Spelling involves the integration of several skills including phonological, morphological and orthographic awareness (Ehri & Wilce 1982; Pijper 2003; Swayer & Joye 2006; Wagner & Torgesen 1987).

Phonological awareness (PA), which refers to the awareness and ability to manipulate sounds of one’s language (Anthony et al. 2003), has been identified as an important contributor to spelling (Bruck & Treiman 1990; Treiman 1993; Vellutino et al. 2004). Specifically, PA enables children to learn phoneme-grapheme correspondences, which is useful for creating accurate spellings (Griffith 1991). For example, Weinrich and Fay (2007) found that phonemic awareness skills were significant in predicting spelling ability in English-speaking children. In the South African context, De Sousa et al.’s (2011) study, which investigated spelling performance by 30 emergent bilingual isiZulu-English and 30 monolingual English children in Grade 2, argued that PA is well represented in spelling. The findings of their study showed that first-language (L1) PA is related to spelling across languages in emergent bilinguals. They assert that an awareness of phonemes and their ability to be segmented and manipulated enhances learners’ decoding skills, which leads to spelling success (De Sousa et al. 2011). Similarly, Diemer’s (2015) study noted that spelling is directly related to phonological representations but is also influenced by grammatical knowledge. She found that spelling in isiXhosa was significantly influenced by PA.

Morphological awareness (MA) has also been shown to play a role in spelling (see Apel & Masterson 2001; Mann & Singson 2003; Nunes & Byrant 2009). Nunes, Bryant and Bindman (1997)’s study showed that in English, scores on a MA task (namely word and sentence analogy) predicted 6–8-year-old children’s successful spelling of the regular past tense verbs as -ed in both words and pseudowords, even after controlling for PA. Furthermore, Deacon, Kirby and Casselman-Bell’s (2009) study showed that MA is a robust variable in predicting spelling outcomes. Results of their study showed that MA measured in Grade 2 determined 8% of the variance in spelling measured two years later, even after controlling for verbal and nonverbal intelligence, PA, short-term memory and rapid automated naming. These results demonstrate that MA has a significant role in children’s spelling development.

In addition, orthographic knowledge is important for spelling in that learners are expected to engage with the orthographic features of a language such as the letter names and sounds or alphabetic, within-word patterns and syllable structure (Fleisch et al. 2017). Therefore, Templeton and Morris (2000) reason that if learners are not able to memorise the complete and accurate orthographic representation of a word, then their ability to spell a word will reflect the orthographic knowledge they were able to utilise in processing the word for spelling. Zaric, Hasselhorn & Nagler’s (2020) study on German children provides further support for the role of orthographic awareness in spelling, with their results showing that word-specific, as well as general orthographic knowledge contributes to spelling performance, over and above intelligence and PA. Similarly, Mpiti (2012) argues that spelling involves visual memory and the knowledge of orthographic rules and conventions (Kress as cited in Pijper 2003; Richards 2001:14; Wagner & Torgesen 1987).

Thus, with respect to the skills required for spelling, children need to simultaneously integrate not only phonological, but also morphological, orthographic, lexical and grammatical knowledge in order to become accomplished spellers (Alcock & Ngorosho 2003). However, little is known about spelling in the Southern Bantu languages and thus the relationship of the above listed metalinguistic skills to spelling in isiXhosa should be interpreted within this constraint.

The knowledge of orthography, defined as the understanding of the conventions used in the writing system of one’s language (Treiman & Cassar 1996), has been shown to facilitate children’s spelling development. Thus, in an alphabetic writing system, spelling involves the process of mapping grapheme to phoneme (Park 2011). Given that isiXhosa is an alphabetic language, this section reviews spelling in alphabetic languages.

Caravoles’ (2004) cross-linguistic study of alphabetic languages highlighted both the similarities and differences between alphabetic orthographies and how these impact spelling development. In particular, the study reviewed English, French and Czech studies to uncover what spelling acquisition characteristics are universal and which are language specific (Caravoles 2004). Caravoles acknowledges that studies focusing on spelling development in languages other than English are significantly rare and, thus, her study provides only preliminary insights into spelling development across languages.

She notes that alphabetic orthographies share one key and possibly universal quality (Caravoles 2004): they are all based on the alphabetic principle, which is the understanding that graphemes represent particular phonemes in spoken language (Caravoles 2004). However, she explains that the extent to which languages adhere to the alphabetic principle will determine the degree to which learners can exploit the principle when spelling (Caravoles 2004:4).

Alphabetic languages exist on a continuum between transparent, in which there is a consistent one-to-one mapping of phoneme to grapheme, and opaque, where there is an inconsistent mapping of phoneme to grapheme. Depending on the type of orthography, learners are required to use different types of orthographic and linguistic knowledges when spelling (Caravoles 2004:4). Only few cross-linguistic studies have examined spelling development in relation to the consistency of the orthography. Concerning the languages of focus in Caravoles’ (2004) study, English is considered opaque and Czech is considered transparent. Similarly, isiXhosa is also considered a transparent language. Caravoles’ review revealed that learners learning to spell in more consistent orthographies tended to perform better than their English counterparts. This was rationalised under the understanding that orthographic consistency promotes faster spelling development and thus learners are able to learn the basic phonological spelling skills and conventional spelling skills with more ease (Caravoles 2004). However, Bigozzi et al. (2016:2) argue that spelling is made more challenging in transparent orthographies, in which the regularity of the orthographic system is higher in grapheme-phoneme relations (forward regularity, e.g. decoding) than it is in phoneme-grapheme relations (backward regularity, e.g. encoding). Caravoles’ study showed that there are clearly differences in spelling development across alphabetic orthographies and recommends that further cross-linguistic longitudinal studies be conducted on languages other than English to confirm or deny the preliminary findings. This provides further motivation for studies of this nature.

Other noteworthy variables mentioned in Caravoles’ (2004) study that seem to play a role in spelling development in alphabetic orthographies include the morphological information encoded in the languages and its impact on the phoneme-grapheme consistency (Caravoles 2004), the presence of borrowed words that have retained their spelling from their host language, archaic spellings that reflect historical word pronunciations, pedagogical practice, home literacy, phonotactic constraints and syllable structure (Caravoles 2004).

There is a growing body of research dedicated to the exploration of patterns in spelling errors in attempts to facilitate our understanding of reading underachievement (Fleisch et al. 2017).

Bahr et al. (2012:6) state that the qualitative evaluation of linguistic patterns of spelling errors is important for two reasons. Firstly, the study of the patterns of spelling errors can enhance our understanding of how learners with language impairments organise their phonological, orthographic and morphological knowledge (Bahr et al. 2012). Secondly, broader knowledge about spelling errors can lead to the creation of individualised instruction and intervention programmes for learners (Bahr et al. 2012). It is for the above-mentioned reasons that spelling errors are the focus of the current study.

Cook’s (1999) classification of errors (namely omission, substitution, insertion and transposition) has been used by researchers to qualitatively categorise spelling errors, especially in studies seeking to understand how to better teach English as a second language (Alhaisoni, Al-Zuoud & Gaudel 2015; Golshan & Reigani 2015; Khider & Othman 2018). Further, these categories have also been used in research on word recognition in English and isiXhosa (Probert & De Vos 2016).

Alcock and Ngorosho’s (2003) study looked at spelling development in Kiswahili with a focus on the patterns of errors made in Kiswahili writing. Kiswahili is a Southern Bantu language which is similar in structure to isiXhosa. While they did not exclusively use Cook’s (1999) classification of errors, the error types identified lend themselves quite closely to those used by Cook but with more precise margins. The results of the 50 real-word spelling task revealed that the most common errors identified, were those containing digraphs followed by the confusion of the letters 〈l〉 and 〈r〉 and the addition and omission of 〈h〉, 〈y〉 and 〈w〉. Another prominent error type identified was the omission of a nasal consonant in a ‘consonant cluster’ (Alcock & Ngorosho 2003). The latter result was reaffirmed in a later study by Alcock & Ngorosho (2007), in which they found that consonant clusters consisting of a nasal consonant in the initial position of a word followed by an additional consonant were more difficult to spell than words that contained either type of consonant in isolation. This result was reasoned to be as a consequence of the lack of salience of consonant clusters in Kiswahili (Alcock & Ngorosho 2007). They concluded that spelling is different to reading in transparent languages in that phonological decoding is a necessary but insufficient skill needed for spelling success (Alcock & Ngorosho 2003), suggesting that there may be other skills more significant for spelling success in African languages (Alcock & Ngorosho 2003). More research is needed on the skills underpinning spelling in African languages.

Research looking specifically at spelling error analysis in South Africa is extremely sparse. Fleisch et al.’s (2017) study, which investigated spelling errors made by English second-language learners, is one of the first to provide an in-depth understanding of the patterns and prevalence of spelling errors in South Africa. It focused on monosyllabic three-letter word spelling errors and provided important insights into the understanding of second-language reading underachievement (Fleisch et al. 2017). The data for their study were drawn from previously conducted learner literacy tests as part of a randomised control trial. Their findings revealed that over two-thirds of the 2500 learners misspelled simple monosyllabic three-letter English words which are assumed by the national curriculum to have been taught in the second term of Grade 2, confirming the large body of research about the reading and writing underachievement of learners (Fleisch et al. 2017). The most prevalent error type identified was associated with the incorrect identification of the middle vowel sound, an indication of weakness in phoneme awareness (Fleisch et al. 2017). The findings also showed low levels of encoding in the spelling test which imply that the learners were failing at the initial ‘learning to read’ foundational skills (Fleisch et al. 2017). Fleisch et al. provided an in-depth analysis of error categories that were present in the learner data, which they later refined to the linguistic categories: alphabetic, L1 interference and vowel error. According to Fleisch et al. (2017), the growing exploration into the identification of error patterns in spelling could prove useful in facilitating our understanding of reading underachievement and how it can be incorporated into pedagogy. This is because the study of learners’ spelling errors provides an opportunity to understand and facilitate the learners’ spelling difficulties (Al-zuoud & Kabilan 2013).

While Fleisch et al.’s (2017) study was instrumental in introducing a body of knowledge dedicated to understanding spelling errors in the South African context, a gap remains in exploring error patterns in Southern Bantu languages, specifically L1 spelling in isiXhosa. In this currently underexplored domain, this research on spelling is important as it adds to the growing understanding of literacy in Southern Bantu languages.

This study is a small-scale, exploratory study, which used a cross-sectional design. The study was conducted at a Quintile 3 primary school on the outskirts of a township in the Eastern Cape. IsiXhosa is used as the language of learning and teaching (LoLT) at the school until the end of Grade 3, after which a transition is made to English as the LoLT. All participants in this study were Grade 3 isiXhosa home-language learners. However, participants’ exposure to and use of English in daily activities varied. Participants comprised 51 Grade 3 home-language isiXhosa learners, which included 21 boys and 30 girls. The average age of the participants was 8 years old.

All learners participated on a voluntary basis after returning informed consent forms signed by their parents or guardians, as well as providing their own verbal assent. All learners’ identities were kept confidential.

A standardised spelling task for the Southern Bantu languages has yet to be developed. Therefore, taking into consideration the literature, and keeping to the phonological and orthographic structure of isiXhosa, an isiXhosa spelling task was developed. This spelling task included two sub-tasks: a real word and pseudoword task.

The spelling task was group-administered by trained research assistants who are isiXhosa home-language speakers. Words were read aloud one by one and then repeated once more. Learners were informed that they would first be spelling the real words followed by the pseudowords. The research assistant informed them when the transition was made. The spelling task consisted of 24 items, which included 12 real words and 12 pseudowords.

The design of the spelling task took into account three considerations, namely frequency, grapheme complexity and word length in syllables (word length). For the real word spelling task, words were sourced from a list of the top 300 high frequency words in isiXhosa, which was drawn from a corpus of 170 000 tokens of isiXhosa words (Rees & Randera 2019). Grapheme complexity was controlled for, as it has been noted that spelling in Southern Bantu languages seems to be more difficult due to the variety of complex graphemes and the agglutinative nature of the language, which makes words longer and denser (Diemer 2015). With respect to grapheme complexity, a range of words including those containing no complex graphemes to those with multiple complex graphemes were included. The third consideration was word length. The length consideration ensured that words ranged from disyllabic to quadrisyllabic. Lastly, to ensure that the task was fair and appropriate, all real words chosen had lexical meaning. The pseudoword spelling task was designed to be symmetrical as far as possible to the real word spelling task. As such, grapheme complexity and word length were controlled for.

Pseudowords were used as participants cannot rely on prior knowledge to spell the words but have to access their own graphemic knowledge of the language. This allows for a true representation of the learners’ encoding abilities. A list of real words and pseudowords used in the spelling task are included as Table 1-A1 and Table 2-A1 of Appendix 1.

Once the task was administered, the spelling subtask was coded using a binary set of codes, that is, correct and incorrect.

A secondary analysis was performed on the results documenting the errors found in the spelling tasks. The errors were coded according to Cook’s (1999) classification of errors (namely omission, substitution, insertion and transposition), with additional error categories included, based on the error types identified in the data. Thus, the error categories used for this study were: blank, illegible, alphabetic (errors in which there were more than three graphemes added or excluded, or which had no phonetic or orthographic relation to the target word), omission (leaving out graphemes), substitution (replacing graphemes with incorrect ones), insertion (adding of incorrect graphemes), transposition (swapping graphemes around) and graphemes written backwards.

Ethical clearance was obtained from the Rhodes University Ethical Standards Committee and the Eastern Cape Department of Basic Education (Nr 2019-0461-452).

The words that were most incorrectly spelled were phandle (33.3%), ngaphakathi (29.45%) and umakhulu (29.45%). The words with the most errors recorded were ngaphakathi (24 errors), phandle (21 errors) and namhlanje (21 errors).

The words that were least incorrectly spelled were oku (10%), amanzi (14%) and amabali (16%). These words also contained the lowest number of errors with five, seven and eight errors made per word. As noted in the methods section, the words ranged in length and grapheme complexity as a means to determine which of these factors influenced the learners’ spelling to a greater extent. Therefore, in order to precisely identify the degree to which word length or grapheme complexity influenced the spelling results, a linear regression was conducted with the number of errors made per word as the predicted variable and word length and grapheme complexity as predictor variables. This is illustrated in Table 2.

The model revealed that word length and grapheme complexity account for 73% of the number of errors made per word (R² = 0.734, p < 0.05). Only grapheme complexity was a significant predictor (β = 0.84, p < 0.001). It is therefore evident that grapheme complexity is a primary factor influencing the production of errors in the real word spelling task.

In terms of the types of errors made by the learners, the most prevalent error types identified were errors of omission (34.7%) and alphabetic errors (25%). Furthermore, the incorrect representation of complex graphemes accounted for 49% of the overall errors made by learners. This supports the view that the difficulty for spelling in isiXhosa lies in the presence of complex graphemes. In particular, learners struggled with the representation of digraphs, aspirated digraphs and nasal blends, with the words with the highest error rates being ngaphakathi (24 errors) and phandle (21 errors), each of which contains these complex graphemes.

In particular, the most common omitted letters were the omission of 〈h〉 in aspirated digraphs such as 〈ph〉, 〈kh〉 and 〈th〉 found in the words phandle, umakhulu and ngaphakathi, and the letter 〈n〉 in nasal blends, consonant blends and digraphs that begin with the letter 〈n〉, such as 〈ndl〉 in phandle, 〈nj〉 in namhlanje, 〈nd〉 in amaqanda and 〈ng〉 in ngaphakathi. There were no significant trends in the substitutions of letters for the real word spelling task.

The learners performed significantly worse on the pseudoword spelling task than the real word spelling task (t = −10.3, p < 0.001), but this is to be expected as the learners would never have encountered these words before and the real words were chosen specifically from a list of high frequency isiXhosa words.

The words that were spelled most incorrectly were ingxembo (86.3%), tshafiba (78.4%) and khuqahleyo (60%). The words with the most errors recorded were ingxembo (60 errors), khuqahleyo (53 errors) and tshafiba (52 errors). The words that were least incorrectly spelled were tuza (21%), lwesha (25%) and ngela (27%). These words also contained the lowest number of errors with 12, 15 and 15 errors made per word. As noted in the real word error analysis, the words ranged in word length and grapheme complexity as a means to determine which of these factors influenced the learners’ spelling to a greater extent.

As was done with the real words, a linear regression was conducted to observe whether word length or grapheme complexity was a stronger predictor of error rates in the pseudowords. The results of the linear regression are reported in Table 3.

The model reaffirmed the results of the real word analysis. The model showed that word length and grapheme complexity combined account for 58% of the variance in the number of errors made per word (R² = 0.576, p < 0.05) with grapheme complexity being the only significant predictor (β = 0.655, p < 0.05) contributing 65% of the variance in error production. These results, therefore, confirm that the presence of complex graphemes in isiXhosa presents a challenge to learners when spelling.

In terms of the types of errors made by the learners, the most prevalent error types identified were omission errors (35%) and substitution errors (34%). Furthermore, the incorrect representation of complex graphemes accounted for 55% of the overall errors made by learners, with the learners struggling mainly with the representation of the nasal blends 〈ngx〉 and 〈mb〉 the aspirated digraph 〈kh〉, the click 〈q〉 and the digraph 〈hl〉, all of which appear in the words with the highest error rates.

Of the 35% of omission errors made, the most common omitted letters were 〈g〉 in the word ingxembo, 〈h〉 in nokhubela and 〈n〉 in inkweseva. Of the 34% of substitution errors made, the most common substitutions were the substitution of 〈dl〉 in ndlashaza for other digraphs such as 〈hl〉 or 〈tl〉, the substitution of 〈mb〉 in ingxembo for 〈g〉, 〈l〉, or 〈ng〉, the substitution of 〈v〉 in inkweseva for 〈mb〉, 〈z〉 or 〈b〉, and the substitution of 〈hl〉 in khuqahleyo for a variety of different letters, for example 〈sh〉 and 〈k〉. Of the 11% of insertion errors, the most common inserted letter was 〈n〉 in the word initial position of tshafiba.