradley1987). Thus, a child's phonological lexicon, or long-term store of phonological representations in their spoken language, in tandem with phonological awareness, establishes the foundation for decoding print to speech during oral reading (Mann & Foy, Reference Mann and Foy2003) and encoding speech to print during spelling (Rickard Liow & Lau, Reference Rickard Liow and Lau2006).

In many unilingual and bilingual settings, adult speakers are diglossic, and use two phonological representations: a standard and a nonstandard (colloquial) form, for a single lexical entry. Typically the nonstandard form is used at home and the standard form is reserved for formal interactions and written communication. Before they enter kindergarten, children from diglossic families are mostly exposed to the colloquial form, so the influence of home language on literacy development is of pedagogical and theoretical interest. This study shows how a comparison of the spelling errors made by diglossic and nondiglossic Malay-speaking children can elucidate models of the cognitive processing and inform classroom practice. Looking at data on children's early spellings of shallow orthographies is especially informative because the mappings between phonology and orthography are less ambiguous than those for English.

Most children find spelling in English is a much more demanding task than reading, yet beginners often try and use letter–sound knowledge, for example, eyes as “ISE,” liked as “LIKT,” kitten as “KITN” (Read, Reference Read1986). If young children make these phonologically plausible encodings before they are taught conventional orthographic units, then they must be relying on speech-based representations (Treiman, Reference Treiman1993). Later, spelling and reading skills tend to be based on a mix of strategies (see Rittle-Johnson & Siegler, Reference Rittle-Johnson and Siegel1999). These strategies change over time (Cataldo & Ellis, Reference Cataldo and Ellis1988), and they depend on orthography–phonology consistency (Caravolas, Reference Caravolas2004; Cheung, Chen, Lai, Wong, & Hills, Reference Cheung, Chen, Lai, Wong and Hills2001; Cook & Bassetti, Reference Cook and Bassetti2005; Durgunoglu & Oney, Reference Durgunoglu and Oney1999; Harris & Hatano, 1999; Perfetti, Rieben, & Fayol, Reference Perfetti, Perfetti, Rieben and Fayol1997; Rickard Liow & Tng, Reference Rickard Liow, Tng, McBride-Chang and Chen2003), classroom teaching methods (Wimmer & Hummer, Reference Wimmer and Hummer1990; Ziegler & Goswami, Reference Ziegler and Goswami2005), print exposure (Brown & Loosemore, Reference Brown, Loosemore, Leong and Joshi1995; Kemp & Bryant, Reference Kemp and Bryant2003), and whether or not the child is learning two languages (Bialystok, Reference Bialystok2001; Bruck, Genesee, & Caravolas, Reference Bruck, Genesee, Caravolas and Blachman1997; Cook & Basetti, 2005; Rickard Liow & Lau, Reference Rickard Liow and Lau2006).

The interplay between these factors and phonological processing is often difficult to unravel once an English-speaking child enters school. This is because teaching methods for literacy skills (whole-word, whole-language, or phonics) can mask underlying cognitive–linguistic processes. However, preschoolers’ emergent spellings in an orthography with unambiguous mappings to phonology could provide unequivocal data on the use of speech-based representations. Rumi, which is widely used in South East Asia for Malay, is one such orthography, and adult speakers of Malay in Singapore and Malaysia are usually diglossic. Unlike English, Malay has syllable timing, short vowel sounds, and very few consonant clusters (Mintz, Reference Mintz1992); and there are no homophones or homophonous letters in the Rumi script. This transparency between the spoken and written forms means that Malay literacy programs emphasise phonological processing (Rickard Liow & Lee, Reference Rickard Liow and Lee2004; Rickard Liow & Poon, Reference Rickard Liow and Poon1998), and reading skills are relatively easy to acquire (cf. Ellis & Hooper, Reference Ellis and Hooper2001; Wimmer & Goswami, Reference Wimmer and Goswami1994; Ziegler & Goswami, Reference Ziegler and Goswami2005). However, the children of parents who are diglossic will be exposed to colloquial Malay almost exclusively, and will therefore develop phonological representations based on the nonstandard pronunciation used at home. If early attempts at writing are based on speech, exposure to nonstandard phonological constructions should influence the kind of spelling errors made by preschoolers. Thus, we predicted that diglossic children's invented spellings of Malay words, dictated with standard phonology, should be consistent with nonstandard phonology used at home if they rely on a speech-based redintegration.

The term redintegration has been used widely in research on memory development (e.g., Gathercole, Pickering, Hall, & Peaker, Reference Gathercole, Pickering, Hall and Peaker2001; Turner, Henry, Smith, & Brown, Reference Turner, Henry, Smith and Brown2004), and is relevant to children's reading ability (Roodenrys & Stokes, Reference Roodenrhys and Stokes2001). It refers to the restoration of the whole of something from fragments using long-term representations to support short-term recall (Hulme, Maughan, & Brown, Reference Hulme, Maughan and Brown1991). We reasoned that if diglossic children rely on their own long-term speech-based representations of words when they are starting to spell, the errors they make for target words dictated using standard pronunciation should show evidence of lexical redintegration from nonstandard phonology. The auditorily presented target word will decay from short-term memory before kindergarten children can finish writing out the letters. To complete their spellings, the children draw upon their own long-term representations (redintegration) and their phonology–orthography mappings for the target will then be consistent with nonstandard pronunciation because this is the form they speak.

The main impetus for our investigation were the questions raised by Treiman, Goswami, Tincoff, and Leevers’ (Reference Treiman, Goswami, Tincoff and Leevers1997) report of dialect effects in native English-speaking school children (chronological age 6 years, 0 months [6;0] to 8;10, with spelling ages ranging from 6;0 to 7;6). Treiman et al. (Reference Treiman, Goswami, Tincoff and Leevers1997) found subtle differences in the writing errors made by matched groups of children from the United States (Detroit) where the dialect is rhotic (/r/ can occur after a vowel in the rime of a syllable) and those made by children attending a school in the United Kingdom (Cambridge) where the dialect is nonrhotic: American children spelled words like hurt with an /r/ as “HRT,” whereas the British children omitted the /r/ and were more likely to spell the same word as “HUT.” Of importance, the observed differences were more marked for low-ability spellers, suggesting continued overreliance on speech-based representations well after school entry.

Although Treiman et al.'s (Reference Reece and Treiman1997) comparison of British and American school children demonstrated that beginners are sensitive to relatively small phonological units, it is not clear whether there were also differences in teaching methods and/or print exposure between the two groups. Moreover, the spellings of the children could have been based on the adult speech during the testing session, rather than the linguistic experience of the child. This is because Treiman et al.'s (Reference Reece and Treiman1997) target words for spelling were dictated by different experimenters, each with the same dialect as the children they tested. In a more recent study of dialect effects in American college students, Treiman (Reference Treiman2004) compared the final consonant spellings of /t/ versus /d/ of African American Vernacular (AAVE) speakers of English with those of white speakers of English, and this time both groups were asked to spell from dictation in both dialects. The results showed that both AAVE and White speakers made more <d> for <t> errors when dictated to by the AAVE speaker. Despite considerable exposure to print, like the weaker spellers in Treiman et al.'s (Reference Reece and Treiman1997) study, these adults continue to rely (too) heavily on phonological processing. Treiman's (Reference Treiman2004) target words were rated as reasonably familiar by participants so what still seems to be missing is a comparison of errors made on words versus nonwords. A lexicality effect (difference between words and nonwords) in terms of the number and/or nature of errors would clarify the role of lexical–phonological representations in spelling.

To resolve the ambiguity about whose speech is encoded by young children, and whether emergent spellings of words reflect lexical phonological representations, we developed lists of high-familiarity Malay words that have nonstandard as well as standard phonological constructions, and a set of control nonwords. Both sets of stimuli were then dictated with same standard pronunciation by the same adult speaker in a two-part study comparing the performance of preschool children from Singapore and Jakarta.

The entire ethnic Malay population in Singapore (approximately 15% of total) is extremely diglossic: standard Malay (SM) is used for reading, writing, and formal settings such as school, but nonstandard Malay (NSM) is used at home. Young children spend most of their time at home with parents and older siblings. This means they are usually exposed to NSM much more than SM from birth until they start kindergarten at age 5 years. The nonstandard form of Malay used in Singapore evolved along the Malacca Straits, between West Malaysia and Sumatra, as an interethnic language to facilitate trade with India and Arabic countries in the West, and China in the East. As in other diglossic communities, the disparity between the nonstandard form and the standard form extends to grammar, vocabulary, and phonology but only the differences in vowel sounds are relevant for this study.

These vowel sound differences are of three kinds: /i/ to /e/ and /u/ to /o/, which occurs in all <i> and <u> vowels occupying the final closed position (e.g., manis “sweet” and kebun “garden”), and /ɑ/ to /ə/, which occurs in all <a> vowels occupying the final open position (e.g., kota “town”). These changes occur in all Malay words with the exception of English loan words in the Malay lexicon (e.g., tema “theme”). See Table 1 and Mintz (Reference Mintz1992) for more details. This means that the pronunciation for manis changes from /mɑnis/ to /mɑnes/ and so the phonologically plausible (but incorrect) spelling would be manes. Similarly, kebun changes from /kəbun/ to /kəbon/ and incorrectly spelt as kebon, and kota changes from /kotɑ/ to /kotə/ and incorrectly spelt as kote.

Table 1.

Contrastive vowel patterns for standard and nonstandard Malay and related phonologically plausible errors in spelling

Both forms of spoken Malay are syllable timed, so these differences are not attributable to stress patterns. Most children would have some exposure to SM if they watch Malay television channels or listen to stories read at home, but they would use NSM almost exclusively in their own conversations with friends and family members. We verified that Singaporean preschoolers tend to use NSM phonology for high frequency words by administering tests of picture naming and oral reading.

The preschoolers living in neighboring Indonesia provide a good comparison for investigating the influence of aural linguistic experience on literacy. In Jakarta, ethnic Malays form the majority (approximately 90%), and SM (called Bahasa Indonesia) is the only official language. There are some lexical and syntactic differences between the SM used in Singapore and Indonesia, stemming from earlier British and Dutch colonization, but both populations use the same Rumi orthography, and the majority of words overlap semantically. The main difference lies in the phonology: NSM is not useful in Jakarta, and so diglossic pronunciation is uncommon even at home. This means that if beginner spellers are using their own speech-based lexical redintegration to support spelling, the children in Singapore will be more likely to make errors that are phonologically plausible for the nonstandard form of Malay (e.g., vowel substitutions for manis, kebun, and kota) than the SM-speaking children in Indonesia. Importantly, these errors should occur even when target words are dictated by the same adult in SM, and they should be infrequent (or absent) for nonwords because these will be less influenced by prior linguistic experience.

Before describing how these hypotheses were tested, a synopsis of literacy instruction for Rumi orthography is appropriate. Most children start learning to read and spell when they enter kindergarten in their fifth year. Teachers usually introduce the five vowels and their sounds with SM phonology, and then the 20 consonants (<x> is not used in Malay). After brief exposure to all the letters in the alphabet, syllable phonology for a range of consonant–vowel (CV) pairs is taught. Children are encouraged to spell and sound out (chant) CV syllables in the classroom, for example, “K, A, /kɑ/. . .”; “D, A, /dɑ/. . .” before combining them into high frequency words, for example, “B, A, /bɑ/, C, A, /cɑ/, /bɑcɑ/. . .” baca (read).

Malay-speaking children seem to acquire reading and spelling skills readily (see Rickard Liow & Lee, 2004, on Malaysian pupils; Winskel & Widjaja, Reference Winskel and Widjaja2007, on Indonesian) and phonological awareness transfers to English even though some correspondences are different (Rickard Liow & Poon, Reference Rickard Liow and Poon1998, on Indonesian pupils). However, the phonology–orthography relationship for the Singapore children from diglossic families is less transparent (especially for the second vowel) because they have been regularly exposed to NSM pronunciation, for example, kota /kotɑ/→kote /kotə/. Anecdotally, teachers in Singapore have reported that pupils’ spellings are adversely affected by home language, and so we expected to find spelling errors that would be phonologically plausible for the nonstandard form of Malay. However, two other explanations needed to be ruled out before specific vowel substitutions could be attributed to phonological representations and early linguistic experience.

First, it might be the case that some children make use of letter names when they do not know individual letter sounds. Treiman (Reference Treiman1994) argued that letter names provide a bridging function in writing for English-speaking children who have not developed sufficient phonemic awareness (see also Reece & Treiman, Reference Reece and Treiman2001; Treiman & Tincoff, Reference Treiman and Tincoff1997). Thus, invented spellings should show <e> for <i> substitutions because the name for letter <e> is pronounced as /i/. Fortunately, it is possible to differentiate letter–name substitutions (LNS) from NSM vowel transcriptions by examining the vowel position. For words that have <i> as the first vowel (e.g., jimat “save”) or the final opened vowel (e.g., saji “serve food”), <e> for <i> errors must be LNS errors. This is because NSM transcriptions can only be made on the final closed vowel in words (e.g., petir “lightning”).

A second possible explanation for the expected vowel substitutions is the influence of English phonology, for example, <o> is pronounced as /u/ in English but not in Malay. In Singapore, Malay-speaking children usually acquire English as a second language in school, but do learn some English words when they are in kindergarten. This means that code mixing, rather than exposure to NSM at home, could be an explanation for kebun being spelled as kebon, and so forth. As with LNS errors, nonstandard vowel transcriptions can be differentiated from the influence of English phoneme–grapheme correspondences by examining the vowel position: for words with <u> as the first vowel (e.g., rumah “house”) or the final opened vowel (e.g., baju “clothes”), <o> for <u> substitutions must be English phoneme–grapheme substitutions rather than NSM substitutions that are made only to the final closed vowel in words (e.g., telur “egg”).

Bearing in mind these two alternative explanations to nonstandard pronunciation at home as the major influence on phonological processing, we designed Malay word and matched nonword spelling tasks suitable for young children in Singapore and Jakarta (see Table 2 for details). In Study 1 we tested whether words dictated in SM to diglossic Singapore preschoolers do result in vowel substitutions that are consistent with NSM phonology rather than letter–name substitutions or English phoneme–grapheme correspondences. For Study 2 we employed three additional oral tasks to confirm that Singaporean children do use NSM phonology when speaking more than children from Jakarta. We then matched the two groups for overall spelling abilities, to show unambiguously that phonological memory representations are influenced by exposure to nonstandard phonology, and these, in turn, affect emergent spellings of diglossic children.

Table 2.

Error coding system and data for Singapore children from diglossic families

STUDY 1: DIGLOSSIC MALAY-SPEAKING CHILDREN'S WORD VERSUS NONWORD SPELLINGS

Word and nonword spelling tests were developed. Both tests comprised items that might elicit NSM second vowel transcriptions (<e> for <i>, <o> for <u>, and <e> for <a>), LNS, and English phoneme–grapheme correspondences (PGCs; see Table 2) to assess whether any vowel errors are related to linguistic experience at home. Putting together the expected lexicality and vowel position effects, two predictions were made for Singaporean children from diglossic families: (H1) They will make significantly more errors involving substitutions of the second vowel in the experimental NSM words than in the matched control words, and (H2) there will be a Lexicality (word vs. nonword) × Item-Type interaction, consistent with support from redintegration, such that more substitutions will be made for the second vowel of NSM words than those made for the first vowel of (potential) LNS words and English PGC words, and there will be an effect of lexicality for NSM targets, such that more vowel transcription errors will be made for words than nonwords, with no effect of lexicality for either LNS or English PGCs targets.

STUDY 2: EMERGENT SPELLINGS OF DIGLOSSIC VERSUS NONDIGLOSSIC CHILDREN

The results from Study 1 confirmed that particular vowel spelling errors made by diglossic kindergarten children in Singapore are consistent with the NSM phonological representation of their own speech. The aim of Study 2 was to compare the errors made by diglossic children with those from a matched group of nondiglossic Indonesian children from Jakarta to check that vowel substitutions are a result of linguistic experience rather than poor encoding skills. First, three oral tasks (word reading, nonword reading, and picture naming) were developed and administered to children in both locations to verify that the use of NSM phonology is greater among Singapore children. Having established that the Indonesian children were not diglossic, a Location × Lexicality interaction was tested (H3). In Study 1, NSM vowel substitutions were made only to lexical entries (i.e., words but not nonwords), so planned comparisons were expected to reveal that Singaporean children do make more NSM spelling errors than Jakarta children for words, but not for nonwords.

GENERAL DISCUSSION

Young children who have had limited exposure to print can invent phonologically plausible spellings by transcribing speech. From Treiman et al.'s (Reference Reece and Treiman1997) comparison of American and British primary school children, it is not clear whether their spelling attempts were based on the phonology of the adult narrator's phonology during the testing session, or compiled from their own phonological representations. From the two experiments reported here, it is now clear that diglossic children base their spellings on their own nonlexical phonological representations because they make phonologically plausible errors for the nonstandard form even when the target words are dictated by an adult using the standard form. This finding represents an important step toward a more complete understanding of how early linguistic experience influences the development of phonological representations, and hence literacy skills.

For Rumi Malay orthography, unlike English, the sound:print relationship is very transparent for the standard pronunciation. However, the phonology of the second vowel changes when some Malay words are spoken with nonstandard pronunciation. We exploited this difference in the pronunciation by diglossic (Singapore) and nondiglossic (Jakarta) families, and compared the nature and number of vowel substitutions made on word and nonword spelling tests by 6-year-old preschoolers. When the words were dictated by the same adult SM speaker to two groups who were matched for spelling proficiency, only the children from diglossic families encoded the vowels so that they were consistent with nonstandard pronunciation. The position of these vowel errors suggests they were not attributable to the use of letter names, or to English code mixing, and equivalent substitutions were very rare for the matched set of nonwords for which redintegration is less likely to occur.

Learning to spell vowels is particularly difficult in English (Stage & Wagner, Reference Stage and Wagner1992; Treiman, Reference Treiman1993), and definitive empirical studies would be complex. This fine-grained analysis of young children's vowel substitutions in Rumi, a shallow alphabetic orthography with short vowels, provides unusually clear evidence that early linguistic experience, such as exposure to nonstandard phonology, influences spelling skills. Target words spoken in SM by an adult were recognized by the diglossic and nondiglossic children, and both groups appear to have used activated stored lexical representations of phonology to reconstruct plausible spellings. A process of redintegration (see Gathercole, Reference Gathercole2006) would be necessary during a spelling task because auditory input decays rapidly (about 2 s) and volitional rehearsal is uncommon before the age of 7 years (Gathercole & Hitch, Reference Gathercole, Hitch, Collins, Gathercole, Conway and Morris1993). For the nondiglossic children, lexically based reconstruction results in a correct spelling because Rumi is so shallow for SM pronunciation. For the diglossic children, the same redintegration process is more error prone because their phonological representations are based on the NSM pronunciation they have learned at home, and hence, the observed vowel substitutions.

Because redintegration (or lexically supported repair) is based on the child's store of known phonological representations, vowel substitutions for matched nonwords were rare. When they did occur (only cetir) it is possible that the children were making an approximation to a phonologically close lexical entry (petir). Repetition accuracy is also associated with the wordlikeness of a nonword (Gathercole, Willis, & Baddeley, Reference Gathercole, Willis and Baddeley1991), and although our nonwords were very wordlike (substitution of a single phoneme), it seems that spellers of transparent Rumi rarely need to make analogies with words. Once diglossic (and dialect-speaking) children are exposed to print regularly, and have had more exposure to standard phonology in the school, their spelling performance would be expected to show steady improvement. However, for high-frequency words, the phonological representations of the nonstandard form are well established and likely to remain strong for diglossic children because they will be regularly used in all informal settings. Moreover, if speech input and output buffers are linked neurologically, as Jacquemot and Scott (Reference Jacquemot and Scott2006) have claimed, the nonstandard form will be reinforced more often than the standard form through speech production as well as compre-hension.

The impact of code differences, arising from parental diglossia and dialect variations, on literacy has been underexplored in all languages, yet it seems that early exposure to a particular phonological form does influence cognitive–linguistic processing. This investigation of children's phonological representations, evidenced by their early attempts at writing in a shallow orthography, revealed vowel spellings that are consistent with the pronunciation they use and hear at home, whether this is standard or NSM. For speakers of SM in Indonesia, reliance on speech-based processing for spelling enhances performance because there is only one phonological form for each word, and the orthography is very shallow. For children from diglossic Malay-speaking families in Singapore, exposure to nonstandard phono-logy during vocabulary acquisition results in subtle deviations in the phonological memory representations needed for literacy in the standard form. This finding has pedagogical implications because early experience of nonstandard pronunciation at home is also commonplace in bilingual as well as diglossic communities. Messbauer and De Jong (Reference Messbauer and De Jong2006) and others have argued that dyslexic children have impairments in the long-term representation of the phonological features of words. Diglossic children do not have impairments as such but their phonological representations of the nonstandard form will be much stronger than those for the standard form before they enter school.

The kinds of errors we observed for Rumi Malay probably reflect a temporary transition phase for pupils learning languages that have transparent phonology–orthography mappings. For more opaque orthographies such as English, phonologically plausible errors may endure in adulthood, cf. Treiman's (Reference Treiman2004) AAVE college students. Despite the strong support for teaching methods based on phonological awareness, optimal spelling development for diglossic and dialect-speaking children may depend on learning particular subsets of word-specific representations (Kreiner & Gough, Reference Kreiner and Gough1990), whether or not the orthography is shallow like Malay, or relatively deep like English.