The Simple View of Reading (Rose 2009, after Gough & Tunmer 1986) model is a primary guide for assessing and teaching reading. The model describes reading comprehension as the product of two areas of expertise: word recognition (decoding) and oral language proficiency. It explains how learners may make progress in one area of reading, but not another, and that, if a learner makes adequate progress in both word recognition and language comprehension, reading will develop adequately. The child who develops good oral language comprehension skills but does not make gains in the word recognition process is at risk for dyslexia.

Researchers Hulme and Snowling (2015) agree that there are three main predictors for decoding (reading) skills:

Early isiXhosa reading instruction exposes learners to a syllabic approach early on in order to combine syllables into common words of two and three syllables (Pascoe & Smouse 2013). African languages, such as isiXhosa, are strongly syllabic, and these learners tend to do better on syllable awareness than phonemic awareness items (Diemer, Van der Merwe & De Vos 2015). However, phonemic awareness still seems to be a stronger predictor of decoding than syllable awareness (Wilsenach 2016).

Irrespective of whether they are monolingual, bilingual or multilingual, learners with dyslexia present with a similar profile despite differences in alphabetic orthography (Chung, Ho & Chan 2011). Joshi, Prakash and Surendranath (2010) reviewed literature on reading and literacy difficulties in bilingual learners and found that almost all participants had difficulty in both languages they were tested in and there would be phonological processing difficulties in both languages if they were dyslexic (Chung et al. 2011; Klein & Doctor 2003). It must be noted, however, that although dyslexia might manifest differently across different writing systems, phonological processing will remain a difficulty (Kovelman, Bisconti & Hoeft 2016). One can therefore make the assumption that a learner who presents with dyslexia traits in isiXhosa as a first language will present with similar difficulties when learning reading and writing in English, especially with phonological processing, in the classroom in later grades. Similarly, reading speed and fluency difficulties might go unnoticed in isiXhosa learners early on, but might emerge in later years. This may be attributed to the fact that dyslexic isiXhosa readers may read accurately but slowly in isiXhosa due to the more transparent orthography of isiXhosa. However, when they read in English (which has an opaque orthography) in later years, they will have both fluency and accuracy difficulties. Since the majority of English reading is only done from Grade 4, it makes it more difficult for teachers to identify reading difficulties early on.

While much research is dedicated to understanding dyslexia in the English-speaking population, there is limited knowledge of the effects of the condition in African languages. Teachers and professionals in South Africa have a difficult time identifying learners at risk for dyslexia and other literacy difficulties reliably, due to insufficient screening tools and a limited knowledge around the development of African languages. A survey with SPLTs in the Western Cape confirmed this, and suggested that assessment materials in dominant local languages will improve confidence in developing intervention strategies (Maphalala, Pascoe & Smouse 2010).

Given that learners in South Africa usually learn in their mother tongue until Grade 4 (with English as an additional language from Grade 1), access to a screening tool for dyslexia in the home language would allow for earlier detection. Schools generally wait for a learner to ‘fail’ at literacy before the child is identified for assessment (Gaab 2017). Gaab (2017) justifies that intensive interventions are most effective in the reception year and first grade. When early learners who are at risk for dyslexia receive explicit, structured and intensive instruction, they have the potential to perform at average reading levels (Torgesen 2004). Gaab recommends several key characteristics to be included in screening batteries: they should be short, quick and easy to administer, be comprehensive (including all the key indicators of dyslexia), be done during or before the reception year, and include a short family history, and be developmentally appropriate (Gaab 2017).

In summary, a dyslexia screener will therefore focus on the main indicators presented by a learner who is at risk for dyslexia. These include phonological awareness, decoding, knowledge of the alphabetic principle (that sound can be represented by a letter or string of letters and letter-sound knowledge), spelling and written output (sentence and paragraph structure). Auditory discrimination and oral fluency (the ability to speak fluently and with ease) are generally intact, but will vary from child to child. Some learners might present with visual or auditory difficulties, or have a double deficit – both fluency and phoneme awareness difficulties, and others only a single deficit (Bowers & Wolf 1993). Listening comprehension should present as adequate, because a child at risk for dyslexia usually presents with a stronger oral language and comprehension, and there will be a clear discrepancy between oral and written language (Kelly & Phillips 2016). The indicators of dyslexia are not normative, but highly individual because of the compensatory skills employed.

As isiXhosa is not widely spoken in the Gauteng area in South Africa, choices for schools were limited. The research team consisted of an educational psychologist, a SPLT proficient in isiXhosa, and an occupational therapist who is also a Foundation Phase remedial teacher. The University of Stellenbosch identified seven schools in Gauteng, two independent schools and five government schools, where the LoLT is isiXhosa. These schools were identified as functional schools with records of good Annual National Assessment (ANA) scores. The schools also had a quintile 4 ranking. A quintile ranking is an indication of the socio-economic status of the school, where a ranking of 1 is a no-fee school that serves a poor school community and quintile 5 schools are fee-paying and represent the least poor school communities (Van Wyk 2015).

The researchers then identified the seven schools using the Gauteng Basic Education Department’s Education Management Information System (EMIS) and contact was made telephonically with the schools. The two independent schools confirmed that their LoLT is English so they were excluded from the selection. The three top ANA performing schools were chosen in an attempt to reduce environmental factors such as poverty as far as possible. The schools were all located in urban informal settlements and had electricity and running water. Classroom sizes varied from 30 to 60 learners per classroom; however, for the larger groups basic resources such as pencils were limited.

Poverty is a reality for most of the learners. The schools provide all learners with a meal. Most of the learners live with a single parent or grandparent in informal settlements or hostels, and ‘home’ remains the Eastern Cape as parents had to move or split up for job opportunities. The social, psychological and emotional stability of these learners may therefore be compromised, which in turn compromises the learners’ true academic potential. One cannot view a child in isolation of the family or their environment (Kelly & Phillips 2016; Mowder 2005; Shonkoff & Meisels 2010). From the parent interviews, it also emerged that two learners experienced trauma after the loss of one or both parents. In the latter case, the interviewee was the learner’s grandmother.

Upon the first visit to the schools, researchers met with the principal, briefed teachers individually on the focus of the research and discussed learner profiles of learners who struggle in literacy. Teachers discussed learners they thought might meet this profile, and thereafter completed the teacher checklist, which took roughly 15 min to complete. Learners with known significant trauma and head injuries were excluded in the selection process. The researchers reviewed the selected learners’ isiXhosa, English and Mathematics books and finalised a shortlist of learners that they deemed may be at risk for dyslexia. Based on their clinical experience, researchers identified 13 learners as potentially at risk for dyslexia across the three schools within the Foundation Phase and Grade 4. The final group of learners was discussed with the principal, who sent out consent letters to the parents.

A control group of seven learners across the three schools were also selected by the teachers, who were asked to identify average performers. A total of 20 learners whose parents consented to the screening was enrolled for the study. From the 20 learners, 3 learners were in Grade 1, 6 learners were in Grade 2, 6 learners were in Grade 3 and 4 learners were in Grade 4. One learner was absent. Of the 19 participants, 13 male and 6 female, 5 were repeating their grade.

During the second visit to the schools, 19 parent interviews were conducted by the clinical psychologist. Running parallel was the learner screening which was done by the isiXhosa-speaking SPLT. A translator was used for three of the interviews. One identified child was absent. Teacher checklists were not administered for the control group. One parent did not arrive for the parent interview.

The Bellavista Dyslexia Screening Tool (BVDST) for English home language learners was adapted and reworked according to the structure of the language, for the young isiXhosa learner in Grades 1–3. The English tool is used for screening learners’ literacy skills and makes use of the administrator’s skills to analyse, find patterns in literacy difficulties and interpret findings. It is not normed. After adapting the tool into isiXhosa, the screening tool was reviewed by a biliteracy specialist working on bilingual and multilingual children’s literacy development. It is not a diagnostic tool, but a screener to guide early identification and intervention for learners at risk of literacy difficulties. It allows for the administrator to make use of examples in order to determine whether the learners understand what is required of them for each item of the tool.

The tool includes screening of the following components:

It is important to note that the tool is designed for an educated or trained assessor who understands the developmental progression of learning and who can make judgements around which items to administer and which items to leave out. Rapid Automised Naming was left out of the assessment, as it forms part of a standardised assessment.

A background history of learning difficulties in the family, as well as a developmental history may contribute to holistic understanding of the learner and the level of risk for dyslexia. A teacher checklist was adapted from Kelly and Phillips (2016), specialists in the assessment and intervention of dyslexic-type difficulties. A parent interview was also developed. The aim of the interview and checklist was to identify difficulties in the area of literacy that consistently presented in the home and classroom, and to triangulate the information with the learners’ performance on the screening tool. The parent interview included questions related to family and medical history, as well as identifying difficulties with reading and spelling. The teacher checklist refers to questions related to reading and spelling, but also to mathematics, as a discrepancy in skills may strengthen the possibility of risk for dyslexia.

Data were captured across the three screening sources (teacher checklist, parent interview and learner screener) and compiled into 10 variables as per Table 1. Capturing across three different sources ensured triangulation of results and therefore increased the reliability of the study. Phoneme awareness and auditory discrimination were excluded from the teacher checklist and these concepts together with sequencing, alphabetic principle and written output were excluded from the parent interview due to complexity of the indicators. The other variables are tracked across the three screening methods.

Each component of the assessment administered was scored. The raw score was then converted into assessment scores: a child who scored in the lower third on the raw scores (33.3%) was ranked as 0 (problem); scores in the second third were ranked as 1 (some difficulty/risk) and scores in the higher third were ranked as 2 (intact). Statistics were run on both the assessment and the raw scores; due to the lower variability of the assessment scores, most statistics were run on the raw scores. After analysing the results, it was found that some of the learners selected as controls (the average performers selected by the teachers) presented with results consistent with a dyslexic or learning difficulty profile. The control learners who presented with dyslexic-type profiles were analysed as part of the risk group. Of the seven control learners, there were only two learners that represented true controls; the other four learners were identified as false negatives. For this reason (such a small control sample), we present all the data grouped together from this point on and have categorised the learners into clearly at risk, at risk, other disorder, and not at risk groups.

Ethical clearance was obtained from the Department of Basic Education. Each parent or guardian of the selected learners, as well as the teachers and their school principal signed and gave informed consent.

Table 2 represents the number of learners for each indicator within these categories.

Based on the triangulation of screening data from the sample of 19 learners (one was absent on the day of assessment), 5 learners were identified as being clearly at risk for dyslexia. Eight learners were identified as being possibly at risk for dyslexia. There were also 5 learners who were identified as being at risk, but their difficulties are better explained by a different condition or learning difficulty, for example when the child’s difficulties were global. Two learners from the control group presented with literacy skills that were intact.

The adapted isiXhosa screener identified learners at risk for dyslexia and was measured against the researchers’ clinical experience. The findings for each child were captured in a personalised report and feedback was given to the parents and Head of Foundation Phase. After completion of the fieldwork and feedback with stakeholders, a 2-h workshop on reading and spelling intervention strategies for at-risk learners was facilitated for all the Foundation Phase teachers and the parents of the selected learners at the three schools.

What did the screening tool show about the not at risk (indicated as control for the purpose of this study) and the at risk learners’ profile, based on the teachers’ categorisation of their learners?

This section looks at average scores for the different domains of functioning across at risk and not at risk children. Figure 1 presents average raw scores for control and at risk children. Given that phonological awareness is the largest category (with a maximum score of 59 on the original tool), it is not surprising that the greatest differential can be seen on this domain. Not at risk children scored 19 on average while at risk children scored 15. Not at risk children also scored more highly on the two domains that should not be affected by dyslexia: Oral/Semantic Fluency and Listening Comprehension.

Figure 2 displays the same information, but using assessment scores (i.e. the three rankings of 0, 1 and 2) rather than raw scores. Additional derived domains now included are Oral/Written Discrepancy (i.e. the discrepancy between Oral and Written language scores), Sequencing, and Written Output. At risk children display lower average scores on most, but not all domains. At risk children have the same average score as not at risk children on Oral/Written Discrepancy and Alphabetic Principle (although the latter was lower when using the more varied raw scores). The greatest discrepancies between the two groups of children are in Phonological Awareness (0.58 lower), Written Output (0.4 lower), Auditory Discrimination (0.3 lower) and Spelling (0.29 lower).

Figure 3 contrasts at risk children with clearly at risk children. Those identified as clearly at risk have similar Oral/Semantic Fluency and higher Listening Comprehension, the two domains unlikely to be affected by dyslexia, and score lower on the other items. On average, then, those identified as clearly at risk are well defined according to the profile of impairment that is known to be associated with dyslexia.

Figure 4 contrasts at risk children (including clearly at risk children) and those identified as not at risk. Those identified as at risk have higher Oral/Semantic Fluency and Listening Comprehension.They also have higher Auditory Discrimination, Sequencing, and Phonological Awareness. As may be expected, when including those who are at risk and not just clearly at risk for dyslexia, the average profile of impairment is not as clearly delineated according to the typical dyslexia presentation.

This section explores how well related the different ‘risk factors’ (Phonological Awareness, Auditory Discrimination, Sequencing, Decoding, Alphabetic Principle, Spelling, Written Output, and Oral/Written Discrepancy) are for children identified as at risk.

Table 3 displays correlations (Spearman’s rho) between the different domains of functioning which should be impaired in dyslexia for children identified as at risk after testing took place. This includes both clearly at risk and at risk children. Unfortunately, due to the very small number of children identified as clearly at risk, the same table cannot be replicated for this group alone. Additionally, since there was no variation in the scores for Auditory Discrimination, these correlations cannot be ascertained. If these children really were all at risk for dyslexia, we should see high correlations between the different domains since they should present with a similar, and typical, profile of impairment. High correlations are defined here as between 1.0 and 0.7, moderate correlations are between 0.4 and 0.6, and low correlations are 0.1 to 0.3.

In Table 3, we see that most correlations are again in the low range. However, we do see very high and statistically significant correlations for Phonological Awareness with Spelling, and Decoding with Knowledge of the Alphabetic Principle, Spelling, and Oral/Written Discrepancy. Sequencing displays low correlations with all domains except spelling. It appears that Decoding and Alphabetic Principle are the most consistent indicators of being at risk of dyslexia in this sample.

After piloting the screening tool, the researchers made further revisions to the content and length of some of the components. It was reviewed by an educational psychologist whose home language is isiXhosa, after which a final version was produced.

Similarly, the teacher checklist was simplified from two pages to one page. Many teachers had difficulty completing the questionnaire accurately. Statements were adapted into simpler questions, and a yes/no option was provided for the teacher to circle, rather than to write a response.

The parent interview was shortened. The order of questions was changed in such a way that the conversation starts with the child’s current functioning and moves back towards birth history, which enabled the parents to report on pressing issues first and therefore build rapport with the interviewer.

There were several limitations to the study. First language isiXhosa speakers in Gauteng are limited in number, and schools with a LoLT in isiXhosa are therefore similarly limited. The schools were not as functional as we had hoped, which made it difficult to rule out environmental factors like poverty and lack of quality instruction. Often, adequate quality instruction in early reading is compromised. The quintile ranking of the schools was not an accurate representation of the school community, thus low socio-economic factors may have had an impact on the results. Classroom sizes ranged from 50 to 60 learners per class. Unfortunately, with such large group sizes it is unlikely that a teacher will have an in-depth knowledge of a child’s learning profile. This was evident in the selection of average performers for the research control group, as many of these learners also presented with literacy difficulties. In retrospect, it may have been more beneficial for the research team to have requested top performers to use as a control group instead of average performers. There were only two learners of the seven control learners that represented true controls.

The selected sample was small, compromising the outcomes. A larger sample within a larger selection of schools would have been beneficial. The tool may benefit from another cycle of piloting, ideally in areas where isiXhosa is the predominant language.

The screening tool requires further refinements in terms of presenting phonological awareness skills more accurately. It is also important to take into consideration that African language readers in general perform lower on phonological awareness tasks (Wilsenach 2019).

The screening tool can only be used by teachers who have a sound understanding of literacy development and literacy difficulties or learning support educators and SPLTs. Critical analysis and age-grade comparison is necessary to make an informed conclusion based on the learner’s performance.