IsiXhosa, like all Bantu languages, has a relatively simple syllable structure based around a (C)V(V) template (Hua 2002). In addition, there are constraints on complex syllable onsets. Although some words appear to have a complex onset orthographically, these are often single consonants corresponding to single phonemes and not a true phonological cluster (1ab). In contrast, English allows for more complex structures than CV, such as CCV and CCCV words (2ab):

The nature of words is language specific (Guthrie 1948; Louwrens & Poulos 2006; Prinsloo 2009). All Bantu languages are agglutinating and consequently a Bantu word includes rich, overt morphology. For example, nouns include noun class prefixes as well as stems whereas verbs include morphological reflexes of subject marking, object marking, tense, aspect, mood, causativity, and negation among others. The result is that Bantu words (particularly nouns and verbs) tend to be much longer than their English equivalents.

Linguistic structure is mediated through orthography. Because linguistic words may differ from orthographic words, learners are faced with language-specific processing challenges when attempting to recognise words in a particular language. These challenges, in turn, presuppose language-specific reading strategies. For example, Nguni languages are written conjunctively and the Tswana and Sotho language groups tend to be written disjunctively. In (3a), a conjunctive writing system yields only one orthographic word also corresponding to one morphological word. The orthographic word ngiyabathanda (3a) is therefore both a morphological word and an orthographic word. In contrast, in disjunctive orthographies, a linguistic word may correspond to a number of orthographic words (3b):

This correspondence between orthographic and morphological words is a characteristic feature of conjunctive orthographies which distinguishes them from disjunctive scripts, such as Northern Sotho (3b) and English (Taljard & Bosch 2006), although all South African Bantu languages have aspects of both conjunctive and disjunctive systems triggered by linguistic context (Louwrens & Poulos 2006).

The reason for the utilisation of different orthographies is based on both historical and phonological considerations. Historically, orthographies were developed by European missionaries. Although prescriptive and colonial suppositions may have influenced their decisions (Van Wyk 1987), the two languages necessitate different orthographies based on their phonological systems (Louwrens & Poulos 2006). In the Nguni languages the presence of phonological processes, such as vowel elision and vowel coalescence make the use of a disjunctive orthography impractical. Example (4a) demonstrates how the two vowels ‘a’ and ‘u’ coalesce into 'o', which makes a disjunctive rendering at odds with the phonetic pronunciation; (4b) demonstrates vowel coalescence between ‘a’ and ‘i’ yielding ‘e’:

Louwrens and Poulos (2006) demonstrate convincingly that Tswana is mostly written disjunctively except in contexts where phonological processes obscure the underlying forms; in these contexts, Tswana is written conjunctively. Thus, orthographic choices mediate linguistic structural considerations. There remain significant questions about how orthographies and language structures conspire to make the reading experience different for different learners. Nevertheless, in South Africa, the relationship between language structure and orthography are not always reflected in pedagogy: teacher training is oriented to English and Afrikaans; indigenous language curriculum documents are often translated directly from English and therefore reflect English teaching concerns (Bikitsha & Katz 2013).

There are limited studies on agglutinating languages which could shed light on how reading occurs in isiXhosa. However, some research on Turkish has been done. The structure of Turkish, like that of isiXhosa is agglutinating and the orthography is conjunctive in character. Durgunoglu and Öney (1999) found that, in Turkish, a phonologically transparent orthography fosters early development of word recognition skills, and that phonological awareness contributes to word recognition in the early stages of reading acquisition. These results were attributed to the phonological and orthographic characteristics of the Turkish language and orthography.

Written languages may differ with respect to orthographic depth and consistency. Orthographic depth refers to the degree to which there are one-to-one correspondences between graphemes and linguistic categories (such as phonemes or syllables) (Aro 2004). The source of opacity within the English orthography partly derives from historical spellings, dialectal speech differences, borrowing, and the fact that there are more phonemes in the spoken language than there are graphemes to represent them (Frost 2005). Consequently, there is a relative lack of one-to-one correspondences between phonemes and graphemes. The second factor, consistency, involves the uniqueness of pronunciation of an orthographic segment. Therefore if two words are spelled similarly but pronounced differently (such as MOTH–BOTH), the letter cluster OTH is considered inconsistent (Frost 2005; Widjaja & Winskel 2004). English has a very high number of words with irregular and unpredictable spellings, included in foundation phase vocabulary (e.g. one, two, four, five, eight, and nine). The irregularity and inconsistency of English orthography results in a deep orthography where readers rely less on grapheme-to-phoneme representation and more so on whole-word processing. In contrast, isiXhosa has a fairly transparent, consistent, alphabetic orthography which is more amenable to sublexical processing.

Orthographic depth affects acquisition and development in reading (Pillunat & Adone 2009; Sprenger-Charolles 2003) and influences the cognitive process of word recognition (Yamashita 2013). According to the Orthographic Depth Hypothesis (ODH), reading in a shallow orthography predisposes readers to rely more on decoding and phonological processing because of the direct and reliable grapheme-phoneme mapping. For deep orthographies, it is hypothesised that readers tend to rely more on orthographic processing because of the opacity in the relationship between grapheme-to-phoneme mapping. Therefore differences in orthographic depth determine the dominant type of information (lexical or sublexical) used in word recognition in the brain (Katz & Frost 1992; Yamashita 2013).

Within a cognitive approach to word recognition, one way of understanding the interaction of orthography and reading is through a dual route model (DRC) of word recognition (Levy et al. 2009; Perre & Ziegler 2008; Simon & Van Herreweghe 2010). The model was originally developed to show differential word recognition in adult learners but can also potentially assist in understanding word recognition among L2 learners.

The model suggests the existence of two distinct processing routes for word recognition – a lexical (direct route) and a sublexical (indirect) route. With respect to the lexical route, the brain recognises a word in its entirety (perhaps as a type of picture or gestalt); the word is accessed directly from the lexicon. Consequently, for the processing of written language, the lexical route processes frequent and orthographically irregular words, such as yacht or one, but fails to process unfamiliar or pseudo-words, such as braston, which have no meaning at all, because, by definition, such words would be novel occurrences (Invernizzi & Hayes 2010; Levy et al. 2009).

In contrast to this, the sublexical route subdivides words into smaller segments and maps those to linguistic constituents, together making up the word in a bottom-up fashion (Eskey & Grabe 1988; Lisson & Wixson 1991). The bottom-up model of reading is concerned with the recognition of individual letters, phonemes, and words and the reading process begins with individual recognition of letter and phonemic counterparts (Eskey & Grabe 1988). According to the bottom-up model, learners learn to read through their ability to sound out letters through grapheme-to phoneme correspondences. Only once the learner has these decoding skills can he/she move onto the meaning of the words in the text (Lisson & Wixson 1991). For top-down processing, the reader starts with a general idea or schema of what should be in the text and uses this in perceiving and interpreting the orthographic text (Eskey & Grabe 1988).

Most typically, it is assumed that graphemes are mapped to phonemes, although Grain Size Theory (Ziegler & Goswami 2005) also suggests the possibility of sublexical processing involving syllables or morphemes. The sublexical route can process pseudo and real words which have a grapheme-to-phoneme representation but would fail to process irregular words which violate this representation (Levy et al. 2009). This is an important distinction which is exploited in our methodology by using pseudo-words to factor out top-down, lexical recognition processes.

Pseudo-words are ‘pronounceable combinations of letters which lack semantic meaning but which can be decoded and pronounced via phonological processing and alphabetic (letter-sound) knowledge’ (Fredrickson, Frith & Reason 1977 cited in Thomson, Crewther & Crewther 2006:290). Zock, brane, and repaki are all examples of English pseudo-words; lebina, warona, and menida are examples of isiXhosa pseudo-words. Pseudo-word reading tasks are commonly used as a measure of phonological processing and word recognition across diverse fields of reading research (Thomson, Crewther & Crewther 2006). According to Siegel (2004) ‘It is a well established fact that reading of pseudo-words is one of the best measures of phonological processing in an alphabetic language’ (cited in Pillunat & Adone 2009:3). Because readers have never encountered a particular pseudo-word before, they cannot rely on top-down recognition processes such as contextual clues or previous sight word familiarity. Rather it requires a reader to recall the phoneme associated with each grapheme and blend the separate sounds together in order to form a pronounceable unit. Pseudo-word reading therefore provides an accurate assessment of the learner’s ability to apply grapheme-to-phoneme knowledge in decoding (Rathvon 2004). However, pseudo-words are not merely an experimental task; they are also an indication of how learners may approach new words with which they are unfamiliar. In the context of school and higher education, ongoing learning involves routinely being exposed to new, unfamiliar words and it is therefore important that learners can cope with them. This makes the pseudo-word paradigm a particularly powerful and illuminating technique.

Transparent orthographies are often associated with faster initial learning trajectories. Wimmer and Goswami (1994) studied reading development in young English and German learners and found that the English learners had substantially more difficulties in decoding tasks than German learners did. These findings were attributed to the transparency of the German orthography. According to Wimmer and Goswami (1994), when learners learn to read using transparent orthographies they make use of the sublexical route, which ultimately ensures successful decoding. However, when approaching reading in a deep orthography, such as English, the learners recognise words via the lexical route. Similarly, Seymour, Aro and Erskine (2003) found that acquisition of a transparent orthography requires access to only one of two decoding strategies, as seen in the sublexical route, whereas acquisition of deep orthographies requires the implementation of both decoding strategies. Likewise, Defior, Martos and Cary (2002), comparing Spanish and Portuguese learners, showed that the accuracy in the pseudo-word reading of Spanish and Portuguese learners was relatively similar to the performance of German learners and much better than that reported for English learners in the aforementioned studies.

The tension between linguistic language structure and orthography results in different written languages presenting very different reading challenges to learners. Important cognitive skills such as phonological and morphological awareness are used in different ways for different language/orthography combinations and emerge in acquisitionally different ways (Bialystok 2002; Geva & Siegel 2000; Goswami 1999; Ota 2010 inter alia). There is thus a very tight relationship between language, orthography, and the particular cognitive strategies used in reading:

In equivalent texts, English will present readers with larger numbers of shorter words in an idiosyncratic and unpredictable orthography; isiXhosa presents readers with smaller numbers of longer, morphologically complex words in a highly regular syllabic pattern and a relatively transparent orthography. Clearly, then, because the nature of words differs between languages, the cognitive problem of word recognition must be particularised too. Therefore, there is no one-size-fits-all approach to fluent reading across languages. Automaticity occurs through at least two neurological pathways and these are conditioned by orthography and language structure. This raises the important question of how these skills are transferred to reading in an L2. Despite this, Bantu languages are notably absent from the literature on word recognition and cross-orthography comparisons. It is this gap that our research attempts to come to grips with; with these issues in mind, our study focuses on, (1) what groups of strategies are used in word recognition and (2) how differences in transparency between two completely different orthographies and language families affect word recognition in bilingual learners.

Data were collected from a sample of 47 Grade 4 isiXhosa home language learners from two different schools. The learners were tested individually at their school by the researcher and an assistant. The first school (which we will refer to as School X) is an isiXhosa-medium school in a coastal town in the Eastern Cape (n = 31). The school is a well-run and well-resourced public school on the outskirts of a township. Learners receive three years of isiXhosa LoLT followed by English as LoLT in their fourth year. Therefore they had been learning in English for approximately 6 months at the time of testing. The second school (School E) is an English-medium school in an inland town in the Eastern Cape (n = 16). School E is a former model C public school and has learners from diverse backgrounds. The selected learners from School E are home language isiXhosa speakers who have been taught in English from the start of their school career. IsiXhosa is taken as an additional language in this school. The learners from this sample, although isiXhosa home language speakers, have only been exposed to written isiXhosa for six months.

It is important to bear in mind that although the learners from both School X and School E have isiXhosa as their L1 and English as their L2, their exposure to the two orthographies is very different. The learners from School X are educated in isiXhosa for three years up until the end of Grade 3, where they make the ‘switch’ to English as the LoLT. However once this ‘switch’ has been made, these learners still speak isiXhosa in their homes and on the playground. The learners from School E however are educated in English, often from preschool age. They are only taught isiXhosa formally as an additional language from Grade 4 and English remains their LoLT. Most of these learners will speak isiXhosa at home, but at school they tend to interact with their friends in English. Thus the two Schools will often be discussed in contrast to one another.

Table 1 provides a breakdown of the two schools involved in this study. The language of first literacy (the language the child first learns to read in), alternatively referred to as the child’s language of learning and teaching, and the learners’ first and second languages are outlined for each school.

The learners completed a set of reading tasks, including both word- and pseudo-word reading, in both English and isiXhosa. Each reading task consisted of a list of 10 words, with three practice words used to familiarise the learners with the task. The number of tokens obtained from each learner is thus 49 in each language. All pseudo-words were pronounceable, adhered to the orthographic conventions of the language being tested, and had been checked with L1 speakers of isiXhosa beforehand. Instructions were given orally by the assistant in both English and isiXhosa to ensure that learners understood what was required of them. To ensure that all children were given the same instructions, the same assistant was used at both schools, with all the children. The assistant received formal training leading up to data collection and was given explicit guidelines on how to phrase the instructions for the learners. Learners were presented with the words and pseudo-words one by one in the form of flashcards and they were then instructed to read the words out loud. Learners’ responses were recorded for more in-depth analysis. Words were randomised in order to secure the most accurate results. Half of the learners were tested in isiXhosa first and the other half were tested in English first so as to avoid any priming effects. The stimulus materials also controlled for word length (number of syllables) and syllable structure (simple vs. complex consonant clusters).

Table 2 provides examples of the stimuli presented to the learners for both isiXhosa and English real- and pseudo-word reading.

The real word reading task investigated the different word recognition strategies which isiXhosa/English bilingual learners adopted when approaching the reading of real words in English and isiXhosa. According to the ODH, (Katz & Frost 1992), different orthographies promote different kinds of word recognition processes. Transparent orthographies like isiXhosa promote sublexical decoding, whereas opaque or deep orthographies like English promote whole-word recognition. One might therefore expect these processes to be evident from the errors made by learners. When there is adherence to alphabetic decoding (grapheme to phoneme mapping) in word recognition, the common errors which occur are those of mispronunciations (i.e. phoneme substitutions, deletions, insertions etc.), whereas the common errors which occur with orthographic (whole word) reading word recognition strategies are lexically primed, real-word substitution errors (i.e. Related but Wrong) (Wimmer & Goswami 1994).

When learners identify and read the real-word stimulus correctly, this shows that they are able to recall and transfer pronunciation already taught to them before the items were encountered. The learner also draws on their knowledge of the orthography to interpret the words correctly. However, it does not necessarily shed any light as to whether the words are being recognised through lexical or sublexical processes. However, when learners make a mistake, the nature of the error may provide an indication of the underlying processes. In contrast, because pseudo-words have not been previously encountered and can only be parsed using sublexical processes, correct reading of pseudo-words demonstrates an ability to decode words sublexically by matching the correct grapheme to a phoneme. Conversely, inability to complete the task shows inability to undertake this mapping.

The learners’ responses in the reading tasks were coded on a 6-point nominal scale (Table 3). The analysis of the results was interpreted using comparative descriptive statistics and Fisher’s exact chi-squared test, which is useful for categorical data in the analysis of determining the significant association (contingency) between two kinds of groups and is used when nominal data are less than five (Fisher 1954).

Table 3 presents the 6-point nominal scale on which the learners’ responses were recorded. This scale is divided into three separate sections – general error types; lexically primed, whole-word recognition error types; and sublexical, decoding error types. An example from the data set is presented alongside each error type.

Having ascertained the strategies used in reading in the language in which literacy was first acquired (i.e. the LoLT), we now turn to the question of whether these reading skills can be transferred to reading another language introduced at a later stage (the language of second literacy). To do this, we investigated the strategies used by School X learners in reading English and School E learners in reading isiXhosa.¹